Cabledriven Parallel Robots 2019

Mechanisms and Machine Science 74
Andreas Pott
Tobias Bruckmann Editors
Cable-Driven
Parallel
Robots
Proceedings of the 4th International
Conference on Cable-Driven Parallel
Robots
Mechanisms and Machine Science
Volume 74
Series Editor
Marco Ceccarelli, Department of Industrial Engineering, University of Rome Tor
Vergata, Roma, Italy
Editorial Board
Alfonso Hernandez, Mechanical Engineering, University of the Basque Country,
Bilbao, Vizcaya, Spain
Tian Huang, Department of Mechatronical Engineering, Tianjin University,
Tianjin, China
Yukio Takeda, Mechanical Engineering, Tokyo Institute of Technology, Tokyo,
Japan
Burkhard Corves, Institute of Mechanism Theory, Machine Dynamics and
Robotics, RWTH Aachen University, Aachen, Nordrhein-Westfalen, Germany
Sunil Agrawal, Department of Mechanical Engineering, Columbia University, New
York, NY, USA
This book series establishes a well-defined forum for monographs, edited Books,
and proceedings on mechanical engineering with particular emphasis on MMS
(Mechanism and Machine Science). The final goal is the publication of research that
shows the development of mechanical engineering and particularly MMS in all
technical aspects, even in very recent assessments. Published works share an
approach by which technical details and formulation are discussed, and discuss
modern formalisms with the aim to circulate research and technical achievements
for use in professional, research, academic, and teaching activities.
This technical approach is an essential characteristic of the series. By discussing
technical details and formulations in terms of modern formalisms, the possibility is
created not only to show technical developments but also to explain achievements
for technical teaching and research activity today and for the future.
The book series is intended to collect technical views on developments of the
broad field of MMS in a unique frame that can be seen in its totality as an
Encyclopaedia of MMS but with the additional purpose of archiving and teaching
MMS achievements. Therefore, the book series will be of use not only for
researchers and teachers in Mechanical Engineering but also for professionals and
students for their formation and future work.
The series is promoted under the auspices of International Federation for the
Promotion of Mechanism and Machine Science (IFToMM).
Prospective authors and editors can contact Mr. Pierpaolo Riva (publishing
editor, Springer) at: [email protected].
Indexed by SCOPUS and Google Scholar.
More information about this series at http://www.springer.com/series/8779

Andreas Pott Tobias Bruckmann
•
Editors
Cable-Driven Parallel Robots

Proceedings of the 4th International
Conference on Cable-Driven Parallel Robots
123
Editors
Andreas Pott Tobias Bruckmann
University of Stuttgart University of Duisburg-Essen
Stuttgart, Germany Duisburg, Germany
ISSN 2211-0984 ISSN 2211-0992 (electronic)

Mechanisms and Machine Science
ISBN 978-3-030-20750-2 ISBN 978-3-030-20751-9 (eBook)
https://doi.org/10.1007/978-3-030-20751-9
© Springer Nature Switzerland AG 2019
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
International Scientific Committee
S. Agrawal, Columbia University

M. Arsenault, Laurentian University
Ph. Cardou, Lavel University
S. Caro, CNRS, LS2N, IRT Jules Verne
M. Carricato, University of Bologna
C. Gosselin, Laval University
M. Gouttefarde, LIRMM, CNRS, Univ. Montpellier
J.-P. Merlet, INRIA
L. Notash, Queen’s University
D. Schramm, University of Duisburg-Essen
A. Verl, University of Stuttgart
v
Preface
In 2012, leading experts from three continents gathered during the “First
International Conference on Cable-Driven Parallel Robots (CableCon 2012)” in
Stuttgart, Germany. This conference initiated a forum for the cable robot com-
munity. Due to the great success, the event was continued by the “Second
International Conference on Cable-Driven Parallel Robots (CableCon 2014)” at the
University Duisburg-Essen in 2014, establishing CableCon as a home for cable
robot researchers.
In 2017, the “Third International Conference on Cable-Driven Parallel Robots
(CableCon 2017)” was organized by the Université Laval in Québec, Canada. This
time, the conference went across the Atlantic Ocean and emphasized its role as an
international event, where researchers from all over the world find a place to
connect with the most recognized experts from the field.
Meanwhile, practical investigations on cable robots are attracting the focus of the
research teams around the world. At the same time, fundamental research still
continues to provide new insights to the physical understanding of cable-driven
parallel robots. This broad variety of research activities is reflected by the content of
this book.
The editors would like to thank the authors for their valuable contributions.
Noteworthy, the strict schedule in the preparation of this book would not have been
feasible without the support of the reviewers and the scientific committee.
All three conferences in the past were organized under the patronage of
International Federation for the Promotion of Mechanism and Machine Science
(IFToMM). This time, the “Fourth International Conference on Cable-Driven
Parallel Robots” takes even place as part of the “15th IFToMM World Congress” in
Krakow, Poland. We are proud that IFToMM decided to welcome the fourth edition
of CableCon at this very special event, which celebrates the 50th anniversary of
IFToMM at the same time.
CableCon 2019 was organized by the University of Duisburg-Essen and the
University Stuttgart together with the organizers of the “15th IFToMM World
vii
viii Preface
Congress”. We would like to express our gratefulness to Tadeusz Uhl and his sup-
porting team – namely Anna Inglot and Marta Polak – for their continuous assistance.
Stuttgart, Germany Andreas Pott

Duisburg, Germany Tobias Bruckmann
Editors
Contents
Part I Design
Planar Cable-Driven Robots with Enhanced Orientability . . . . . . . . . . . 3
M. Vikranth Reddy, N. C. Praneet and G. K. Ananthasuresh
Chain Driven Robots: An Industrial Application Opportunity.
A Planar Case Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Guillermo Rubio-Gómez, David Rodríguez-Rosa,
Jorge A. García-Vanegas, Antonio Gonzalez-Rodríguez,
Fernando J. Castillo-García and Erika Ottaviano
Non-slipping Conditions of Endless-Cable Driven Parallel Robot
by New Interpretations of the Euler-Eytelwein’s Formula . . . . . . . . . . . 23
Takashi Harada and Koki Hirosato
Analysis of Cable-Configurations of Kinematic Redundant
Planar Cable-Driven Parallel Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Koki Hirosato and Takashi Harada
Improving cable length measurements for large CDPR using
the Vernier principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Jean-Pierre Merlet
Part II Kinematics and Static

Stiffness of Planar 2-DOF 3-Differential Cable-Driven Parallel
Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Lionel Birglen and Marc Gouttefarde
Stability Analysis of Pose-Based Visual Servoing Control
of Cable-Driven Parallel Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Zane Zake, Stéphane Caro, Adolfo Suarez Roos, François Chaumette
and Nicolò Pedemonte
ix
x Contents
Practical Stability of Under-Constrained Cable-Suspended

Parallel Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Dragoljub Surdilovic and Jelena Radojicic
Singularity Characteristics of a Class of Spatial Redundantly
actuated Cable-suspended Parallel Robots and Completely
actuated ones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Lewei Tang, Xiaoyu Wu, Xiaoqiang Tang and Li Wu
Kinetostatic Modeling of a Cable-Driven Parallel Robot using
a Tilt-Roll Wrist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Saman Lessanibahri, Philippe Cardou and Stéphane Caro
Static Analysis of a Two-Platform Planar Cable-Driven Parallel
Robot with Unlimited Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Thomas Reichenbach, Philipp Tempel, Alexander Verl and Andreas Pott
Part III Workspace

Calculation of the cable-platform collision-free total orientation
workspace of cable-driven parallel robots . . . . . . . . . . . . . . . . . . . . . . . 137
Marc Fabritius, Christoph Martin and Andreas Pott
Workspace Analysis of Cable Parallel Manipulator for Side
Net Cleaning of Deep Sea Fishing Ground . . . . . . . . . . . . . . . . . . . . . . . 149
Liping Wang, Haisheng Li, Zhufeng Shao, Zhaokun Zhang
and Fazhong Peng
Identifying the largest sphere inscribed in the constant orientation
wrench-closure workspace of a spatial parallel manipulator driven
by seven cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Ambuj Shahi and Sandipan Bandyopadhyay
A Bounding Volume of the Cable Span for Fast Collision Avoidance
Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
M. Lesellier and M. Gouttefarde
Computation of the interference-free wrench feasible workspace
of a 3-DoF translational tensegrity robot . . . . . . . . . . . . . . . . . . . . . . . . 185
Marc Arsenault
Antipodal Criteria for Workspace Characterization of Spatial
Cable-Driven Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Leila Notash
Contents xi
Part IV Control
Robust Adaptive Control of Over-Constrained Actuated Cable-Driven
Parallel Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Alireza Izadbakhsh, Hamed Jabbari Asl and Tatsuo Narikiyo
Model Predictive Control of Large-Dimension Cable-Driven Parallel
Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
João Cavalcanti Santos, Ahmed Chemori and Marc Gouttefarde
Linearised Feedforward Control of a Four-Chain Crane
Manipulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Michael Stoltmann, Pascal Froitzheim, Normen Fuchs, Wilko Flügge
and Christoph Woernle
An experimental study on control accuracy of FAST cable robot
following zigzag astronomical trajectory . . . . . . . . . . . . . . . . . . . . . . . . 245
Hui Li and Mingzhe Li
Part V Motion Planning

Path Planning of a Mobile Cable-Driven Parallel Robot
in a Constrained Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Tahir Rasheed, Philip Long, David Marquez-Gamez and Stèphane Caro
Development of Emergency Strategies for Cable-Driven Parallel
Robots after a Cable Break . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Roland Boumann and Tobias Bruckmann
A Conditional Stop Capable Trajectory Planner for Cable-Driven
Parallel Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Patrik Lemmen, Robin Heidel and Tobias Bruckmann
Part VI Advanced Cable Modeling

Modeling of Elastic-Flexible Cables with Time-Varying Length
for Cable-Driven Parallel Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Philipp Tempel, Dongwon Lee, Felix Trautwein and Andreas Pott
Static and dynamic analysis of a 6 DoF totally constrained cable
robot with 8 preloaded cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Damien Gueners, B. Chedli Bouzgarrou and Hélène Chanal
Slackening Effects in 2D Exact Positioning in Cable-Driven
Parallel Manipulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Erika Ottaviano, Andrea Arena, Vincenzo Gattulli
and Francesco Potenza
xii Contents
Part VII Calibration and Identification

Automatic Self-Calibration of Suspended Under-Actuated
Cable-Driven Parallel Robot using Incremental Measurements . . . . . . . 333
Edoardo Idá, Jean-Pierre Merlet and Marco Carricato
Eye-on-Hand Calibration Method for Cable-Driven Parallel Robots . . . 345
Nicolas Tremblay, Kaveh Kamali, Philippe Cardou, Christian Desrosiers,
Marc Gouttefarde and Martin J.-D. Otis
On the automatic calibration of redundantly actuated cable-driven
parallel robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
Han Yuan, Yongqing Zhang and Wenfu Xu
Towards a Precise Cable-Driven Parallel Robot - A Model-Driven
Parameter Identification Enhanced by Data-Driven Position
Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
Marcus Hamann, Pauline Marie Nüsse, David Winter
and Christoph Ament
Part VIII Applications

Design, Implementation and Long-Term Running Experiences
of the Cable-Driven Parallel Robot CaRo Printer . . . . . . . . . . . . . . . . . 379
Andreas Pott, Philipp Tempel, Alexander Verl and Frederik Wulle
A Dual Joystick-Trackball Interface for Accurate and Time-Efficient
Teleoperation of Cable-Driven Parallel Robots within Large
Workspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Kwun Wang Ng, Robert Mahony and Darwin Lau
Active Vibration Damping of a Cable-Driven Parallel Manipulator
Using a Multirotor System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
Yue Sun, Matthew Newman, Arthur Zygielbaum and Benjamin Terry
Reproduction of Long-Period Ground Motion by Cable Driven
Earthquake Simulator Based on Computed Torque Method . . . . . . . . . 415
Daisuke Matsuura, Taishu Ueki, Yusuke Sugahara, Minoru Yoshida
and Yukio Takeda
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Part I
Design
Planar Cable-Driven Robots with Enhanced Orientability
M. Vikranth Reddy1, N. C. Praneet2, and G. K. Ananthasuresh3

[1-2-3]
Indian Institute of Science, Bengaluru-560012, India
[email protected], [email protected], [email protected]
Abstract. Cable-driven robots consist of a moving platform to which cables are attached and
pulled using ground-mounted motors. Even though there are many advantages of these over con-
ventional robots, there is a serious limitation of restricted orientations of the platform. Further-
more, the orientation depends on the external load. In this work, the inherent limited orientation
attribute of planar cable robots is demonstrated with analysis and simulations. Also presented is
a concept and implementation of complete orientability of the end-effector attached to the moving
platform of a planar cable-driven robot by using only one additional cable. The new concept, by
design, gives rise to a moment load on the moving platform, which does not yet appear to have
been considered in cable-driven robots. Implications in this analysis and path planning are dis-
cussed. Working prototype of a planar three-cable robot with an extra cable to enhance the ori-
entability was built and tested. Additionally, for a given configuration and loading, cable tensions
were pre-calculated and stored in a structured database for expeditious execution of the path and
orientation of the robot.
Keywords: Planar cable-driven robot, orientations.
1 Introduction
Cable-driven robots are structurally similar to parallel robots with the fundamental dif-
ference that cables can only pull the moving platform but cannot push it. Due to this
feature, well-known results in robotics for trajectory planning and control must be mod-
ified to reflect the constraints of non-negative cable tensions [1]. The limitations of
these robots are restricted orientations of the moving platform and a constraint of non-
negative tensions. The past two decades have seen rapid growth in this area, which gave
rise to a variety of applications. Few applications of these include automated cranes [2-
4], ultra-high speed robots [5], and virtual reality interfaces [6]. Numerous research
studies have been reported for determining the conditions for obtaining the workspace,
stiffness, singularities, etc. They include NIST Robocrane [2], workspace analysis
methodology that can be applied for optimal design of the planar cable robots [7], con-
troller designs [8], and workspace analysis [9-11], and constraints on tensions are used
to check the feasible orientations of a rectangular platform suspended by cables in a
square workspace [12]. To the best knowledge of the authors, limited work has been
done on the orientation aspect of the moving platform of the cable-driven robots. In this
work, limited orientations of the moving platform of the cable-driven robots are dis-
cussed and a concept to obtain the complete orientability of the end effector mounted
on the moving platform with an extra cable and a linear spring is presented. The new
concept, by design, gives rise to a moment load on the moving platform. To confirm
© Springer Nature Switzerland AG 2019 3

A. Pott and T. Bruckmann (eds.), Cable-Driven Parallel
Robots, Mechanisms and Machine Science 74,
https://doi.org/10.1007/978-3-030-20751-9_1
4 M. V. Reddy et al.
the theory with experiments, an experimental test bed of a planar cable-driven robot
with three-cables and an extra cable to enhance the orientability of the end effector is
fabricated.
The organization of this paper is as follows: Section 2 highlights the restricted ori-
entations of the moving platform in workspace and a procedure is described to check
the feasible orientations of a planar cable robot. A concept for providing complete ori-
entability of the end effector mounted on a moving platform is presented in Section 3.
Implementation of this concept in a prototype along with results are presented in Sec-
tion 4. Discussion and closing remarks are in Section 5.
2 Limited orientability of the moving platform in cable-

driven robots
In this section, a model of a three cable and a four cable robot is discussed to highlight
the limited orientations of the platform.
2.1 Three-cable planar robot

A three-cable planar robot consists of a triangular platform constrained to move in a
plane that is connected by three cables to a base platform at three fixed points. Let us
consider a reference frame (O, X , Y ) fixed to the centroid of the base called the base
frame. Coordinates of the fixed points on the base frame and the contact points on the
moving platform are denoted by ( x fj , y fj ) and ( xmj , ymj ) respectively (where j =1, 2,
3). Selecting a reference point P ( x, y ) on the moving platform and a direction Y c fixed
relative to the moving platform, we obtain the moving frame ( P, X c, Y c) . Its orientation,
denoted by I , is the relative angle between the fixed Y axis and the moving Y c axis
with the centroid of the base frame as the reference. T j is the angle between the j th
cable and the X axis of the base frame, measured counter-clockwise. Fig. 1 shows the
representation of such a model. The procedure that is used to check the non-negative
tensions at a given point ( x, y ) for a given orientation I is taken from [12] and is
briefly presented below.
For the system to be in static equilibrium, we need to balance cable forces in X and
T
Y directions with the given external load f ª¬ Fx Fy M z º¼ , and also balance mo-
ment about the Z axis.
n
Fx ¦T j cos T j
j 1
n (1a)
Fy ¦T
j 1
j sin T j
n § xmj · § T j cos T j ·
Mz ¦ R¨ ¸u¨ ¸ (1b)
j 1 © ymj ¹ © T j sin T j ¹
Planar Cable-Driven Robots with Enhanced Orientability 5
Fig. 1. Model of the three-cable planar robot
§ cos I
sin I ·
where R ¨ ¸ is the rotation matrix; ( xmj , ymj ) are the initial coordinates
© sin I
cos I ¹
of the moving platform (where j = 1, 2, 3); T j is the tension in j th cable;
c ·
§ y fj ymj c ·
§ xmj § x· § xmj ·
Tj tan 1 ¨ ; and ¨ ¨ ¸ R¨ y ¸ .
¨ x xc ¸¸ c ¸¹
© fj mj ¹ © ymj © y¹ © mj ¹
To determine the tensions t [T1 T2 T3 ]T , Eqs. 1(a)-1(b), can be rewritten in ma-
trix form as follows.
Lt f (2)
§ ·
¨ ¸
¨ cos T1 cos T 2 cos T3 ¸
where L ¨ sin T1 sin T 2 sin T 3 ¸ (3)
¨ ¸
¨ § xm1 · § cos T1 · § xm 2 · § cos T 2 · § xm 3 · § cos T 3 · ¸
¨R¨ ¸u¨ ¸ R¨ ¸u¨ ¸ R¨ ¸u¨ ¸¸
© © ym1 ¹ © sin T1 ¹ © ym 2 ¹ © sin T 2 ¹ © ym 3 ¹ © sin T3 ¹ ¹
When rank (L) =3, we calculate the tensions as
t = L-1f (4)
If all the elements in t are non-negative, then the considered orientation at the given
point is feasible. If L is rank deficient, Eq. 4 is not applicable. It is possible have a
solution to Eq. (2) only if f is in the left nullspace of L and its components are all
non-negative [12]. The procedure to obtain non-negative tensions is described next.
L_aug = > L f @ (5)
rank (L) rank (L_aug)

t p = LT (LLT )-1 (-f)
(6)
N nullmatrix L
t = t p + ND
t p is the particular solution and its elements can be any real value; N is the null matrix
and its elements are real values; D is an arbitrary scalar when rank (L) =2 and is a vector
of 2 elements when rank (L) =1.
For example, consider the configuration as shown in Fig. 2. The coordinates of the
base frame and the platform (initially) are (0, 5.77), (5, -2.89), (-5, -2.89) and (0, 1.15),
T
(1, -0.58), (-1, -0.58). For an external force f >0 1 0@ and for an orientation
I 1q at a point ( x, y ) (2, 2) , tensions are computed using the Eqs. 1(a)-3, and 4(or
5 and 6). For this case, rank (L) =3. Using Eq. 4, the calculated values of tensions in the
cables are t [3.126 9.687 8.673]T . As the tensions in the cables are all non-nega-
tive for I 1q , it is considered as a feasible orientation. We get a range of such feasible
orientations for this point by varying I from > 90q,90q@ with an increment of 1º while
ensuring that there is no interference between the cables. The feasible orientations for
this example range from 1º-14º and are indicated by the yellow sector shown in Fig. 2.
For I 15q , the calculated tensions in the cables are t [1.11 0.25 0.03]T . Hence,
I 15q is not a feasible orientation. The plot illustrated in Fig. 3 indicates the region of
feasibility in orientations where tensions in the cables are non-negative.
For chosen limits of the workspace, we exhaustively search and determine the fea-
sible orientations at each point within the limits following the aforementioned proce-
dure. Feasible orientations at various points for two different configurations of the pla-
nar three-cable robot are shown in Fig. 4, indicated by green sectors in the circles. The
center line in the circle indicates 0º w.r.t the Y axis. Green sectors depicted on the left
and right to the center line indicate positive and negative orientations respectively.
Fig. 2. Tensions in the cables and orientation of the platform at a point ( x, y ) (2, 2) for a
rotation of 1º are depicted.
Fig. 3. Orientations feasible at (2, 2)
Fig. 4. Orientations are depicted when a force F = [0 -1 0]T (left) and F = [0 1 0]T (right) is applied
on the moving platform. Coordinates of the base frame are indicated in red.
2.2 Four-cable robot

A four-cable robot model is developed similar to the three-cable model. The platform
here is a quadrilateral suspended at four points by the cables. The notations used for
this model hold the same meaning as they did in the three-cable model. In this case, L
is (3 u 4) matrix and t is (4 u 1) vector. The procedure stated in Section 2.1 is used to
determine the feasible orientations with the modification that only Eqs. 5-6 are used to
compute the values of tensions in the cables for any rank (L). Results of simulations
for two such configurations are presented in Fig. 5. From Fig. 5, it is observed that the
range of orientations of the moving platform can be improved by varying the geome-
tries of the platform for the same workspace. However, it is still restricted. We propose
a concept to achieve complete orientability with the help of one extra cable and a trans-
lational spring which is detailed in the next section.
3 Complete orientability in cable robots

We now introduce a concept to achieve complete orientability of the end effector, using
an extra cable and a translational spring.
Fig. 5. A Force F = [0 -1 0]T is applied on the moving platform. Orientations are depicted for (i)
(left) configuration with square platform in a square workspace and (ii) (right) rectangular plat-
form in a square workspace. Note that for the same workspace, the rectangular platform has more
orientability.
3.1 An extra cable to enhance orientation

We propose to fit the end-effector to the moving frame using a revolute joint. The end-
effector is modelled as a spool with a helical groove. An extra cable is wound over the
spool along its groove and it is connected to a rotary actuator on one end and to a linear
spring on the other. Fig. 6 illustrates such a model, where the end-effector is mounted
on a triangular platform which is connected by three cables to the base frame at three
fixed points. The cable is kept in pre-tension by elongating the spring to a set limit. This
is done to reduce the backlash error arising at the interface of the cable and the end-
effector. When actuated, the difference in tensions between the two ends of the cable
results in the rotation of the winch at any given point in the workspace. The pre-loaded
spring acts as an additional actuator and assists in rotating the end effector back to its
initial orientation when the cable is released by the rotary actuator. As a result, complete
orientability of the end effector is realized. Furthermore, the design of the revolute joint
includes the flexibility of arresting it to the moving platform thereby giving rise to an
external moment load on the platform.
Fig. 6. (Left)Model for enhancing orientation. (Right) Spool with a helical groove with the cable
wound over it.
3.2 Re-calculated tensions for orienting in a given location

Ts1 and Ts 2 are the tensions in the cables connected to the actuator and the spring re-
spectively, as shown in Fig. 6. The resolved components of these tensions act along the
X and Y axes, which contribute to the external forces. Thus, tensions in the cables
have to be re-calculated taking these new external forces into count. Within the feasible
range of orientations at a point, there exists a unique orientation satisfying the given
objective function such as minimizing or maximizing sum of tensions. Hence, for a
given objective function and chosen path (finite number of points), there is a unique
locus of such orientations. Path planning of the planar cable robot is done by command-
ing the platform to move along such a locus. The procedure for doing this is briefly
presented next. We consider the centroid of the moving platform to be the reference
point for any specified path. First, for a given path, when there is no change in the
orientation of the end effector, Ts1 and Ts 2 are equal in magnitude and opposite in di-
rection. L is calculated at every point in the path as presented in Section 2.1. Ts1 and
Ts 2 are determined from the following equation:
Ts1 Ts 2 k ('x ) (7)
where 'x is the pre-set elongation of the spring to provide the required pre-tension in
the cable and k is the stiffness of the spring. Next, forces on the moving platform are
re-calculated using:
Fx _ new Fx Ts1 cos(T s1 ) Ts 2 cos(T s 2 )
(8)
Fy _ new Fy Ts1 sin(T s1 ) Ts 2 sin(T s 2 )
T s1 and T s 2 are the angles made by the cables w.r.t the X axis on the actuator and
spring sides respectively. Lastly, quantities in Eq. 4 (or 5 and 6) are re-calculated.
When the end effector is given certain rotation, we re-calculate the tensions with the
following modifications:
i. When the rotary actuator is winding the cable, quantities in Eq. 4 (or 5 and 6)
and 8 are re-calculated using:
Ts k ('x c)
Ts1 Ts 2 (e PT w ) (9)
Ts 2 Ts
where P is the coefficient of friction between the end-effector and the cable; T w the
wrapping angle of the cable on the end-effector; and 'x c is the elongation of the spring.
ii. When the rotary actuator is releasing the cable, the quantities in Eq. 4 (or 5
and 6) and 8 are re-calculated using:
Ts k ( 'x c)
Ts1 Ts 2 / (e PT w ) (10)
Ts 2 Ts
Path Planning. Fig. 7 presents the consolidation of three frames of simulation. The
platform is programmed to go along a vertically linear path from (0, -4) to (0, 4) w.r.t.
the base frame. The first frame depicts the end effector at (0, 0) oriented at 0º. The
second frame depicts the end effector undergoing a rotation of 30º at (0, -4). With the
orientation fixed at 30º, the platform moves vertically upward to (0, 4). In the third
frame, the end effector is rotated by -30º at (0, 4). The platform is then made to return
to its initial position as shown in the first frame.
Fig. 7. A Force of [0 5.59 0]T is acting on the triangular platform.
4 Prototype
A prototype of a planar cable robot was developed to confirm the theory by experi-
menting with three and four cables for various geometries of the moving platform. The
cable robot is an open-loop system containing electronic and mechanical modules. The
electronic module consists of an Apple MacBook Pro computer, an Arduino MEGA
2560 microcontroller board, micro-stepping motor drivers, and stepper motors. The
simulation code is written in MATLAB 2018b and the results are downloaded on to the
microcontroller board. The results of the simulations are encoded by the microcontrol-
ler board before being fed to micro-stepping motor drivers which then convert the input
command signals into the power necessary for energizing the stepper motor windings.
The mechanical module consists of a base plate, a movable platform suspended by the
cables, a spring, end effector, and pulleys secured to the shaft of the motors (Refer
Table 1 for specifications of the setup). Cable stoppers are used as the fixed points on
the base frame through which the cables are passed from pulleys mounted on the motors
to the platform. These fixed points are connected by the lines depicted in Fig. 8(a). A
photograph of the prototype is shown in Fig. 8.
Table 1. Specifications of the prototype
Component/ Part Specifications

Computer Apple MacBook Pro ± i5 (2016)
Microcontroller board Arduino MEGA 2560 board
Micro-stepping motor
TB6560
driver
NEMA ± 23 , holding torque = 1.89Nm at 2.8A/ per
DC stepper motor
phase
Power supply 12V 5A 60W SMPS unit
Mild steel ( length = 1m, breadth = 1m, thickness =
Base plate
2.2mm)
Aluminum alloy ( base = 180mm, height = 180mm, thick-
Triangular platform
ness = 8mm)
Alloy steel ( length = 121.24 mm, breadth = 70mm, thick-
Rectangular platform
ness = 15mm)
Cables Synthetic tennis gut (nylon) , diameter =1.56mm
Aluminum (Outer diameter = 42.60mm , width = 32mm,

Pulley
pitch = 4mm)
End effector (spool Aluminium ( pitch = 4mm, outer diameter = 42.60mm,
with helical groove) width = 40mm)
Spring Spring steel, D2 grade, stiffness = 3.2517N/cm
Fig. 8. a) Prototype of the three-cable robot b) Pre-loaded spring c) End effector with cable wound
d) Triangular platform e) Pulley mountable on motor shaft f) Cable stopper g) Motor driver
4.1 Three-cable robot with complete orientability

An end-effector modelled as a spool with a helical groove that can freely rotate about
its axis is mounted on a triangular platform. For our experiments, we considered two
windings of the cable over the end effector with one end of the cable wound on a pulley
mounted on the shaft of a DC stepper motor, and the other end hooked to a translational
spring. Using the method presented in Section 3, the end effector can be rotated by any
angle at any point in the workspace. The path considered in Section 3.2 is depicted in
Fig. 9 where the end-effector is rotated by an angle of 30°when it arrives at the end
points of a vertically linear path (from (0, -4) to (0, 4)). A 10% error was observed in
the rotation of the end effector. We attribute this error to the lack of sufficient friction
between the cable and the smooth surface of the pulley.
5 Closure
Inherent restricted orientation attribute of the moving platform in planar cable robots is
explained with the aid of simulations results for three- and four-cable robots. A concept,
involving the use of one extra cable and a translational spring, to achieve complete
orientability of the end effector is explained in Section 3. Experimental values of ori-
entations obtained deviated from the orientation values considered in Section 3.2 by
10%. This may be attributed to slip of the cable on the spool and other reasons that will
be considered in our future work. Closed-loop control is also likely to improve the ac-
curacy of position and orientation.
Fig. 9. Enhanced orientability of the end effector. Note that in image 2 and 3 there is a rotation
of about 27º (indicated by the black pointer attached to the end-effector).
References
1. Oh, S. R., Agrawal, S. K.: Cable suspended planar robots with redundant cables: Controllers
with positive tensions. IEEE Transactions on Robotics, 21(3), 457-465.
2. Albus, J., Bostelman, R., Dagalakis, N.: The NIST robocrane. Journal of robotic sys-
tems, 10(5), 709-724.
3. Yang, L. F., Mikulas, M. Jr.: Mechanism synthesis and 2-D control designs of an active
three cable crane. In 33rd Structures, Structural Dynamics and Materials Conference (p.
2342).
4. Shiang, W. J., Cannon, D., Gorman, J.: Optimal force distribution applied to a robotic crane
with flexible cables. In Robotics and Automation, 2000. Proceedings. ICRA'00. IEEE Inter-
national Conference on (Vol. 2, pp. 1948-1954). IEEE.
5. Kawamura, S.: Development of an ultrahigh speed robot FALCON using wire drive sys-
tem. Robotics and Automation, 215-220.
6. Eichstadt, F., Campbell, P., Haskins, T.: Tendon suspended robots: Virtual reality and ter-
restrial applications (No. 951571). SAE Technical Paper.
7. Fattah, A., Agrawal, S. K.: On the design of cable-suspended planar parallel robots. Journal
of mechanical design, 127(5), 1021-1028.
8. Alp, A. B., Agrawal, S. K.: Cable suspended robots: design, planning and control. In Robot-
ics and Automation, 2002. Proceedings. ICRA'02. IEEE International Conference on (Vol.
4, pp. 4275-4280). IEEE.
9. Barette, G., Gosselin, C. M.: Kinematic analysis and design of planar parallel mechanisms
actuated with cables. In Proceedings of ASME Design Engineering Technical Confer-
ence (pp. 391-399).
10. Fattah, A., Agrawal, S. K.: Workspace and design analysis of cable-suspended planar paral-
lel robots. In ASME 2002 International Design Engineering Technical Conferences and
Computers and Information in Engineering Conference (pp. 1095-1103). American Society
of Mechanical Engineers.
11. Fattah, A., Agrawal, S. K.: Design of cable-suspended planar parallel robots for an optimal
workspace. In Proceedings of the workshop on fundamental issues and future research di-
rections for parallel mechanisms and manipulators (pp. 195-202).
12. Roberts, R. G., Graham, T., Lippitt, T.: On the inverse kinematics, statics, and fault tolerance
RIFDEOHဨVXVSHQGHGURERWV Journal of Robotic Systems, 15(10), 581-597.
Chain Driven Robots: An Industrial Application
Opportunity. A Planar Case Approach
Guillermo Rubio-Gómez1, David Rodríguez-Rosa1, Jorge A. García-Vanegas2, Anto-

nio Gonzalez-Rodríguez1, Fernando J. Castillo-García1, and Erika Ottaviano3[0000-0002-
7903-155X]
1 University of Castilla-La Mancha. Avda. Carlos III. Real Fábrica de Armas, TO 45071, Spain
2 University of Ibagué. Carrera 22 Calle 67 B/Ambalá. Ibagué, Tolima, Colombia
3
University of Cassino and Southern Lazio, Viale dell'Università, 03043 Cassino, Italy
[email protected]
Abstract. This work presents Chain-Driven Parallel Robots replacing cables by

chains. The use of conventional sprockets adds some important advantages with
regards to Cable-Driven Parallel Robots. The most important ones are: a) no
drum is required; b) no cable plasticity limitation must be imposed; c) using coun-
terweights the manipulator can move the required payload with low motorization.
In this paper some design considerations for allowing an accurate positioning and
maximizing the robot workspace are presented. As example, a 2 Degrees-of-
Freedom planar manipulator has been designed and built. The robot can com-
mand a 60 kg payload into a 0.8m ൈ 1.8m workspace using only two 150W DC
motor.
Keywords: Chain-Driven Robot, Parallel Robot, Industrial Applications.
1 Introduction
The use and applications of Cable Driven Parallel Manipulators (CDPMs) are increas-
ing in the last decades thanks to the main advantages with respect their classical coun-
terparts. These advantages refer to the very large workspace, relatively lightweight
structure, overall reduction of the costs related to the main components of the system.
Application of CPDMs for high-speed pick and place manipulators taking advantage of
the low-inertia are reported in [1, 2]. Applications related to the large workspace in-
clude displacement of materials over large areas, positioning systems for heavy objects,
rescue operations, service, assistance, rehabilitation, [3, 4, 5, 6].
Main components of the CDPMs are the fixed frame (or structure), the end-effector,
the cables, the actuation and transmission system. Each of these elements should be
correctly modelled in order to get a good mechanical design and realistic simulations
RI WKH FDEOH V\VWHP¶V EHKDYLRU, [7, 8, 2]. One of the challenges in the study of the
CDPMs is the cable and its correct modeling, since it is the main tool for operating
payloads and influences the kinematic, static, dynamic analyses and control.

14 G. Rubio-Gómez et al.
Several studies on cable-driven robots deal with their kinematics [9] under static or
quasi-static conditions [10, 1]. Even in this condition is immediately unclear that assur-
ing cables to be in tension affects the workspace and the available operational wrench
of such manipulators, which turns out to be quite different from that of conventional
parallel mechanism. As clearly reported in the relevant literature, cables cannot be mod-
eled as rigid, contrary to conventional limbs in parallel manipulators. This feature has
led to model the stiffness of cable-actuated mechanisms. Models including sagging ca-
bles have been discussed in [11, 12, 13, 14], all of those models have advantages but
possess drawbacks. The main issue is computational complexity and computational
time, vs accurate modeling [14].
CDPMs are often classified referring to the end-effector DOFs that can be controlled.
When the CDPR is in a crane configuration, and gravity acts like an additional cable
[7] not all the end-effector DOFS can be controlled. Underconstrained cable robots of-
fer several advantages with respect to fully constrained ones. A smaller number of ca-
bles means reduced number of actuators, overall costs and setup time, improved ease
of assembly and a lower likelihood of cable interference. These features make under-
constrained CDPRs well suited for all those applications in which not all the end-effec-
tor DOFs can be controlled but having the advantage of decreasing complexity, such as
rehabilitation tasks, rescue operations, positioning of objects over large workspaces,
[15]. Besides the interest towards fully constrained robots the suspension of the end-
effector has recently drawn attention in under-constrained CDPRs [16, 2].
Since cables can exert forces only when they are taut, additional actuators can be
used with respect to the number of degrees of freedom (DOFs) to be controlled to pre-
vent cables from becoming slack. This is often the case for redundantly actuated cable-
driven robots. One of the main difficulties in the modelling and control of suspended
manipulators is the cable behavior that influences correct positioning, kinematics and
dynamics. The kinematic problem has been addressed solving simultaneously kine-
matic compatibility and equilibrium equations, as proposed in [7, 17, 13, 18]. Conven-
tional displacement analysis problem was defined as geometrico-static as in [7] also
expressing some connections between stability and energy [10].
As a completely new approach, in this paper we propose the use of chain instead
cables to drive and locate the end-HIIHFWRU7RWKHDXWKRUV¶NQRZOHGJHWKHXVHRIFKDLQ
instead cables in CDPMs has never been proposed before, therefore, we are addressing
the study of Chain Driven Robot (CDRs).
The use of chain instead cables may be effective for a number of advantages that
make them suitable for displacing large payloads and gaining accurate positioning.
In traditional CDPMs, in order to modify the cable length, cables are usually wound
around a spool and the spool is rotated to release or retract the cable through suitable
actuation system mounted on the same axis of the spool. The anchor position is the
point at which the cable leaves the spool. The change in diameter of the spool due to
the cable being wound results in the change in location of anchor position. When the
cable is retracted the diameter of the spool gets larger and when the cable is let out the
diameter gets smaller. In addition, as the cable travels along the axis of spool, while it
is retracted or released, it moves out of the plane of mobile platform. These two issues
are usually avoided in practical mechanical design of a cable-based manipulator by
Chain Driven Robots: An Industrial Application… 15
placing a pulley between the spool and the cable attachment point on the mobile plat-
form, which gives uncertainties in the determination of the cables length and position-
ing, as reported in [2]. Additional problems are related to the axial and transversal flex-
ibility of the cable, as previously reported. The use of chain prevents from problems
related to the drum and pulleys. Roller chain drives are traditionally and widely used in
various high-speed, high-load and power transmission applications, because their main
characteristics of high reliability and durability, on the contrary of cables they are in-
extensible, there is no slippery.
In this paper, we evaluate the use of chain driven robot for industrial application and
detail the design requirements for a proper design.
2 Chain-Driven Robot vs Cable-Driven Robot
As introduced in Section 1, chain-driven robots present several important advantages

in comparison to cable-driven ones. The main advantages, which have been considering
for the design, are:
x No drum is required: Using chain and sprockets, we avoid the use of any mech-
anism for roll in and out cable (like drums, see Fig. 1a and 1b).

a) b) c)
Fig. 1. Drum vs. Sprocket
x Cable plasticity limitation: When parallel robots are commanded by means of

cable all transmission elements (pulleys or drums) must be designed considering
the minimum radius to avoid cable plasticity. In the case of using chains the
restriction is regarded to the size of the chain link. In this sense with cables,
when payload increase, cable size also increases, and this requires that all drums
and pulleys must be larger for avoiding cable plasticity. With chain, any com-
patible sprocket can be applied without plasticity limitation (see Fig. 1c, where
both, small and large sprockets can couple a heavy chain).
x Counterweight for decreasing torque requirements: As chain and sprocket are
coupled, counterweight can be used, and motor torque needed, ߬௜ , notoriously
decreases (see Fig. 2).
On the other hand, using chains for commanded the end-effector, has several prob-
lems, still opened:
x Dynamics model implications of using chains owing to the non-negligible
weight of chains.
x Minimum tension of chains for ensuring an accurate positioning (sagged

chains).
x Using chains for commanded low payloads.
x Spatial designing for connecting frame to end-effector by means of chains and
swivel sprockets.
Wi
end-effector counterweight
Fig. 2. Counterweight for decreasing needed torque.
3 Preliminary Prototype
3.1 Prototype design
In this paper a 2 degrees-of-freedom planar chain-driven robot has been designed and
built. Reflexive pulleys have been included in end-effector for an accurate kinematic
model [2]. Figure 3 shows represent the robot scheme where ܹ and ‫ ܪ‬are the width
and height of the frame and, ‫ ݓ‬and ݄ the width and height of the end-effector, which
planar pose is ሾ‫ݔ‬௘ ǡ ‫ݕ‬௘ ሿ் . The mass of both counterweights has been noted as ݉ଵ and
݉ଶ , respectively, since the mass of end-effector is ݉௘ . Finally, all sprockets radius is
‫ݎ‬, since the motors angular positions are ߙଵ and ߙଶ . Note that motor 1 (ߙଵ ) commands
the pair of chains on the left so they are parallel and the same with motor 2 (ߙଶ ) and the
pair of chains on the right. This scheme maintains constant the angle of end-effector
[1].
t
r
Ś T1
T2
T1 T2
ǁ
D1 D2
me
Ś
,
m1 m2
[x e , ye ]
Fig. 3. Scheme of 2DOF planar chain-driven robot

3.2 Experimental platform
The experimental parameters are summarized in Table 1.
Table 1. Parameters of experimental platform
Parameter Value
ܹ (m) 1.680
‫( ܪ‬m) 2.310
‫( ݓ‬m) 0.320
݄ (m) 0.140
‫( ݎ‬m) 0.050
The values of end-effector mass, ݉௘ , and counterweights, ݉ଵ and ݉ଶ will be con-

figured for maximizing workspace ensuring an accurate positioning. Figure 4 shows
the final aspect of the experimental platform.

a) b)

c) d)
Fig. 4. Experimental platform
The robot is motorized by two 150W DC motor RE40 Maxon Motor coupled to a
74:1 gearbox. The nominal values of the motor are 0.177 Nm as nominal torque (13.098
Nm at the gearbox output) and 6940 rpm as nominal speed (93.78 rpm at the gearbox).
3.3 Tension Analysis
In this section we analyze the minimum tension of chain for an accurate positioning.
Figure 5 represents different end-effector pose for a 2 kg payload. Note that sagging
chains are present is different end-effector position when chain tension is lower than a
limit which depends of the chain properties.

Fig. 5. End-effector poses for a 2kg payload
For determining the minimum tension allowed for accurate positioning we have
blocked the sprockets at a position corresponding to a centered end-effector pose with
known chains angle (ߠଵ and ߠଶ in Fig. 3). For ensuring non-sagging chains, we have
started the experiments with a very-large payload, ݉௘ ൌ30 kg, and we have decreased
it until we note that sagged effect is present. This lower limit payload corresponds to
the lower limit tension (ܶଵ and ܶଶ in Fig. 3):
௠೐ ௚
ܶ௜ǡ௠௜௡ ൌ (1)
ଶୱ୧୬ሺఏ೔ ሻ
where ܶ௜ǡ௠௜௡ is the minimum tension of chain ݅. The sagged effect has been detected by
measuring the variation of the end-effector pose allowing a maximum distance varia-
tion of 1 mm. Figure 6 summarizes the experiments, where the end-effector pose has
been established for ߠଵ ൌ ͳ͵ͷι and ߠଶ ൌ Ͷͷι. The resulting payload for non-sagged
effect was ݉௘ ൌ10.20 kg, which correspond to ܶଵǡ௠௜௡ ൌ ܶଶǡ௠௜௡ =70.81 N.

a) b) c) d)
Fig. 6. Minimum tension determination
3.4 Workspace Analysis
Once, minimum tension for accurate positioning has been determined, the end-ef-
fector mass, ݉௘ , together with the mass of counterweights, ݉ଵ and ݉ଶ , can be obtained
for maximizing the workspace for a given maximum torque of motor/gearbox set
(߬௠௔௫ ൌ13.098 Nm). The torque of motor will be exerted for moving end-effector. On
the other hand, motors should move the end-effector in the opposite direction of coun-
terweight effect. These both conditions can be written as:
ȁ߬௜ ȁ ൌ ȁ‫ ݎ‬൉ ሺܶ௜ െ ݉௜ ݃ሻȁ ൏ ߬௠௔௫ (2)
being ߬௜ the torque exerted by motor ݅.

The optimal workspace is obtained with ݉௘ ൌ ͸ͶǤͺ Kg and ݉ଵ ൌ ݉ଶ ൌ ͶͲǤ͹ Kg.
(see Fig. 7a). If counterweights are not properly designed and are lower than the optimal
one, workspace is reduced as in Fig. 7b. If end-effector mass is not properly designed
and is lower than the optimal one, workspace is reduced as in Fig. 7c.

a) b) c)
Fig. 7. Workspace analysis
Finally, Fig. 8 represents the maximum rectangular workspace which is 0.810 m ൈ

1.86 m. Note that the designed robot can manipulate a payload of 64.8 Kg into a 0.810
m ൈ 1.86 m workspace only with 2 motors of 150W.
Fig. 8. Maximum rectangular workspace
4 Results
4.1 Experiments setup
A Graphical User Interface (GUI) has been designed on MATLAB. This GUI allows
establishing the manoeuvres parameters such as initial and final pose or time profile of
the trajectory. The GUI send messages to a CAN-BUS node used as a bridge between
computer and the controller of each motor (Maxon Motor EPOS 2 70/10). Finally, the
controllers execute the motor orders and, when the manoeuvre finishes, return the con-
trol to the GUI for following movements (see Fig. 9).
Fig. 9. Hardware architecture of the robot

4.2 First movements
For the first experiments, a 30 Kg payload (2 ൈ 15 Kg) has been placed at the end-
effector, since both counterweights have been set to 15 Kg. Using the hardware archi-
tecture shown in Fig. 9, some controlled movements have been executed. To achieve
the desired end-effector movements the Inverse Kinematic Transform of the robot has
been developed as in [2]. The experimental results shown as the 2 ൈ 150W motorization
is able to move the required payload into the workspace although no positioning accu-
racy has been still evaluated.
Fig. 2. First experiments setup.
5 Conclusions
In this paper a chain-driven parallel manipulator has been presented, replacing cables
by chains for commanding the movements of a payload. This new kind of manipulator
can be used for industrial applications, for example, pick and place. The main ad-
vantages of these manipulators are:
x No drum is required to be designed.
x The size of the pulleys, now sprockets, can be reduced because no plasticity
limitation should be considered.
x The use of counterweights reduces the needed torque for moving the payload.
On the other hand, the implications of using chains over the dynamics model should
be explored.
In this paper several design considerations have been presented, such as the trans-
mission design, based on commercial sprockets, or the end-effector and counterweight
design for maximizing the feasible workspace of the robot. As example, a 2 Degrees-
of-Freedom planar manipulator has been designed and built. The robot can command a
60 kg payload into a 0.8m × 1.8m workspace using only two 150W DC motor.
Acknowledgements
This work was partially supported by EU Call RFCS-2017 through the research project
DESDEMONA (grant agreement number 800687).
References
1. Castelli, G., Ottaviano, E., Gonzalez, A.: Analysis and simulation of a new Cartesian cable-
suspended robot. Proc. IMechE Part C: J. Mechanical Engineering Science 224: 1717{1726
(2010). doi:10.1243/09544062JMES1976
2. Gonzalez-Rodriguez, A., Castillo-Garcia, F.J., Ottaviano, E., Rea, P., Gonzalez-Rodriguez,
A.G.: On the effects of the design of cable-Driven robots on kinematics and dynamics mod-
els accuracy. Mechatronics 43: 18{27 (2017). doi:10.1016/j. mechatronics.2017.02.002
3. Albus, J., Bostelman, Dagalakis N.: The nist robocrane. Journal of Robotic Systems10 (5),
709{724 (1993).
4. Castelli, G., Ottaviano, E., Rea, P.: A Cartesian Cable-Suspended Robot for improving end-
users' mobility in an urban environment. Robotics and Computer-Integrated Manufacturing
30(3): 335{343 (2014). doi:10.1016/j.rcim.2013.11.001
5. Havlik, S.: A cable-suspended robotic manipulator for large workspace operations. Comput.
Aided Civil Infrastruct. Eng. 15(6): 56{68 (2000).
6. Nan, R., Peng, B.: A chinese concept for 1km radio telescope. Acta Astronautica 46 (10-
12): 667{675 (2000).
7. Abbasnejad, G., Carricato, M.: Direct geometrico-static problem of underconstrained cable-
driven parallel robots with n cables. IEEE Transactions on Robotics 31 (2): 468{478 (2015).
8. Du, J., Agrawal, SK.: Dynamic Modeling of Cable-Driven Parallel Manipulators with Dis-
tributed Mass Flexible Cables. J. Vib. Acoust. 137(2): 1{8 (2015). doi:10.1115/1.4029486
9. Bosscher, P., Ebert-Uphoff, I.: Disturbance robustness measures for underconstrained cable-
driven robots. Proc. IEEE Int. Conf. Robot. Autom., Orlando, FL: 4205{4212 (2006).
10. Carricato, M., Merlet, J-P.: Stability analysis of underconstrained cable-driven parallel ro-
bots. IEEE Trans. on Robotics 29(1):288{296 (2013).
11. Kozak et al.: Static Analysis of Cable-Driven Manipulators with Non-Negligible Cable
Mass. IEEE Transactions on Robotics 22(3):425{433 (2006).
doi:10.1109/TRO.2006.870659
12. Merlet, J.-P.: A generic numerical continuation scheme for solving the direct kinematics of
cable-driven parallel robot with deformable cables. In IEEE International Conference on
Intelligent Robots and Systems: 4337{4343 (2016).
13. Ottaviano, E., Gattulli, V., Potenza, F.: Elasto-Static Model for Point Mass Sagged Cable-
Suspended Robots. Advances in Robot Kinematics 2016. Springer Proceedings in Advanced
Robotics, vol 4. Springer, Cham (2018).
14. Pott, A, Tempel, P.: A Uni_ed Approach to Forward Kinematics for Cable-Driven Parallel
Robots Based on Energy. Advances in Robot Kinematics (ARK): 401{409 2018.
doi:10.1007/978-3-319-93188-3 46.
15. Merlet, J.-P., Daney, D.: A portable, modular parallel wire crane for rescue operations. IEEE
International conference on robotics and automation, Anchorage, Alaska, 3{8 May 2010:
2834{2839 (2010).
16. Collard, J.F., Cardou, P.: Computing the lowest equilibrium pose of a cable-suspended rigid
body. Optimization and Engineering 14(3): 457{476 (2013).
17. Merlet, J.-P., dit Sandretto, A.: The Forward Kinematics of Cable-Driven Parallel Robots
with Sagging Cables. In Cable-Driven Parallel Robots: 3{15 (2014) Springer.
18. Pott, A.: An algorithm for real-time forward kinematics of cable-driven parallel robots. In
Advances in Robot Kinematics: 529{538 (2010), Springer.
Non-slipping Conditions of Endless-Cable
Driven Parallel Robot by New Interpretations of
the Euler-Eytelwein’s Formula
Takashi Harada1 and Koki Hirosato2

1
Kindai University, 3-4-1 Kowakae Higashiosaka Osaka 577-8502, Japan,
[email protected],
WWW home page: https://sites.google.com/site/parallelmech/
2
Kindai University, Graduate School of Science and Engineering Research
Abstract. Non-slipping conditions of endless-cable driven parallel robot

(E-CDPR) which enables unlimited rotation of the hand are discussed
in this paper. Instead of fixing the end of the cable to the pulley and
the winch, endless-cable (loop-cable) is turned around the endless-pulley
and the endless-winch. Because friction forces between the cable and the
drum transfer the cable tension, slipping of the cable which is dominated
by the well-known Euler-Eytelwein’s formula is taking into consideration
of the statics of the E-CDPR. In this paper, a new interpretation of the
Euler-Eytelwein’s formula is proposed by using a graph that the non-
slipping condition is expressed as an area in the cable tensions. Equations
of the statics of the endless-pulley and the endless-winch are superim-
posed on the graph, then the non-slipping conditions of the E-CDPR are
derived.
Keywords: endless-Cable, friction drive, the Euler-Eytelwein’s formula,

statics
1 Introduction
Cable-driven parallel robots (CDPRs) have excellent characteristics such as high
speed and large translational workspace [1]. There are a lot of researches and ap-
plications of the CDPSs using these characteristics [2][3]. However, the rotational
workspace of the CDPR is limited by the collisions of the cables. Additional ro-
tational DOF is one of the most promising ideas for solving this problem[4]
[5][6][7]. In this paper, a novel CDPR which enables unlimited rotation about
one axis is proposed. Our idea is similar to the drum-embedded CDPR in [4],
but be different in that the drum is rotated unlimitedly by a novel endless-cable
driven mechanism. Instead of fixing each end of the cable to the pulley and the
winch, looped endless-cable is turned around the endless-pulley and the endless-
winch. Because friction forces between the cable and the drum transfer the cable
tension, slipping of the cable is taking into consideration of the statics of the
endless-cable driven parallel robot (E-CDPR). The Euler-Eytelwein’s formula
[8][9] is well known as when the ratio of the cable tensions Tt /Ts between the

24 T. Harada and K. Hirosato
tight side Tt and the slack side Ts is less than eμθ , then the cable does not slip on
the drum. Where, μ is the coefficient of friction between the cable and drum ma-
terials, and θ is the total contact angle swept by all turns of the cable, measured
in radians. In this paper, a new interpretation of the Euler-Eytelwein’s formula
is proposed by using a graph that the non-slipping condition is expressed as an
area in the cable tensions. Equations of the statics of the endless-pulley and the
endless-winch are superimposed on the graph, then the non-slipping conditions
of the E-CDPR are derived.
2 Endless-cable driven parallel robot
Fig. 1. Schematic image of Endless-cable driven parallel robot: OCTABE
Figure 1 represents six-dof eight-cables driven parallel robot, dubbed OC-

TAVE [7]. The moving part of OCTAVE generates six-dof motions, i.e., three
translational motions along the xyz axis and three rotational motions around
the xyz axis. An endless-pulley is embedded inside the moving part for expand-
ing the rotational workspace around the z axis. Instead of fixing the end of the
cable to the pulley or the moving part, two of eight-cables, C1 -C2 ,. . . ,C7 -C8
are looped then be turned around the endless-pulley. Thus OCTAVE is driven
by four loop-cables (endless-cables). All endless-cables are wraped around the
identical endless-pulley. The driving side of the loop-cable is comprised of two
endless-winches and a tensioner by constant pressure spring for the purpose
winding the cable extra length when the moving part moves. These mechanisms
enable the endless-pulley to unlimited rotation. Force of the tensioner acts as a
bias torque for the motors driving the winch, and it effectively utilizes the mo-
tor torque range on the negative side to the cable tension on the side from the
winch to the pulley. In addition, the bias force makes the cable wrapped around
Non-slipping Conditions of Endless-Cable Driven Parallel Robot… 25
the winch less slippery. Thus, OCTAVE totally has seven-dof mechanism, which
means that OCTAVE has one kinematic redundancy for six-dof hand attached
at the tip of the endless-pulley. The seven-dof mechanism of Octave is controlled
by eight-cables, which means that OCTAVE has one actuation redundancy like
as the conventional CDPRs.
3 Statics of the endless-pulley

3.1 single-drum and double-drums
Fig. 2. Options of the endless-pulley
There are two typical types of the endless-pulley; i.e., the single-drum (non-
grooved) and the double-drums (grooved) as shown in Fig. 2 (a) and (b). Figure
2 (a) illustrates an endless-pulley of single-drum is driven by a cable by the
frictions between the drum and the cable. The cable is several turned around
the drum such as U-shaped capstan drum. The generated torque τp of the pulley
is given by the cable tensions ta and ta as
τp = rp (ta − tb ) (1)
rp represents the radius of the drum.
In the case of the double-drums as shown in Fig. 2 (b), the cable is alternately
several turned around and between the main grooved-drum and the idle grooved-
drum. When the winding number of the cable is large, the double-drum is suit-
able because the cable is stably guided by the grooves. Assuming that the gen-
erated torque of the idle pulley is zero, one obtains,
ta1 = ta , ta2 = tb1 , ...., ta(i) = tb(i−1) , ...., tb(n) = tb (2)
Thus, the generated torque of the pulley is given as

n n

τp = rp tai − tbi = rp (ta − tb ) (3)
i=1 i=1
The relation between the tensions of the cable and the generated torque of the
pulley becomes same as a single-drum pulley of Eq. (1).
Fig. 3. Schematic images of cable-driven pulleys
3.2 A new interpretation of the Euler-Eytelwein’s formula

The Euler-Eytelwein’s formula [8][9] is well known with Fig.3 (a) as when the
ratio of the cable tensions Tt /Ts between the tight side Tt and the slack side
Ts (Tt > Ts ) is less than eμθ , then the cable does not slip on the drum. The
non-slipping condition of the cable on the drum is given by
Tt
< eμθ (4)
Ts
Where, μ is the coefficient of friction between the cable and drum materials,
and θ is the total contact angle swept by all turns of the cable, measured in
radians. When the pulley rotates at high speed, the centrifugal force of mv 2 acts
on the cable then the frictional force decreases. In this research, it is assumed
that a nylon cable with a diameter of 3 mm (7 g/m) is operated at a maximum
speed of 30 m/s. At that time, bias tension of approximately mv 2 = 60 N is
applied for compensating the influence of the centrifugal force. For convenience,
the influence of the centrifugal force is omitted in this paper. The contact angle
θ of the double-grooved-drums of Fig. 2 (b) can be made larger, thus be grippy
than the single-drum.
In this subsection, we expand the understanding of the Euler-Eytelwein’s
formula and propose a new interpretation of the formula for the E-CDPR. In
the case of the E-CDPR, the endless-pulley must generate not only positive
(CCW direction) torque τp 0 but also negative (CW direction) torque τp < 0
while changing the tight and slack sides of the cables. Thus the non-slipping
condition of the cable on the drum for the E-CDPR is given as
⎧
ta
⎨ < eμθ (ta ≥ tb )
⎪
⎪
tb (5)
⎪ t a 1
⎩ > μθ (ta < tb )
⎪
tb e
In the case of the double-drums pulley, the total contact angle θ is given by the
sum of the contact angle swept by all turns of the cable on the driven pulley.
Rewite Eqs.(1) and (3) in vector form as
ta
τp = rp −rp = JT t (6)
tb
where JT represents the transposed Jacobian. The cable tension t is derived by

inverting Eq. (6) as

1 √1
2rp
t = JT + τp + hσ = τp + 2 σ (7)
− 2r1p √1
2
JT + represents the pseudo-inverse of JT , which gives the minimum norm solution

of Eq. (6). h represents the vector spanning the null space of JT , and σ is an
arbitrary scholar. This situation is illustrated in Fig. 3 (b).
Relations of the non-slipping conditions of Eq. (5), generated torque of the
pulley by the cable tensions of Eq. (6), and the cable tensions for desired torque
of Eq. (7) are newly interpreted by the graph in Fig.4. The vertical axis and
the horizontal axis represent cable tensions ta and tb , respectively. The non-
slipping conditions of Eq. (5) is represented by the light gray area in Fig. 4.
Note here that the area includes both (ta ≥ tb ) and (ta < tb ) in Eq. (5). The
inclinations of the upper and lower border lines lu and lb of the non-slipping area
are represented by eμθ and 1/eμθ , respectively. If μ, the coefficient of friction, gets
larger, i.e., materials or surface texture of the pulley and/or cable are selected
less slippy, or θ, the total contact angle between the cable and drum, gets larger,
i.e., the numbers of turns of the cable around the drum are increased, then
the inclinations of the upper borderline lu of eμθ is increased and that of lower
borderline lb of 1/eμθ is decreased, thus the borders of the non-slipping area is
expanded from lu to lux and from lb to lbx as shown in the figure.
Fig. 4. Statics of the endless-pulley on the new interpretation of the E-E’s formula
Here and the former part of the following subsection, we focus on the area of
τp ≥ 0 (ta ≥ tb ). Line l represents the constant torque line for the cable tensions
of Eq. (7). Cable tensions at the point (tb , ta ) on the line generate the same
torque of τp at the pulley. Cable tension at the point A on the line corresponds
to the minimum-norm solution of Eq. (6), namely, t = JT + τp . As shown in Fig.4,
cable tensions at point A are not only out of the non-slipping area but also one
of the cable tension ta is not a positive value. Unit vector h and scholar σ in Eq.
(7) correspond to the unit direction vector of the line of the constant torque and
the distance from the minimum-norm solution along the line, respectively. In the
physical meanings, hσ corresponds to the bias tensions which do not affect the
output torque of τp . In the case of the E-CDPR, the bias tensions of cables can be
controlled so as not only the cable tensions are not negative values, but also the
cable satisfies the non-slipping condition of Eq. (5). When the proportionality
constant σ of the bias tensions is gradually increased from zero to the positive
value, the corresponding point of the tensions on the line moves from the point
A(σ = 0)√
to the right upper direction. When the point crosses over the point
B(σ = 2r2p ), both cable tensions ta and tb become positive but still be out of the
non-slipping area.
3.3 Non-slipping condition of the endless-pulley

When the point crosses over the point C in Fig. 4, cable tensions enter the non-
slipping area. After the point C, cable tensions keep in the non-slipping area
even if the proportionality constant σ of the bias tensions is set larger value.
The non-slipping condition of σ in τp ≥ 0 (ta ≥ tb ) is given as,
√
2τp eμθ + 1

σ> (8)
2rp eμθ − 1
In the case of τp < 0 (ta ≥ tb ), the constant torque line, the minimum-norm
solution, the border of the positive tensions, and the border of the non-slipping
conditions are represented by l , A , B , and C in Fig. 4. The non-slipping
condition of σ is given as,
√
2τp eμθ + 1

σ>− (9)
2rp eμθ − 1
The new interpretation of the Euler-Eytelwein’s formula using Fig. 4 is con-

venient not only for understanding the non-slipping conditions of the endless-
pulley of the E-CDPR but also for designing the non-slipping condition of the
endless-winch which is discussed in the following section.
4 Statics of the endless-winch

4.1 Endless-winch system in the E-CDRP
Figure 5(a) represents the endless-winch system in the E-CDPR, which is also
represented in Fig 1. The endless-winch system is comprised of two endless-
winches A and B, an endless-cable and a tensioner. In Fig 5, endless-winch is
comprised of the double-drums in which one drum is driven by a motor. The

tensioner is comprised of two fixed pulleys Pa and Pb , a moving pulley Pm guided
by a slider and pulled by a constant force spring. ts represents the constant
pulling force by the spring. Two motors and the tensioner generates the cable
tensions ta and tb which drive the endless-pulley of the previous section.
Fig. 5. Endless-winch system and its simplified model
Figure 5(b) represents the simplified model of an endless-winch for the static
analysis. tw represents the tension to the endless-pulley which corresponds to
the tensions ta or tb in Fig. 5(a). The double-drums is replaced by a single-drum
of both values of coefficient of friction μ between the cable and drum materials,
the total contact angle θ swept by all turns of the cable, and the radius rp of the
drum are equivalent. tb represents the bias tension by the tensioner. In the case
of Fig 5(a) in which the endless-cable is pulled by one moving pulley, tb equals
to fs /2. τw represents the torque of the motor.
4.2 Statics and non-slipping condition of the endless-winch

The goal of this section is to derive the relationships between the range of the
cable tension tw , the range of the motor torque τw and bias tension tb . These
relationships are useful for designing the endless-winch system.
The simplified model of the endless-winch in Fig. 5 (b) looks similar to the
model of the endless-pulley in Fig. 3. Indeed, both model are dominated by the
Euler-Eytelwein’s formula. However, the input and the output of each model are
different. In the case of the endless-pulley of Eq. (1), two cable tensions are the
input and the generated torque of the endless-pulley is the output. On the other
hand, in the case of the endless-winch, bias tension tb by the tensioner and the
motor torque τw are the input and the cable tension tw to the endless-pulley is
the output which is given by
τw
tw = + tb (10)
rw
In general, motor can generate both positive (CCW) and negative (CW) torque.
The range of the motor torque τw of the winch is defined as
−τw0 ≤ τw ≤ τw0 (11)
Negative torque of the motor generates negative (pushing) cable tension to the
endless-pulley, which cannot be supported by the flexible body of the cable. It
means that the negative range in Eq. (11) is not valid for the CDPR. In order
to effectively use of the motor torque for CDPRs, bias tensions for the winch
are commonly used in the previous researches [10]. The tensioner of our research
acts not only as the bias force for the effective use of the motor but also as an
additional force for the non-slipping between the endless-cable and the endless-
winch. The cable slipping phenomena of the endless-winch can be expressed by
Fig. 6. Statics of the endless-winch on the new interpretation of the E-E’s formula
the new interpretation of the Euler-Eytelwein’s formula as shown in Fig. 6. In

the figure, the horizontal axis and the vertical axis represent bias tension tb and
tension tw of the cable exported from the endless-winch, respectively. Hereafter,
tw is called as winch tension. The non-slipping area of the cable tensions is
indicated by light gray same as Fig. 4. When the bias tension tb = tb0 (= 0),
range of the winch tension is expressed by the hollow arrow on the line of tb = tb0
in the figure. Not only the winch tension below zero is not only valid but also
the hollow arrow exists out of the non-slipping area. When the bias tension tb is
increased, the hollow arrow which expresses the range of the winch tension moves
to the right upper side as shown in the figure. The upper and the bottom edge
of the hollow allow lying on the line l and l , respectively. These lines correspond
to the constant torque lines in Fig. 4. When the bias tension tb = tb1 (= τw0 /rw ),
the minimum and maximum value of the winch tensions become tw1n = 0 and
tw1x = 2τw0 /rw , respectively. In this case, the range of the winch motor seems to
be effectively used for the winch tension. However, as shown in Fig. 6, the lower
part of the hollow arrow is out of the non-slipping area. Under this condition,
1 τw0
if the winch tension below tw1b = eμθ rw is applied, the endless-cable slips on
the endless-winch. The part of the hollow arrow in the non-slipping area is filled
with dark gray as shown in the figure. When the bias tension is given as tb = tb2 ,
the range of the winch tension becomes inside the non-slipping area as shown in
the figure. At that time, the minimum and maximum value of the winch tensions
become tw2n and tw2x , respectively. tb2 , tw2n and tw2x are given as,

τw0 eμθ
tb2 = (12)
rw eμθ − 1
⎧
⎪ τw0 1
⎨ tw2n =
⎪
⎪
rw eμθ − 1
(13)
⎪
⎪ τw0
⎩ tw2x = tw2n + 2
⎪
rw
Note here that in the case of the non-slipping condition of the endless-winch,
minimum winch tension greater than zero. As shown in Fig. 6, the bias tension
tb is set as greater than tb2 , the range of the winch tension keeps inside the non-
slipping area, but the minimum and the maximum value of the winch tension
are got larger. At that time, the non-slipping condition of the bias tension tb of
the endless-winch and range of the winch tension tw are given as

τw0 eμθ
tb ≥ (14)
rw eμθ − 1
τw0 τw0
tb − ≤ tw ≤ tb + (15)
rw rw
5 Statics of general E-CDPR
In this section, the non-slipping condition of the endless-pulley is extended for

general E-CDPRs such as OCTAVE in Fig. 1, the endless-pulley is driven by more
than two endless-cables as shown in Fig. 2 (c). The inverse statics equation of
the E-CDPR from the generalized force f of m dof mechanism to the generalized
cable tension t of (m + 1) cables comprised of n = (m + 1)/2 endless-cables is
given as
t = JT + f + hσ (16)
Where, JT + , h and σ represent pseudo-inverse of the transposed Jacobian ma-

trix, (m + 1) × 1 vector which spans null space of JT and an an arbitrary scholar,
respectively. Equation (16) is divided into each endless-cable as
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ta1 ha1
⎢ tb1 ⎥ ⎢ A1 ⎥ ⎢ hb1 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ .. ⎥ ⎢ .. ⎥ ⎢ . ⎥
⎢ . ⎥ = ⎢ . ⎥ f + ⎢ .. ⎥ σ (17)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ tan ⎦ ⎣ ⎦ ⎣ han ⎦
An
tbn hbn
Statics of the ith (i = 1, . . . , n) endless-cable is given as
ti = Ai f + hi σ
tai hai (18)
ti ≡ , hi ≡
tbi hbi
Fig. 7. Non-slipping conditions of endless-pulley of the E-CDPR
The non-slipping condition of each endless-cable of the E-CDPR can be ex-

plained by the new interpretation of the Euler-Eytelwein’s formula as shown in
Fig. 7. The horizontal and the vertical axis of the figure represent each ten-
sion of ith endless-cable of tai and tbi , respectively. The minimum-norm solu-
tion t = JT + f of Eq. (16) is projected to each endless-cable tension’s space as
ti = Ai f of Eq. (18), which corresponds to the tensions on the point Ai in Fig.
7. In the simple case of a single endless-pulley with a single endless-cable of Eq.
(7), angle ϕhi of the unit vector h with respect to the horizontal axis equals π/4.
In general, μθ in the Euler-Eytelwein’s formula greater than zero, then eμθ > 0,
thus one obtaines
1
tan−1 μθ < ϕhi < tan−1 eμθ (19)
e
Under the condition of Eq. (19), it is obvious from Fig. 4 that the cable tensions
Fig. 8. Slipping conditions of endless-pulley of the E-CDPR
can be shifted on the constant torque line then into the non-slipping area with
appropriate bias tensions. After the tensions in the non-slipping area, they keep
inside the area when the bias tensions are set larger value. Same situations may
occur in the case of the general E-CDPR, but there exist some differences. Here,
the E-CDPR satisfies the wrench-closure condition [11][12], all elements of the
vector h in of Eq. (16) share the same signum. In convenient, the signum of each
element of h is set as positive. At that time, the bias vector hi σ in Eq.(18) can
be set as its direction be going to right upper side. However, vector hi is not an
unit vector nor its direction angle ϕhi = tan−1 (hai /hbi ) is not equals to π/4 as
illustrates in Fig. 7. If, ϕhi < tan−1 (1/eμθi ) or ϕhi > tan−1 eμθi , then the cable
tensions cannot be shifted in the non-slipping area as shown (a) and (b) in Fig.
8, or cable tenshions is shifted outside of the non-slipping area when the bias
tenshion hi σ is set as large value as shown (c) in Fig. 8. θi is the total contact
angle swept by all turns of the ith endless-cable.
From the above discussions, the sufficient condition of the endless-cables’
non-slipping on the endless-pulley is given as
1
tan−1 < ϕhi < tan−1 eμθi , (i = 1, · · · , n) (20)
eμθi
At the end of this section, σ in Eq. (17) for the non-slipping condition is
derived. In the case of the E-CDPR, σ is a common value for all endless-cables.
Firstly, as shown in Fig. 7, each σic at the point Ci that the cable tensions come
into the non-slipping area is derived. If the cable tensions at the point Ai is in
the non-slipping area, σic is set as zero. Then, σx , the minimum value of σ that
satisfies the non-slipping condition for all endless-cables is given as
σx = max{σci , i = 1, · · · , n} (21)
6 Conclusions
Non-slipping conditions of E-CDPR which enables unlimited rotation of the hand
were discussed in this paper. A new interpretation of the Euler-Eytelwein’s for-
mula was proposed by using a graph that the non-slipping condition is expressed
as an area in the cable tensions. Equations of the statics of the endless-pulley
and the endless-winch are superimposed on the graph, then, the non-slipping
conditions of general E-CDPR were derived.
Acknowledgement
This work was supported by JSPS KAKENHI Grant Number 18K04068.
References
1. Pott, A.: Cable-Driven Parallel Robots, Theory and Application. Springer, (2018).
doi:10.1007/978-3-319-76138-1
2. Cable-Driven Parallel Robots, Mechanisms and Machine Science 13, Editors,
Bruckmann, T., Pott, A., Springer (2013). doi:10.1007/978-3-642-31988-4
3. Cable-Driven Parallel Robots, Mechanisms and Machine Science 53, Edi-
tors, Goselin, C., Cardou, P., Bruckmann, T, Springer (2017). doi:10.1007/
978-3-319-61431-1
4. Fortin-Côté, A., Faure, C., Bouyer, L., McFadyen, B.J., Mercier, C., Bonenfant,
M., Laurendeau, D., Cardou, P., Gosselin, C.: On the Design of a Novel Cable-
Driven Parallel Robot Capable of Large Rotation About One Axis. in Cable-Driven
Parallel Robots, Mechanisms and Machine Science 53, pp. 390-401, Springer (2017).
doi:10.1007/978-3-319-61431-133
5. Miermeister, P., Pott, A.: Design of Cable-Driven Parallel Robots with Multi-
ple Platforms and Endless Rotating Axes. Interdisciplinary Applications of Kine-
matics, Mechanisms and Machine Science 26, pp. 21-29 Springer, (2015). doi:
10.1007/978-3-319-10723-3 3
6. Pott, A., Miermeister, P: Workspace and Interference Analysis of Cable-Driven
Parallel Robots with an Unlimited Rotation Axis. in Advances in Robot Kinemat-
ics, pp. 341-350 Springer, (2018). doi:10.1007/978-3-319-56802-7 36
7. Makino, T., Harada, T.: Cable Collision Avoidance of a Pulley Embedded Cable-
Driven Parallel Robot by Kinematic Redundancy. 4th International Conference
on Control, Mechatronics and Automation, pp. 117-120, Barcerona (2016). doi:
10.1145/3029610.3029620
8. Meriam, J.J, Kraige, L.G.: Engineering Mechanics, Statics (Seventh Edition). pp.
377-378, John Wiley & Sons (2006). ISBN:978-0-470-61473-0
9. Popov, V.L.: Contact Mechanics and Friction, Physical Principles and Applica-
tions. pp. 148-149, Springer (2010). doi:10.1007/978-3-642-10803-7
10. Zitzewitz, J.V., Fehlberg, L., Bruckmann, T., Vallery, H.: Use of Passively Guided
Deflection Units and Energy-Storing Elements to Increase the Application Range
of Wire Robots. Cable-Driven Parallel Robots, Mechanisms and Machine Science
12, pp. 167-184, Springer (2013). doi:10.1007/978-3-642-31988-4 11
11. Gouttefarde, M., Gosselin, C.: L., Analysis of the Wrench-Closure Workspace of
Planar Parallel Cable-Driven Mechanisms. IEEE Transactions on Robotics, VOL.
22, No. 3, pp. 434-445 (2006). doi:10.1109/TRO.2006.870638
12. Loloei, A.Z., Taghirad, H.D.: Controllable Workspace of Cable-Driven Redundant
Parallel Manipulators by Fundamental Wrench Analysis. Transactions of the Cana-
dian Society for Mechanical Engineering, Vol. 36, No. 3, pp. 297-314 (2012)
Analysis of Cable-Configurations of Kinematic
Redundant Planar Cable-Driven Parallel Robot
Koki Hirosato1 and Takashi Harada2

1
Kindai University, Graduate School of Science and Engineering Research
3-4-1, Kowakae Higashiosaka Osaka 577-8502, Japan,
[email protected]
2
Kindai University
Abstract. A redundant planar three-dof cable-driven parallel robot is

proposed and analysis of its cable-configuration is discussed in this paper.
Loop cables and constant force springs unlimitedly rotate the endless-
pulley which is embedded inside the moving part. The angle of the
hand is redundantly given by the sum of the angles of the moving part
frame and the endless-pulley. Three-dof hand is controlled by four-dof
mechanism using five-cables. This means that the proposed CDPR is a
novel cable-driven parallel robot which simultaneously has the kinematic
and actuation redundancies. Tactical design and control for the cable-
configurations are proposed as the robot satisfies the wrench-closure con-
dition.
Keywords: Cable-driven, Redundantly Acutuated, Cable configurations,

Wrench-closure condition, Kinematic redundancy
1 Introduction
Cable-Driven parallel robot (CDPR) has a large translational workspace when
using long cables. However, collisions between cables and singular configurations
restrict its rotational workspace [7]. In order to enlarge the rotational workspace
of CDPRs, additional rotational mechanism such as crank [6] or pulley [1] [3] is
embedded inside the moving part.
In this paper, a novel planar CDPR with a pulley is embeded inside the
moving part as an additional rotational degrees of freedom (dof) is proposed.
The pulley embedded mechanism of the proposed CDPR is similar to [1], but be
different in that;
– The pulley embedded in the moving part is driven by loop-cables which are
tensioned by constant springs. This mechanism enables unlimited rotation
of the hand which is attached at the end of the pulley. Hereinafter, this
mechanism is called endless-pulley.
– Not only the angle of the endless-pulley, but also the angle of the moving
part frame are actively controlled by the cables. Angle of the hand is given
by the sum of the angle of the moving part frame and the endless-pulley,
thus, the proposed CDPR has kinematic redundancy.

36 K. Hirosato and T. Harada
Three-dof hand is controlled by four-dof mechanism (translations along x

and y direction, rotation of the moving part φm and additional rotation of the
endless-pulley φp ) which is driven by five-cables. The five-cables are spanned as
a star-hexagon which looks like the family crest of the Japanese diviner Abe no
Seimei. So the proposed CDPR is dubbed SEIMEI. The kinematic redundancy
is applied not only for enlarging the rotational range of the hand, but also for
a tactical design of the cable configurations and control for the wrench-closure
condition of SEIMEI. For the n-dof CDPR driven by n + 1 cables, if and only
if the transposed Jacobian matrix (TJM) has full rank and all elements of the
null space vector of the TJM share the same signum, then the CDPR satisfy the
wrench-closure condition [2] [4] [10]. At that time, tensions of all cables can be
positive by adding positive bias tensions, which are projected to the null space
of the TJM.
The remainder of the paper proceeds as follows. In Sect. 2, design and mod-
eling of SEIMEI are introduced. In Sect. 3, equations of kinetostatics of SEIMEI
are derived. In Sect. 4, tactical designs of cable-configurations and control for
the wrench-closure condition of SEIMEI is discussed. In Sect. 5, derived cable-
configurations in Sect. 4 were confirmed the wrench-closure condition by numer-
ical tests.
2 Mechanical design of unlimited rotatable CDPR

2.1 Planar three-dof by five cables CDPR “SEIMEI”
Schematic image of SEIMEI is shown in Fig. 1. In convenient, base frame and
the moving part (MP) frame are formed of regular pentagon. Pbi and Pmi at
each vertex of these pentagons represent points of ith cable port on the base
frame and the MP frame, respectively.
An endless-pulley is embedded into the MP frame as its rotational axis be
fixed to the MP frame via a rotational pair. Five cables control position x, y,
angle φm of the MP frame and angle φp of the endless-pulley. Σb represents the
coordinate frame fixed on the base, Σm and Σp are those on the MP frame, and
on the endless-pulley respectively. The length of the ith cable is defined as li ,
which corresponds to the distance between Pbi and Pmi . As shown Fig.1(b), a
hand is attached to the endless-pulley. Angle θ of the hand is redundantly given
by sum of the angle φm of the MP frame and the angle φp of the endless-pulley
as
θ = φ m + φp (1)
Thus, SEIMEI is categorized into the kinematic redundant parallel robot. Even
if collisions between cables and singular configuration occur in a specific config-
uration, the kinematic redundancy contributes to avoid them by changing angle
φm of the MP frame, then changing the direction of the cables without changing
position x, y and angle θ of the hand. Instead of fixing each end of the cable to the
winch and the endless-pulley, cable-loop mechanism is applied to SEIMEI, which
Analysis of Cable-Configurations of Kinematic Redundant Planar… 37
Fig. 1. Planar three-dof CDPR “SEIMEI”

enables unlimited rotation of the hand. The cable-loop mechanism is comprised
of loop cable, winches, constant force spring and slider. Figure 2 illustrates 3D
sketch of the cable-loop mechanism. The loop cable from the winch j via the
port Pbi is imported inside the MP via the port Pmj , next the cable is wound
several turns around the endless-pulley, then the cable is exported from the MP
via the port Pmi , then imported in the frame via the port Pbi , wound several
turns around the other winch i. After that the loop cable is guided by the two
fixed pulleys and tensioned by constant force spring via a moving pulley on the
slider. Finally, the loop cable is wound several turnes around the winch j. While
fixing the position and angle of the MP frame, the cable can run then turn the
endless-pulley. When the MP moves as shown by dashed line in Fig.2, exces-
sive length of the loop cable is pulling by the constant force spring. Note here
that the cable is driven by frictions between the cable and the winch or the
endless-pulley, that may occurs slip when the tension of the cable is getting to
be small. Bias tension for the cable by the constant force spring contributes the
cable to non-slipping. The neighboring two cables C1-C2 and C3-C4 are set as
loop-cables as shown in Fig.1 (a). The rest cable C5 in Fig.1 (a) is jointed to
center of the MP frame by using a rotational pair. This makes specific elements
of the TJM equal 0, thus the analysis of the cable configurations can be simple.
Fig. 2. Cable-loop mechanism

2.2 Modeling
Fig. 3. Kinematic model of SEIMEI Fig. 4. Relation of cable length
Kinematic model of SEIMEI around the MP is schematically shown in Fig.

3. p is position vector of the MP in the base coordinate frame Σb . bi and vi
represent the position vector of ith cable port of the frame and unit direction
vector of the ith cable in Σb , respectively. si is the position vector of the cable
port Pmi in Σb . li represents the length of the ith cable from the port Pbi to the
port Pmi . The vector loop equation of the ith cable is given as
p + s i = bi + li v i (2)
Differentiate both sides of Eq.(2) with respect to time, one obtains
l˙i = (vi × si )φ̇m + vi T ṗ (3)
As mentioned in Sect.2.1, while the position and angle of the MP are fixed, the
loop cable can be run then turns endless-pulley as shown in Fig.4. di represents
the running length of the loop cable on the ith winch.
Relations between the cable length li , running length di and angle of the
endless-pulley φp , and their differential with respect to time are given as Eq.(4)
and Eq.(5), respectively.
d i = l i + ri φ p (4)
d˙i = l˙i + ri φ̇p (5)
where ri = ±rp (rp : radius of the endless-pulley), manner of the signum is given
as Fig.5. The signum is determined by the winding direction of the cable to the
pulley, which is named the pulley winding direction in this paper.
Fig. 5. Signum of ri
By substituting Eq.(3) into Eq.(5), one obtains
d˙i = ri φ̇p + (vi × si )φ̇m + vi T ṗ (6)
Kinematics and statics of SEIMEI are given as
d˙1
⎡ ⎤ ⎡ ⎤
φ̇p
⎢ .. ⎥
⎣ . ⎦ = Jdx ⎣ φ̇m ⎦ (7)
d˙5 ṗ
f x = Jdx T f d (8)
f x and f d represent the wrench of the hand and cable tensions, respectively. Jdx T
represents the transposed Jacobian matrix (TJM) of SEIMEI which is given as
⎡ ⎤
r1 . . . r 5
T
Jdx = ⎣ a1 . . . a5 ⎦ (9)
v1 . . . v5
where ai = |vi ||si |sinθi , θi is the angle from vi to si .

3 Expand the null space vector of the transposed

Jacobian matrix
3.1 wrench-closure condition
Inverse statics from the wrench f x to the cable tensions f d is given by inverting
Eq.(8) as
+
f d = Jdx T f x + hdx σdx (10)
T+
Jdx , hdx and σdx represent the pseudo-inverse of the TJM, 5 × 1 vector spans
the null space of the TJM and an arbitrary scholar, respectively. hdx σdx cor-
responds to the bias tension on the cable. Because Jdx T (hdx σdx ) = 0, the bias
tension doesn’t influence to the wrench f x of the hand. In order to the following
analysis, rewriting the part of each column of the TJM as
⎡ ⎤
r1 . . . r 5
r . . . r5
Jdx T = ⎣ a1 . . . a5 ⎦ = 1 = w1 . . . w5 (11)
t1 . . . t 5
v1 . . . v5
The null space vector hdx of the TJM is derived by the Cramer’s rule [5] as
⎡ ⎤ ⎡ ⎤
det([ w5 w2 w3 w4 ]) det([B1 ])
⎢ det([ w1 w5 w3 w4 ]) ⎥ ⎢ det([B2 ]) ⎥
⎢ ⎥ ⎢ ⎥
hdx = ⎢⎢ det([ w1 w2 w5 w4 ]) ⎥ = ⎢ det([B3 ]) ⎥ (12)
⎥ ⎢ ⎥
⎣ det([ w1 w2 w3 w5 ]) ⎦ ⎣ det([B4 ]) ⎦
−det([ w1 w2 w3 w4 ]) −det([C])
The wrench-closure condition for the n-dof CDPR driven by n + 1 cables is
given as follows;
i) rank (Jdx T ) = n(full-rank)
ii) all elements of hdx share the same signum
When a CDPR satisfies the wrench-closure condition, all cable tensions are con-
trolled to be positive value by applying an appropriate bias tension hdx σdx .
3.2 Cable configurations

A lot of researchers can be found related to the wrench-closure condition of
CDPRs, e.g., wrench-closure configuration, wrench-closure workspace [8] [10],
and cable-configurations for satisfying the wrench-closure condition [2] [9]. It
is difficult to find these conditions analytically, so one-by-one numerical search
used to be applied in the previous researcher [11].
In this paper, an analytical approach for cable-configurations that satisfy the
wrench-closure condition are proposed. Each element of the null space vector of
the TJM is given by corresponding minor of the TJM. Each minor is converted
to polynomial by the cofactor expansion, then be related to three kinematic
parameters about cable-configurations of the robot. Strategies of design and
control for the three kinematic parameters of SEIMEI are proposed so as all
polynomials, i.e., all elements of hdx , share the same signum.
3.3 Cofactor expansion of the null space vector of the transposed

Jacobian matrix
Each elements of the null space vector hdx of Eq.(12) is given by the correspond-
ing minor of the TJM of Eq.(11). As an example, row 5 of hdx is expanded about
ri as

r1 r2 r3 r4
det([ w1 w2 w3 w4 ]) = det
t1 t2 t3 t 4
= r1 det([ t2 t3 t4 ]) − r2 det([ t1 t3 t4 ])
+ r3 det([ t1 t2 t4 ]) − r4 det([ t1 t3 t3 ]) (13)
Furthermore, det([ t2 t3 t4 ]) of the first term is expanded as

a2 a3 a 4
det([ t2 t3 t4 ]) = det
v2 v3 v4
= a2 det([ v3 v4 ]) − a3 det([ v2 v4 ]) + a4 det([ v2 v3 ]) (14)
Each element of hdx is converted to polynomial of twelve terms which are com-
posed by ri , ai , det([ vi vj ]). These valuables are related to the kinematic pa-
rameters about the cable-configurations of SEIMEI.
4 Determine cable configurations

4.1 Three kinematic parameters
In this section, we evaluate strategies of design and control for the three kine-
matic parameters so as SEIMEI satisfy the wrench-closure condition. Basic idea
of the strategy is to find kinematic conditions that the signums of specific com-
binations of the twelve terms keep their signums during SEIMEI moves in the
specific area. As mentioned in Sect.2.1, four of five cables form a pair, not formed
the cable is jointed to center of the MP. This makes specific elements r5 and a5
of the TJM equal 0, that enables the analysis to be simple.
Cross product by directions vectors of two cables ; det([vi vj ])

For the simplicity, size of the MP is assumed to be zero in this subsection.
Det([ vi vj ]) denotes cross product of unit direction vectors of cables, vi and
vj . As shown in Fig.6, this corresponds to sinθij (θij is the angle from vi to vj )
because vi and vj are unit direction vectors. Signum of det([ vi vj ]) changes
when the MP across the line which connect cable port Pbi and Pbj of the base
frame. The following strategy a) is applied.
a) A specific limited workspace is determined so as the cross products of two
cable’s direction vectors are not changed when the moving part is in this
area. By doing this, signums of the valuables are fixed.
This area is surrounded by diagonal lines of the base frame, and it is in the shape
of the inverse pentagon as shown in Fig.7. Signums of all det([ vi vj ]) do not
change in the limited workspace.
Fig. 6. The angle from vi to vj Fig. 7. Limited workspace

The moving part pulling direction; ai
ai = |vi ||si |sinθi is given by the cross product of the cable unit direction vector
vi and position vector si of the cable ports Pmi in the Σb of the base coordinate
frame. θi corresponds to the angle from vi to si . Signum of ai are determined by
value of angle θi as shown in Fig.8. Possible initial configurations of all cables

Fig. 8. Signum of ai
of SEIMEI about the MP pulling direction are represented in Fig.9. Each cable
pair C1-C2 and C3-C4 is consisted by a loop cable. Remained cable C5 is jointed
to the center of the MP via a rotational pair. At that time, s5 equals zero vector,
thus a5 equals zero, then the analysis becomes simpler. The signum is determined
by the pulling direction of the cable to the MP, which is named the MP pulling
direction in this paper. The MP pulling direction changes when the MP moves as
shown in Fig.10. Note here that SEIMEI has kinematic redundancy as Eq.(1).
The hand of SEIMEI can be rotated by the endless-pulley while keeping the
MP pulling direction. Namely, the following strategy is applied for control of

SEIMEI.
b) During the motions of SEIMEI, the MP pulling direction of each cable to

the moving part frame is actively controlled not to be changed by using the
kinematic redundancy. By doing this, signums of ai is fixed.
Fig. 10. Changes the MP pulling direction
Fig. 9. Cable configurations about ai
The pulley winding direction; ri
As mentioned in Fig.5, the signum of ri is determined by the pulley winding

direction. Remained cable C5 is jointed to the center of the MP and not to wind
to the endless-pulley, thus ri equals zero, then the analysis becomes simpler.
Possible configurations about the pulley winding direction are showed in Fig.11.
Signum of ri is only affected by winding direction, therefore it is fixed even if the
MP moves anywhere in the limited workspace and the MP frame and the endless-
pulley has angles. Cable configurations which make all terms of polynomial which
are composed by kinematic parameters share same signum are derived. This is
the strategy c).
c) Under the conditions applied strategies a) and b), the pulley winding direc-
tion is determined so as to satisfy the wrench-closure condition.
4.2 Verification of cable configurations of SEIMEI
In this subsection, cable configurations which all elements of the null space vector
of TJM share the same signum are derived. For example, the cable configuration
whose signums of kinematic parameters is based on Table 1 is verified.
Fig. 11. Cable configurations about ri

Table 1. Combination of signum
r1 r2 r3 r4 r5 a1 a2 a3 a4 a5
+ − − + 0 − + − + 0
One obtains cofactor expand of det([B1 ]) of hdx as
det ([B1 ]) = det([ w5 w2 w3 w4 ])

= r5 det([ t2 t3 t4 ]) − r2 det([ t5 t3 t4 ]) + r3 det([ t5 t2 t4 ]) − r4 det([ t5 t2 t3 ])
= r5 (a2 sinθ34 − a3 sinθ24 + a4 sinθ23 ) − r2 (a5 sinθ34 − a3 sinθ54 + a4 sinθ53 )
+ r3 (a5 sinθ24 − a2 sinθ54 + a4 sinθ52 ) − r4 (a5 sinθ23 − a2 sinθ53 + a3 sinθ52 ) (15)
Taking r5 = 0 and a5 = 0 into account, Equation (15) is simplified as

(−)
det([ w5 w2 w3 w4 ]) = rp a2 (+) (sinθ53 + sinθ54 (−) )
+ rp a4 (+) (sinθ53 (−) − sinθ52 (+) )
− rp a3 (−) (sinθ54 (−) + sinθ52 (+) ) (16)
It is evident from substituting signums of Table.1 for ri and ai that 1st and 2nd
terms, rp a2 (sinθ53 + sinθ54 ) and rp a4 (sinθ53 − sinθ52 ) are negative regardless of
the value of elements. The formula in the parenthesis of 3rd term is rearranged
as Eq.(17).
sinθ54 + sinθ52 = sinθ54 + sinθ(54−24)

= sinθ54 + sinθ54 cosθ24 − cosθ54 sinθ24
= sinθ54 (−) (1 + cosθ24 ) − cosθ54 sinθ24 (+) (17)
Signum of cosθij is changed when the MP across the circle line whose diameter
is given by the line from Pbi to Pbj . Thus additionally limited workspace is
defined. This area is showed the hatched part in Fig.12, and be similar to the
star-shaped which is given by trimming the circle off the limited workspace in

Fig.7. Signums of all cosθij is fixed when the MP is in this area. Thus signums
of cosθ24 and cosθ54 are positive and 3rd term has negative value. Value of three
terms are negative, therefore signum of row 1 of hdx is negative. Signum of
det([Bi ]) (i = 2 . . . 4) and −det([C]) are regarded as negative in the same way.
All elements of hdx have negative value, cable configuration showed in Table
1 satisfies the wrench-closure condition.
Fig. 12. Additionally limited workspace

Fig. 13. Wrench-closure workspace
5 Simulation
This section investigates whether derived cable configuration in Sect.4 satisfies
the wrench-closure condition. Various parameters are given later, radius of the
pulley rp equals 5mm, distance between center of the MP and cable port Pmi
equals 10mm, distance between center of the base frame and cable port Pbi
equals 100mm. Figure 13 shows the wrench-closure workspace. The black dot
line represents the boundary of the wrench-closure workspace when the angle φm
of the MP frame equals 0◦ , the hatched part represents the limited workspace.
Proposed additionally limited workspace is conservative, and actual workspace of
SEIMEI is more larger in that additionally limited workspace is in the boundary
of the wrench closure workspace. The cable-configuration derived in Sect.4.2
satisfies the wrench-closure condition, and proposed tactical design and control
for cable configuration are valid.
6 Conclusion
A redundant planar three-dof cable driven parallel robot “SEIMEI” was pro-
posed. SEIMEI is different from conventional CDPR in that, having cable loop
mechanism and kinematic redundancy. Cable-loop mechanism rotates the endless-
pulley embedded in the MP unlimitedly. SEIMEI can control the angle of the
endless-pulley and the MP frame actively by having kinematic redundancy. Tac-

tical design of cable-configurations and control for the wrench-closure condition
were proposed. The cable-configuration which satisfies the wrench-closure condi-
tion is derived by using the “strategy” and kinematic redundancy. Derived cable
configuration were confirmed the validity by numerical tests.
Acknowledgement
This research was supported by JSPS KAKENHI Grant Number 18K04068.
References
1. Fortin-Côté, A., Céline, F., Bouyer, L., J.McFadyen, B., Mercier, C., Bonenfant,
M., Laurendeau, D., Cardou, P., Gosselin, C.: On the Design of a Novel Cable-
Driven Parallel Robot Capable of Large Rotation About One Axis. Cable-Driven
Parallel Robots, pp.390-401. Springer (2017)
2. Gouttefarde, M., Gosselin, C.: Analysis of the Wrench-Closure Workspace of Pla-
nar Parallel Cable-Driven Mechanisms. IEEE TRANSACTIONS ON ROBOTICS,
vol.22, pp. 434-445 (2006)
3. Makino, T., Harada, T.: Cable collison avoidance of a pulley embedded cable-driven
parallel robot by kinematic redundancy. 4th International Conference on Control,
Nechatronics and Automation, pp. 117-120, Barcelona (2016)
4. Marc Arsenault.: Optimization of the prestress stable wrench closure workspace of
planar parallel three-degree-of-freedom cable-driven mechanisms with four cables.
2010 IEEE International Conference on Robotics and Automation, pp. 1182-1187,
Alaska (2010)
5. McColl, D., Notash, L.:WORKSPACE ENVELOPE FORMULATION OF PLA-
NAR WIRE-ACTUATED PARALLEL MANIPULATORS. Transactions of the
Canadian Society for Mechanical Engineering, vol.33, No.4, pp. 547-560 (2009)
6. Miermeister, P., Pott, A.: Design of Cable-Driven Parallel Robots with Multiple
Platforms and Endless Rotating Axes. Interdisciolinary Applications of Kinemat-
ics, pp. 21-29. Springer (2014)
7. Pott, A.: Cable-Driven Parallel Robots. Springer (2017).
8. Pott, A., Kraus, W.: Determination of the Wrench-Closure Translational
Workspace in Closed-Form for Cable-Driven Parallel Robots. IEEE International
Conference on Robotics and Automation, pp. 882-887, Stockholm (2016)
9. Seon, J.A., Park, S., Ko, S.Y., Park, J.O.: Cable Configuration Analysis to Increase
the Rotational Range of Suspended 6-DOF Cable Driven Parallel Robots. 2016
16th International Conference on Control, Automation and Systems, pp. 1047-
1052, Korea (2016)
10. Sheng, Z., Park, J.P., Stegall, P., Arrawal, S.K.: ANALYTIC DETERMINATION
OF WRENCH CLOSURE WORKSPACE OF SPATIAL CABLE DRIVEN PAR-
ALLEL MECHANISMS. ASME 2015 International Design Engineering Technical
Conference & Computers and Information in Engineering Conference, DETC2015-
47976, Boston (2015)
11. Tadokoro, S., Nishioka, S., Kimura, T., Hattori, M., Takamori, T., Maeda, K.: On
fundamental Design of Cable Configurations of Cable-Driven Parallel Manipulators
with Redundancy. Transactions of the Japan Society of Mechanical Engineers,
series C, vol.66, No.647, pp.2247-2254 (2000)
Improving cable length measurements for large CDPR
using the Vernier principle
Jean-Pierre Merlet
J-P. Merlet, HEPHAISTOS project, Université Côte d’Azur, Inria, France

[email protected]
Abstract. Cable lengths is an important input for determining the state of a

CDPR. Using drum with helical guide may be appropriate for small or medium-
sized CDPR but are problematic for large one. Another issue is the initialization
of the cable length as measurements based on drum rotation are incremental.
We propose to address automatic initialization and improvement of cable length
measurements by using regularly spaced color marks on the cable combined with
color sensors in the mast of the CDPR. We show that this disposition allows one
to automate the initialization issue and then how it allows to get regularly accurate
estimation of the cable length by using the Vernier principle.
Keywords: cable length, accuracy, initialization, calibration
1 Introduction
Measuring the cable lengths of CDPR with rotary winch is usually done by measuring
the rotation of the drum with an encoder. Coiling the cable on the drum is usually done
in two manners: the drum may have a spiral guide and a guiding mechanism moves
synchronously the cable in front of the free part of the spiral or the cable is just coiled
on the fly on the drum. An alternate to using rotary actuators is the use of linear actuator
with a pulley mechanism that amplifies the stroke of the actuator [Merlet(2008)] but we
will not consider this case in this paper. The spiral-guided mechanism is the one used
in many cases for small to medium sized CDPR such as IPANEMA [Pott et al(2012)]
or COGIRO [Gouttefarde et al(2012)]. It leads to a one-to-one and fixed relationship
between the cable length and the drum rotation and if elasticity can be neglected it
provides an accurate measurement of the cable length but it has drawbacks:
– the drum may accept only a single layer. This limits the available total cable length
as increasing the drum radius will both increases the length measurement inaccu-
racy due to error in the drum rotation measurement and the necessary motor torque
for a given cable tension. Hence such a mechanism may not be appropriate for very
large CDPR that involves coiling several dizains or hundreds meters of cable
– the friction of the cable on the spiral guide increases the cable wear and cable
elasticity is not taken into account
– according to the cable tension the cable may jump to another part of the guide

48 J.-P. Merlet
Another issue for such a winch system is that the drum encoder usually provides only
a relative measurement. When starting the CDPR it is therefore necessary either to cal-
ibrate the platform’s pose [Alexandre(2012)], [Alexandre et al(2013)], [Chen(2004)],
[Baczynski(2010)], [Miermeister and Pott(2012)] or to measure the initial length of the
cable, both tasks being tedious. To the best of the author knowledge an automatic deter-
mination of the initial cable lengths has not yet been presented. We propose a method
that allows both to obtain the initial cable lengths and provides regularly information
on the cable length.
2 Approach
Our method is directly inspired by a method we have implemented successfully on our

MARIONET-ASSIST CDPR [Merlet(2010)]. This CDPR uses synthetic cables and we
have glued several aluminium foils on the cables at known distance from the platform
cable attachment point B. At the winch level the cable goes through two electrically
isolated Delrin guides that have been covered with aluminium foils, the mid-point be-
tween the guides being the output point A of the winch. Each time a cable foil goes
through the guides an electric contact is established providing a boolean information
for the control computer. Using this event a semi-automated procedure may be used for
initialization (the robot stops when detecting a foil and the operator input manually the
corresponding cable length) and, in operation, for updating the current estimation of the
cable length. In this paper we propose to improve this method by using colored marks
on the cable and several color sensors in the mast that constitute the support structure
of the CDPR (figure 1) for benefiting from a Vernier scale. Color sensor consists in leds
that provides a constant illumination and receptors that are sensitive to a particular color
(figure 1). Such sensors are inexpensive and our tests have shown that they can reliably
detect at least the three RGB colors. They can easily be integrated in the support mast
in small non transparent boxes with a circular opening for the cable (figure 1), the color
sensor being protected from external illumination.
support
color sensor
Fig. 1. A color sensor and the principle of color marks on the cable and color sensor in the mast
Improving cable length measurements for large CDPR… 49
3 The initialization problem: a first approach
A major problem for CDPR is to determine automatically what are the cable lengths
at the start of the operation. The basic idea of our approach for automatizing this ini-
tialization process is the detection by a color sensor of n successive color marks when
coiling (or uncoiling) the cable, while having disposed the colors marks on the cable
in such a way that there is only a single occurrence of any n successive colors on the
cable. For example if we use 3 possible colors R, G, B for the mark, then there will be,
for example, a single GBR sequence on the whole cable, the other sequences being any
triplet among the set (R, B, G), possibly with repetition (for example GGG or GGB).
Let us assume that a given color sequence (n1 , n2 , n3 ) has been detected and that dl is the
known distance between A and the color sensor. As a given color sequence (n1 , n2 , n3 )
is unique on the cable the detection of n3 gives us the corresponding mark number on
the cable and consequently the distance d between B and the mark. The cable length
ρ = ||AB|| is therefore obtained as ρ = d − dl . Hence there is a one-to-one relationship
between all possible sequences (n1 , n2 , n3 ) and ρ. Therefore the initialization method
consists simply in uncoiling the cable until we have at least n marks between a color
sensor and the B point and then coiling the cable until the color sensor has detected n
marks, such a process being easy to automatize. A faster initialization process will be
presented in section 5.3.
4 Number of marks
As will be presented in the next sections we also intend to use the marks to update
the current cable length. Intuitively it makes sense to have a large number of marks
on the cable for that purpose, but our initialization process imposes the constraint of
having a single occurrence of n successive marks, whatever is the color combination.
We will now investigate the influence of this constraint on the maximal number of
marks on a cable. Assume that we use marks with k different colors: our problem is
to determine the maximal number of marks m that we may have on the cable so that
all subsets of n successive marks are different in term of color value. In other words
we have to determine the largest color sequence that satisfy this constraint. This prob-
lem is well known in combinatorial theory and such a sequence is called a De Bruijn
sequence [Bruijn(1975),Bruijn(1946)]. For a cable with k possible mark values that
has all n combination of the mark values a single time in the sequence, the sequence
length (i.e. the number of marks) is kn for a cyclic sequence. So for n = k = 3 we have
a sequence of length 27. But as our sequence is not cyclic we may add 2 additional
marks (but not 3, which will be contradictory) so we may have up to 29 marks on the
cable. An example of such sequence (the colors are indicated by the number 1, 2, 3)
is: [12312112213222323313111333212]. Using this coding the detection of 3
successive colors by a color sensor allows one to determine the location of the last mark
on the cable and consequently the cable length.
For n = 4 the De Bruijn sequence has a length of 34 = 81. and we may add 3
additional marks while keeping the property so that we will end up with a total of 84
marks on the cable. For k = 4 colors and a sequence of n = 3 marks the De Bruijn
50 J.-P. Merlet
sequence has a length of 64 to which we may add 2 marks for a total of 66 marks while
for a sequence of n = 4 marks we will have a total of 259 marks on the cables. We have
now to examine how the marks may be used to determine the cable length.
5 Measuring cable lengths

The height of the tower will be denoted h (in the examples we will assume h = 15)
and we assume that there is a color sensor every dm meters starting from the base of
the tower so that we have kmax sensors in the tower with kmax = f loor(h/dm ), where
f loor is the largest integer lower than h/dm . The cable sensor will be numbered from
1 to kmax , the sensor kmax being the highest in the tower. We have kmax1 marks on the
cable that are assumed to be distributed regularly on the cable, the distance between
two successive marks being denoted by dc meters.The marks are numbered from 1 to
1 , mark k1 1
kmax max being the one the closest to B. The distance between mark kmax and
the attachment point B, called the dead length, will be denoted by b. When the cable
is completely uncoiled we assume that the first mark on the cable is located at the kmax
sensor. The total length Lt of the cable between the base of the tower and B is
1
Lt = kmax dm + (kmax − 1)dc + b (1)
If we assume that the k1 -th mark is located at the k-th color sensor, then the cable length
L between the color sensor and B is:
1
L = (kmax − k1 )dc + b (2)
and the length ρ of the cable between A, B is then obtained from L by subtracting the
distance h − kdm between the k-th sensor and the top of the tower:
ρ = (kmax
1
− k1 )dc + b − (h − kdm ) (3)
Consequently the lowest value ρmin for ρ is obtained for k1 = kmax

1 and k = 1 with the
value ρmin = b − h + dm
We will choose b = h−dm so that this minimum is 0 in order to have always positive
ρ when a mark lies in front of a color sensor. Consequently ρ is obtained as:
ρ = (kmax
1
− k1 )dc + (k − 1)dm (4)
By taking all possible values for k, k1 we get all ρ that will correspond to a mark
detection by a color sensor, that we will call a sensor event, and is defined by a triplet
(k1 , k, ρ) corresponding respectively to the mark number, to the sensor number and the
1 ] there will be
corresponding cable length.. As k, k1 lie respectively in [1, kmax ], [1, kmax
at most kmax kmax different ρ. This is an upper bound as the same ρ may possibly be
1
obtained for different pairs (k1 , k). For these values of ρ we will get a sensor event that
will allow us to determine the cable length between A, B.
Sorting by increasing value all possible values of ρ and taking the differences be-
tween successive values will provide the various cable length changes Δ ρ between two
sensor events. Note that Δ ρ cannot exceed dm which correspond to the case where a
given mark is seen successively by the same color sensor. Between two sensor events
the cable length will be interpolated from the measurement of the drum rotation with
some uncertainty because of modeling error on the coiling process. Clearly there is an
interest of having the largest possible number of significant sensor event (i.e. one that
provides an update on the cable length), a relatively flat distribution of the Δ ρ with
a low average value. Another interest of a flat distribution is that we may relate two
successive sensor events (which will provide a change Δ ρ in the cable length) to the
corresponding drum rotation Δ θ in order to obtain an estimate of the mean drum radius
that will be updated at each sensor event. This estimate will then be used to determine
the cable length between two sensor events.
5.1 Significant sensor events
Clearly we are interested in having the largest number of different ρ. Consequently

we should avoid having two sensor events for the same ρ. In other words we have to
determine if we may have Δ ρ equal to 0 being given dm , dc . Consider now 2 sensor
events defined by the triplets (k1 , k, ρ), (k1 , k , ρ ). Using equation (4) we may calculate
the Δ ρ = |ρ − ρ | as
Δ ρ = |(k1 − k1 )dc + (k − k )dm |
There is a symmetry in this relation as, being given the events (k1 , k), (k1 , k ) we will get
the same Δ ρ for the events (k1 − kmax
1 ,k
max − k ,), (kmax − k1 , kmax − k ). Let us assume
1
that the ratio dm /dc is a rational number p/q and consider the rational number
p1 k1 − k1
=
q1 (k − k )
which is such that |p1 | is the lowest possible value for |k1 − k1 | while |q1 | is the minimal
value of |k − k |. We have
−p1 p
Δ ρ = dc |(k − k )( + )| (5)
q1 q
As dc is a known positive constant the minimum of Δ ρ will be obtained when |(k −

k )( −p p
q1 + q )| is minimal. As |(k − k )| ≥ |q1 | the minimum of Δ ρ is obtained for the
1
rational p1 /q1 that is the closest to p/q.

Equation (5) is essential to assert the distribution of the Δ ρ. For example for kmax
1 =
29 (29 marks on the cable) and dm = 2.4 (which leads to kmax = 6) and dc = 2 we
get dm /dc = 1.2 = 6/5. If we set p1 = 6, q1 = 5 we get k1 − k1 = 6 leading to k1 =
k1 + 6 ∈ [7, 29] and k1 ∈ [1, 23]. We have also k − k = 5 and as k cannot exceed 6 we
get k = 1, k = 6. The ρ obtained for the pair (k1 ∈ [7, 29], 6) will be the same than for
the pair (k1 − 6, 1) so that we will get 23 identical ρ. As the maximum number of ρ is
29 × 6 = 174 there will be 174-23=151 different ρ available. .
However we are not only interested in discarding the sensor events that will give
the same ρ but also in the set of Δ ρ that are lower than dm . Hence we have to find the
positive rationals p1 /q1 with a denominator at most equal to kmax − 1 and lower or equal
1
to kmax that will lead to |Δ ρ| that are lower of equal to dm . For that purpose we consider
52 J.-P. Merlet
the following theorem that is used for studying the Farey sequences1 :
Theorem 1: Let s = a/b and t = c/d > s and if there is no rational between s and
t with a denominator that is lower than the largest b or d, then bc − ad = 1 and the
a+c
rational with smallest denominator between s and t is b+d .
Let us illustrate the application of this theorem in our previous example for dm = 2.4
and dc = 2. We have t = p/q = 6/5 and we are looking for s that is smaller than t
so that there is no rational between s,t with a denominator that is smaller than a, 5.
According to the theorem we must have 6b − 5a = 1 that is clearly satisfied for a =
b = 1. Then according to the theorem the next rationals having the lowest denominator
are 7/6, 13/11 and so on with increasing denominator. As the denominator are larger
than 5, then the closest valid rational p1 /q1 to p/q, that is lower than p/q, is 1/1. As
b = k − k we get the minimal Δ ρ as dc × (−1 + 6/5) = dc /5 = 0.4. For any ρ obtained
for the pair (k1 , k), then the ρ obtained for the pair (k1 + 1, k + 1) will differ from the
previous one by 0.4.
We may now examine the closest valid rational p1 /q1 to p/q, that is larger than
p/q. Using the theorem we get the condition 5c − 6d = 1 that is fulfilled by c = 5, d = 4
leading also to Δ ρ = 0.4. For being more systematic we may use the following theorem
Theorem 2: let two rationals a/b and c/d and let u = p/q the rational closest to c/d
and larger than c/d with denominator lower or equal to n. Let k be the largest integer
such that k ≤ (n + b)/d. The value of p, q are given by
p = kc − a q = kd − b
We will use this theorem by starting from the lowest possible successive value for
a/b, c/d which are 1/(kmax − 1), 1/(kmax − 2) and construct the full Farey sequence or
order 5 corresponding to .dm = 2.4 and dc = 2. We get the following pairs (p/q, |Δ ρ|):
(1/1 or 5/4, 0.4), (4/3, 0.8), (3/2, 1.2), (2/1, 1.6), (7/5, 2). Note that Δ ρ = dm is always
possible by setting k1 = 1, k = kmax , k = kmax − 1. But although we have obtained what
are the possible values of Δ ρ we have not established their frequency for a given config-
uration. The dead length is b = 12.6 and measurements for ρ between 2 and 68 will be
obtained. We get 142 successive ρ that differs by 0.4, 2 that differs by 0.8, 1.2, 1.6 and
1 that differ by 2. The Δ ρ distribution as a function of the cable length ρ is provided in
figure 2. As may be seen, apart at the extremity, the Δ ρ is everywhere equal to 0.4 with
a mean value of 0.442 and a variance of 0.04571.
Let us now consider that dm = 1.3, dc = 2 so that dm /dc = 13/20 and kmax = 11.
We get the following pair (p/q, |Δ ρ|): (2/3, 0.1), (5/8, 0.4), (1/2, 0.6), (1/1, 0.7), (3/4,
0.8), (5/7, 0.9), (7/10, 1), (4/7, 1.1) and Δ ρ = 0.1 may always be obtained . This is
confirmed by the calculation with 216 measurement differing by 0.1, 72 by 0.4, 12 by
0.5, 12 by 0.6, 4 by 0.7 and 1 for 1.3. The value of b is 13.7 and we get measurements
for ρ between 1.3 and 69 (figure 2). The mean value of the Δ ρ is 0.213 and its variance
is 0.0322. It may be observed that although we have almost doubled the number of
1 A Farey sequence of order n are the irreducible rationals between 0 and 1 whose denominator
is lower or equal to n
Fig. 2. On the left the distribution of Δ ρ as a function of ρ for dm = 2.4, dc = 2 with 29 marks
and 6 color sensors. On the right the distribution for dm = 1.3, dc = 2 with 29 marks and 11 color
sensors.
color sensors compared to the previous case we still get a large number of cases with
Δ ρ = 0.4 so that the choice of dm = 1.3 may not be optimal. Optimal configuration is
addressed in the next section.
5.2 Optimal configuration
Optimal choice of sensor distance dm for a given dc
A key point for realizing such a measurement system is first to determine what could
be the number kmax of color sensors and then select the dm for the best performance. For
a given kmax , dm should lie in the range ]h/(kmax +1), h/kmax ]. Consider for example that
h = 15, kmax = 6, dc = 2. We may draw the curve of the mean and variance values of the
Δ ρ as a function of dm , figure 3. As may be seen the mean value of Δ ρ is an increasing
function of dm while there is a discontinuity point for dm = 2.4 with a sudden increase
of the mean value but also a sudden decrease of the variance.
To understand this behavior we have to consider how many minimal Δ ρ we will
obtained for a given rational ratio dm /dc = p/q. For that purpose we have to look at
the rational p1 /q1 which has a denominator q1 lower or equal to kmax and that is the
closest to p/q. The larger the sets of possible k1 , k are the larger will be the set of
1 −
minimal Δ ρ. As the values of k1 , k are restricted to lie respectively in the range [1, kmax
p1 ],[1, kmax − q1 ] the larger the sets will be if p1 , q1 are close to 1.
Let us assume that q is lower than kmax . According to theorem 1 the closest rational
p1 /q1 that is lower than p/q should satisfy pp1 − q1 q = 1. If we set p1 = q1 = 1 we get
the constraint (A) p − q = 1 so that dm = pdc /(p − 1) Hence dm is a decreasing function
of p that cannot be greater than kmax = 6. If we look at all possible value for p between
1 and 6, then we found out that (A) may be satisfied only for p = 6. The corresponding
dm /dc ratio is 6/5 = 2.4 and for this ratio we will get the maximal number of Δ ρ that
are at the minimum (in this case Δ ρ = 0.4 as seen on figure 2). However for this value
54 J.-P. Merlet
Fig. 3. The mean value and variance of Δ ρ as a function of the distance dm between the color
sensor for dc = 2 and 29 color marks
the mean value is not that good: other value of dm may lead to a sequence of .Δ ρ whose
minimum is lower than 0.4.
Let us look now at dm = 2.35, a value that presents the second minimal value for
the variance. For this value we get 23 Δ ρ = 0.25, 140 Δ ρ = 0.35, 2 Δ ρ = 0.6, 0.95,
1.3, 1.65 and 1 Δ ρ = 2. Here we have a larger number of Δ ρ equal to 0.25 or 0.35 than
compared to the value 0.4 obtained for dm = 2.4. Hence apparently for a given dc we
shall consider as possible optimal values of dm the one having the lowest variances.
We may also consider increasing dc to a larger value for CDPR having large cable
length. For example if we have 29 marks with dc = 5, dm = 2.2 (6 sensors) we get a
measuring range of 151 meters with the following pairs of Δ ρ and their number: (0.6,
112), (1, 27), (1.6, 30). If we move to dc = 10 and the same number of sensors and dm ,
then the measuring range increases to 291 meters with (1, 28), (1.2, 56), (2.2, 88).
Influence of the number of marks
Although the number of sensors in a system is important, these sensors are inexpen-
sive and require a low level of maintenance while the marks will require more attention.
We may thus consider having less than the maximum number of marks imposed by the
the initialization process, but this will impose to have larger dc in order to have a suffi-
cient total cable length.
As seen in the previous section the ratio dm /dc = p/q that has the maximum of
lowest Δ ρ is such the closest rational with a denominator lower or equal to q should
be p1 /q1 = 1/1. Using theorem 1 we get p − q = 1 if p1 /q1 is lower than p/q so
that dm = dc p/(p − 1) provided that p − 1 is lower or equal to the number of sensors
f loor(h/dm ). If p1 /q1 is greater than p/q we get q − p = 1 so that dm = dc p/(p + 1)
Therefore being given dc we are able to find all the valid dm .
Assume that we will use only 20 marks instead of 29 but set dc to 3 in order to still
have a large total cable length. For dc = 3 table 1 provides some examples of dm leading
to interesting set of Δ ρ with their distribution.
dm ρmax number of sensors Δρ number

4.5 66 3 1.5-3 40 1
4 65 3 1-2-3 55 2 1
3.75 68.25 4 0.75 -1.5-2.25-3 73 2 2 1
2.5 69.5 6 0.5-1-1.5-2-2.5 111 2 2 2 1
2.4 69 6 0.6-1.2-1.8-2.4 97 2 2 1
2.15 67.75 6 0.4-0.45-0.85 34 54 27
1.9 68.4 7 0.3-0.5-0.8-1.1-1,9 72 34 26 4 1
1.1 70.2 13 0.1-0.2-0.3-0.5-0.8-1.1 32 85 126 6 6 3
Table 1. Minimal Δ ρ measurement and their number for various dm being given dc = 3 and 20
marks on the cable
It may be seen that the best compromises for 3 sensors is dm = 4 with an increase
of accuracy for 4 sensors with dm = 3.75, while dm = 2.5 is optimal for 6 sensors.
Note that we have added a line with dm = 2.15 that correspond to the lowest variance
beside the one leading to the maximal number of minimal Δ ρ as it offers interesting
performances. As may be seen from this table increasing the number of sensors may
have a large benefit on the measurement. For 13 sensors we get 243 changes for Δ ρ in
the range [0.1, 0.3].
We may now consider going into the opposite direction by increasing the number
of marks in order to increase the accuracy of the system, at the price of requiring the
reading of 4 marks to initialize the cable length (although this argument will be invali-
dated if we use the initialization procedure described in section 5.3). Consider first that
dc = 1.5 so that we have 40 marks. We consider the case where the variance is minimal
and table 2 summarizes the result. As it seems difficult to position the sensor with an
dm ρmax number of sensors Δρ number

2.4 69.75 6 0.3-0.6-0.9-1.5 189 12 4 1
2.25 70.5 6 0.75-1.5 89 1
1.6875 70.3125 8 0.1875 . . . 305
1.8 71.1 8 0.3-0.6-0.9-1.2-1.5 209 2 2 2 1
1.83 71.31 8 0.15-0.18-0.33 . . . 102 140 69
1.66 71.78 9 0.16-0.22-. . . 312 31
1.35 72 11 0.15-. . . 392
1.27 71.2 11 0.11-0.12-0.23-. . . 136 175 118
Table 2. Minimal Δ ρ measurement and their number for various dm being given dc = 1.5 and 40
marks on the cable
56 J.-P. Merlet
accuracy of half a centimeter, the optimal solution for dm seem to be 2.4, 1.8, 1.83, 1.66,
1.35. Table 3 summarizes good design solutions for cable length in the range of 60-70
meters.
number of marks dc dm number of sensors Δρ number

12 5 3.75 4 1.25 43
15 4 3 5 1 58
20 3 1 3 1 55
20 3 3.75 4 0.75 73
20 3 2.5 6 0.5 111
29 2 2.4 6 0.4 142
29 2 2.35 6 0.25-0.35 23-140
40 1.5 2.4 6 0.3-0.6 189-12
40 1.5 1.83 8 0.15-0.18-0.33 102-140-69
40 1.5 1.66 9 0.16-0.22 312-31
40 1.5 1.35 11 0.15 392
60 1 2.4 6 0.3 289
60 1 1.75 8 0.25 363
60 1 1.3 11 0.1 0.2 375 272
84 0.7 1 15 0.1 639
Table 3. Best system arrangement for various dc (distance between marks) and number of sensors
(dm = distance between color sensors)
5.3 The initialization problem: a second approach

We have proposed in section 3 a first approach to determine the initial cable length based
on the detection of n successive marks by one sensor, the coding of the cable being such
that there is a single occurrence of a given sequence of n colors on the cable. For the
initialization process we therefore need to coil the cable by ndc . However we may get a
faster initialization process by taking into account that we have several color sensors in
the mast so that we may use all event detection event for performing the initialization
process. Clearly we aim at determining the cable length by coiling the cable by less
than ndc .
If we have n different colors on the cable and m marks on the cable so that m is
maximal, then we will have roughly m/n marks of the same color. For example for
n = 3, m = 29 and the marking presented in section 3 we have 10 marks of color 1, 2
and 9 marks of color 3. Remind that if the color sensor ns detects mark ms , then the
cable length ρ is given by
ρ = ns dm − h + L0 − (ms − 1)dc (6)
where L0 is the total length of the cable. After starting the coiling there will be a first
sensor detection event that provide ns while ms is not known. However the detected
color provides a limited choice for ms (at most 10 in our example) and consequently
a limited set {ρ1 , . . . , ρl } of possible ρ. For each of the ρi we consider the set of all
possible values of ρ that are lower than ρi obtained for all possible values of ns , ms
and we then sort this set by decreasing value of ρ. The first element of the ordered set
provides what will be the next sensor detection event if the current value of the ρ is ρi .
Hence we get a list L of (ρi , nis , ci ) where nis is the sensor number and ci is the mark
color for the next event. When a new sensor event occurs we check the coherence of
the ns and the color with each nis , ci : each element of L that is not coherent with the
sensor number or color is removed from L . This leads to a new list that contains all
possible values of the current value of ρ and prediction about the next sensor event and
we repeat the process. As the coherence test allows one to reduce the size of L , we
shall end up with a list with a single element that is the current value of ρ.
For example we consider the case where we have 29 marks separated by 2 meters
with 3 different colors and 6 color sensors separated by 1.6 meter. If the cable length is
ρ = 21.5 when starting the initialization process, then the first detection event is color 3
on sensor 4. Using color 3 we obtain the list L ={(45.4, 5, 1), (29.4, 5, 2), (21.4, 5, 2),
(17.4, 5, 3), (15.4, 5, 1), (11.4, 5, 1), (3.4, 5, 3), (1.4, 5, 3)}. The second sensor event is
color 2 for sensor 5 meaning that among the previous list only the value 21.4 and 29.4
are coherent, leading to L = {(29, 6, 2), (21, 1, 2)}. As the next event has color 2 and
sensor 1 we may discard 29 and we will have determined that ρ = 20.2, after coiling
only 1.3 meter of cable. More generally figures 4 shows how much cable should be
coiled before getting the current ρ value as a function of the initial value of ρ and the
number of detection events that are necessary to identify this value.
Fig. 4. On the left the coiling amount that is necessary to identify the current ρ as a function of
the ρ value. On the right the number of sensor events that are necessary to identify the current ρ
as a function of the ρ value (29 marks, 6 sensors, dc = 1.6, dm = 2).
As may be seen on these figures the average amount of coiling distance for the
identification of the current ρ is 1.069 meter (variance: 0.054, min: 0.4, max: 1.95) for
an initial ρ between 10 and 50 meter, much better than the value of 6 meters required
58 J.-P. Merlet
for the method proposed in section 3. If we use dm = 1.83, dc = 1.5, 8 color sensors and
40 marks, then in average we need only 0.622 meter of coiling (variance: 0.03, min:
0.34, max: 1.29) for determining the current ρ.
6 Conclusion
In this paper we propose a setup for both allowing an automated determination of the
initial cable lengths and improving the cable length measurements. The method is sim-
ple and allows one to obtain very good estimation of the current cable lengths especially
for large CDPR (for example being able to provide the cable length for any change of
0.4 meter for a 60 meters cable). It allows also to get an estimate of the drum radius
at each sensor event, thereby allowing to improve the estimation of the cable length
between two sensor events. The method is robust: being given the sensor price it is pos-
sible to use sever color sensors within a given sensor box. Furthermore it is possible to
detect sensor failure based on the absence of a color signal during a significant change
of cable length. We intend also to explore if this system may not allow to estimate the
cable elastic deformation by comparing the difference between the distance between
marks (that is known at rest) and the distance observed between sensor events. On-line
calibration may also benefit from being able to fix the cable lengths at known values.
References
[Baczynski(2010)] Baczynski J, Baczynski M (2010) Simple system for determining starting
position of cable-driven manipulator. In: IEEE Int. Conf. on Computer Information Systems
and Industrial Management Applications (CISIM), Cracow, pp 102–106
[Bruijn(1946)] Bruijn N (1946) A combinatorial problem. Indagationes Math 8:461–467
[Bruijn(1975)] Bruijn N (1975) Acknowledgment of priority to c. flye sainte-marie on the count-
ing of circular arrangements of 2n zeros and ones that show each n-letter word exactly once.
Indhove: Technische Hogeschool Eidenhoven
[Chen(2004)] Chen Q, et al (2004) An integrated two-level self-calibration method for cable-
driven manipulator. IEEE Trans on Robotics 10(2):380–391
[Gouttefarde et al(2012)] Gouttefarde M, et al (2012) Simplified static analysis of large-
dimension parallel cable-driven robots. In: IEEE Int. Conf. on Robotics and Automation,
Saint Paul, pp 2299–2305
[Merlet(2008)] Merlet JP (2008) Kinematics of the wire-driven parallel robot MARIONET using
linear actuators. In: IEEE Int. Conf. on Robotics and Automation, Pasadena
[Merlet(2010)] Merlet JP (2010) MARIONET, a family of modular wire-driven parallel robots.
In: ARK, Piran, pp 53–62
[Miermeister and Pott(2012)] Miermeister P, Pott A (2012) Auto calibration method for cable-
driven parallel robot using force sensors. In: ARK, Innsbruck, pp 269–276
[Pott et al(2012)] Pott A, et al (2012) IPAnema: a family of cable-driven parallel robots for in-
dustrial applications. In: 1st Int. Conf. on cable-driven parallel robots (CableCon), Stuttgart,
pp 119–134
[Alexandre(2012)] Alexandre dit Sandretto J, Daney D, Gouttefarde M (2012) Calibration of a
fully-constrained parallel cable-driven robot. In: RoManSy, Paris, pp 12–41
[Alexandre et al(2013)] Alexandre dit Sandretto J, et al (2013) Certified calibration of a
cable-driven robot using interval contractor programming. In: Computational Kinematics,
Barcelona
Part II
Kinematics and Static
Stiffness of Planar 2-DOF 3-Differential
Lionel Birglen1 and Marc Gouttefarde2

1
Department of Mechanical Engineering, Polytechnique Montréal, QC, Canada
2
LIRMM, University of Montpellier, CNRS, Montpellier, France
[email protected], [email protected]
Abstract. Planar 2-degree-of-freedom (DOF) 3-differential Cable-Driven

Parallel Robots (CDPRs) consist of a point-mass end-effector driven by
a number of cables. Each cable is divided into four segments, three of
them being connected to the point-mass end-effector by means of routing
pulleys. This paper deals with the stiffness analysis of such planar 2-DOF
3-differential CDPRs. Based on the usual linear spring cable elongation
model, the expression of the stiffness matrix is derived. The stiffness and
workspace of several examples of planar 2-DOF 3-differential CDPRs are
then compared. The results of these comparisons illustrate that the stiff-
ness of planar CDPRs can be significantly improved by means of pulley
differentials.
Keywords: Cable-driven parallel robots, differential pulley actuation,

stiffness analysis
1 Introduction
This paper deals with cable-driven parallel robots (CDPRs) whose cable seg-
ments are not each driven by a single actuator but are coupled through differ-
entials. Coupling and transmitting the actuation torque of a single actuator to
several cable segments has been shown in previous works to be a possible so-
lution to extend the workspace of CDPRs by purely passive mechanical means
[1–3], i.e. without relying on additional actuators or relocating the latter on a
structural frame. Compared to conventional CDPRs, more cable segments then
cluttered the workspace but this should not be a critical issue for planar and
suspended CDPRs. In the specific example studied here, a single actuator simul-
taneously controls the lengths of three cable segments going from the ground to
the point-mass end-effector of a planar robot and the coupling between these
cable segments is implemented through pulley differentials, similarly in principle
to the well-known block and tackle. The present paper presents and quantifies
the stiffness of such planar 2-DOF 3-differential CDPRs. Such differential cou-
plings will be shown to provide a noticeable improvement in the stiffness of the
CDPR.
As shown in Fig. 1, each cable of the CDPR consists of four successive seg-
ments Bi P , P Ei , Ei Mi , and Mi P . Three of these cable segments link the point

62 L. Birglen and M. Gouttefarde
Fig. 1. A planar 2-DOF 3-Differential CDPR where each cable consists of four segments
Bi Pi , Pi Ei , Ei Mi , and Mi Pi , three of these segments connecting the end-effector P to
the ground points Bi , Mi and Ei .
mass P , which plays the role of the CDPR 2-DOF end-effector, to the ground.
In the remainder of this paper, these three segments are referred to as the active
cable segments. The other one, Ei Mi , extends between points Ei and Mi which
are both fixed to the ground. In other words, cable i exits the base first at point
Bi , and then goes in this order through points P , Ei , Mi to finally be attached
to point P . Hence, these pulley drives are referred to as 3-differential since each
one has three “active” outputs. Figure 2 shows this cable routing. In order to
implement such a mechanism, pulleys need to be used at points P , Ei , and Mi .
However, as a first approximation, the diameters of these pulleys are neglected
(considered to be null) and pulley friction is ignored (no friction). The latter
assumption implies that the cable tension is constant all along the cable, i.e.,
the four cable segments have the same tension and each active segment applies
a force of identical magnitude on the end-effector. Let n be the number of cables
used to drive point P (n = 3 in the case of Fig. 1). This paper deals with the
case of planar 2-DOF 3-differential CDPRs with n ≥ 2 cables. The mass of the
cables is neglected and all the cable segments are considered to be straight line
segments.
Stiffness and vibration analyses of usual CDPRs, where the driving cables are
directly attached to the end-effector, have been proposed in a number of works,
e.g. [4–11]. As previously mentioned, the main difference with these previous
works is that the present paper deals with the stiffness analysis of planar point-
Stiffness of Planar 2-DOF 3-Differential Cable-Driven Parallel Robots 63
Fig. 2. Cable routing in a single differential: The cable starts at point Bi where the
actuated drum is located, goes to P (the end-effector), then to Ei where it is redirected
to point Mi in order to be finally attached to point P .
mass CDPRs with 3-differential pulley drive. To the best of our knowledge, the
only previous work on stiffness analysis of differential-pulley planar CDPRs is
[12] where the case of cables with two active segments is considered. Moreover,
the stiffness analysis in [12] only considered stiffness in a single direction and did
not mention the trade-off between the size of the workspace and the improvement
in stiffness inherent to pulley differentials as will be shown here.
This paper is organized as follows: First, Section 2 presents the Jacobian
and wrench matrices of differential-pulley-driven planar point-mass CDPRs with
three active cable segments. Then, in Section 3, the stiffness matrix of these
mechanisms is derived based on the usual linear spring cable elongation model.
Finally, Section 4 reports simulations of the resulting stiffness of several examples
of planar 2-DOF 3-differential CDPRs.
2 Wrench and Jacobian Matrices
The active length of the cable i in each differential routing is defined as li =

−−→ −−→ −−→
lBi P +lEi P +lMi P where lBi P = Bi P , lEi P = Ei P and lMi P = Mi P are the
strained lengths of the active cable segments Bi P , Ei P and Mi P , respectively.
Length li is the sum of the cable segment lengths that change when the position
of point P changes. It should not be confused with the total length of the cable
which includes the length of the cable segment between points Ei and Mi . Let us
T
define the vector l of the active cable lengths li as l = [l1 , l2 , . . . , ln ] . Moreover,
T
let l̇ be the time derivative of l, p = [px , py ] be the position vector of point P
in the fixed reference frame (0, x, y), and ṗ the velocity of point P . A Jacobian
matrix J then maps ṗ (resp. dp) into l̇ (resp. dl)
⎡˙ ⎤
l1
⎢ l˙2 ⎥
l̇ = ⎢ . ⎥ = Jṗ ⇐⇒ dl = Jdp (1)
⎢ ⎥
⎣ .. ⎦
l˙n
with
uTB1 + uTE1 + uTM 1
⎡ ⎤
J=⎣
⎢ .. ⎥
(2)
. ⎦
uTBn + uTEn + uTM n n×2
where uBi , uEi and uM i are the unit vectors directed along the cable segments
Bi P , Ei P , and Mi P , pointing from the base points Bi , Ei , and Mi , to point
T
P , respectively. The vector of cable tensions ti is denoted t = [t1 , t2 , . . . , tn ] .
Pulley friction being neglected, the cable tension ti is constant all along the
cable. The force f applied by the n cables to the point-mass P is then given by
f = Wt (3)
where the 2 × n wrench matrix W is defined as W = −J . T
3 Stiffness Analysis
The elastic potential energy of cable i is given by
1 2
Vi =ki (lti − l0i ) (4)
2
where lti is the total length of cable i
lti = li + lEi Mi (5)
with lEi Mi the length of the passive cable segment Ei Mi and li = lBi P + lEi P +
lMi P the active length of cable i, as defined in Section 2. l0i is the corresponding
unstrained length, l0i = l0Bi P +l0Ei P +l0Ei Mi +l0Mi P . The elongation is supposed
to be uniform along the cable length and the cable is assumed to be a linear spring
with ti = ki (lti − l0i ), ki = AE
l0i , where A and E are the cable cross-sectional
area and elastic modulus, respectively.
The total elastic potential energy V of the 3-differential CDPR is then the
sum of the elastic potential energies of all its n cables
n n
1 2
V = Vi = ki (lti − l0i ) (6)
i=1
2 i=1
Computing the derivative of Equation (6) yields

n
n
n

dV = dVi = ki (lti − l0i ) dlti = ki (lti − l0i ) dli (7)
i=1 i=1 i=1
where dlti = dli according to Eq. (5) and to the fact that dlEi Mi = 0 (the
distance between the ground points Ei and Mi is constant). Then, defining the
diagonal matrix D and vectors lt and l0 as follows
⎡ ⎤ ⎡ ⎤
⎡ ⎤ lt1 l01
k1 ⎢ lt2 ⎥ ⎢ l02 ⎥
D = ⎣ . . . ⎦ , lt = ⎢ . ⎥ and l0 = ⎢ . ⎥ (8)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ .. ⎦ ⎣ .. ⎦
kn
ltn l0n
Eq. (7) can be written

T
dV = (lt − l0 ) Ddl (9)
According to Eq. (1), dl is equal to Jdp so that according to Eq. (9)
⎡ ∂V ⎤
dV ⎢ ∂px
⎥ = JT D (lt − l0 )
⎥
=⎢ (10)
dp ⎣ ∂V ⎦
∂py
where it can be noted that D (lt − l0 ) = t since ti = ki (lti − l0i ). The stiffness
matrix K of the 3-differential CDPR is defined as the second derivative of the
elastic potential energy, namely
∂2V ∂2V
⎡ ⎤
d2 V ⎢ ∂p2 ∂px ∂py ⎥

x
K= = (11)
⎢ ⎥
dp2 ⎣ ∂2V ∂2V ⎦
⎢ ⎥
∂py ∂px ∂p2y
Taking the derivative of Eq. (10) with respect to p, the following classic
expression of the stiffness matrix is obtained
dl dJT dJT
K = JT D + D (lt − l0 ) = JT DJ + t (12)
dp dp dp
where, once again, the equality dlt = dl has been used. It is worth noting that this
stiffness matrix K can only be used in local stiffness analyses because Eq. (12)
is obtained from Eq. (10) by assuming that l0 is constant. In other words, the
matrix K is defined at an equilibrium, defined by both the position p and the
cable tensions t, and it can be used to analyze the stiffness of the robot in the
vicinity of this equilibrium. In this local analysis, the actuators are assumed to be
locked or their position controlled to a constant value. Note also that, according
dV
to Eq. (3) and (10), = JT t = −f , and thus
dp

d dV d(−f )
K= = ⇐⇒ Kdp = d(−f ) (13)
dp dp dp
Hence, in the vicinity of an equilibrium defined by p and t, K maps an infini-

tesimal displacement dp in the position of point P to the corresponding infini-
tesimal change in the force −f , where f is the force applied by the cables to
point P (Eq. (3)).
The approach used above to derive the stiffness matrix is a rather classic one,
but care must be taken to use in Eq. (12) the Jacobian matrix defined in Eq. (1),
where each row of this matrix is equal to the sum uTBi + uTEi + uTM i of the unit
vectors directed along the three straight line segments delineated by the active
cable segments.
Referring to the right-hand side of Eq. (12), the stiffness matrix is seen to
be the sum of two matrices. In the next section of this paper, for simplicity and
due to space limitations, only the influence on stiffness of the matrix JT DJ will
be considered.
4 Results
Fig. 3. Example planar 2-DOF 3-differential CDPR with 3 cables (Case 1)
This section reports simulations of the stiffness of six planar 2-DOF 3-

differential CDPRs with n = 3 cables, across their respective workspaces. These
six examples are defined as Cases 1 to 6 in Tables 1 and 2. Case 1 is shown in
Fig. 3 together with its Wrench-Closure Workspace (WCW) [13] and Wrench-
Feasible Workspace (WFW) [14]. Here, the WFW is defined as the set of po-
sitions of point P such that any force f of magnitude less or equal to 0.1 N
can be generated by the cables at P with tensions ti , i = 1 . . . n, satisfying
tmin ≤ ti ≤ tmax where tmin = 0.1 N and tmax = 1 N.
Case 2 defined in Table 1 is a planar 2-DOF 3-differential CDPR where the
three active cable segments are superposed. It is obtained from Case 1 by taking
Bi ≡ Ei ≡ Mi for i = 1, 2 and 3. Case 3 in Table 1 defines a CDPR without
pulley differential where each cable consists of only one segment from Bi to P . It
is obtained from Case 1 by keeping segment Bi P and removing the other cable
segments P Ei , Ei Mi , and Mi P .
Case 4 defined in Table 2 is another planar 2-DOF 3-differential CDPR. It
is shown in Fig. 4 where its WFW is defined as above for Case 1. Case 5 is a
planar 2-DOF 3-differential CDPR obtained from Case 4 by superposing the 3
Case 1 Case 2 Case 3

x (m) y (m) x (m) y (m) x (m) y (m)
B1 cos(π/3) sin(π/3) cos(π/3) sin(π/3) cos(π/3) sin(π/3)
E1 cos(2π/3) sin(2π/3) cos(π/3) sin(π/3) × ×
M1 0 1 cos(π/3) sin(π/3) × ×
B2 −1 0 −1 0 −1 0
E2 cos(−2π/3) sin(−2π/3) −1 0 × ×
M2 cos(−5π/6) sin(−5π/6) −1 0 × ×
B3 cos(−π/3) sin(−π/3) cos(−π/3) sin(−π/3) cos(−π/3) sin(−π/3)
E3 1 0 cos(−π/3) sin(−π/3) × ×
M3 cos(−π/6) sin(−π/6) cos(−π/3) sin(−π/3) × ×
Table 1. Geometries of the first three examples of planar 2-DOF 3-differential CDPRs.
The crosses × indicate that the corresponding point Ei or Mi does not exist, i.e., Case
3 defines a CDPR without pulley differential where each cable consists of only one
segment from Bi to P .
Case 4 Case 5 Case 6

x (m) y (m) x (m) y (m) x (m) y (m)
B1 1 0 1 0 1 0
E1 cos(2π/3) sin(2π/3) 1 0 × ×
M1 cos(π/3) sin(π/3) 1 0 × ×
B2 cos(2π/3) sin(2π/3) cos(2π/3) sin(2π/3) cos(2π/3) sin(2π/3)
E2 cos(−2π/3) sin(−2π/3) cos(2π/3) sin(2π/3) × ×
M2 −1 0 cos(2π/3) sin(2π/3) × ×
B3 cos(−2π/3) sin(−2π/3) cos(−2π/3) sin(−2π/3) cos(−2π/3) sin(−2π/3)
E3 1 0 cos(−2π/3) sin(−2π/3) × ×
M3 cos(−π/3) sin(−π/3) cos(−2π/3) sin(−2π/3) × ×
Table 2. Geometries of the last three examples of planar 2-DOF 3-differential CDPRs.
Case 6 defines a CDPR without pulley differential where each cable consists of only
one segment from Bi to P .
Fig. 4. Example planar 2-DOF 3-differential CDPR with 3 cables (Case 4)
active segments of each cable and Case 6 is a CDPR without pulley differential
obtained from Case 4 by removing, for each cable, all segments but Bi P .
As mentioned at the end of Section 3, the subsequent stiffness analysis is
based on considering that the stiffness matrix is K = JT DJ. To “normalize”
the stiffness coefficient with respect to specific values, we use E = 1 Pa and
A = 0.001 m2 . Note that these values, together with the tmin and tmax values
defined above, do not correspond to actual physical values since, for instance,
steel wire ropes typically have elastic modulus E of tens of GPa. However, it
is not important for the purposes of this section where stiffness comparisons
between several 2-DOF 3-differential CDPRs are made. Moreover, in the calcu-
lation of D, for simplicity, we assume that the unstrained length of each cable
segment is equal to its strained length. For a given position of the end-effector P ,
the two eigenvalues of the 2 × 2 stiffness matrix K can then be calculated. The
minimum (resp. maximum) stiffness values shown in the left part (resp. right
part) of Fig. 5 are defined as the minimum (resp. maximum) of these two eigen-
values. The values of the ratios of the minimum to the maximum eigenvalues
over the WCW of Case 1 are shown in Fig. 6.
The means of the minimum stiffness values, maximum stiffness values and
ratio of minimum to maximum stiffness values over the WCW of the planar
2-DOF 3-differential CDPR examples defined in Tables 1 and 2 are given in
Table 3. The mean of the minimal stiffness in Case 1 is 2.4 times larger than in
Case 3, where Case 3 corresponds to a classic CDPR with no pulley differential.
The means of the maximum stiffness and stiffness ratio in Case 1 are both 1.4
times larger than in Case 3. Comparing Case 2 to Case 3, the minimum and
maximum stiffness values are 3 times larger for Case 2 while the stiffness ratios
are identical. Hence, planar 2-DOF 3-differential CDPRs (Cases 1 and 2) present
a larger stiffness over their workspace than the corresponding CDPR without
Fig. 5. Minimum and maximum stiffness values over the WCW of Case 1
Case min (N/m) max (N/m) ratio

1 0.0018 0.0041 0.43
2 0.0022 0.0086 0.31
3 0.00074 0.0029 0.31
4 0.0012 0.0016 0.79
5 0.0022 0.0086 0.31
6 0.00074 0.0029 0.31
Table 3. Means of the minimum stiffness values (min), maximum stiffness values (max)
and ratio of minimum to maximum stiffness values (ratio) over the WCW of the 2-DOF
3-differential CDPR examples
pulley differential (Case 3). Similar results are obtained by comparing Cases 4
and 5 to Case 6. The cases with superposed cable segments (Cases 2 and 5) have
larger minimum and maximum stiffness values than the ones with distinct cable
segments (Cases 1 and 4). Note that Case 4 has the best stiffness ratio.
Finally, Table 4 shows a comparison of the size of the WCW and WFW in
each case. The WFW is defined for all the six cases as it has been defined above
for Case 1. From these examples, it can be concluded that the cases of 2-DOF
3-differential CDPRs with superposed cable segments (Cases 2 and 5) lead to
the largest stiffness improvements but with a marginal gain on the sizes of the
WCW and WFW, as compared to the cases without pulley differentials (Cases 3
and 6). In comparison, the stiffness improvement is less significant in the cases of
2-DOF 3-differential CDPRs with distinct cable segments (Cases 1 and 4) with,
however, a slightly better improvement in the WCW and WFW sizes.
Fig. 6. Ratio of minimal to the maximal stiffness values over the WCW of Case 1
5 Conclusion
This paper presented the stiffness analysis of planar 2-DOF CDPRs with pulley
differentials where a single actuator simultaneously controls the lengths of three
cable segments going from the ground to the point-mass end-effector. Based on
the usual linear spring cable elongation model, the expression of the stiffness ma-
trix has been derived. The stiffness and workspace of several examples of planar
2-DOF 3-differential CDPRs have then been compared. These comparisons show
that the stiffness of such planar CDPRs can be significantly improved compared
to more common CDPRs without pulley differentials, without decreasing the
size of the CDPR workspace. Additionally, the presented example comparisons
Case WCW WFW

1 37.6% 29.5%
2 32.4% 27.3%
3 32.4% 24.7%
4 33.6% 22.6%
5 32.4% 27.3%
6 32.4% 24.7%
Table 4. WCW and WFW coverage of the total area occupied by the CDPR examples.
This total area is the one delimited by the circles shown in Fig. 3 and 4.
also showed that a trade-off may exist between designing a CDPR with pulley
differentials to improve stiffness and designing one to increase the workspace
size.
References
1. H. Khakpour, L. Birglen, and S. Tahan, “Synthesis of differentially driven planar
cable parallel manipulators,” IEEE Transactions on Robotics, vol. 30, no. 3, pp.
619–630, June 2014.
2. H. Khakpour and L. Birglen, “Workspace augmentation of spatial 3-dof cable par-
allel robots using differential actuation,” in 2014 IEEE/RSJ International Confer-
ence on Intelligent Robots and Systems, Sept 2014, pp. 3880–3885.
3. H. Khakpour, L. Birglen, and S.-A. Tahan, “Analysis and optimization of
a new differentially driven cable parallel robot,” Journal of Mechanisms and
Robotics, vol. 7, no. 3, pp. 034 503–034 503–6, Aug 2015. [Online]. Available:
http://dx.doi.org/10.1115/1.4028931
4. S. Behzadipour and A. Khajepour, “Stiffness of cable-based parallel manipulators
with application to stability analysis,” ASME J. Mech. Des., vol. 128, pp. 303–310,
jan 2006.
5. X. Diao and O. Ma, “Vibration analysis of cable-driven parallel manipulators,”
Multibody System Dynamics, vol. 21, pp. 347–360, 2009.
6. M. Arsenault, “Workspace and stiffness analysis of a three-degree-of-freedom spa-
tial cable-suspended parallel mechanism while considering cable mass,” Mechanism
and Machine Theory, vol. 66, pp. 1–13, 2013.
7. H.Yuan, E. Courteille, and D. Deblaise, “Static and dynamic stiffness analyses of
cable-driven parallel robots with non-negligible cable mass and elasticity,” Mecha-
nism and Machine Theory, vol. 85, pp. 64–81, 2015.
8. H.Yuan, E. Courteille, M. Gouttefarde, and P.-E. Hervé, “Vibration analysis of
cable-driven parallel robots based on the dynamic stiffness matrix method,” Jour-
nal of Sound and Vibration, vol. 394, pp. 527–544, 2017.
9. M. Anson, A. Alamdari, and V. Krovi, “Orientation workspace and stiffness opti-
mization of cable-driven parallel manipulators with base mobility,” ASME Journal
of Mechanisms and Robotics, vol. 9, 2017.
10. S. Abdolshah, D. Zanotto, G. Rosati, and S. K. Agrawal, “Optimizing stiffness
and dexterity of planar adaptive cable-driven parallel robots,” ASME Journal of
Mechanisms and Robotics, vol. 9, 2017.
11. H. Jamshidifar, A. Khajepour, B. Fidan, and M. Rushton, “Kinematically-
constrained redundant cable-driven parallel robots: Modeling, redundancy analysis,
and stiffness optimization,” IEEE/ASME Transactions on Mechatronics, vol. 22,
no. 2, pp. 921–930, April 2017.
12. C. Nelson, “On improving stiffness of cable robots,” in Cable-Driven Parallel Robots
- Proceedings of the 3rd International Conference on Cable-Driven Parallel Robots,
ser. Mechanisms and Machine Science, vol. 53. Netherlands: Springer Netherlands,
2018, pp. 331–339.
13. M. Gouttefarde and C. M. Gosselin, “Analysis of the wrench-closure workspace of
planar parallel cable-driven mechanisms,” IEEE Transactions on Robotics, vol. 22,
no. 3, pp. 434–445, June 2006.
14. P. Bosscher, A. T. Riechel, and I. Ebert-Uphoff, “Wrench-feasible workspace gen-
eration for cable-driven robots,” IEEE Transactions on Robotics, vol. 22, no. 5, pp.
890–902, Oct 2006.
Stability Analysis of Pose-Based Visual Servoing
Control of Cable-Driven Parallel Robots
Zane Zake12 , Stéphane Caro13 , Adolfo Suarez Roos2 , François Chaumette4 ,

and Nicolò Pedemonte2
1
Laboratoire des Sciences du Numérique de Nantes, UMR CNRS 6004,
1, rue de la Noë, 44321 Nantes, France,
[email protected],
2
IRT Jules Verne, Chemin du Chaffault, 44340, Bouguenais, France
[email protected],
3
Centre National de la Recherche Scientifique (CNRS),
1, rue de la Noë, 44321 Nantes, France,
[email protected]
4
Inria, Univ Rennes, CNRS, IRISA, Rennes, France,
[email protected],
Abstract. Cable-driven parallel robots are robots with cables instead

of rigid links. The use of cables introduces advantages such as high pay-
load to weight ratio, large workspaces, high velocity capacity. Cables also
bring drawbacks such as bad accuracy when the robot model is not ac-
curate. In this paper, a visual servoing control is proposed in order to
achieve high accuracy no matter the robot model precision. The stability
of the solution is analyzed to determine the tolerable perturbation limits.
Experimental validation is performed both in simulation and on a real
robot to highlight the differences.
Keywords: cable-driven parallel robots, visual servoing, stability
1 Introduction
In cable-driven parallel robots (CDPRs) rigid links are substituted by flexible

cables. This substitution leads to advantages such as: (i) large workspace (WS),
(ii) reconfigurability [1], and (iii) high payload to weight ratio. However, the
accuracy of CDPRs is to be improved. Previously, the following methods to
improve accuracy have been studied: (i) increasing the complexity of the CDPR
model in model-based control [2]; using proprioceptive sensors, such as (ii) force
sensors to measure cable tensions [3] or (iii) angular position sensors to measure
cable angles [4]; using exteroceptive sensors, namely vision sensors [5] [6] [7].
The increasing popularity of vision-based control is due to its high robust-
ness to unexpected change in the environment. The two main approaches of
such control are: image-based visual servoing (IBVS) and pose-based visual ser-
voing (PBVS) [8]. The latter is used when information from the image and some
additional knowledge about the object (usually its model) allows us to estimate

74 Z. Zake et al.
the 3D pose of the object in the camera frame Fc . The control consists of min-
imizing the difference between the acquired object pose and the desired one at
each iteration. IBVS is used when estimation of the 3D pose is not possible.
Here, visual features such as 2D image coordinates or image moments are used
instead. The control then consists of minimizing the error in the image space by
comparing the desired and the current visual features.
Vision-based control on CDPRs is not yet well researched. It is surprising
given the challenge to achieve accurate model-based control, since it requires
to predict complex aspects like cable elongation and sagging, and pulley effect.
Furthermore, to find the solutions to the Forward Kinematic problem for CD-
PRs is a tedious task that requires a good knowledge of actual cable lengths and
tensions, acquired by proprioceptive sensors. Although the addition of proprio-
ceptive sensors can increase the robot accuracy, vision-based control is inherently
robust to modeling errors and uncertainties, and avoids the need of computing
the Forward Kinematic problem. This is attained by actively perceiving either
the MP with a static camera in the so-called eye-to-hand configuration [5] [6], or
the object of interest with a moving camera mounted on MP in the eye-in-hand
configuration [7]. The robot is then actuated according to what is perceived.
Dallej et al. used multiple static cameras facing the MP as well as observing
cable exit points to determine their sag [5]. Multiple control schemes were pro-
posed. Similarly, Chellal et al. used 6 infra-red sensors to determine the MP pose
with high accuracy [6]. Remy et al. used the eye-in-hand configuration with a
single camera [7]. Their task was to control a spatial CDPR with three transla-
tional degrees of freedom. Finally, a control scheme was proposed in [9] to control
all six degrees of freedom while still using only one camera mounted on the MP.
One of the main characteristics of any control scheme is its stability. It was
found in [9] that the PBVS control of a CDPR is highly robust to modeling er-
rors. However due to the wide range of such errors and their combined effect on
the system it was not possible to determine their maximum values. In this paper,
possible sources of errors are studied separately to find their maximum values
for different MP working range. Such a separated analysis does not allow us to
evaluate the interaction between different errors. For this reason, a second eval-
uation is done, assuming a constant non-negligible perturbation. These values
are experimentally validated both in simulation and on a CDPR prototype.
This paper is organized as follows. Section 2 presents the control scheme.
Stability condition is established in Section 3. A case study is shown in Section 4.
Finally, the conclusions are drawn in Section 5.
2 Vision-Based Control of a CDPR

2.1 CDPR Kinematics
The schematic of a spatial CDPR in a suspended configuration is shown in Fig. 1.
Given that the camera is mounted on the MP, the transformation matrix
p
Tc between the camera frame Fc and the MP frame Fp is constant. On the
contrary, the transformation matrices b Tp and c To change with time.
Stability Analysis of Pose-Based Visual Servoing Control… 75
i cble
moving pltform
li
Bi p
bse
bi c cmer
Op
ai c object
p Oc
c
o
b
p Oo
b o
Ob
Fig. 1. Schematic of a spatial CDPR with eight cables, a camera mounted on its MP
and an object in the WS.
# »
The length li of the ith cable is the 2-norm of the vector Ai Bi pointing from
cable exit point Ai to cable anchor point Bi , namely,
# »
li = Ai Bi (1)

2
with
# »
li b ui = b Ai Bi = b bi − b ai = bRp p bi + b tp − b ai (2)
# »
where b ui is the unit vector of b Ai Bi that is expressed as:
b# »
Ai Bi b
b − ba b
R p b − b a + b tp
b
ui = # »
= i# »i = p i # »i (3)
b Ai Bi
b b
Ai Bi Ai Bi

2 2 2
The cable velocities l˙i are obtained upon differentiation of Eq. (2) with re-
spect to (w.r.t.) time:
l̇ = A b vp (4)
where b vp is the Cartesian velocity of the MP expressed in Fb , l̇ is the cable veloc-

ity vector, and A is the Forward Jacobian matrix of the CDPR, defined as [10]:
uT1 (bRp p b1 × b u1 )T
⎡b ⎤
A = ⎣ ... .. (5)
⎢ ⎥
. ⎦
u m ( Rp b m × u m )
b T b p b T
2.2 Pose-Based Visual Servoing
The control scheme proposed in this paper is shown in Fig. 2. An image is

retrieved from the camera and processed with a computer vision algorithm,
that returns the current pose of the object s. It is compared to a previously
76 Z. Zake et al.
known desired pose s∗5 and an error vector e is defined as e = [eTt eTω ]T , where
∗
et = c to − c to∗ = [ex ey ez ]T and eω = uθ, u being the axis and θ the angle of
the rotation matrix cRc∗ .
b
s e c
vc vp l qm

L d

1=r! CDPR
s Camera
+ image
s Computer Vision
algorithm
Fig. 2. Control scheme for pose-based visual servoing of a CDPR
To decrease the error e, an exponential decoupled form is selected ė = −λe

with a positive adaptive gain λ, that is computed at each iteration, depending
on the current ||e||2 [7].
The relationship between ė and the Cartesian velocity of the camera c vc ,
expressed in Fc , is defined as:
ė = Ls c vc (6)
where Ls is the interaction matrix and it is defined in [8]:

Finally, the instantaneous velocity of the camera in its own frame is expressed
as a function of the pose error as follows:
c
vc = −λ L−1
s e (7)
where L−1 s is the inverse of the estimation of the interaction matrix Ls . Note
that the inverse is directly used, because for PBVS Ls is a (6 × 6)–matrix that
is of full rank [8].
2.3 Kinematics and Vision

To combine the modeling shown in Sects. 2.1 and 2.2, the MP twist b vp is
expressed as a function of camera velocity c vc :
b
vp = Ad c vc (8)
where Ad is the adjoint matrix that takes the following form [11]:
b
Rc b
tc b
Rc
Ad = × (9)
03 b
Rc
where b tc = bRp p tc and p tc is the vector pointing from Op to Oc .

5
In this paper, a superscript ∗ denotes the desired value, e.g. desired object pose s∗
∗
and c∗ in c to∗ refers to desired camera frame Fc∗ in which the object is in the
desired pose s∗
3 Stability Condition
One of the main characteristics of any system is its stability. It is a measure to

assess the effects of estimation quality. That is, how coarse can the estimation
be for the system to still converge to its goal [12]. Lyapunov analysis is used to
determine the stability of the closed-loop visual servoing system.
From Eqs. (4), (6) and (8) the model is the following:
ė = Ls A−1 †
d A l̇ (10)
Upon injecting (8) and (7) into (4), the output of the control scheme, that
is, the cable velocity vector takes the form:
l̇ = −λ A Ad L−1
s e (11)
where A and Ad are the estimations of A and Ad , respectively.

The following closed-loop equation is obtained from (10) and (11):
ė = −λ Ls A−1 † −1
d A A Ad Ls e (12)
From (12), the system stability criterion is defined as:
Π = Ls A−1 † −1
d A A Ad Ls > 0, ∀t (13)
Π > 0 is a sufficient condition to obtain global asymptotic stability (GAS).

It can be seen from the closed-loop equation (12) that if Π is positive definite,
then the control scheme will ensure an exponential convergence of the error e
to 0. However, if it is negative then the error e will increase and the system may
diverge from the goal. Indeed, (13) is only a sufficient condition, therefore the
stability of the system is uncertain once the condition is not held.
3.1 Estimated Parameters
Given the closed loop equation (12), the following variables are estimated and
can therefore affect the system stability:
– ŝ – object pose in Fc is computed from image features, so it will not be the

exact pose s
– p Tc – the pose of the camera in the MP frame Fp . We have an idealistic
model, however due to manufacturing imprecisions, the actual camera pose
will be a little bit different from the modeled pose.
– p Bi – the Cartesian coordinates of cable anchor points to the MP, expressed
in Fp . Due to mechanical solution of the anchor points, the actual point is
a point located on a sphere around the nominal point.
– b Ai – the Cartesian coordinates of cable exit points, expressed in the base
frame Fb . We are using the simplified CDPR model, that does not include
pulleys actually located at the exit points.
78 Z. Zake et al.
pulley
base
cable
moving platform
camera
prilTags
(a) (b)
Fig. 3. CDPR prototype: (a) ACROBOT; (b) V-REP model of ACROBOT
– b Tp , b Tp is estimated by exponential mapping:

c
(b Tp )t+Δt = (b Tc )t+Δt c Tp = (b Tc )t exp( vc ,Δt) c
Tp (14)
Since velocity c vc , necessary for exponential mapping, is computed from the

object pose measurement ŝ, which we admit to be different from s, then
b
Tp = b Tp . Furthermore, initial b Tp is only coarsely known6 .
4 Case Study
The stability condition (13) is applicable to any CDPR with a PBVS control.
However, this analysis includes model-related parameters, thus it should be done
for each CPDR separately. Hence, in the following sections the chosen CDPR is
presented and results of stability analysis are shown.
4.1 ACROBOT and Simulation in V-REP

ACROBOT For this paper, a CDPR prototype, named ACROBOT and shown
in Fig. 3(a), is used. Its WS is a 1 m3 cube. The robot is assembled in a suspended
configuration. A simple webcam AUTOPIX MT4018 is mounted on the MP in
the eye-in-hand configuration facing the ground.
To simplify the computer vision part, AprilTags [14] are used instead of
real objects. They are especially convenient to use in combination with ViSP
library [15], because the latter contains functions that allow to recognize the
tags and retrieve their 3D pose.
6
For this CDPR, the initial pose was defined at the center of the WS, which itself was
measured by hand with a measurement error of ±2 cm along X and Y axes, resp.
Fig. 4. The representation of a cable and its pulley in V-REP
V-REP Model The V-REP simulation environment [16] was used in order to
create a dynamic model of ACROBOT including the vision sensor. This gives
us the capacity to use the same software to control the real and the virtual
hardware. As a result we can speed up the development time, test and debug
our algorithms in simulation (with a perfectly known ground truth), then use
the real robot for final verifications.
To create a dynamical simulation, the pulleys and cables are modeled as a
sequence of joints and mass objects, as shown in Fig 4. The current model does
not take into account the pulley diameter and the cable sag. The pulleys are
represented as a vertical revolute passive joint followed by a small spherical mass
and a horizontal revolute passive joint. The cables are modeled as a sequence
of prismatic joint, cylindrical mass, prismatic joint, cylindrical mass and a final
spherical joint attaches the cable to the MP. The first prismatic joint is used to
change the cable length. The second prismatic joint is responsible for the cable
behavior through a specific joint control callback script, which models the cable
forces as either an elastic spring, when in tension, or an element transmitting
zero force, when in compression.
To have a stable simulation some model design rules need to be consid-
ered [17]. In our tests, the Vortex physical engine was used.
4.2 Numerical Analysis
It is not possible to express analytically the pseudo-inverse of the Jacobian A† ,

thus stability analysis is only possible in numerical form. In this paper, we will
only study two parameters, namely p Tc and bAi .
The Cartesian coordinates of bAi are known, we need to find out the range of
perturbation that does not destabilize the system. In Fig. 5(a) the perturbation
range is defined by the radius rAi of the sphere centered at point Ai . We want to
find the maximum value rAi,max so that for any point within a sphere of radius
rAi ≤ rAi,max the system is stable.
Unlike cable exit points, camera pose in Fp could in fact be changed, if
necessary. Therefore, here we first define the range, where the camera could
80 Z. Zake et al.
Rpc
i ri
Op
rpc
(a) (b)
Fig. 5. Perturbation ranges for: (a) cable exit points Ai ; (b) camera pose in Fp
be positioned, which is defined in Fig. 5(b) by radius Rpc of the largest sphere
around origin Op of Fp . Then for any camera position, the perturbation range
is defined by the small sphere radius rpc . Our goal is to find rpc,max so that if
rpc ≤ rpc,max and camera position is within the sphere defined by radius Rpc ,
then the system will be stable.
Finally, we transform pRc into axis-angle representation θpc u. We define the
perturbation as δθpc , that can be made about any unit vector uδ . Therefore, our
goal is to find δθpc,max so that |δθpc | ≤ δθpc,max for any uδ , u and θpc .
To find the range of perturbation, we define the system as stated in Table 1.
Then we portray two distinct cases: (i) The system is assumed to be ideal,
no perturbations other than the one we are studying; (ii) the system has small
perturbations in all parameters. The perturbation values are chosen either based
on the mechanical errors (such as pBi and bAi ) or as 5 to 10 percent of the actual
parameter value. For the latter, we also distinguish the results depending on the
desired range of motion of the MP in Fb expressed as Rbp .
The perturbation limits that we have established based on condition (13) are
shown in Table 2. Here are the observations:
– None of the parameters are affected by the actual value of p tc or pRc . This
is especially surprising for the perturbations on these parameters defined by
rpc and δθpc .
– To keep the full motion range Rbp and the full rotational range bRp , pertur-
bations on bAi must not be larger than 0.01 m.
– In the ideal case, the perturbation values for camera pose in Fp are large. In
fact these perturbations do not appear separately. Indeed, all the considered
perturbations are affecting the system at the same time, not independently.
For this reason it was important to do the second part of this study while
considering perturbations in all parameters.
– When the perturbations, defined in Table 1, are added to the system, the
latter remains stable for all tested motion range of MP.
– However, once the MP motion range is reduced, it is possible to increase the
perturbation levels without making the system unstable.
– As soon as Rbp is reduced, rpc increases significantly, as well as δθpc . This
allows to conclude that perturbations in the respective variables have little
effect on the stability of the system.
– There is a considerable increase in rAi as well. This means that the modeling
accuracy of exit points is not important. Even at WS borders the tolerated
perturbation easily covers the corresponding model inaccuracies. That is,
even though pulleys that are present on the cable exit points are not modeled,
the small changes in the actual exit point location on the pulley are covered
by the perturbation rAi = 0.01 m and do not affect the system stability.
Table 1. Variable and perturbation ranges
variable family variable value perturbation value
Rst = 0.5 m rst = 0.05 m

object pose in Fc
◦
|θω | ≤ 50 |δθω | ≤ 2.5◦
rpc = 0.01 m
camera pose in Fp −0.032 −0.026 −0.01 180◦ 0◦ 180◦
δθpc = 3◦
cable anchor In eight corners of MP of
rBi = 0.008 m
points pBi size 0.1 × 0.1 × 0.05 m
cable exit Two exit points at each top corner
rAi = 0.01 m
points bAi of ACROBOT
Rbp is equal to 0.5 m, 0.3 m and 0.1 m, resp. rbp = 0.02 m
MP pose in Fb Rotation about global axes: 45◦ about Z, 5◦ about Z axis, 3◦
20◦ about Y, 20◦ about X about Y and X axes
Table 2. Perturbation change depending on MP motion range
Condition Camera pose in Fp Cable exit points bAi

Ideal robot, no other perturba-
rpc = 0.5 m |δθpc | ≤ 55◦ rAi = 0.01 m
tion in the system, Rbp = 0.5 m
Minimal perturbations from
rpc = 0.03 m |δθpc | ≤ 3◦ rAi = 0.01 m
Table 1, Rbp = 0.5 m
rpc = 0.5 m |δθpc | ≤ 16◦ rAi = 0.09 m
rpc = 0.8 m |δθpc | ≤ 24◦ rAi = 0.18 m
4.3 Experimental Validation

Some experiments have been conducted to validate the theoretical results 7 . First,
we assume that the rotational error of the camera in Fp and about Z axis is out
7
Please also see the accompanying video at https://youtu.be/tfiTDlp1ZIY
82 Z. Zake et al.
400 0.6
0.4
300
0.2
200 0
-0.2
100
-0.4
0 -0.6
0 200 400 600 0 10 20 30 40
(a) (b)
0.8
400 0.6
300 0.4
200 0.2
100 0
0 -0.2
0 200 400 600 0 10 20 30
(c) (d)
0.8
400 0.6
300 0.4
200 0.2
100 0
0 -0.2
0 200 400 600 0 2 4 6
(e) (f )
0.8
400 0.6
300 0.4
200 0.2
100 0
0 -0.2
0 200 400 600 0 5 10
(g) (h)
Fig. 6. CDPR behavior depending on added perturbations. (a) and (b): the AprilTag
trajectory and error e over time in V-REP simulation with δθ = 85◦ ; system is not
stable (only center-point trajectory shown). (c) and (d): the AprilTag trajectory and
error e over time in V-REP with δθ = 55◦ ; system is stable; (e) and (f ): the AprilTag
trajectory and error e over time on ACROBOT with δθ = 55◦ ; robot does not converge
because AprilTag leaves the camera field of view. (g) and (h): the AprilTag trajectory
and error e over time on ACROBOT with δθ = 16◦ ; system is stable.
of bounds of stability, i.e. δθpc = 85◦ . As shown in Fig. 6(a) the robot diverges
from its goal position and error oscillates and slowly increases (Fig. 6(b)), the
system is not stable.
If δθpc = 55◦ , the V-REP model will successfully reach the target as shown
in Fig. 6(c), though the trajectory is far from optimal (a straight line). Some
error oscillations can be observed in Fig. 6(d). The same rotational error on the
actual robot is shown in Fig. 6(e). The initial behavior is similar, but AprilTag
leaves the image and the task is failed. This is not surprising, given the additional
uncertainties of the actual robot and the lower image quality.
Finally, δθpc was set to its maximum for a noisy system with range of motion
reduced to Rbp = 0.3 m, i.e. δθpc = 16◦ . Once implemented on ACROBOT, we
see that though the trajectory is perturbed, the MP reaches its targeted pose
(Fig. 6(g)). Indeed, error e converges to zero without oscillations (Fig. 6(h)).
To summarize, for an ideal robot the range of perturbation on a single param-
eter is very large. On the contrary, as soon as we acknowledge that all the system
parameters are noisy, then each individual perturbation has quite a limited range
within the bounds of stability.
5 Conclusions
This paper proposed a method to analyze the stability of Pose-Based Visual Ser-
voing (PBVS) control of Cable-Driven Parallel Robots (CDPRs). A general sta-
bility criterion was introduced. The stability of ACROBOT, a CDPR prototype
located at IRT Jules Verne, was analyzed. Two CDPR model-related parameters
were studied and their maximum perturbation range was found both for ideal
robot and for a noisy one. A dynamic CDPR model in V-REP was presented.
It was found that for an ideal system any one parameter could be highly
perturbed without making the system unstable. This result was successfully val-
idated in simulation. When adding a large perturbation (validated in simulation)
to ACROBOT, the system does not converge. However, if the perturbation is
kept within the corresponding (noisy system) range, the robot will be able to
complete its task.
For ACROBOT, the tolerated perturbation on cable exit points is rather
large. This is beneficial, because it allows to avoid adding pulley kinematics to
the model. Indeed, cable exit point variations on the pulley are smaller than the
tolerated perturbation, which does not affect the stability of the system.
The added perturbations affect the trajectory to the goal and the system’s
ability to actually reach it. However, perturbations in the robot model do not
affect the final accuracy. This is because the final pose is considered to be reached
only when the error between the current and the desired object pose becomes
smaller than the threshold defined in the visual servoing loop.
Future work will deal with the implementation of PBVS control on a large
semi-industrial CDPR prototype as well as its stability analysis. It is also of
interest to consider real objects instead of AprilTags to validate this approach
for more realistic use cases.
84 Z. Zake et al.
Acknowledgment This work is supported by IRT Jules Verne (French Institute

in Research and Technology in Advanced Manufacturing) in the framework of
the PERFORM project.
References
1. L. Gagliardini, S. Caro, M. Gouttefarde, and A. Girin, “Discrete Reconfiguration
Planning for Cable-Driven Parallel Robots”, in Mechanism and Machine Theory,
vol. 100, pp. 313–337, 2016.
2. V. L. Schmidt, “Modeling Techniques and Reliable Real-Time Implementation of
Kinematics for Cable-Driven Parallel Robots using Polymer Fiber Cables”, Ph.D.
dissertation, Fraunhofer Verlag, Stuttgart, Germany, 2017.
3. E. Picard, S. Caro, F. Claveau, and F. Plestan, “Pulleys and Force Sensors Influ-
ence on Payload Estimation of Cable-Driven Parallel Robots”, in IROS, Madrid,
Spain, October, 1–5 2018.
4. A. Fortin-Côté, P. Cardou, A. Campeau-Lecours, “Improving Cable-Driven Paral-
lel Robot Accuracy Through Angular Position Sensors”, in IROS, pp. 4350–4355,
2016.
5. T. Dallej, M. Gouttefarde, N. Andreff, R.Dahmouche, and P. Martinet, “Vision-
Based Modeling and Control of Large-Dimension Cable-Driven Parallel Robots”,
in IROS, pp. 1581–1586, 2012.
6. R. Chellal, L. Cuvillon, and E. Laroche, “A Kinematic Vision-Based Position Con-
trol of a 6-DoF Cable-Driven Parallel Robot, in Cable-Driven Parallel Robots,
pp. 213–225, Springer, Cham, 2015.
7. R. Ramadour, F. Chaumette, and J.-P. Merlet, “Grasping Objects With a Cable-
Driven Parallel Robot Designed for Transfer Operation by Visual Servoing”, in
ICRA, pp. 4463–4468, IEEE, 2014.
8. F. Chaumette, S. Hutchinson, “Visual servo Control I. Basic Approaches”, in IEEE
Robotcs & Automation Magazine, vol. 13, no. 4, pp. 82–90, 2006.
9. Z. Zake, F. Chaumette, N. Pedemonte, S. Caro, “Vision-based Control and Stabil-
ity analysis of a Cable-Driven Parallel Robot”, submitted to ICRA 2019.
10. A. Pott, “Cable-Driven Parallel Robots: Theory and Application”, vol. 120.,
Springer, 2018, pp. 52–56.
11. W. Khalil, E. Dombre, “Modeling, Identification and Control of Robots”,
Butterworth-Heinemann, 2004, pp. 13–29.
12. H. K. Khalil, Nonlinear systems 2nd ed., Macmillan publishing Co., New York 1996.
13. C.R. Johnson, “Positive definite matrices”, in The American Mathematical
Monthly, vol. 77, no. 3, pp. 259–264, 1970.
14. E. Olson, “AprilTag: A robust and flexible visual fiducial system”, in ICRA,
pp. 3400–3407, IEEE, 2011.
15. É. Marchand, F. Spindler, F. Chaumette, “ViSP for visual servoing: a generic
software platform with a wide class of robot control skills”, in IEEE Robotics &
Automation Magazine, vol. 12, no. 4, pp. 40–52.
16. E. Rohmer, S. P. Singh, M. Freese, “V-REP: A versatile and scalable robot simu-
lation framework”, in IROS, pp. 1321–1326, 2013.
17. V-REP, Designing Dynamical Simulations. http://www.coppeliarobotics.com/
helpFiles/en/designingDynamicSimulations.htm
Practical Stability of Under-Constrained Cable-
Suspended Parallel Robots
Dragoljub Surdilovic and Jelena Radojicic
Fraunhofer Institute for Production Systems and Design Technology IPK-Berlin, Department
Robotics and Automation, Pascalstr. 8-9, 10587 Berlin, Germany, E-mail:
[email protected]
Abstract This paper, motivated by the development of a novel gait rehabilitation

system, presents a mechanical approach for the dynamic modelling and analysis of
equilibrium stability of under-constrained cable suspended parallel robots. These
types of cable robots exhibit interesting characteristics of self-motion in the Jaco-
bian null-space. Modelling and understanding of this motion is essential for their
applications. It is demonstrated that both a wrench consistency test and proof of
stability conditions, derived for real robots with a pulley mechanism, play a cru-
cial role for practical equilibrium stability assessment. Thereby dynamic simula-
tion of the null-space motion help to analyse robustness of the equilibrium against
perturbations. Several examples with a 4-4 type robots illustrate the theoretical
analysis.
Key Words: cable-driven parallel robots, under-constrained cable suspended

structures, equilibrium stability analysis, gait rehabilitation
1. Introduction
A specific class of cable-driven parallel robots (CDPR) with all cables located
above the common platform, referred to as force-supported or cable-suspended
parallel robots (CSPR), has been recently addressed in numerous researches [1-2]
focusing on their advantages for implementing simple, large span, fast moving,
lightweight and heavy-duty active overhead mechanisms. A major problem with
cable-suspended structures is ensuring tension forces for any admissible motion,
as well as realizing a stable static equilibrium (rest poses) and a desired rest to rest
motion. It is well recognized that gravitation force of a cable platform and payload
plays an essential role in stabilizing CSPR, acting as an additional virtual wire that
ensures cables tension. In contacts with environment, however, the external pro-
cess/contact forces become also relevant and may jeopardize stability and cause
undesired, even uncontrollable motion. The contact applications of CSPR have
been not intensively investigated in the past, despite some attractive presentations
such as grinding of large objects with NIST RoboCrane [3], or transportation and
assembly tasks performed by cooperative aerial towing [4-5] (CSPR with cables
attached to aerial robots). This paper deals with structurally under-constrained
CSPRs, where the platform is suspended with less than 6 cables which are often
applied in practice [1, 2]. It is worth mentioning that CSPR with 6 or more cables
86 D. Surdilovic and J. Radojicic
can also become under-constrained, when the Cartesian wrench-space spanned by

taut cables become less than 6 [2].
The stability questions concerning mechanical equilibrium of under-
constrained CSPRs have been well investigated in various systems using numer-
ous approaches, such as analysis of net forces and torques, momentum, virtual
displacements and velocities, or potential energy gradient. The stability of a
pose/configuration have been mainly investigated in the context of solving direct
and inverse kinematic problems. In order to determine a feasible configuration,
without cable slack or overloaded cables (aerial robots), the kinematic analysis,
typically for general CDPR, cannot be considered independently of the system dy-
namic, i.e. quasi-static analysis. Consequently these problems have been referred
to as geometric(o)-static analysis [6]. In general case with fixed cable length, rela-
tive motion of an under-constrained CSPR platform (with less than 6 taut cables)
around a possible stable equilibrium remains still possible, under action of exter-
nal perturbation forces. Thereby the robot can switch to other equilibrium in an
unpredictable way. This motion has been referred to as self-motion in singular
configuration of parallel robots [7], however it is also characteristic for under-
constrained CSPRs in which all Cartesian DOFs cannot be controlled. A typical
example is so called 2-2 CSPR configuration, in which by analogy with the cou-
pler of an overhead four-bar planar mechanism [4, 6], the platform can move in
the working plane with fixed cables length. Moreover, the “cranks” of this CSPR
are not rigid, the cables may become slack or the platform can move outside the
plane, which is likely to occur in practice.
Several researches [6, 8-9] have tried to analyse these effects in under-
constrained CSPR by solving the direct kinematic problem for different standard
structures with 2 to 5 cables, computing all solutions for given parameter sets and
analysing their stability for a given payload. The background for stability analysis
of under-constrained CSPRs has been established by Carricato and Merlet [6]
based on static forces virtual works around an equilibrium (i.e. pose candidate). In
[4] an energetic approach has been applied considering that an equilibrium pose
corresponds to the minimum in the potential energy. The Hessian of payload po-
tential energy expressed in terms of independent angular rotations around equilib-
rium has been used to derive complex analytical expressions and analyse stability
(eigenvalues of Hessian matrix). Collard and Cardou [10] have pursued a similar
approach trying to compute lowest equilibrium pose of CSPR by minimizing po-
tential energy subject to complex algebraic constraints.
In previous work the wire-robot motion effects occurring in commonly applied
pulley elements (e.g. due to pulley rolling and wire coil) have been neglected [12].
In this paper we will derive dynamic equilibrium models of a general under-
actuated CSPR system, taking also into account the pulley motion effects, and ap-
ply them for stability assessment and analysis of robustness against perturbations.
A novel algorithm has been proposed for efficient numerical finding of potential
equilibrium configurations and assessing their stability. The stability of under-
constrained CSPR has been considered in the domain of practical stability (notion
introduced by LaSalle and Lefschetz [13]) of dynamical systems, providing a
Practical Stability of Under-Constrained Cable-Suspended Parallel Robots 87
practical framework for assessing stability and robustness of an equilibrium

against perturbations. This work is motivated by a further development of CSPR
technology for gait rehabilitation by simplification an over-constrained system
STRINGMAN [14] to a CSPR structure with cable implemented by pneumatic ar-
tificial muscles (Fig. 1), which provide an intrinsic safety to patient. Analysis and
understanding of possible system motion and equilibrium stability require com-
plete dynamic models, including human dynamics (MATMAN model developed
in Matlab with 40 DOFs using bio-mechanical human data), that will be in the pa-
per simplified by considering only possible trunk motion.
Fig. 1 4-4 CSPR with pneumatic artificial muscles as wires
Fig. 2 MATMAN model (left) and simplified configuration used in the paper
2. Kinematic Analysis
A model of under-constrained CSPR with n-wires ( n 6 ) is given in (Fig. 3).
The position of i-th wire platform attachment point B i is defined by
pi a i Li p bi (1)
where constant vector a i defines “attachment” Ai of pulley (the bearing block of
the pulley is mounted around fixed cable axis e i , in front of the pulley), b i is
platform attachment Bi position vector wrt local platform frame, p is the position
vector of the platform reference frame and Li Ai Ci CiTi l i , where l i is the
effective wire-length vector, while Ci and Ti denote centre of the pulley and actual
wire tangent points (instant center of rotation of cable l i , i.e. point Bi) (Fig. 3).
Fig. 3 Cable-suspended robot structure

In general the wire performs a complex composite motion that can be decom-
posed into transferred motion, representing rotation (rolling motion of the pulley)
of the entire wire plane {AiBiCiTi} around the fixed pulley axis ei, and relative mo-
tion in the wire plane (Fig. 4). The relative motion consists of relative translation
i.e. change of the relative length in the actual cable direction (due to cable control,
i.e. via a winch that are not presented in Fig. 3-4), and a relative rotation of the
wire around the pulley (i.e. instantaneous center of rotation Ti that changes its po-
sition during winding). Differentiating (1) twice with respect to time the expres-
sions, absolute velocities and accelerations of wire end-point (Bi) are obtained
vi p i ωe i u Li ω r i u l i l i v p ω p u bi (2)
ai i
p ε e i u Li ε r i u l i ω e i u (ω e i u L i ) ω e i u (ω r i u l i ) ω r i u (ω r i u l i )
2(ω e i ω r i ) u l i l i a p ε p u b i ω p u (ω p u b i )
(3)
where ω e i and ε e i denote pulley rotation velocity and acceleration around e i
(Fig. 4), ω r i and ε r i relative wire rotation velocity and acceleration around wire-
plane normal n i , l i and l i are linear wire relative velocity and acceleration due
to cable length changes, v p and ω p , a p and ε p are platform linear and angular
velocities and acceleration vectors respectively. Relative velocity components and
their directions are shown in (Fig. 4). The projections of the velocity and accelera-
tion vectors (3, 4) into wire-length vector direction, defined by unit vector
l i 0 l i li , i.e. scalar multiplication of these equations by l i 0 yields the magni-
tudes of wire linear relative velocity and acceleration respectively
l i0 bi ]t p
T T
li [l i 0 (4)
& & & & & &

Ze 2 [( ei u L i ) n i ][( ei u n i ) li 0 ] Zr 2 li [l i 0 T l i 0 b i ] t p l i 0 ω p b i ω p
T T
li i i
(5)
x p ω p ]T is the platform twist vector, where xp denotes plat-

T T
where t p [v p
form pose (both position and orientation), while b i indicates the skew-symmetric
3x3 matrix formed from the elements of the vector b i in order to represent the
vector product in the matrix form. The first two components of the right-hand side
of (5) represent the projections of centrifugal accelerations components (corre-
sponding to the pulley and relative wire rotations), while the remaining parts de-
fine projections of platform tangential and centrifugal accelerations into the wire
directions. The expressions for vectors of angular pulley and wire relative rota-
tions in terms of platform twist vector, are obtained by scalar multiplication of (2,
3) by vectors n l n i u l i 0 and n i respectively
i
Ze
1 ªn T n i b i ºt p ; Zr
1
& & & [l i 0 n i
T T T
l i 0 n i b i ]t p (6)
i
(ei u L i ) n i «¬ i »¼ i
li
Fig. 4 Velocity vectors components (in wire-pulley plane)

The relationship between relative wire velocity, defining the cable length varia-
tions, and platform twist vector is defined by the wire-robot Jacobian J nx 6
ª l10 l i 0 l n 0 º
l Jt p ; JT «b l » (7)
«¬ 1 10 b i l i 0 b n l n 0 »¼
where l [l 1 l i l n ]T . The time derivative of CSPR Jacobian is obtained by dif-

ferentiating (7)
ª l l l º
Jt p J t p ; J T
10 i0 n0
l « l l b l l b l » (8)
b l
«¬ 1 10 b b b n n0 »
1 10 i i0 i i0 n n0
¼
where the time derivatives of the constant intensity vectors l i 0 and b i (consider-
ing an ideal rigid platform and cables) are
l ω ei u l i 0 ω ri u l i 0 ; b i ω p u bi (9)
i0
and substituting (6) and (9) yields
J t p J tp [J1 Ji J n ]T
T T T T T
(10)
where J is a n(6)x1(6) block matrix (numbers outside parenthesis define block

matrix dimension, while within parenthesis the dimension of each block-matrix
element has been given), is the Kronecker’s product (each block-element of J
T
is multiplied by t p ), and the block element Ji 6 x 6 has the form
ª 1 1 1 1 º
« l n i l i0 l i0 n i & & n i l i0 l i0 n i b i & &
T T T T
& n i l i0 e i & n i l i0 e i b i » (11)
« ( e u L i ) ni
li (e i u L i ) n i »
Ji i i
« 1 b n l l T n 1 1
b i n i l i0 l i0 n i b i & &
1 »
& & b i n i l i0 e i b i b i l i0 »
T T T
« l i i i 0 i 0 i (e& u L ) n& i i i 0 i
bnl e
¬ i i i i
li (e i u L i ) n i ¼
3. Stability of equilibrium
The equilibrium between internal cable tension forces, grouped in the wire tension
vector f [f1 f i f n ]T with elemental forces acting along l i 0 (i=1,…,n), and
Cartesian wrench w [FT MT ]T , involving gravity and external net static
force/torque acting on platform frame, is defined by the balance of virtual works
of these forces on virtual displacements compatible with the constraints around an
equilibrium
GL G l T f G xT w 0 (12)
> T T
@
where G x G pT G o , Gp and Go are possible relative translational and ro-
tational virtual displacements of platforms around an equilibrium that are compat-
ible with the constraints, G l >G li @ . G L represents variation of a Lagrange func-
tion, where the first term characterizes the work of internal cable forces
(representing Lagrange constraints multipliers), and the second one is variation of
potential energy (virtual work of gravitation forces and torques). Substituting
G l JG x p in (12), yields the known Jacobian mapping between internal cable
forces and corresponding external wrench, defining a static equilibrium condition
w JT f (13)
The second order variation of L is given by
G 2 L G 2 lT f G 2 xTp w; G 2l JG 2 x p G JG x p (14)
Where the variation of Jacobian based on (10) may be written in the form
GJ G xTp J G x pT [J1 Ji J n ]T

T T T
(15)
Substituting (13) and (15) in (14) yields
G 2L G xTp JT G x p f G 2 xTp J T f G 2 xTp w G xTp JT G xp f G xTp JT $ f G x p

(16)
where the product “ $ ” defines the multiplication between block-vectors and ma-
trices applying standard rules on block-elements, i.e. the multiplication of raw
block-vector JT and f yields
n
JT $ f ¦ Ji f i
T
(17)
i 1
and J i is given in (11) for a general CSPR with pulleys. Finally
G 2 L GxTp HG x p (18)
with 6x6 Hessian matrix

n
JT $ f ¦ Ji f i
T
H (19)
i 1
Orthogonal projection of (18) into null-space of robot Jacobian N null J pro-

vides symmetric reduced Hessian [6]
NT HN NT §¨ ¦ J i f i ·¸N
n
T
HN (20)
©i 1 ¹
A sufficient condition for a stable equilibrium of CSPR consists in H N being pos-
itive definite, i.e. having all positive eigenvalues Oi (i=1,…,6-n).
4. Practical tasks and stability

The stability assessment based on (20) requires determination of equilibrium
configurations for known cable-lengths li and static external wrench w (e.g. given
payload w G in the free space), satisfying equilibrium condition (13). However,
this is not a trivial task. This problem, referred to as direct geometrico-static prob-
lem DGP, has been intensively investigated in [6, 8-9] etc., specifically for 4-4
CSPR robots in [15]. The inverse geometrico-static problem IGP assumes known
platform poses [6]. Thereby mostly analytic approaches based on exact arithmetic
elimination procedure, by analogy with overhead rigid parallel robot structures, al-
so using the same SW packages, have been applied to find distinct real solutions
of high-order polynomials (with degrees 12, 156, 216 and 140, respectively, for
two, three, four and five taut cables) obtained from algebraic constraints equa-
tions. However, finding all solutions of direct DGP is computationally expensive
and difficult for real-time applications (though assessing the stability may be
compatible with RT requirements). Recent improvements [9, 11] have focused on
reduction of computational burdens.The problem in real CSPR with pulleys be-

come considerably more complex and cannot be solved in closed form.
In the following, alternative numeric approaches with additional tasks and sta-
bility assessment methods, based on dynamic model and above analysis, are pro-
posed. As common in dynamic systems we can obtain a static equilibrium condi-
tion in a considered pose from the dynamic model by nullifying the motion terms.
Assuming for the case of simplicity that the platform-reference frame and cen-
ter of mass coincide P=C, the dynamic model of a CSPR may be written as
Mt p t Tp Ctp w G J T f (21)
ªm p E 0 º Cv ½ ª0 0 º
« 0 ® ¾
I p »¼ «0 I »
M C Cv 06 x 6 CZ
¬ ¯CZ ¿ ¬ p¼
E and 0 denote 3x3 unity and zero matrices respectively, mp and Ip are platform
mass and inertia, C is a 2(6)x1(6) block vector of centrifugal forces and torques,
T
and t p ª v T ωT º is a 6x3 matrix composed of skew-symmetric velocities.
«¬ p p»
¼
Nullifying twist vectors t p t p 0 in (25) yields
wG JT f (22)
This condition (22) in an arbitrary pose x p and for a given static wrench w G
(IGP), different from articulated robots, or fully- and over-constrained CDPRs, in
general cannot be fulfilled. Therefore x p will be referred to as quasi-equilibrium.
For the considered under-constrained CSPRs, the above mapping defines an over-
determined system of linear equations where J T 6 xn and n<6. It is well known
that in general case, for an arbitrary w , this system of equation cannot be satisfied,
and has no solution f , unless w falls in the span of J T column space (so called
consistent case).
The consistency of a known wrench (in general case including also static exter-
nal forces) may be proven by projection into null-space of J and testing if
NT w 0 (23)
where 0 is zero (6-n)x1 vector, has been satisfied. The gravitation wrench w G
provides a consistent solution (real-equilibrium) only in specific, often obvious
symmetric configurations (when all cables and gravitation vector cross in a point).
An arbitrary w , including also external contact steady-state forces and torques, is
in general inconsistent for the mapping (23). Approximate solutions based on min-
imizing J T f w , such as least-squares (LS) optimal solution often applied in
practice.
fˆ J T # w
1
JJ T Jw (24)
provides the projection onto column space of J T
J T fˆ
1
w proj J T JJ T Jw z w; NT w z 0 (25)
The non-zero projection NT w causes the platform motion in the null-space de-
spite fixed cable-lengths until a real equilibrium satisfying (13) has been reached.
Therefore a new task for under-constrained CSPR referred to as pose-stabilization
in a given quasi-equilibrium pose x p will be defined. It consists of computing (6-
n) modified, i.e. additional 'w wrench components, keeping the remaining ele-
ments of the initial (gravitation, external static) wrench, thus that entire wrench
become consistent in x p for the mapping (13), ensuring an equilibrium like in
other robotic systems. This will computed by solving
NT w 'w NT w cons 0 (26)
Which 6-n components of w have to be updated, is based on analysis of the

null-space matrix N elements. In an ideal case the additional wrench components
should act as much as possible in the null-space (elements of N close to 1). After
determination of w cons and computation of cable forces from (13, 22), the stability
of the equilibrium with the new wrench can be evaluated based on (20).
However, an open problem is how to realize this modified wrench w cons in
practice. Innovative solutions, using additional platform actuators, such as propul-
sion fan forces, as proposed in [16], provide feasible approaches and require fur-
ther investigations. In the crane systems for assembly of prefabricated modules in
construction, with similar suspended fixed-cable structures, often additional stabi-
lizing forces/ torques are realized by human workers by pulling ropes attached to
the platform (module). The stabilization tasks may be used to estimate human
stress during stabilization in a desired assembly pose. In the considered CSPR
case (Fig. 1-2), these external wrench components may be realized by patient own
efforts trying to stabilize posture. Intermittent perturbation forces, causing the new
wrench to become inconsistent with an initial equilibrium pose can move the robot
from one to other equilibrium poses without changing active cables length.
In real systems for known (measured) fixed cables length only, we cannot solve
the direct kinematic problem exactly, since it is not unique in under-actuated
CSPRs. Based on cable-force measurements also external wrench cannot be com-
pletely estimated, only its projection (25). The information about actual uncontrol-
lable and unobservable (in cable space) platform position can be only resolved by
means of additional external sensors (e.g. mechanization of attached IMUs). Then
the inverse kinematic may be computed (providing actual effective cables length li
and pulley winding, i.e. actual Ti). For a known payload w G , in the next step the
consistency condition (22) has to be proven, and if it is satisfied the actual pose
x p indeed represents a real-equilibrium. Its stability can be assessed by compu-

ting eigenvalues of H N in real time (20).
In simulation and design experiments, the following numerical algorithms ap-
pears to be practicable. When the cables are fixed (locked drives) and cannot
change their lengths, the motion around a quasi-equilibrium is constrained by
**
l Jt p 0; l Jt p J t p 0 (27)
Equations (26, 30-32) can be now assembled into single matrix equation
ªM J T º t p ½ T
°t p C t p w G JT f ½
°
« »® ¾ ® ¾ (28)
¬J 0 ¼ ¯ λ ¿ °̄ Jt p °
¿
where λ is nx1 Lagrange constraint multiplayer vector, representing virtual cable
forces that prevent the motion in the cable direction and keep the active cable
length constant. Computation of the constrained dynamic model (28) is not trivial,
due to numerical instability caused by differentiation of constraints (27) (con-
straint violation problem). Therefore methods for stabilization of solutions [17]
have to be applied. The order of the model (28) can be reduced by projection into
null-space and simulation of motion in the null space until right side becomes ze-
ro.
NT MNt pN NT t Tp CNt pN N T w G (29)
These methods help to find a real static equilibrium x peq starting from an initial
“quasi-stable” pose x p . However, dynamic stability remains to represent a prob-
lem and the system is prone to oscillations around the equilibrium. Transition into
another equilibrium in the neighborhood is also thereby possible Additional dissi-
pative wrench Dt p (D is damping matrix) has to be introduced in the models (28,
29) to bring the system into steady state. However, the open question is again how
to realize dissipation effects in the null-space of real system, since the wire (con-
trol) forces, acting in the column-space of J T have no effects on the null-space
motions.
The presented numerical approach can be efficiently applied to find real equi-
librium solutions. Moreover, it provides the possibility to analyze dynamic stabil-
ity, of an equilibrium including robustness to intermittent perturbations w pert and
required possible external control actions to stabilize CSPR in equilibrium pose. If
we assume that perturbations are bounded w pert d G , and define deviation fields
Q around a stable equilibrium x peq , and a set Q0 of initial configurations x p 0 then
the following definition of practical stability [18] applied to CSPRs is:
Definition 1: If every solution x p of (28, 29) satisfies x p (x p 0 , t0 , t ) Q for

x p 0 Q 0 with w pert d G , then the equilibrium solution ( x peq , t p tp 0 ) to
(28, 29) is called practically stable with respect to G , Q0 and Q.
The notion of practical stability is relevant to number G and sets Q0 and Q.
For a practical CSPR, Q is the permissible set and Q0 is initial state set that should
be specified. The proposed models and the stability analysis provide background
for assessing the number G characterizing robustness of equilibrium.
5. Examples
The novel 4-4 CSPR for balance and gait rehabilitation (Fig. 2) has been con-
sidered as an example. As mentioned above, in order to model human body gravi-
ty forces and bottom reactions during balance tests, including unsupported fall ex-
periment, a rigid body human model (MATMAN) has been developed. However,
this will not be further considered in this paper, focusing on stability of the sup-
porting cable robot. Therefore only trunk has been included in the analysis as a
common platform, assuming various external net forces and torques that may be
outcomes of human efforts and ground reaction. The parameters of the considered
under-actuated CSPR are given in the Table 1.
Initially the stability of the equilibrium in symmetric steady-state configuration
(Fig. 5 left) has been analyzed. Obviously the external gravitation wrench
w G >0 0 294.3 0 0 0@ is consistent to (13) and falls into column span
T
of J T (22), providing the tension solution fi 95.98 (N) in each cable. The equi-
librium is stable with eigenvalues of null-space Hessian H N λ >26.48 55.4@ .
However, if the relative attachments b i are selected bellow the center of mass
(Fig. 5 right), the equilibrium becomes obviously unstable λ > 31.5 5.3@
(the consistency is further ensured).
Cable a b e r
1 [-0.65 -2.11 2.28] [0.1 -0.1 0.1] [0 -1 0] 0.015
2 [-1.8 -2.11 2.28] [-0.1 -0.1 0.1] [0 -1 0] 0.015
3 [-0.65 -0.29 2.28] [0.1 0.1 0.1] [0 1 0] 0.015
4 [-1.8 -0.29 2.28] [-0.1 0.1 0.1] [0 1 0] 0.015
Table 1: Geometric cables parameters of the robot with platform mass and inertia matrix
m=30 (kg) and Ic=diag[1.25 2.67 1.42] (kgm2), respectively, r is the pulley radius (m)
In an asymmetric configuration (Fig. 6 left) wrench w G is inconsistent to the

mapping (16), a solution for f doesn’t exist. In this configuration we can solve
stabilization task by modifying w G into a virtual consistent wrench w con involv-
ing the same gravitation force components from w G and adding two additional
torques, around x and y axes respectively, computed from (26). Basically an exter-
nal torque w conx 10.73 (Nm) around lateral trunk x-axis is only needed to virtu-
ally stabilize the equilibrium.
Fig. 5 Consistent stable (left) and unstable (right) configurations

Without this stabilization torque, the current w G , i.e. its projection on the null-
space (mainly torque component around lateral axis x) causes rotation in the null-
space in the sagittal plane until a new stable and for w G consistent configuration
has been reached (Fig. 6 right). This new really stable equilibrium configuration is
determined by simulation using dynamic model (29), and has been defined by a
relative rotation ox 0.195 (rad) around lateral x-axis wrt. the initial pose.
A completely asymmetric real-equilibrium pose consistent with w G obtained
also by simulation model (29) starting from vertical initial posture is presented in
(Fig. 7). For the considered robot all consistent stable poses in a workspace of in-
terest that can be reached by the trunk rotations and pelvis (body) inclination are
presented in (Fig. 8). However, different to stable symmetric configuration (Fig.
6), the practical stabilities bounds ( G and Q) of these poses are relatively small,
and the platform is prone to switch to other neighborhood equilibrium poses. The
estimation of these practical stability measures is the topic of the current work.
6. Conclusion
This paper has presented the detailed modeling and analysis of practical stabil-
ity of general under-constrained CSPRs with pulley elements. The practical stabil-
ity analysis includes both consistency tests of the applied wrenches (gravity and
external platform forces and torques) and assessment of eigenvalues of Hessian
projected on the null-space of robot Jacobian. For quasi-stable configurations a
stabilization task has been proposed, which can find attractive applications in fu-
ture CSPRs systems with additional drives embedded in the platform. Despite
these conditions have been fulfilled, a dynamic motion-oscillations in the null-
space due to perturbation effects can destabilize robot and bring it in another
steady-state stable pose. The practical stability framework has been proposed in
conjunction with the null-space dynamic model to assess robustness of equilibri-
um against perturbations.
For the gait rehabilitation robot under development this self-motion may be
very useful to support patient efforts to stabilize trunk posture and to make train-
ing of balance more efficient based on “assist as needed, less as possible” para-
digm. Thereby patient fall is prevented by fixed-controlled cable lengths. However

in other under-constrained CSPRs the self-motion can be quite undesirable. An ex-
ternal damping is needed to dynamically stabilize the null-space motion using ad-
ditional actuators, since the cable drives act in the Jacobian column space. The as-
sessment of practical dynamic stability of an equilibrium in rehabilitation and
transportation applications is a subject of the on-going work.
Fig. 6 Initial (inconsistent, left) and final stable (right) configurations
Fig. 7 Asymmetric gravity wrench consistent configuration
Fig. 8 Stable configurations in the trunk-pelvis workspace
7. References
[1] B. Zi and S. Qian, Design, Analysis and Control of Cable-Suspended Parallel Ro-
bots and Its Application, Springer-Verlag, 2017.
[2] J-P. Merlet, On the robustness of cable configurations of suspended cable-driven

parallel robots, Proc. 14th IFTOMM World-Congress, Taipei, October 2015.
[3] J. Albus, R. Bostelman and N. Dagalakis, The NIST Robocrane, Journal of Robotic
Systems, Vol. 10, No. 5, 1993, pp.709-724.
[4] N. Michael, S. Kim, J. Fink and V. Kumar, Kinematics and Statics of Cooperative
Multi-Robot Aerial Manipulation with Cables, Proceedings of ASME 2009 IDETC/CIE
Conference, San Diego, pp. 83-91.
[5] J. Fink, N. Michael, S. Kim and V. Kumar, Planning and control for cooperative
manipulation and transportation with aerial robots, The Int. Journal of Robotic Research,
30(3), 2015, pp.324-334.
[6] M. Carricato and J-P. Merlet, Stability Analysis of Underconstrained Cable-Driven
Parallel Robots, IEEE Transaction on Robotics, Vol. 29, No.1, February 2013, pp. 288-296.
[7] Karger, A. and Husty, M.L. (1998). Classification of all self-motions of the original
Stewart-Gough platform. Computer-Aided Design 30(3), 1998, pp. 205–215.
[8] G. Abbasnejad and M. Carricato, Direct Geometrico-Static Problem of Undercon-
strained Cable-Driven Parallel Robots with n-Cables, IEEE Transaction on Robotics, Vol.
31, No. 2, April 2015, pp.468-478.
[9] A. Berti, J-P. Merlet and M. Carricato, Solving the direct geometrico-static problem
of underconstrained cable-driven parallel robots by interval-analysis, The Int. Journal of
Robotic Research, Vol. 36, 2017, pp. 723-729.
[10] J-F. Collard and P. Cordou, Computing the lowest Equilibrium Pose of a Cable-
Suspended Rigid Body, Optim. Eng. 14, 2013, pp. 457–476.
[11] G. Liwen, X. Huayang and L. Zhihua, Kinematic analysis of cable-driven parallel
mechanisms based on minimum potential energy principle, Advances in Mechanical Engi-
neering, Vol. 7(12), 2015, pp. 1-11.
[12] A. Pott, Influence of pulley kinematics on cable-driven parallel robots, In Latest
Advances in Computation Kinematics (Ed. J. Lenarcic), Springer, 2012, pp. 197-204.
[13] J. LaSalle and S. Lefshetz, Stability by Lyapunov’s Direct Method, Academic
Press, New York, 1961.
[14] D. Surdilovic. R. Bernhardt, T. Schmidt and J. Zhang, STRING-MAN: A Novel
Wire Robot for Gait Rehabilitation, in Advances in Rehabilitation Robotics, Lecture Notes
in Control and Information Sciences, Vol. 306, 2004, pp. 413-424.
[15] M. Carricato and G. Abasnejad, Direct Geometrico-Static Analysis of Under-
Constrained Cable-Driven Parallel Robots with 4 Cables, Cable-Driven Parallel Robots
(Eds. T. Bruckmann and A. Pott), Springer, 2013, pp. 269-285.
[16] G. Stepan, A. Toth, L. Kovacs, G. Bolmsjo, G. Nikoleris, D. Surdilovic, A. Conrad,
A. Gasteratos, , N. KyriakoulisR, J. Canou, T. Smith, W. Harwin, R. Loureiro, R. Lopez
and R. Moreno, ACROBOTER: a ceiling based crawling, hoisting and swinging service ro-
bot platform, Proceedings of beyond gray droids: domestic robot design for the 21st century
workshop at HCI 2009.
[17] O.A. Bauchau and A. Laulusa, Review of Contemporary Approaches for Constraint
Enforcement in Multibody Systems, Journal of Computational and Nonlinear Dynamics,
3(1), 2008, pp. 1-8.
[18] X-S. Yang, Practical stability in dynamical systems, Chaos, Solitons and Fractals,
Pergamon, 11 (2000), pp. 1087-1092.
Singularity Characteristics of a Class of Spatial
Redundantly actuated Cable-suspended Parallel Robots
and Completely actuated ones
Lewei Tang1, Xiaoyu Wu1, Xiaoqiang Tang2, and Li Wu3

1
College of Mechanical Vehicle Engineering, Hunan University, Changsha, 410082, China
2 Department of Mechanical Engineering, Tsinghua University, Beijing, 100084, China
3
Department of Project Engineering &Technology Center, GAC FIAT CHRYSLER Automo-
biles Co., Ltd., Changsha, 410100, China
[email protected]
Abstract. This paper aims to study singularity characteristics of a class of spa-

tial redundantly actuated Cable-Suspended Parallel Robots(CSPRs) and com-
pletely actuated ones with pairwise cables. This study focuses on the CSPRs
with purely three translational degrees of freedom using redundant actuators or
complete actuators. One class of CSPRs is able to perform the translational
movement with pairwise cables as parallelograms. There are two types of sin-
gularity to be discussed, which result from dynamic equations of CSPRs and
parallelogram pairwise cables configurations. To assure three-translational dofs
without the rotation of the end-effector, the matrix formed by normals of pair-
wise cables should maintain in full rank. In the case study, one type of CSPRs
with a planar end-effector is discussed to clarify and conclude the singularity
features. The results show that in some configurations of CSPR there exists sin-
gularity of completely actuated CSPR but redundantly actuated ones are able to
fulfill the three-translational-dof movement.
Keywords: Cable-suspended Parallel Manipulator, Singularity analysis, Re-

dundant Actuator.
1 Introduction
Cable mechanism has been used in a large range of engineering field due to its flexi-
ble feature and suitable strength with lightweight structure such as construction [1],
transmission unit of anthropomorphic manipulator [2], leg exoskeleton [3], material
manipulation [4] and astronomical observation [5]. Compared with rigid link parallel
robots, the cable-driven parallel robot employs flexible cables as force transmission
and position constraint parts to manipulate the end-effector in workspace.
Since cables can be wound around winches, cable robot features large-scale work-
space and ease of reconfiguration. As shown in Fig. 1, the workspace below line p2p3
is divided into four areas A1, A2, A3 and A4, where the area above line p2p3 is not
available because it is not fully controllable. A1 is formed as a polygon with the pul-

100 L. Tang et al.
leys as vertices, which is referred to as static workspace or force-closure workspace

[6]. A2 is the area limited by lines p1p2, p3p4 and p1p4, where only the gravity can
exert force downward so that in this area the mechanism with this configuration is
also named as cable-suspended parallel robot. The other areas as A3 and A4 is de-
fined as the dynamical workspace with considering inertial effect from acceleration
of end-effector. Group of Gosselin has presented several literature regarding dynam-
ical workspace of CSPRs [7-9].
p2 p3
p(xo,yŽ)
A1
G p4
p1
A3 A2 A4
Fig. 1. Workspace of Cable-driven Parallel Robots.
A gripper or grasping mechanism is fixed on the end-effector to perform pick-and-

place movement in general. It is difficult to assemble grippers or other tools on the
end-effector as a mass point. To solve this issue, several new designs of CSPRs are
presented by Behzadipour in [10], where the parallelogram structure with pairwise
cables is adopted to constrain rotational DOFs and perform only translational DOFs
such as BetaBot and planar cable-based manipulators. Pairwise cables in parallelo-
gram are able to constrain the rotational DOF around the normal of plane formed by
these cables.
Considering the gravity of the end-effector, a class of CSPRs similar to Be-
taBot is proposed by Vu [11] in place of the rigid link spine to extend the work-
space with only translational DOFs. Due to this parallelogram design, there is no
need to limit the end-effector as a mass point, which is beneficial for designing
end-effector structures with different tools.
Gosselin[12] first studied point-to-point dynamic trajectory planning method of
two and three DOFs cable-suspended robots. Additionally, some papers focus on
redundantly-actuated cable-suspended robots other than completely-actuated
ones. In the paper [8], a planar two-DOF redundantly actuated cable-suspended
parallel robot with the end-effector as a mass point was studied on dynamic tra-
jectory planning and its characteristics had been discussed in details.
In this paper, a class of new cable-suspended robot with actuator redundancy is
studied by considering structure of the end-effector with parallelogram pairwise
cables attached. Compared to the structure mentioned in [11], redundant actuation
is introduced in cable-suspended robots to alleviate the issues of both system vi-
Singularity Characteristics of a Class of Spatial Redundantly… 101
bration and abrupt tensions. Moreover, the redundant cables are able to exert ad-
ditional tensions with various combinations and better performance can be ob-
tained[13].
Singularity is another issue to work on for new mechanisms before trajectory
planning since singular positions should be avoided as static state during motion.
In the range of achievable workspace, there may exist some singular positions
where the end-effector can not be controllable. Therefore, before discussing the
trajectory planning of this class of CSPRs, singularity analysis should be finished.
Many researchers had studied the singularity issue of rigid parallel mechanisms.
Zlatanov defined constraint singularity of parallel mechanisms, and noted that
constraint singularity only occurs in constrained parallel mechanisms[14]. Screw
theory is also used in singularity analysis of 3-dof planar parallel mecha-
nisms[15], where the singular loci of PPMs with parallelograms have been dis-
cussed. Liu proposed a geometric treatment to identify singularities of parallel
manipulators[16]. Since CSPRs are driven by cables, tension should be consid-
ered as a constrain for these mechanisms. In this paper, a class of CSPRs with
parallelogram structure is studied with redundant actuators on singularity analysis
and the results are compared with CSPRs using complete actuators.
The rest of this paper is organized as follows. Section 2 introduces a spatial re-
dundantly-actuated cable-suspended parallel robot with parallelogram pairwise
cables as the research object. In what follows, Section 3 provides the kinematic
and dynamic modeling of the cable-suspended robots with redundant actuators.
Singularity analysis is implemented in Section 4. Next, one type of CSPRs is dis-
cussed in Section 5. Finally, the conclusion is drawn in Section 6.
2 Spatial Redundantly-actuated CSPR Description with

Parallelogram Pairwise Cables
For a spatial redundantly-actuated CSPR, one end of pairwise cables as parallelo-

gram (as shown in Fig. 2) is attached to pulleys and the other end of cables is
anchored at the end-effector. Each pair of cables is employed as a parallelogram
during motion because one servo motor is able to be applied to control the exten-
sion and retraction of pairwise cables in synchronization. The global coordinate
system is defined with z-axis along the opposite direction of gravity acceleration
in this paper.
Due to the parallelograms[11], rotational degrees of freedom around the nor-
mal of each plane of parallelogram pairwise cables can be forbidden so that rota-
tional DOFs of the end-effector can be constrained in space at any non-singularity
position.
102 L. Tang et al.
Pulleys
Parallelogram
... Cables
Attachment
points
End-effector
Fig. 2. Parallelogram Pairwise Cables of CSPR.
3 Kinetic Modelling
Kinematic model of the CSPR is illustrated in Fig. 3, where the relative coordi-
QDWHV\VWHP2¶-[¶\¶]¶LVGHILQHGZLWKWKHRULJLQDWWKHFHQWHURIPDVVRIWKHHQG-
effector. A12 is denoted as the middle point of line A1A2 (attached points of pul-
leys at the parallelogram) and a12 is the middle point of line a1a2 (anchors of par-
allelogram on the end-effector ). The position vector of two origins is defined as
p, where A and a are position vectors in global and relative coordinate systems
respectively. At any non-singular position, this robot is able to realize three trans-
lational-dof motion using the parallelogram design. Thus, the length of cable 1&2
is denoted as
A 1&2 A p a . (1)
As shown in Fig. 4, the wrench (force and moment) is exerted on the end-effector
by the pairwise cables such as cable 1&2. Tensions of cable 1&2 are referred to
as t1 and t2 with the directions along each cable. Since A1A2a2a1 is constrained as
a parallelogram during motion, the directions of tensions of pairwise cables are
identical. Forces exerted on the end-effector by tensions of cable 1&2 are (t1+t2),
where the moments are (a1xt1+a2xt2) with respect to the center of mass of the
end-effector. The force and moment from tensions on the end-effector are denoted
as
F12 t1 t 2 (t1 t2 )l1 , (2)
M12 a1 u t1 a2 u t 2 (t1a1 t2a2 ) u l1 , (3)
where l1 is defined as the unit vector of t1.
Fig. 3. Kinematic Modeling. Fig. 4. Dynamical Modelling.
Assume that the tensions of pairwise cables have the same value as t1=t2. Consid-
ering the acceleration of the end-effector, the dynamic modeling of this CSPR is
obtained
n
2¦ t(2i 1)l i g ,
mp (4)
i 1
¦t
i 1
( 2 i 1) a ( 2i 1)(2i ) u l i 0, (5)
where m is the mass of the end-effector, ‫݌‬ሷ ൌሾ‫ݔ‬ሷ ǡ ‫ݕ‬ሷ ǡ ‫ݖ‬ሷ ሿ் ,܏ ൌ ሾͲǡͲǡ െሿ் (g is the
gravitational acceleration) and n is the number of pairwise cables. This CSPR is
referred to as the completely actuated mechanism with three-translational-DOF as
n is equal to three, while the redundantly actuated CSPR is denoted as n is larger
than three.
4 Singularity Analysis
For this class of redundantly-actuated CSPR, singular positions may exist over the
workspace. When it happens, the three-translational-dof of the end-effector can
not be maintained. In this section, singularity analysis is proceeded to classify the
feasible positions to fulfill the requirement of translational motion.
Accordingly, at least three pairs of cables are required to keep taut to control
motion of this CSPR. Each pair of cable with parallelogram structure restrains the
rotational degree of freedom around normal to the parallelogram. In order to con-
strain all rotational dofs, the first singularity of this robot is deduced as
Rank([(a2-a1) xl1,..., (a2i-a(2i-1))xli,..., (a2n-a(2n-1))xln ])<3. (6)
The second singularity of this robot is from Equ. (5), which is denoted as for
{a12xl1, ..., a(2i)(2i-1)xli,..., a2n(2n-1)xln } there is no plane to separate these vectors in
104 L. Tang et al.
one side of a plane and no vectors in the other. Item a(2i)(2i-1)xli is referred to the
normal to the plane formed by {a(2i)(2i-1), li} (i=1,2,...,n).
In conclusion, two different singularity have been discussed in this section. In next
section, the singularity of one type of CSPRs is presented in details.
5 Case Study
The type of CSPR is shown in Fig. 5 with a rectangle end-effector, where detailed
dimensions of structure are illustrated. The global coordinate system O-xyz is located
at the origin of the center of the rectangle with eight pulleys, where the relative coor-
dinate system O'-x'y'z' is defined at the center of the rectangle as the end-effector.
Assume that the origin O' of relative coordinate system in the global coordinate
system is (x,y,z). Then, the unit tension direction of each pairwise cables are de-
duced as l1=[ x, b-c+y, z ]T/l1, l2=[ a-d+x, y, z ]T/l2, l3=[ x, c-b+y, z ]T/l3, l4=[ d-
a+x, y, z]T/l4, where li (i = 1,2,3,4) is the length of pairwise cables.
Based on Equ.(6), singular positions of this CSPR with a rectangle end-effector is
obtained as
det([(1,0,0) T u l1 ,(0,1,0) T u l 2 ,(1,0,0) T u l 3 ]) 0,
° T T T
°det([(1,0,0) u l1 ,(0,1,0) u l 2 ,(0,1,0) u l 4 ]) 0,
® T T T (7)
°det([(1,0,0) u l1 ,(1,0,0) u l 3 ,(0,1,0) u l 4 ]) 0,
°det([(0,1,0) T u l ,(1,0,0) T u l ,(0,1,0) T u l ]) 0.
¯ 2 3 4
Then, the singular positions for this rectangle end-effector of CSPR is deduced as
z 2 (c b) 0,
°
® 2 (8)
°̄ z (d a) 0.
The results show that one singular plane z=0 always exists regardless of the value of
{a,b,c,d}. The plane z=0 is also considered as the upper limitation of feasible work-
space of this CSPR. For a special case where c and d are equal to b and a respective-
ly in Fig. 6, the rotational-dof around z-axis is uncontrollable, which leads to all the
positions within the static workspace (composed of the space under the upper pul-
leys) are singular.
Another singularity of this CSPR is introduced as below
det([(0,1,0) T u l1 ,(1,0,0) T u l 2 ,(0,1,0) T u l 3 ])=0,
° T T T
°det([(0,1,0) u l1 ,(1,0,0) u l 2 ,(1,0,0) u l 4 ])=0,
® T T T
(9)
°det([(0,1,0) u l1 ,(0,1,0) u l 3 ,(1,0,0) u l 4 ])=0,
°det([(1,0,0) T u l ,(0,1,0) T u l ,(1,0,0) T u l ])=0.
¯ 2 3 4
Thus, each three items in the set are in the same plane which also means all the four
items are in the same plane.
Fig. 5. Rectangle End-effector. Fig. 6. CSPR with a Rectangle End-effector (a=d, b=c)
Fig. 7. Second Singularity Analysis.

Based on the configuration of this CSPR, the four items in the set are denoted as the
normals of red and yellow planes so that in every three items there must be two of
them parallel to each other. As a result, the second singularity is degenerated to de-
termine whether there is a line that can include all vectors in one side. From this
configuration, all the vectors span a plane. It means that the second singularity does
not happen for this type of CSPRs with tensions along four cables maintained in a
range of positive interval.
If one pair of cables is ignored or missed for this CSPR, then redundant actuator is
removed. Therefore, this CSPR with three pairwise cables is reconfigured to be a
completely actuated one. The first singularity is the same according to the aforemen-
tioned analysis. However, the second singularity is changed because three vectors
can only span half of the plane. Therefore, the variant CSPR with three pairwise
cables has a feature with one rotational DOF added.
106 L. Tang et al.
One conclusion is drawn for this planar rectangle end-effector of CSPRs that the
first singularity is the same for both completely actuated and redundantly actuated
CSPRs. Additionally, for the planar end-effector of CSPRs with all parallelogram
pairwise cables, the second singularity is able to be avoided if lines of pairwise
symmetric attachment points of the end-effector can form a closed polygon.
6 Conclusions
This paper discusses the singularity analysis of cable-driven parallel robots with
three-translational-dof. To constrain the rotational dofs, pairwise cables as parallelo-
grams are adopted and the parallelism is maintained during the movement. The first
singularity happens if the parallelism structure is not able to limit the rotational dofs
in space. In the available workspace, the end-effector is required to fulfill the dynam-
ical modelling, where the moment equilibrium produces the second constraint singu-
larity.
To validate the results, the singularity analysis of one CSPR is performed with a
planar end-effector. The results show that the first singularity exists when at least
two pairwise cables are in the same plane. In some special cases, it also happens in
the CSPR with corresponding parameters (for example, a=d and c=b in Fig.6 ). The
second singularity configuration is determined if there is no plane to separate these
vectors {a12xl1, ..., a(2i)(2i-1)xli,..., a2n(2n-1)xln } in one side of a plane and no vectors in
the other. The second singularity is avoided with redundantly actuated symmetric
structure and it emerges if one pair of cables is missing.
Acknowledgments
This work is supported by the Fundamental Research Funds for the Central Universi-
ties (Grant No. 531107051087) and Natural Science Funds of Hunan Province in
China (Grant No. 2018JJ3050).
References
1. P. Bosscher, R. L. Williams, L. S. Bryson and D. C. Lacouture: Cable-suspended robotic
contour crafting system, Automation in Construction, 17 (1), 45-55(2007).
2. Y. Kim: Anthropomorphic low-inertial high-stiffness manipulator for high-speed safe in-
teraction, IEEE Transactions on Robotics, 33(6) 1358-1374(2017).
3. X. J, A. Prado and S. K. Agrawal: Retraining of human gait - are lightweight cable-driven
leg exoskeleton designs effective? IEEE Transactions on Neural Systems and Rehabilita-
tion Engineering, 26(4) 847-855(2018).
4. L. Barbazza, F. Oscari, S.Minto and G. Rosati: Trajectory planning of a suspended cable
driven parallel robot with reconfigurable end effector. Robotics and Computer-Integrated
Manufacturing, 48, 1-11(2017).
5. X.Q. Tang, Z. F. Shao: Trajecotry generation and tracking control of a multi-level hybrid
support manipulator in FAST. Mechatronics, 23(8), 1113-1122(2013).
6. C. B. Pham, S. H.Yeo, G. Yang and M. S. Kurbanhusen: Force-closure workspace analysis

of cable-driven parallel mechanisms. Mechanism and Machine Theory, 41, 53-69(2006).
7. C. Gosselin, P. Ren, S. Foucault: Dynamic trajectory planning of a two-dof cable-
suspended parallel robot. 2012 IEEE International conference on Robotics and Automa-
tion, 1476-1481(2012).
8. L. Tang, X. Tang, X. Jiang, C. Gosselin: Dynamic trajectory planning study of planar two-
dof redundantly actuated cable-suspended parallel robots. Mechatronics, 30, 187-
197(2015).
9. X. Jiang, C. Gosselin: Dynamical point-to-point trajectory planning of a three-dof cable-
suspended parallel robot. IEEE transactions on Robotics, 32(6), 1550-1557 (2016).
10. S. Behzadipour, A. Khajepour: Cable-based robot manipulators with translational degrees
of freedom. Industrial robotics: theory modelling control, 211-236(2006).
11. D. Vu, E. Barnett, A. Zaccarin, C. Gosselin, On the design of a three-DOF cable-
suspended parallel robot based on a parallelogram arrangement of the cables. Cable-driven
parallel robots, Springer, 319-330(2018).
12. Gosselin, C., Foucault, S: Dynamic Point-to-Point Trajectory Planning of a Two-DOF Ca-
ble-Suspended Parallel Robot. IEEE Transactions on Robotics, 30(3), 728-736(2014).
13. C. Gosselin, M. Grenier: On the determination of the force distribution in overconstrained
cable-driven parallel mechanisms. Meccanica, 46(1), 3-15(2011).
14. D. Zlatanov, I.A. Bonev, C. Gosselin: Constraint singularities of Parallel Mechanisms.
2002 IEEE International Conference on Robotics and Automation, 496-502(2002).
15. I.A. Bonev, D. Zlatanov, C. Gosselin: Singularity analysis of 3-dof planar parallel mecha-
nisms via screw theory. Journal of Mechanical Design, 125, 573-581(2003).
16. G. Liu, Y. Lou and Z. Li, Singularities of parallel manipulators: a geometric treatment.
IEEE transcations on robotics and automations, 19(4), 579-594(2013).
Kinetostatic Modeling of a Cable-Driven Parallel
Robot using a Tilt-Roll Wrist
Saman Lessanibahri1,2, Philippe Cardou3 , and Stéphane Caro2,4

1
École Centrale de Nantes, Nantes, 44321 France,
2
Laboratoire des Sciences du Numérique de Nantes (LS2N), UMR CNRS 6004,
Nantes, 44300, France,
3
Département de génie mécanique, Université Laval, Québec, QC, Canada,
4
Centre National de la Recherche Scientifique (CNRS), Nantes, 44321, France,
Emails: [email protected]
Abstract. This paper introduces the concept of a new Cable-Driven

Parallel Robot (CDPR) applicable for tasks requiring large rotation and
translation workspaces, e.g., camera orienting devices, visual surveillance
over vast areas and tomography scanners. The manipulator consists of
a fixed frame and a tilt-roll wrist attached to a Moving-Platform (MP).
The MP is an under-constrained and articulated mechanism connected
to the fixed frame with six cables. The power is transmitted directly
from the motors fixed on the frames to the tilt-roll wrist through cable
loops. While most of the CDPRs only provide a limited range of rotation
of the MP, the proposed manipulator is applicable for a relatively large
and singularity-free rotation workspace thanks to its articulated MP.
Keywords: novel mechanism, cable loop, tilt-roll wrist, large orienta-

tion workspace
1 Introduction
There exists a wide variety of applications requiring large rotation workspace,

e.g., search and rescue, motion simulation and entertainment. Cable-Driven Par-
allel Robots (CDPRs) have drawn their interest of industry thanks to their
fundamental advantages and capabilities, such as high payload-to-weight ra-
tios, large translational workspaces and high-speed motions. They are parallel
robots with cables instead of rigid links and they can provide large translational
workspaces. They generally cannot cover large rotation workspaces due to colli-
sions between their moving parts. This paper introduces a hybrid robot with a
large translational and rotation workspace consisting in a CDPR connected to a
tilt-roll wrist. The manipulator combines the advantages of CDPRs, i.e., a large
translational workspace, with those of tilt-roll wrists, namely, large amplitudes
of rotation one. Solution to overcome the limited rotation workspace of CDPRs
could consist in mounting a standard tilt-roll wrist on the Moving-Platform (MP)
of a CDPR, but articulation of the tilt-roll wrist could be a challenge. Adding

110 S. Lessanibahri et al.
motors to the MP for actuating the wrist increases the weight of the MP and
reduces the dynamic performance of the manipulator. Therefore, a pair of bi-
actuated cable circuits, namely, cable loops are employed in the design of the
CDPR in order to transmit the power from the motors fixed on the frame to the
tilt-roll wrist. In general, the workspace of parallel robots can expand by con-
necting other parallel or serial mechanisms to them. This usually complicates
the design and reduces the dynamic performance of the manipulator unless, the
power is transmitted directly to the end-effector without locating motors on the
MP. By employing remote actuation, we avoid mass and entanglement of a long
power line tethering the MP to the ground. Moreover, because of the mentioned
reasons, remote actuation minimizes the size of the MP. This leads to a better
dynamic performance due to the lower inertia of the MP. Hence, coupling two
parallel mechanisms is proposed in order to exploit their combined workspaces
through cable loops. Many applications require pointing the end-effector over
a wide range of directions such as, camera surveillance, laser scanning, tomo-
graphic scanner, etc.
Cable loops have been addressed in several papers [7], [13] and it was em-
ployed in [10] for actuation of a hoist mechanism embedded in a MP. In the latter
the large rotational amplitudes provided by a cable loop was used to actuate a
one Degree of Freedom (DoF) mechanism, while here we employ two cable loops
to actuate a tilt-roll wrist having a large rotational workspace. H. Khakpour et
al. in [4],[5] and [6] introduced differentially driven cable parallel robots using
cable differentials. They proved that, by replacing single-actuated cables with
differential cables the static and wrench feasible workspaces of CDPRs can be
extended.
The study of the concept of the hybrid CDPR providing a large translational
and rotational workspace is organized as follows. Section 2 introduces the archi-
tecture of the under-constrained manipulator with eight DoF. Section 3 describes
the kinetostatic model of the tilt-roll wrist. The kinetostatic model of the overall
manipulator is presented in Sec. 4. Section 5 investigates the static workspace
of the CDPR and the last section concludes the paper.
l0
w0 P0
h0
z0 y
0
x0
F0
Fig. 1: Schematic of the CDPR using a Tilt-Roll Wrist

Kinetostatic Modeling of a Cable-Driven Parallel Robot using a Tilt-Roll Wrist 111
2 Manipulator Architecture
Here, we present the architecture of the manipulator, which is a hybrid robot

consisting of a base frame and a tilt-roll wrist connected to the MP of a CDPR.
The MP is suspended by six cables, i.e., two bi-actuated cable circuits, also
named cable loops, and four uni-actuated cables as illustrated in Fig. 1.
A cable loop forms a circuit by connecting two actuators while passing
through two anchor points on the MP and coiling about a gear on the tilt-roll
wrist. This arrangement of cables is used for two distinct purposes. The former
is to locate the MP within its workspcae and the latter is to actuate the tilt-roll
wrist connected in series to the MP. In other words, two motions can be induced
by the cable loop depending on the relative rotation of its two actuators. The
first one is the displacement of the MP for identical inputs to the two motors.
The second motion is the rotation of gears P3 and P4 about their axes (z3 and
z4 ), shown in Fig. 2, when the two actuators rotate in opposite directions. This
leads to an unlimited rotation of P3 and P4 about their respective rotation
axes. Figure 3 describes the main geometrical parameters of the MP as well as
the tilt-roll wrist. A cable loop connected to two motors and drums is highlighted
in Fig. 3. Cable loop consists of two segments each connecting the drums to their
anchor points on the MP with independent tensions and velocities.
The MP frame F1 is located on the top of the MP at point P. 0 p is the
vector pointing from the origin of F0 to the origin of F1 . 0 R1 is the rotation
matrix from frame F0 to frame F1 and it is expressed as three sequence of
rotations, i.e., 0 R1 = Rz (ψ)Ry (χ)Rx (φ). α and β, shown in Fig. 4, characterize
the two-DoF rotational motions of the tilt-roll wrist.
3 Kinetostatic Model of the Tilt-Roll Wrist
Kinetostatic model of the tilt-roll wrist is investigated in this section. The

section-view of the MP and the wrist is shown in Fig. 2. Their main compo-
nents are denoted as Pi , i = 1, ..., 5. The MP is represented with P1 , carrier
of the wrist is denoted as P2 , input gears connected to the cable loops C12 and
C56 being P3 and P4 , respectively. P5 is the end-effector.
Figure 4 shows the parametrization of the tilt-roll wrist in terms of an-
gles. The rotation angles of P2 , P3 and P4 about z2 -axis are denoted as
α = (z2 , x5 ), θ3 = (y1 , x3 ) and θ4 = (y1 , x4 ), respectively. The roll an-
gle of the end-effector is defined as β = (z2 , x5 ). The gear train ratio, μ is
expressed as:
r5 N5
μ= = (1)
r3 N3
where N3 and N5 are the number of teeth of gears P3 and P5 . The pitch radius
of ith gear is denoted as ri and i = 3, 4, 5. It should be noted that r3 = r4 . The
linear velocities at gear contact points C35 and C45 are expressed as follows:
1
vc35 = −θ̇3 r3 1 y2 (2)
z1
F1
P x1
P1
C12 C1 C56
P2 P4
r3 r4
O2
H5
P3 C35 r5 C45 h2
C2
P5
w5
Fig. 2: Section-view of the moving-platform and tilt-roll wrist
1
vc34 = −θ̇4 r4 1 y2 (3)
with θ̇3 and θ̇4 being the angular velocities of P3 and P4 , respectively.
1
vH5 = −r3 α̇1 y2 (4)
1
vH5 is the linear velocity of point H5 which belongs to P5 expressed in F1 .
Furthermore, the angular velocity of the end-effector P5 with respect to F1 and
expressed in F1 takes the form:
1
ω5/1 = α̇1 z2 − β̇ 1 x2 (5)
Knowing the linear velocity 1 vH5 of point H5 in F1 , the linear velocities 1 vc35
and 1 vc45 of points c35 and c45 in F1 can be expressed as:
1
vc35 = 1 vH5 + 1 h5 − 1 c35 × 1 ω5/1

(6)
1
vc45 = 1 vH5 + 1 h5 − 1 c45 × 1 ω5/1

(7)
From Eqs. (2)-(5) and Eqs. (6)-(7):
−θ̇3 r3 1 y2 = −r3 α̇1 y2 + r5 1 z2 × (α̇1 z2 − β̇ 1 x2 ) (8)

−θ̇4 r4 1 y2 = −r4 α̇1 y2 − r5 1 z2 × (α̇1 z2 − β̇ 1 x2 ) (9)
Upon simplification of Eqs. (8) and (9), we obtain:
θ̇3 = α̇ + μβ̇ (10)

θ̇4 = α̇ − μβ̇ (11)
A7
A2 A12 A1 A4 A8
l1 Cable loop
A3 2rc
l2
u12 A5
u8 A6
B1 z1 B8
l3 B2 B12 y1
F1
B3 P x1 B7
B4 B6
B5 B56
2rc p
y3
F3
x3 x2
O3 z3 y2 F2 y4 F4
O2 z2 x4
z0 y O4 z4
0
x0
F0
y5 O
5 F5
x5
z5
Fig. 3: Schematic of the moving-platform with an embedded camera
y2 z1 z1 y5
y3 y2
y4
x3 x4
θ3 x2 θ4 x2 x5
α α β
y1 y1
z2 , z3 , x1 z2 , z4 , x1 z5 z2
Fig. 4: Angles α, β, θ3 and θ4

As a result, the output rotational velocity vector q̇T R of the tilt-roll wrist is
expressed as a function of its input velocity vector θ̇ in a matrix form as follows:
q̇T R = JT R θ̇ (12)
with JT R is the kinematic Jacobian matrix of the wrist expressed as:
1/2 1/2
JT R = (13)
1/2μ −1/2μ
q̇T R = [α̇, β̇]T and θ̇ = [θ̇3 , θ̇4 ]T .

Let m = [mα , mβ ]T be the output moments of the tilt-roll wrist and τ =
[τ3 , τ4 ]T be the input torque vector of the wrist. From Eq. (12),
τ = −JTT R me (14)
with me = −m being the external moments applied by environment to the tilt-

roll wrist. By rewriting Eq. (14) we can express the static equilibrium of the
tilt-roll wrist as follows:
me = WT R τ (15)
with WT R being the wrench matrix of the wrist:
WT R = −J−T
TR
(16)
Besides me is expressed as:

T
me = m2 gh2 cos α m2 gw5 cos α sin β (17)
m2 indicates the mass of P5 and g = [0, 0, −g]T is the gravity acceleration with
g = 9.81 m.s−2 . As shown in Fig. 2, h2 and w5 denotes the distances between
O2 and C2 along z5 and y5 , respectively. C2 is the Center of Mass (CoM) of the
end-effector.
4 Kinetostatic Model of the Manipulator

In this section we present the kinetostatic model of the overall manipulator.
In order to define the manipulator Jacobian and its wrench matrix, we first
introduce the loop-closure equations of the CDPR, which are given by
0
li = 0 ai − 0 p − 0 R1 1 bi , i = 1, 2, ..., 8 (18)
where 0 li is the ith cable vector, i.e., the Cartesian coordinate vector pointing
from point Bi to point Ai . Points Ai and Bi stand for the ith cable exit point
and anchor point, respectively. The former point is the location of the ith pulley
fixed to the ceiling and the latter is the connection point between the cable and
the MP. 0 ai = [aix , aiy , aiz ]T , 1 bi = [bix , biy , biz ]T and 0 p = [px , py , pz ]T
are the Cartesian coordinate vector of points Ai , Bi and P, respectively. ti ,
i = 1, 2, ..., 8, stands for the ith cable tension vector. ti = ti 0 ui and its
magnitude is expressed as ti = ti 2 , i = 1, ..., 8. 0 ui denotes the ith cable unit
vector namely,
0
li
0
ui = , i = 1, 2, ..., 8 (19)
li
li being the ith cable length.
In the modeling, one simple virtual cable substitutes each cable loop. The
latter virtual cables replace each cable loop with Ai and Bi , i = 12, 56, as
their exit point and anchor points, respectively. As long as the MP is far from
points A1 and A2 , then A1 B1 and A2 B2 can be assumed parallel and the virtual
cable model replaces the first cable loop. This analogy also holds for second
cable loop. Therefore, the effect of first cable loop onto the MP is the force
passing through the midpoint B12 between B1 and B2 along the unit vector u12
of segment A12 B12 . t12 = t1 + t2 is the tension resultant in the first cable loop.
The Cartesian coordinates of the exit and anchor points in both the cable loops
are:
a12 = (a1 + a2 )/2 (20)
a56 = (a5 + a6 )/2 (21)
and,
b12 = (b1 + b2 )/2 (22)
b56 = (b5 + b6 )/2 (23)
equilibrium of the external forces applied on the MP, is formulated as follows:

ti 0 ui + mg = 0, i = 12, 3, 4, 56, 7, 8 (24)
mg is the weight of the MP. m denotes total mass of the MP and the spherical
wrist. The equilibrium of moments applied onto the MP about point P expressed
in frame F0 takes the form:

ti 0 R1 1 bi × 0 ui + m(0 c − 0 p) × 0 g = 0, i = 1, ..., 8

(25)
with c being the Cartesian coordinates vector of the CoM of MP:
m 1 c 1 + m2 c 2
c= (26)
m1 + m2
c1 stands for the CoM of components P1 to P4 of mass of m1 and c2 denotes
the CoM of P5 of mass of m2 . Therefore, overall mass of the MP is expressed
as follows:
m = m 1 + m2 (27)
The input tilt-roll wrist torques τ3 and τ4 are a function of the cable tension
difference in cable loops C12 and C56 , respectively:
τ3 = rc (t1 − t2 ) (28)
τ4 = rc (t6 − t5 ) (29)
rc being the radius of the groove made in P3 and P4 to house the two cable
loops. From Eq. (15) and Eqs. (24) to (29) the static equilibrium model of the
manipulator is expressed in a matrix form as:
Wt + wg = 08 (30)
where 08 is a eight-dimensional zero vector and the wrench matrix W takes the
following form:
⎡0
u12 0 u12 0 u3 0 u4 0 u56 0 u56 0 u7 0 u8
⎤
⎢ 0 d 1 0 d 2 0 d 3 0 d 4 0 d 5 0 d 6 0 d 7 0 d8 ⎥
W=⎢ ⎣ rc −rc 0 0 −rc rc
⎥ (31)
0 0 ⎦
μrc −μrc 0 0 μrc −μrc 0 0
with,
0
di = 0 R 1 1 b i × 0 u i , i = 1, ..., 8 (32)
t is the vector containing the tensions exerted by the eight actuators to the
cables.
t = [t1 t2 t3 t4 t5 t6 t7 t8 ]T (33)
wg is the eight-dimension gravity wrench vector applied on the MP and tilt-roll
wrist, namely,
wg = [m0 gT m(0 R1 1 c × 0 g)T mTe ]T (34)

It should be noted that, in Eq. (30) the friction between cable-loops and P3
and P4 is neglected. Hereafter, we can express the differential kinematics of
the manipulator that formulates the relation between the output twist q̇ =
T T
ṗT , ω T , q̇TT R of the manipulator and the cable vector l̇ = l˙1 , ..., l˙8 :
Jq̇ = l̇ (35)
Velocity of the origin of frame F1 with respect to F0 is defined as ṗ and the

angular velocity of the MP with respect to F0 is defined as ω . Jacobian matrix
J is calculated based on the well-known kineto-static duality:
J = −WT (36)
5 Workspace Analysis
The section deals with the workspace analysis of the CDPR using a tilt-roll wrist
under study.
The set T forms the feasible cable tensions as a box in eight-dimensional space:
T = {t ∈ R8 : tmin ≤ t ≤ tmax }. (37)
where, tmin and tmax are the lower and upper bounds of admissible cable ten-
sions. Static Workspace (SW), namely, S is set of MP poses and tilt-roll wrist
configurations satisfying the static equilibrium of the manipulator with admis-

sible cable tensions.
S = {(0 p, 0 R1 , qT R ) ∈ R3 × SO(3)× ∈ R2 : ∃t ∈ T , Wt + wg = 08 } (38)
where SO(3) is the group of proper rotation matrices. From Eq. (38) we can
derive two subsets S1 and S2 based on the constrained orientations of the
MP and the tilt-roll wrist. The former subset, namely, S1 is a set for a given
orientation of the MP and the wrist, i.e., the translational SW of the manipulator
with constant orientation of 0 R1 and qT R .
S1 = {0 p ∈ R3 |(0 R1 , α, β) given : ∃t ∈ T , Wt + wg = 08 } (39)
S2 amounts to the SW of the manipulator for a constant orientation of the MP
while the tilt and roll is free to rotate.
S2 = {0 p ∈ R3 |0 R1 given : −π ≤ α & β ≤ π : ∃t ∈ T , Wt + wg = 08 } (40)
Here, we consider CREATOR prototype introduced in [10] in order to trace
S1 and S2 for different orientations of the moving-platform. The main dimen-
sions of the prototype are given in Table 1. The specifications listed in Table 1
are based on the preliminarily manufactured prototype shown in Fig. 6. The
weight of the wrist highly depends on the material of the parts P2 , P5 and
the end-effector. Because ABS plastic is employed for P2 , the overall weight of
the MP can be significantly reduced by substituting steel bevel gears with nylon
ones. The prototype frame is 4 m long (l0 ), 3.5 m wide (w0 ) and 4 m high (h0 ).
The MP under study is suspended by six cables as shown in Fig. 1. The SW
Table 1: Parameters of CREATOR prototype

Parameter Abbreviation Value
1
Vector of CoM associated to P1 -P4 [m] c1 [0, 0, −0.15]T
2
Vector of CoM associated to P5 [m] c2 [−h2 , w5 cosβ, −w5 sinβ]T
Mass of P1 -P4 [kg] m1 4
Mass of P5 [kg] m2 0.6
Radius of the groove made in P3 and P4 [m] rc 0.1
Distance between O2 and C2 along z5 [m] h2 0.23
Distance between O2 and C2 along y5 [m] w5 0.03
Maximum admissible cable tension [N] tmax 128
Minimum admissible cable tension [N] tmin 0
illustrated in Fig. 5 refers to constant orientation of the MP for different given

orientation of MP. Cable loops are plotted in magenta and single-actuated cables
are represented in red. It is apparent that ψ has significant effect on the size of
the SW as well as the position of anchor points and exit points. According to the
obtained results, the largest SW (fig. 5b) is associated with ψ = 0◦ . The gravita-
tional external wrench due to the varying CoM of the wrist, namely C2 has effect
S1 with α = π/2 and β = 0 (home configuration of the wrist)

S2 with −π ≤ α ≤ π, −π ≤ β ≤ π
C4 C7 C4 C7
C3 C8 C3 C8
C12 C56 C12 C56
(a) SW with φ = 0◦ , χ = 0◦ and ψ = −30◦ (b) SW with φ = 0◦ , χ = 0◦ and ψ = 0◦
C4 C7 C4 C7
C3 C8 C3 C8
C12 C56 C12 C56
(c) SW with φ = 0◦ , χ = 0◦ and ψ = 45◦ (d) SW with φ = 0◦ , χ = 0◦ and ψ = 60◦
C4 C7 C8 C4 C7 C8
C3 C3
C56 C56
C12 C12
(e) SW with φ = 0◦ , χ = 30◦ and ψ = 30◦ (f) SW with φ = 0◦ , χ = −20◦ and ψ =

−30◦
Fig. 5: S1 and S2 for different orientations of the moving-platform

on the SW. Therefore, two cases are considered. The former case (shown with
green volume in the figure) considers the minimum external wrench applied onto
the MP by the wrist, i.e., the tilt-roll wrist is in home configuration (α = π/2
and β = 0). The latter case computes the SW while the P5 exerts the maximum
external wrench due to gravity on the MP and the wrist with blue volume in the
figure. It should be noted that, the gravitational external wrench of the wrist is
a function of α and β. The figure shows the variation of S1 and S2 for different
orientation of MP. The largest volumes for S1 and S2 are 25.3 m3 and 14.15 m3 ,
respectively, as shown in Fig. 5(b). On the other hand, the smallest volume of
SW is associated to Fig. 5(d) with 13 m3 of S1 and 8.15 m3 of S2 .
It appears to be a trade-off between translation and large rotation workspaces
due to the cable-loops in the fully-actuated CDPR. It is noteworthy that trans-
lation and orientation workspaces become larger when employing two additional
actuators.
Fig. 6: Prototype of the CDPR with an embedded tilt-roll wrist
6 Conclusions
A Cable-Driven Parallel Robot (CDPR) using a Tilt-Roll (T&R) wrist has been
introduced in this paper. The robot combines the advantages of CDPR in terms
of large translational workspace with those of tilt-roll wrist in terms of large ro-
tational amplitudes. The robot uses cables to transmit power directly from fixed
motors on the ground to the tilt-roll wrist. In comparison with conventional
cable-driven parallel robots and cable-driven suspended robots, the proposed
concept adds singularity-free and large tilt-roll motions to the end-effector. This
leads to a trade-off, however, between translational and rotational workspaces
due to tension coupling in cable-loops for fully-actuated CDPRs. The kineto-
static model of the proposed manipulator was studied. The static workspace of
the manipulator was also defined and analyzed. The detailed design, prototyping
and dynamic modeling of the cable-driven parallel robot equipped with a tilt-roll
wrist are left for future work. Moreover, the study of the over-actuated CDPRs
with the embedded tilt-roll wrist will be carried out in order to maximize their
workspaces.
References
1. S. Bai and J. Angeles. The design of a gearless pitch-roll wrist. In Robotics
and Automation, 2005. ICRA 2005. Proceedings of the 2005 IEEE International
Conference on, pages 3213–3218. IEEE, 2005.
2. F. Guay, P. Cardou, A. L. Cruz-Ruiz, and S. Caro. Measuring how well a structure
supports varying external wrenches. In New Advances in Mechanisms, Transmis-
sions and Applications, pages 385–392. Springer, 2014.
3. K. Hyodo and H. Kobayashi. Kinematic and control issues on tendon controlled
wrist mechanism. In Robotics, Mechatronics and Manufacturing Systems edited by
T. Takamori, K. Tsuchiya, pages 89–94. Elsevier, 1992.
4. H. Khakpour and L. Birglen. Workspace augmentation of spatial 3-dof cable par-
allel robots using differential actuation. In Intelligent Robots and Systems (IROS
2014), 2014 IEEE/RSJ International Conference on, pages 3880–3885. IEEE, 2014.
5. H. Khakpour, L. Birglen, and S.-A. Tahan. Synthesis of differentially driven planar
cable parallel manipulators. IEEE Transactions on Robotics, 30(3):619–630, 2014.
6. H. Khakpour, L. Birglen, and S.-A. Tahan. Analysis and optimization of a new
differentially driven cable parallel robot. Journal of Mechanisms and Robotics,
7(3):034503, 2015.
7. T. N. Le, H. Dobashi, and K. Nagai. Configuration of redundant drive wire mech-
anism using double actuator modules. ROBOMECH Journal, 3(1):25, 2016.
8. T. N. Le, H. Dobashi, and K. Nagai. Kinematical and static force analysis on
redundant drive wire mechanism with velocity constraint modules to reduce the
number of actuators. ROBOMECH Journal, 3(1):22, 2016.
9. S. Lessanibahri, P. Cardou, and S. Caro. Parasitic inclinations in cable-driven
parallel robots using cable loops. Procedia CIRP, 70:296–301, 2017.
10. S. Lessanibahri, P. Cardou, and S. Caro. Kinetostatic analysis of a simple cable-
driven parallel crane. In Proceedings of the ASME 2018 International Design Engi-
neering Technical Conferences & Computers and Information in Engineering Con-
ference IDETC/CIE 2018, Quebec city, Canada, August 28, 2018.
11. K. Nagai, T. N. Le, Y. Hayashi, and K. Ito. Kinematical analysis of redundant
drive wire mechanisms with velocity constraint. In Mechatronics and Automation
(ICMA), 2012 International Conference on, pages 1496–1501. IEEE, 2012.
12. A. Platis, T. Rasheed, P. Cardou, and S. Caro. Isotropic design of the spherical
wrist of a cable-driven parallel robot. In Advances in Robot Kinematics 2016, pages
321–330. Springer, 2018.
13. P. Racioppo, W. Saab, and P. Ben-Tzvi. Design and analysis of reduced degree of
freedom modular snake robot. In ASME 2017 International Design Engineering
Technical Conferences and Computers and Information in Engineering Conference,
pages V05BT08A009–V05BT08A009. American Society of Mechanical Engineers,
2017.
14. A. L. C. Ruiz, S. Caro, P. Cardou, and F. Guay. Arachnis: Analysis of robots
actuated by cables with handy and neat interface software. In Cable-Driven Parallel
Robots, pages 293–305. Springer, 2015.
aiiB+ MHvbBb Q7 hrQ@SHi7Q`K SHM`
*#H2@.`Bp2M S`HH2H _Q#Qi rBi? lMHBKBi2/
_QiiBQM
h?QKb _2B+?2M#+? - S?BHBTT h2KT2H- H2tM/2` o2`H- M/ M/`2b SQii
AMbiBimi2 7Q` *QMi`QH 1M;BM22`BM; 7Q` J+?BM2 hQQHb M/ JMm7+im`BM; lMBib UAaqV-
lMBp2`bBiv Q7 aimii;ì-
i?QKbX`2B+?2M#+?!BbrXmMB@bimii;ìX/2-
?iiT,ffrrrXBbrXmMB@bimii;ìX/2f
#bi`+iX *QKT`BM; +#H2@/`Bp2M T`HH2H `Q#Qib Ub?Qì *.S_b Q` +@

#H2 `Q#QibV rBi? +QMp2MiBQMH T`HH2H K2+?MBbKb- *.S_b ?p2 /pM@
i;2b BM i2`Kb Q7 ~2tB#BHBiv- /vMKB+b- M/ rQ`FbT+2 bBx2X AM ;2M2`H- i?2
`QiiBQMH +T#BHBiB2b Q7 T`HH2H K2+?MBbKb rBi?Qmi Mv //BiBQMH +@
imiQ` bvbi2K `2 HBKBi2/X h?Bb TT2` T`2b2Mib M TT`Q+? 7Q` i?2 /2bB;M
M/ MHvbBb Q7 *.S_ r?B+? HHQrb M mMHBKBi2/ `QiiBQM #Qmi QM2
tBb #v +imiBM; bQH2Hv +#H2bX h?2 mMHBKBi2/ `QiiBQM +QMbBbib Q7 `2H@
iBp2 TQbBiBQMBM; #2ir22M KmHiBTH2 THi7Q`KbX h?2 FBM2KiB+b- biiB+ 7Q`+2
/Bbi`B#miBQM- M/ i?2 rQ`FbT+2 Q7 THM` +#H2 `Q#Qi `2 MHvx2/X
7Q`KmHiBQM Q7 i?2 bi`m+im`2 Ki`Bt 7Q` +#H2 `Q#Qib rBi? KmHiBTH2 THi@
7Q`Kb Bb ;Bp2M- HHQrBM; iQ mb2 +QMp2MiBQMH H;Q`Bi?Kb 7Q` +H+mHiBQM Q7
7Q`+2 /Bbi`B#miBQMbX h?2 T2`7Q`K2/ bBKmHiBQM b?Qrb /Bz2`2Mi +?`+i2`@
BbiB+b Q7 i?2 7Q`+2 /Bbi`B#miBQM M/ rQ`FbT+2 Q7 i?Bb M2r ivT2 Q7 +#H2
`Q#Qi +QMi``v iQ +QMp2MiBQMH QM2bX 6BMHHv- i?2 +QM+HmbBQM b?Qrb i?i
i?2 /2ti`Qmb rQ`FbT+2b Q7 i?2 BMp2biB;i2/ *.S_ Bb MQM2KTivX
E2vrQ`/b, +#H2@/`Bp2M T`HH2H `Q#Qi- mMHBKBi2/ `QiiBQM- KmHiBTH2 THi@

7Q`Kb
R AMi`Q/m+iBQM
h?2 #2M2}ib Q7 +#H2@/`Bp2M T`HH2H `Q#Qib Ub?Qì *.S_b Q` +#H2 `Q#QibV `2
?B;? ~2tB#BHBiv- H`;2 rQ`FbT+2- M/ ?B;? /vMKB+bX h?Bb mM/2`HB2b HB;?ir2B;?i
+QKTQM2Mib HBF2 +#H2b- r?B+? +MƘbTM Qp2` H`;2 /BbiM+2b rBi? +QKT`#Hv HQr
KbbX h?2 TQbBiBQM M/ Q`B2MiiBQM Q7 i?2 2M/ 2z2+iQ` U11V Q` KQ#BH2 THi7Q`K
+M #2 +?M;2/ #v KMBTmHiBM; H2M;i?b Q7 +#H2b ;mB/2/ pB TmHH2vb M/ rQmM/
QM rBM+?2bX
>Qr2p2`- /m2 iQ i?2 BM?2`2Mi T`QT2ìB2b Q7 +#H2 #2BM; #H2 iQ QMHv i`MbKBi
i2MbBH2 7Q`+2b M/ /m2 iQ +QHHBbBQM BMi2`72`2M+2- i?2 iQiH Q`B2MiiBQM rQ`FbT+2 Q7
+#H2 `Q#Qi Bb HBKBi2/X ++Q`/BM; iQ :Qmii27`/2 2i HX UkyR8V- 72bB#H2 THi7Q`K
Q`B2MiiBQMb `2 ±45◦ #Qmi i?2 ?Q`BxQMiH t2b- Q` ±105◦ #Qmi i?2 p2ìB+H tBb-
`2[mB`BM;- ?Qr2p2`- bT2+BH /2bB;MX

122 T. Reichenbach et al.
H`;2` `QiiBQM +T#BHBiv Bb `2[mB`2/ 7Q` ?M/HBM; ibFb 2X;X- TB+F@M/@TH+2-

BMbT2+iBQM- Q` TBMiBM; TTHB+iBQMbX lT iQ MQr- mMHBKBi2/ `QiiBQM +M QMHv #2
+?B2p2/ #v ii+?BM; M //BiBQMH +imiQ` iQ i?2 KQ#BH2 THi7Q`KX 1bb2MiBHHv-
KFBM; i?2 `Q#Qi ?v#`B/ b2`BH@T`HH2H KMBTmHiQ`X h?Bb `2[mB`2b iFBM; i?2
//BiBQMH Kbb M/ K2/B bmTTHv BMiQ ++QmMi r?2M /2HBM; rBi? FBM2KiB+b
M/ /vMKB+b Q7 i?2 bvbi2KX 6mì?2`KQ`2- i?2 #2?pBQ` Q7 *.S_b Bb HBKBi2/ #v
HQr2` KtBKmK TvHQ/- HQr2` /vMKB+b- M/ bKHH2` rQ`FbT+2 +QKT`2/
iQ +#H2 `Q#Qi rBi?Qmi THi7Q`K@bB/2 +miiBM; bvbi2KX SQbbB#H2 +QHHBbBQMb
#2ir22M +#H2b- THi7Q`K- M/ K2/B bmTTHv Kmbi HbQ +QMbB/2`2/ /m`BM; /2bB;MX
hQ pQB/ i?2 /2b+`B#2/ /`r#+Fb Q7 M //BiBQMH +imiBM; bvbi2K QM i?2
THi7Q`K- +#H2@/`Bp2M mMHBKBi2/ `QiiBQM r2`2 7pQ`#H2X
6B`bi- a2+iBQM k ;Bp2b b?Qì bii2 Q7 i?2 ì 7Q` +#H2 `Q#Qib rBi? KmHiBTH2
THi7Q`Kb M/ QmiHBM2b i?2 `2b2`+? Bbbm2X h?2M- a2+iBQM j `2pBb2b i?2 FBM2KiB+b
Q7 bBM;H2 THi7Q`K *.S_- r?BH2 a2+iBQM 9 BMi`Q/m+2b THM` +#H2 `Q#Qi rBi?
irQ THi7Q`Kb DQBM2/ #v +`MFb?7i K2+?MBbKX q2 T`QpB/2 i?2 FBM2KiB+b
7Q`KmHiBQM M/ rQ`FbT+2 MHvbBb Q7 i?2 K2+?MBbK BM i?2 bK2 b2+iBQMX GbiHv-
a2+iBQM 8 +QM+Hm/2b i?2 +QMi`B#miBQM rBi? BMp2biB;i2/ `2bmHibX
k JmHiBTHi7Q`K *#H2 _Q#Qib
PM2 TQbbB#H2 bQHmiBQM 7Q` bm+? `QiiBQM@2M#HBM; K2+?MBbK rBi?Qmi //BiBQMH

+imiQ` bvbi2K Bb +`MFb?7i ;2QK2i`v b b?QrM BM 6B;m`2 RX "b2/ QM i?Bb
T`BM+BTH2- Bi Bb TQbbB#H2 iQ BKTH2K2Mi M mMHBKBi2/ `QiiBQM mbBM; QMHv +#H2b
rBi? bBM;H2 THi7Q`K b BMBiBHHv /2b+`B#2/ #v JB2`K2Bbi2` M/ SQii UkyR8V
M/ SQii M/ JB2`K2Bbi2` UkyR3VX qBi?BM i?2b2 BMBiBH TT`Q+?2b- irQ bTiBH
`Q#Qi /2bB;Mb rBi? irQ M/ i?`22 THi7Q`Kb- `2bT2+iBp2Hv- r2`2 ;Bp2M- rBi? i?2
bBKTHB}+iBQM Q7 +QKKQM +#H2 ii+?K2Mi TQBMi T2` M+?Q`X >2`2- i?2 7Q`+2
/Bbi`B#miBQM rb MHvx2/ mbBM; i?2 +HQb2/@7Q`K M/ .vFbi` K2i?Q/ 7Q` +QK@
TH2i2 `2pQHmiBQM i bT2+B}+ 11 TQbBiBQMX 6mì?2`KQ`2- i?2 r`2M+?@72bB#H2 iQiH
Q`B2MiiBQM rQ`FbT+2 rb BMp2biB;i2/ mbBM; i?2 ?mHH K2i?Q/X AM //BiBQM- ;2M@
2`HBx2/ KQ/2H Q7 KmHiBHBMF *.S_ mbBM; i?2 +#H2@`QmiBM; Ki`B+2 rb ;Bp2M
#v Gm 2i HX UkyRjVX
h?2 +?`+i2`BbiB+b Q7 i?2 MQM@;2M2`B+ +#H2 `Q#Qi /2bB;M- b?QrM BM 6B;m`2 R
7Q` 2tKTH2- +?M;2 /Bb+QMiBMmQmbHv B7 bT2+B}+ ;2QK2i`B+ +QM/BiBQMb `2 MQi 7mH@
}HH2/X AM i?Bb +b2- QM2 ;2QK2i`B+ +QM/BiBQM Bb i?2 +QKKQM +#H2 ii+?K2Mi
TQBMiX h?2 K2+?MB+H +QMbi`m+iBQM Q7 bQH2Hv QM2 +QKKQM +#H2 ii+?K2Mi TQBMi
QM i?2 THi7Q`K M+?Q`b Bb #`2Hv TQbbB#H2 pQB/BM; Mv +#H2@+#H2 M/ +#H2@
THi7Q`K +QHHBbBQMbX AM T`+iB+2- iQ HHQr 7Q` `2HiBp2 KQiBQM Q7 irQ #Q/B2b rBi?
`2bT2+i iQ 2+? Qi?2`- #2`BM; 2H2K2Mi M22/b iQ #2 //2/X
M 2tKTH2 Q7 #HH #2`BM; rBi? +#H2 ii+?K2Mi TQBMib r?B+? +M #2 mb2/
b #2`BM; 2H2K2Mi Bb b?QrM BM 6B;m`2 k rBi? i?2 +#H2 7Q`+2b M/ /B`2+iBQMb fj,i
7Q` i?2 i@i? +#H2 Q7 i?2 j@i? THi7Q`KX h?Bb BKTHB2b bQK2 bTiBH 2ti2Mi Q7 i?2
K2+?MB+H TB2+2 mHiBKi2Hv H2/BM; iQ i?2 +#H2b MQi T2`72+iHv BMi2`b2+iBM; BM
QM2 TQBMi #mi HQM; i?2 T`QD2+i2/ #2`BM; tBb c#2`-D Ub22 6B;m`2b k+ M/ k/VX
Static Analysis of a Two-Platform Planar Cable-Driven Parallel Robot… 123
UV *`MFb?7i ;2QK2i`v rBi? QM2 +QKKQM U#V *`MFb?7i ;2QK2i`v rBi? #2`BM; 2H@
+#H2 ii+?K2Mi TQBMi T2` M+?Q` QM 2K2Mib M/ KmHiBTH2 THi7Q`KbX
bBM;H2 THi7Q`KX
6B;X R, *QM+2Ti Q7 *.S_ rBi? +`MFb?7i ;2QK2i`v iQ +?B2p2 M mMHBKBi2/

`QiiBQMX h?2 +#H2 7Q`+2b fj,i `2 b?QrM BM `2/- r?2`2 j Bb i?2 THi7Q`K MmK#2`
M/ i i?2 i@i? +#H2 Q7 i?2 j@i? THi7Q`KX h?2 `272`2M+2 +QQ`/BMi2 bvbi2Kb Q7 i?2
+#H2 ii+?K2Mi TQBMib M/ THi7Q`Kb `2 /`rM #Hm2 rBi? i?2B` HQ+H +QQ`/BMi2
bvbi2Kb Kj - (j = 1, . . . , q) M/ i?2 2M/@2z2+iQ` +QQ`/BMi2 bvbi2K K11 X
>2M+2- 7mì?2` /2;`22b Q7 7`22/QK U.P6V BM i?2 7Q`K Q7 TbbBp2 DQBMib `2 //2/
iQ i?2 bvbi2KX
AM i?Bb +b2- i?2 bBM;H2 THi7Q`K Bb `2TH+2/ #v b2`BH +?BM Q7 KmHiBTH2
THi7Q`Kb r?B+? Bb b?QrM BM 6B;m`2 R#X h?2 mMHBKBi2/ `QiiBQM Q7 i?2 11 +M
#2 /2b+`B#2/ b `2HiBp2 KQiBQM Q7 KmHiBTH2 THi7Q`Kb- ?Qr2p2`- i?2 THi7Q`K
i?i Bb mb2/ iQ T2`7Q`K i?2 `2pQHmiBQM ?b iQ #2 2++2Mi`B+ iQ i?2 `QiiBQM b?7ib
tBbX *QMb2[m2MiHv- i?Bb BKTHB2b i?i i?2 ?v#`B/ `Q#Qi bi`m+im`2 b b2`BH +?BM
rBi?BM T`HH2H `Q#QiX
hQ mb2 i?2 /2bB;M Q7 *.S_ rBi? KmHiBTH2 THi7Q`Kb 7Q` 7Q`+2 /Bbi`B#miBQM-
rQ`FbT+2 MHvbBb M/ +QMi`QH H;Q`Bi?Kb- i?2 //BiBQMH .P6 ;2M2ì2/ #v i?2
THi7Q`Kb Kmbi #2 iF2M BMiQ ++QmMi BM i?2 FBM2KiB+b M/ 7Q`+2 /Bbi`B#miBQM
+H+mHiBQM H;Q`Bi?KbX AM i?Bb TT2`- r2 T`2b2Mi M TT`Q+? iQ KQ/2HBM; M/
MHvxBM; bBKTHB}2/ THM` +#H2 `Q#Qi b2imT rBi? M 2M/H2bb `QiiBM; tBb
rBi? +QMbB/2ìBQM Q7 `2pQHmi2 DQBMi rBi? i?2 +#H2 /B`2+iBQM BMi2`b2+iBM; BM
+QKKQM tBbX
UV *QM+2Ti rBi? #HH #2`BM;b b KmHiBTH2 U#V hQT pB2r Q7 #HH #2`BM; ii+?K2Mi
THi7Q`Kb
U+V G27i pB2r Q7 #HH #2`BM; ii+?K2Mi U/V _B;?i pB2r Q7 #HH #2`BM; ii+?K2Mi
6B;X k, 1tKTH2 Q7 THi7Q`K@bB/2 +#H2 ii+?K2Mi rBi? #HH #2`BM; M/

;`QKK2ib iQ KQp2 KmHiBTH2 THi7Q`Kb ;BMbi 2+? Qi?2`X
j AMp2`b2 EBM2KiB+b M/ aiiB+ 1[mBHB#`BmK Q7 *#H2

_Q#Qib rBi? aBM;H2 SHi7Q`K
AM ;2M2`H- +#H2 `Q#Qi KQp2 Bib THi7Q`K #v rBM/BM; M/ mMrBM/BM; Bib +#H2bX
"2+mb2 Q7 i?2 T`HH2H bi`m+im`2 M/ i?2 +#H2 #2?pBQ`- r?B+? QMHv HHQrb i2MbBH2
7Q`+2b- `2/mM/MiHv +QMbi`BM2/ *.S_ ?b iQ 7mH}HH i?2 2[miBQM n + 1 > m-
rBi? i?2 bvbi2K ?pBM; n .P6 M/ m +imi2/ +#H2b Uo2`?Q2p2M Ukyy9VVX h?2
BMp2`b2 FBM2KiB+b 7Q` i?2 i@i? +#H2 H2M;i? H2/b iQ
li = li = ai − r − R bi , URV
r?2`2 li Bb i?2 +QMi`QHH#H2 H2M;i? Q7 i?2 i@i? +#H2- ai i?2 rBM+? TQbBiBQM- r i?2
THi7Q`Kb 11 TQbBiBQM- R i?2 THi7Q`K `QiiBQM Ki`Bt- M/ bi i?2 THi7Q`K
M+?Q` BM HQ+H THi7Q`K +QQ`/BMi2bX LQ`KHBxBM; i?2 BMp2`b2 FBM2KiB+b 7Q` 2+?
+#H2 i rBi? ;Bp2M THi7Q`K TQb2 (r, R) ∈ a1(3) H2/b iQ i?2 bQ +HH2/ bi`m+im`2
Ki`Bt A - r?B+? /2b+`B#2b i?2 +#H2 /B`2+iBQM ui = llii M/ irBbi rBi? `2bT2+i
iQ `272`2M+2 +QQ`/BMi2 bvbi2K ++Q`/BM; iQ o2`?Q2p2M Ukyy9V M/ "`m+FKMM
2i HX Ukyy3VX
u1 · · ·

um
A = , UkV
b01 × u1 · · · bm × um
0
r?2`2 b0i = R bi X
h?2 bi`m+im`2 Ki`Bt A +M #2 mb2/ 7Q` L2riQM@1mH2` 1[miBQMb Q7 biiB+
bvbi2K
A f + wT = 0 , UjV
r?2`2 wT Bb i?2 p2+iQ` Q7 TTHB2/ r`2M+?2b QM i?2 THi7Q`K M/ f Bb i?2 p2+iQ`
Q7 i?2 MQKBMH +#H2 7Q`+2bX >Qr2p2`- B7 i?2 bvbi2K Bb `2/mM/MiHv +QMbi`BM2/-
1[miBQM UjV Bb mM/2`@/2i2`KBM2/X h?2 biiB+ 2[mBHB#`BmK BM 1[miBQM UjV +M #2
/2b+`B#2/ b Lp @MQ`K QTiBKBxiBQM T`Q#H2KX h?2 biiB+ 7Q`+2 /Bbi`B#miBQM Q7 i?2
+#H2 7Q`+2b `2 +H+mHi2/ rBi? QTiBKBxiBQM H;Q`Bi?Kb BX2X- HBM2` M/ MQMHBM2`
T`Q;`KKBM; Ua?BM; 2i HX UkyyyVV- [m/ìB+ T`Q;`KKBM; UGB 2i HX UkyRjVV-
o2`?Q2p2MǶb :`/B2Mi J2i?Q/ Uo2`?Q2p2M Ukyy9VV- .vFbi` J2i?Q/ U>bbM M/
E?D2TQm` UkyydVV M/ *HQb2/@6Q`K M/ AKT`Qp2/ *HQb2/@6Q`K J2i?Q/ USQii
2i HX UkyyNVc SQii UkyR3VVX
h?2 ;2QK2i`B+ /2b+`BTiBQM Q7 i?2 rQ`FbT+2b Bb /2}M2/ #v J2`H2i UkyyeV
M/ :Qmii27`/2 2i HX UkyydV- r?2`2 i?2 rQ`FbT+2 Q7 T`HH2H `Q#Qib Bb +QK@
KQMHv /2b+`B#2/ b i?`22@/BK2MbBQMH bm#b2i Q7 i?2 ;2M2`HHv bBt@/BK2MbBQMH
rQ`FbT+2X PM2 /BbiBM;mBb?2b #2ir22M i?2 i`MbHiBQM- Q`B2MiiBQM- BM+HmbBQM Q`B@
2MiiBQM- KtBKmK- iQiH Q`B2MiiBQM- M/ /2ti2`Qmb rQ`FbT+2X >Qr2p2`- i?2
//BiBQMH .P6 BMi`Q/m+2/ #v i?2 `2pQHmi2 DQBMib Q7 i?2 *.S_ rBi? KmHiBTH2
THi7Q`Kb M/ +`MFb?7i 2MH`;2 i?2 ;2M2`HHv bBt@/BK2MbBQMH QT2ìBQM bT+2
iQ ?B;?2` /BK2MbBQMH r?B+? /2T2M/b QM i?2 MmK#2` Q7 `2pQHmi2 DQBMibX AM Qm`
+b2 Q7 THM` +#H2 `Q#Qi rBi? irQ THi7Q`Kb- i?2 iQiH Q`B2MiiBQM rQ`FbT+2
Bb i?`22@/BK2MbBQMH bm#bT+2 r?B+? +M #2 pBbmHBx2/ bi`B;?i 7Q`r`/X
9 SHM` *#H2 _Q#Qi rBi? hrQ SHi7Q`Kb M/ lMHBKBi2/

_QiiBQM
*QMbB/2`BM; KQ`2 i?M QM2 THi7Q`K 7Q` *.S_- #Qi? THi7Q`Kb FBM2KiB+HHv
/2T2M/ QM 2+? Qi?2` rBi? M //BiBQMH +QMbi`BMiX THM` /2bB;M Q7 *.S_
rBi? KmHiBTH2 THi7Q`Kb M/ M mMHBKBi2/ `QiiBQM Bb b?QrM BM 6B;m`2 jX h?2 ;2@
QK2i`v Q7 i?2 `Q#Qi Bb b?QrM BM 6B;m`2 j b +QM+2Ti /`rBM;- r?BH2 6B;m`2 j#
b?Qrb i?2 KQ/2H2/ `Q#Qi bi`m+im`2 b THQiX SHi7Q`K A rBi? i?2 +QQ`/BMi2
bvbi2K KA Bb +QMi`QHH#H2 rBi? i?2 +#H2 H2M;i? l,i /`rM `2/X PM2 +M /2b+`B#2
i?2 BMp2`b2 FBM2KiB+b 7Q` 2+? +#H2 Q7 THi7Q`K A mbBM; 1[miBQM URVX SHi@
7Q`K B rBi? +QQ`/BMi2 bvbi2K KB Bb HBMF2/ iQ THi7Q`K A #v }t2/ /BbiM+2
k = k = k RΦ (ϕΦ ) ex - 7Q` 2tKTH2 rBi? HBMF;2 M/ `2pQHmi2 DQBMib BM i?2
+2Mi2` Q7 Kbb Q7 2+? THi7Q`KX aBKBH`Hv iQ THi7Q`K A- THi7Q`K B +M #2
+QMi`QHH2/ rBi? i?2 +#H2 H2M;i? lB,i X 6Q` 2tKTH2- THi7Q`K B +M #2 THM`
KQ/2H Q7 #HH #2`BM;- b b?QrM BM 6B;m`2 kX
UV :2QK2i`v Q7 i?2 THM` *.S_X
U#V ai`m+im`2 JhG" THQi Q7 i?2 THM` *.S_X
6B;X j, *QM+2Ti Q7 THM` *.S_ rBi? mMHBKBi2/ `QiiBQM M2;H2+iBM; +#H2 M/
THi7Q`K +QHHBbBQMbX
>2M+2- QM2 +M +QMi`QH THi7Q`K A rBi? i?2 `2/ +#H2b M/ +M +QMi`QH THi@
7Q`K B rBi? i?2 #Hm2Ƙ+#H2bX HbQ- #Qi? THi7Q`K A M/ THi7Q`K B +M `Qii2
#Qmi i?2 Q`B;BM Q7 i?2 +QQ`/BMi2 bvbi2Kb rBi? i?2 M;H2b ϕ M/ ϕ" rBi? i?2
`QiiBQM Ki`B+2b R (ϕ ) M/ R" (ϕ" )- `2bT2+iBp2HvX "2+mb2 Q7 i?2 }t2/ /Bb@
iM+2 k #2ir22M i?2 irQ THi7Q`Kb- THi7Q`K B +M QMHv KQp2 +B`+mH`Hv #Qmi
THi7Q`K AX h?Bb i`D2+iQ`v Bb /`rM /b?2/ BM 6B;m`2 j- rBi? i?2 M;H2 ϕΦ
7Q` i?2 `QiiBQM Ki`Bt RΦ (ϕΦ )X 6mì?2`KQ`2- i?2 11 TQbBiBQM Bb /2}M2/ BM
i?2 Q`B;BM Q7 +QQ`/BMi2 bvbi2K KA X 6Q` i?2 THM` +b2- +#H2@THi7Q`K- +#H2@
+#H2- M/ THi7Q`K@THi7Q`K +QHHBbBQMb `2 M2;H2+i2/X 6QHHQrBM;- Bi Bb bbmK2/
i?i 2+? THi7Q`K +QKT2Mbi2b Bib Kbb Q` HQ/- `2bT2+iBp2Hv- ?2M+2 /vMKB+
+QmTHBM; 7Q`+2b rBHH HbQ #2 M2;H2+i2/ M/ MQ //BiBQMH 2ti2`MH r`2M+?2b `2
TTHB2/ iQ i?2 11X h?2 HBMF;2 #2ir22M i?2 THi7Q`Kb Bb KQ/2H2/ rBi?Qmi KbbX
h?2 FBM2KiB+ KQ/2HBM;- 7Q`+2 /Bbi`B#miBQM +H+mHiBQMb M/ rQ`FbT+2 MHvbBb

Bb T2`7Q`K2/ rBi? JhG" kyR3X
9XR EBM2KiB+b
++Q`/BM; iQ 1[miBQM URV- i?2 +QMi`QHH#H2 +#H2 H2M;i? lB,i 7Q` 2+? +#H2 i Q7
THi7Q`K B +M #2 /2b+`B#2/ b
l",i = di − r − R" ei − RΦ k , U9V
r?2`2 di Bb i?2 /BbiM+2 #2ir22M i?2 `272`2M+2 +QQ`/BMi2 bvbi2K K0 BM bT+2
M/ i?2 +#H2 QmiH2i TQBMi Q7 THi7Q`K B- ei i?2 /BbiM+2 #2ir22M i?2 `272`2M+2
+QQ`/BMi2 bvbi2K KB M/ i?2 /BbiH ii+?K2Mi TQBMi Q7 THi7Q`K B- k i?2 HBMF
#2ir22M THi7Q`K A M/ THi7Q`K B M/ RB i?2 Q`B2MiiBQM Q7 THi7Q`K BX h?2
mMHBKBi2/ `QiiBQM Q7 i?2 HBMF k Bb +mb2/ #v i?2 `2HiBp2 KQp2K2Mi #2ir22M THi@
7Q`K A M/ THi7Q`K BX PM2 +M /2b+`B#2 i?2 bi`m+im`2 Ki`Bt B Q7 THi7Q`K B
b
u",1 · · ·

u",m
B = 0 , U8V
e1 × u",1 · · · e0m × u",m
l
r?2`2 e0i = R" ei - u",i = l",i i?2 /B`2+iBQM Q7 i@i? +#H2 Q7 THi7Q`K BX
",i
>2M+2- i?2 +#H2 /B`2+iBQM Q7 i?2 THi7Q`Kb `2 /2i2`KBM2/ rBi? i?2 bi`m+im`2
Ki`B+2b A - 1[miBQM UkV- M/ B - 1[miBQM U8V- i?2 bi`m+im`2 Ki`Bt Q7 i?2
2MiB`2 THM` KmHiBTH2 THi7Q`K bvbi2K D +M #2 r`Bii2M b

A 0

D =
0 B
u,1 · · · u,m 0 ··· 0
⎡ ⎤
UeV
⎢b0 × u,1 · · · b0m × u,m 0 ··· 0
=⎣ 1 ⎦.
⎥
0 ··· 0 u",1 ··· u",m
0 ··· 0 e01 × u",1 · · · e0m × u",m
h?2 ;2M2`HBx2/ +QQ`/BMi2b Q7 i?2 THM` *.S_ rBi? KmHiBTH2 THi7Q`Kb

M/ mMHBKBi2/ `QiiBQMb `2, r - R (ϕ )- R" (ϕ" )- RΦ (ϕΦ )X h?mb- bvbi2K ?b 8
.P6X
9Xk 6Q`+2 .Bbi`B#miBQM MHvbBb

++Q`/BM; iQ 1[miBQM UjV- i?2 L2riQM@1mH2`@1[miBQM 7Q` i?2 biiB+ 2[mBHB#`BmK
Bb
f,1
⎡ ⎤
⎢ ··· ⎥
⎢f,m ⎥
⎢ ⎥
w T,
⎥ + w T, " = 0 ,
⎢ ⎥
D ⎢ UdV
⎢ f",1 ⎥
⎣ ··· ⎦
wT,.
f",m
f.
r?2`2 f,i Bb i?2 b+H` pHm2 Q7 i?2 i@i? +#H2 7Q`+2 QM THi7Q`K A M/ f",i i?2
b+H` pHm2 Q7 i?2 i@i? +#H2 7Q`+2 QM THi7Q`K BX 6Q` i?2 THM` *.S_- b?QrM
BM 6B;m`2 j- i?2 MmK#2` Q7 +#H2b Bb mj = 4 7Q` i?2 j@i? THi7Q`KX h?2 TTHB2/
r`2M+? wT,. +QMbBbib Q7 i?2 irQ r`2M+?2b wT, M/ wT, " Q7 2+? THi7Q`KX 6Q`
i?2 +#H2 7Q`+2 +H+mHiBQMb- QMHv i?2 Kbb Q7 i?2 THi7Q`Kb Bb iF2M BMiQ ++QmMiX
>2M+2- i?2 biiB+ 2[mBHB#`BmK +M #2 bQHp2/ rBi? i?2 H;Q`Bi?Kb bmKK`Bx2/
BM a2+iBQM kX M 1tKTH2 Q7 i?2 7Q`+2 /Bbi`B#miBQM 7Q` THM` *.S_ rBi? irQ
THi7Q`Kb M/ irQ TQbBiBQMb M/ Q`B2MiiBQMb `2 b?QrM BM 6B;m`2 9X h?2 7Q`+2
/Bbi`B#miBQM 7Q` 2+? +#H2 rBi? `2bT2+i iQ i?2 `QiiBQM #2ir22M THi7Q`K A M/

THi7Q`K B rBi? ϕΦ = 0◦ iQ 360◦ i i?2 +2Mi2` TQbBiBQMb r,1 = 0, 0 M/

rBi? i?2 THi7Q`K Q`B2MiiBQMb ϕ,1 = 0◦ - ϕ",1 = 0◦ M/

r,2 = 0, 0.75
ϕ,2 = 3 - ϕ",2 = −3◦ `2 +QKTmi2/X 6Q` +#H2 7Q`+2 +H+mHiBQMb- i?2 BKT`Qp2/
◦
+HQb2/@7Q`K USQii- kyR3- TX N9V QM i?2 TQb2b ;Bp2M M/ 7Q` QM2 7mHH `2pQHmiBQM Q7
THi7Q`K B #Qmi THi7Q`K A Bb mb2/X h?2 `2pQHmiBQM Q7 B Bb /Bb+`2iBx2/ rBi?
`2bT2+i iQ A BMiQ 33 bi2Tb- 7Q` 2+? bi2T i?2 7Q`+2 /Bbi`B#miBQM Bb bQHp2/X 6Q`+2
HBKBib M/ Kbb T`K2i2`b iQ +H+mHi2 i?2 r`2M+?2b wT, M/ wT," `2 HBbi2/
BM h#H2 RX
h#H2 R, *#H2 7Q`+2 HBKBib M/ BM2ìBH T`K2i2`b mb2/ 7Q` rQ`FbT+2 +H+mH@
iBQMX
S`K2i2` oHm2 lMBi
fKBM 5 N
fKt 500 N
m 20 kg
m" 10 kg
qBi? i?2 *HQb2/@6Q`K J2i?Q/- SQii 2i HX UkyyNV- i?2 +#H2 7Q`+2b `2 +QK@
Tmi2/ mbBM;
f. = fJ − D + (wT,. + D fM ) , U3V
r?2`2 f. Bb i?2 +QHmKM p2+iQ` Q7 i?2 +#H2 7Q`+2b f M/ f" - fJ i?2 p2`;2
72bB#H2 7Q`+2 p2+iQ` rBi? fJ,i = 12 (fKBM + fKt ) M/ D + = D(D D)−1 i?2
JQQ`2@S2M`Qb2 ;2M2`HBx2/ Ki`Bt BMp2`b2 Q7 i?2 bi`m+im`2 Ki`Bt D X
h?2 7Q`+2 /Bbi`B#miBQMb BM 6B;m`2 9 b?Qr i?2 7Q`+2b f Q7 THi7Q`K A 7QHHQrb
HBM2- #2+mb2 i?2 +#H2b `2 +QKT2MbiBM; 7Q` ;`pBiiBQMH 7Q`+2b Q7 i?2 THi7Q`K
Kbb r?BH2 THi7Q`K A Bb bi2/vX h?2 +#H2 7Q`+2b Q7 THi7Q`K B `2 T2`BQ/B+-
`2bmHiBM; 7`QK i?2 7mHH `2pQHmiBQMX >Qr2p2`- i?Bb +#H2 `Q#Qi /2bB;M ?b 3 +#H2b
M/ 8 .P6X h?mb- i?2 /BK2MbBQM Q7 MmHHbT+2 Q7 i?2 +#H2 7Q`+2b Bb jX qBi?
1[miBQM UeV- i?2 /BK2MbBQM Q7 MmHHbT+2 2[mHb k #2+mb2 Q7 M2;H2+iBM; i?2
+QmTHBM; 7Q`+2bX
6B;X 9, 6Q`+2 /Bbi`B#miBQM Q7 i?2 *.S_ 2tKTH2 7Q` M mMHBKBi2/ `QiiBQMc f,i
M/ f",i `2 i?2 +#H2 7Q`+2b rBi? `2bT2+i iQ +#H2b l,i M/ l",i b?QrM BM 6B;m`2 jX
9Xj qQ`FbT+2 MHvbBb
6Q` i?2 rQ`FbT+2 MHvbBb Q7 i?2 THM` *.S_ rBi? mMHBKBi2/ `QiiBQM- i?2
b2`+? `2;BQM Q7 TQbbB#H2 TQb2b Bb /Bb+`2iBx2/X >2`2- HH TQbbB#H2 +QK#BMiBQMb Q7
TQb2 M/ Q`B2MiiBQMb `2 biQ`2/ BM HBbi Q7 TQb2bX "2+mb2 Q7 i?2 }p2 ;2M2`HBx2/
+QQ`/BMi2b r - R (ϕ )- R" (ϕ" ) M/ RΦ - i?2 TQb2 HBbi +QMbBbib Q7 nqbi2Tb b2i Q7
TQbBiBQMb M/ Q`B2MiiBQMbX >2`2- nbi2Tb Bb i?2 MmK#2` Q7 /Bb+`2iBxiBQM TQBMib
BM i?2 b2`+?BM; `2 M/ q i?2 MmK#2` Q7 ;2M2`HBx2/ +QQ`/BMi2bX h?mb- i?2
rQ`FbT+2 rBHH #2 b?QrM b /Bb+`2i2 7Q`K rBi? nbi2Tb MmK#2` Q7 ;`B/ TQBMibX
h?2 T`K2i2`b mb2/ iQ MHvx2 i?2 rQ`FbT+2 `2 HBbi2/ BM h#H2 kX
;BM- 7Q` 2+? TQbBiBQM M/ Q`B2MiiBQM i?2 BKT`Qp2/ +HQb2/@7Q`K SQii UkyR3-
TX N9V Bb 2pHmi2/ iQ +QKTmi2 i?2 7Q`+2 /Bbi`B#miBQM 7`QK i?2 biiB+ 2[mBHB#`BmK
BM 1[miBQM UdVX HH TQbBiBQMb M/ Q`B2MiiBQMb r?B+? pBQHi2 i?2 +#H2 7Q`+2 HBKBib
`2 MQi BM i?2 `2+?#H2 rQ`FbT+2 Q7 i?2 *.S_X h?2 rQ`FbT+2 Bb pBbmHBx2/
rBi?BM i?2 `Q#Qi 7`K2 +QMbB/2`BM; i?2 7QHHQrBM; +?`+i2`BbiB+b Ub22 6B;m`2 eVX
>2`2- pHB/ TQb2b `2 TBMi2/ v2HHQr- HB;?i M/ /`F ;`22MX h?2 v2HHQr TBMi2/
`2 b?Qrb i?2 KtBKmK rQ`FbT+2 Q7 i?2 TQbBiBQM r (x, z) rBi? Mv 11 M/
THi7Q`K Q`B2MiiBQMb URΦ - R M/ R" VX >Qr2p2`- i?2`2 KB;?i #2 TQbBiBQMb
r r?2`2 bQK2 Q`B2MiiBQMb Q7 i?2 THi7Q`K TQb2 R - R" - RΦ `2 BMpHB/ M/
UV SHi7Q`K Q`B2MiiBQM 7Q` HB;?i ;`22M M/ /`F ;`22M rQ`FbT+2 b2i iQ ϕ = 0◦ M/
ϕ" = 0◦ - `2bT2+iBp2HvX
U#V SHi7Q`K Q`B2MiiBQMb 7Q` i?2 HB;?i M/ /`F ;`22M rQ`FbT+2 b2i iQ ϕ = 3◦ M/
ϕ" = −3◦ - `2bT2+iBp2HvX
6B;X 8, h`MbHiBQMH rQ`FbT+2 rBi? BMpHB/ TQb2b `2/- pHB/ TQb2b v2HHQr M/
;`22MX u2HHQr rQ`FbT+2 Bb KtBKmK rQ`FbT+2 rBi? R - R" - RΦ X GB;?i ;`22M
Bb i?2 BM+HmbBQM Q`B2MiiBQM rQ`FbT+2 M/ HHQrb Mv Q`B2MiiBQM RΦ #2ir22M
THi7Q`Kb A M/ BX
h#H2 k, S`K2i`BxiBQM Q7 +#H2 `Q#Qi mb2/ 7Q` TQb2 HBbi ;2M2ìBQMX

S`K2i2` oHm2
nbi2Tb 33
q 5
r,x −2.4 m iQ 2.4 m
r,z −2.4 m iQ 2.4 m
ϕ −45◦ iQ 45◦
ϕ" −45◦ iQ 45◦
ϕΦ 0◦ iQ 360◦
MQi `2HiBM; iQ i?2 72bB#H2 rQ`FbT+2X h?2 /`F ;`22M `2 b?Qrb i?2 iQiH
Q`B2MiiBQM rQ`FbT+2 Q7 i?2 *.S_ r?2`2 #Qi? Q`B2MiiBQMb Q7 THi7Q`K A M/
THi7Q`K B `2 }t2/ M/ mMHBKBi2/ `QiiBQM Bb TQbbB#H2X
qBi?BM i?2 HB;?i ;`22M rQ`FbT+2- THi7Q`K Q`B2MiiBQMb R M/ R" `2 }t2/
iQ bT2+B}+ pHm2 M/ i?2 11 +M `2+? 2+? TQb2 rBi? Mv 11 Q`B2MiiBQM RΦ X
h?Bb rQ`FbT+2 2t2KTHB}2b i?2 /2ti`Qmb rQ`FbT+2 r?B+? Bb bT2+BH FBM/ Q7 i?2
iQiH Q`B2MiiBQM rQ`FbT+2X AM //BiBQM- i?2 /`F ;`22M rQ`FbT+2 Bb bm#b2i Q7
i?2 HB;?i ;`22M- r?B+? BM im`M Bb bm#b2i Q7 i?2 v2HHQr rQ`FbT+2bX *QKT`BM;
6B;m`2 8 rBi? 6B;m`2 8#- i?2 iQiH Q`B2MiiBQM rQ`FbT+2 U/`F ;`22MV +?M;2b
rBi? i?2 THi7Q`K Q`B2MiiBQMb R M/ R" X
AM //BiBQM iQ i?2 irQ /BK2MbBQMH pBbmHBxiBQM Q7 i?2 rQ`FbT+2 b?QrM BM
6B;m`2 8X h?2 iQiH Q`B2MiiBQM rQ`FbT+2 Q7 i?2 TQbBiBQM r (x, z) rBi? `2bT2+i
iQ i?2 THi7Q`K Q`B2MiiBQM R (ϕ ) Q` R" (ϕ" ) Bb b?QrM BM i?`22@/BK2MbBQMH
;`T? Ub22 6B;m`2 eVX >2`2- i?2 rQ`FbT+2 ?mHH rb ;2M2ì2/ rBi? i?2 #QmM/`v
TQBMib Q7 i?2 iQiH Q`B2MiiBQM rQ`FbT+2 M/ i`BM;mHiBQMX Ai Bb bbmK2/ i?i
pHB/ TQb2b rBi? TQbbB#H2 mMHBKBi2/ `QiiBQM rBi? i?2 ;Bp2M THi7Q`K Q`B2M@
iiBQMb R (ϕ ) M/ R" (ϕ" ) rBHH #2 BMbB/2 i?Bb ?mHHX >Qr2p2`- i?2`2 KB;?i #2
TQbBiBQMb M/ Q`B2MiiBQMb r?B+? `2 MQi `2+?#H2 BMbB/2 i?2 ?mHH #2ir22M i?2
/Bb+`2iBxiBQM TQBMibX h?Bb FBM/ Q7 rQ`FbT+2 ;2QK2i`v Bb /2b+`B#2/ b bi`BTX
h?2 b?QrM ;`T? +M #2 BMi2`T`2i2/ b i?2 i`MbHiBQMH `QiiBQMH +T#BHBiv 7Q`
i?2 irQ THi7Q`Kb r?2`2 i?2 *.S_ +M T2`7Q`K +QKTH2i2 `2pQHmiBQMX 6Q` 2t@

KTH2- bbmKBM; i?2 11 TQbBiBQM r = −1.5, −1.5 - i?2 `QiiBQM +T#BHBiB2b Q7

THi7Q`K A Bb bKHH2` +QKT`2/ iQ i?2 `QiiBQM +T#BHBiB2b i TQb2 r = 0, 0 X

>Qr2p2`- biBHH #Qi? 11 TQbBiBQMb HHQr M mMHBKBi2/ `QiiBQMX "2+mb2 i?Bb Bb

THM` +QM+2Ti- +#H2@+#H2- +#H2@THi7Q`K M/ THi7Q`K@THi7Q`K BMi2`72`2M+2
Kmbi #2 M2;H2+i2/X bbmKBM; bTiBH /2bB;M- i?2b2 BMi2`72`2M+2 +QmH/ `2/m+2
i?2 iQiH Q`B2MiiBQM rQ`FbT+2X >Qr2p2`- +`MFb?7i THi7Q`K /2bB;M r?B+?
Bb b?QrM BM 6B;m`2 R M/ 6B;m`2 k Bb bbmK2/ iQ ?p2 KBMBKH MmK#2` Q7
+QHHBbBQMb Q` BMi2`72`2M+2- `2bT2+iBp2HvX 6mì?2` `2b2`+? rBi? `2bT2+i iQ +QHHBbBQM
BMi2`72`2M+2 Bb biBHH BM T`Q;`2bbX }`bi TT`Q+? rb K2MiBQM2/ BM SQii M/
JB2`K2Bbi2` UkyR3VX
UV SQbBiBQM r (x, z) rBi? `2bT2+i iQ i?2 U#V SQbBiBQM r (x, z) rBi? `2bT2+i iQ i?2
THi7Q`K Q`B2MiiBQM R (ϕ ) M/ Mv ϕ" X THi7Q`K Q`B2MiiBQM R" (ϕ" ) M/ Mv ϕ X
6B;X e, qQ`FbT+2 ?mHH rBi? `2bT2+i iQ i?2 11 TQbBiBQM M/ THi7Q`K Q`B2MiiBQMb
BM i?`22 /BK2MbBQMH THQiX qBi?BM i?2 ?mHH Bi Bb bbmK2/ i?i 7mHH `QiiBQM Bb
TQbbB#H2X
8 *QM+HmbBQM
q2 T`2b2Mi i?2 KtBKmK- BM+HmbBQM Q`B2MiiBQM- M/ iQiH Q`B2MiiBQM rQ`FbT+2
Q7 THM` 1_2h 8@+#H2 `Q#Qi rBi? irQ THi7Q`Kb HBMF2/ i?`Qm;? HBMF;2
K2+?MBbKX hQ +?B2p2 i?Bb- i?2 FBM2KiB+b `2 /2}M2/ M/ bQHp2/ 7Q` i?2 biiB+b
+b2 vB2H/BM; i?2 7Q`+2 /Bbi`B#miBQM 7Q` ;Bp2M mMHBKBi2/ `QiiBQM TQb2X
h?2M- pHB/ 7Q`+2 /Bbi`B#miBQMb `2 +QKTmi2/ mbBM; i?2 BKT`Qp2/ +HQb2/@7Q`K
H;Q`Bi?K M/ 2tKTH2b Q7 i?2 7Q`+2 /Bbi`B#miBQM Q7 irQ /Bz2`2Mi TQb2b rBi? mM@
HBKBi2/ `QiiBQM `2 T`2b2Mi2/X Ai im`Mb Qmi i?i i?2 +#H2 7Q`+2b Q7 i?2 #b2
THi7Q`K A +QKT2Mbi2 Bib Kbb bQH2Hv- M/ i?2 +#H2 7Q`+2b Q7 THi7Q`K B +QK@
T2Mbi2 7Q` ;`pBiiBQMH 7Q`+2b M/ +`2i2 i?2 +B`+mH` KQiBQMX
lbBM; i?2 2ti2M/2/ 7Q`KmHiBQM 7Q` i?2 7Q`+2 /Bbi`B#miBQM- r2 BMi`Q/m+2/
i?2 /2ti`Qmb rQ`FbT+2 b bT2+BH FBM/ Q7 i?2 iQiH Q`B2MiiBQM rQ`FbT+2
TTHB+#H2 iQ i?Bb M2r ivT2 Q7 +#H2 `Q#QiX b `2bmHi- i?Bb }`bi TT`Q+? `2p2Hb
i?i i?2 /2ti`Qmb rQ`FbT+2 Q7 *.S_ Bb MQM2KTivX Ai +M #2 BM72``2/ i?i
i?Bb ivT2 Q7 rQ`FbT+2 ?b [mBi2 /Bz2`2Mi +?`+i2`BbiB+b iQ i?2 rQ`FbT+2 Q7
+QMp2MiBQMH +#H2 `Q#Qib M/ Bib b?T2 Bb ì?2` mMBMimBiBp2X
AM i?2 7mim`2- r2 rBHH }M/ 7mì?2` THM` *.S_b bi`m+im`2b rBi? mMHBKBi2/
`QiiBQMX 6mì?2`KQ`2- i?2 +QmTHBM; 7Q`+2b #2ir22M i?2 THi7Q`Kb rBHH #2 +QM@
bB/2`2/ r?B+? T`Q##Hv 2MH`;2 i?2 iQiH Q`B2MiiBQM rQ`FbT+2 #2+mb2 Q7 i?2
//BiBQMH /BK2MbBQM Q7 MmHHbT+2X HbQ- KQ/2Hb Q7 bTiBH *.S_b rBHH #2 /2`Bp2/
+QMbB/2`BM; i?2 //BiBQMH .P6bX h?2 mi?Q`b `2K`F- i?i i?2 TT2` /2b+`B#2
bBKTHB}2/ THM` +QM+2Ti r?B+? +MMQi `2HBx2/ b T`QiQivT2- #2+mb2 Q7 +#H2@
+#H2- +#H2@THi7Q`K M/ THi7Q`K@THi7Q`K +QHHBbBQMbX h?2 ;QH Q7 i?Bb rQ`F Bb
iQ 2ti2M/ i?2 FBM2iQbiiB+ /2b+`BTiBQMb 7Q` bTiBH KmHiBTHi7Q`K +#H2@/`Bp2M
T`HH2H `Q#Qi U*.S_V rBi? mMHBKBi2/ `QiiBQMX
+FMQrH2/;K2Mib h?Bb rQ`F rb bmTTQì2/ #v i?2 :2`KM _2b2`+? 6QmM@

/iBQM U.6:@T`QD2+i MmK#2`, j83R9kdyRV i i?2 lMBp2`bBiv Q7 aimii;ìX
_272`2M+2b
"`m+FKMM- hX- JBF2HbQMb- GX- "`M/i- hX- >BHH2`- JX- a+?`KK- .X, qB`2 `Q#Qib Tì A,
EBM2KiB+b- MHvbBb /2bB;MX AM, _vm- CX>X U2/XV S`HH2H JMBTmHiQ`b- L2r .2p2HQT@
K2MibX A@h2+? 1/m+iBQM M/ Sm#HBb?BM; Ukyy3V
:Qmii27`/2- JX- *QHH`/- CX6X- _B2?H- LX- "`/i- *X, :2QK2i`v b2H2+iBQM Q7 `2/mM/MiHv
+imi2/ +#H2@bmbT2M/2/ T`HH2H `Q#QiX A111 h`Mb+iBQMb QM _Q#QiB+b jRUkV- 8yRĜ8Ry
UkyR8V
:Qmii27`/2- JX- J2`H2i- CXSX- .M2v- .X, q`2M+?@72bB#H2 rQ`FbT+2 Q7 T`HH2H +#H2@
/`Bp2M K2+?MBbKbX AM, A111 AMi2`MiBQMH *QM72`2M+2 QM _Q#QiB+b M/ miQKiBQM-
TTX R9NkĜR9Nd UkyydV
>bbM- JX- E?D2TQm`- X, JBMBKmK@MQ`K bQHmiBQM 7Q` i?2 +imiQ` 7Q`+2b BM +#H2@#b2/
T`HH2H KMBTmHiQ`b #b2/ QM +QMp2t QTiBKBxiBQMX AM, A111 AMi2`MiBQMH *QM72`2M+2
QM _Q#QiB+b M/ miQKiBQM- TTX R9N3ĜR8yj UkyydV
Gm- .X- P2iQKQ- .X- >H;Km;2- aXEX, :2M2`HBx2/ KQ/2HBM; Q7 KmHiBHBMF +#H2@/`Bp2M
KMBTmHiQ`b rBi? `#Bi``v `QmiBM; mbBM; i?2 +#H2@`QmiBM; Ki`BtX A111 h`Mb+iBQMb
QM _Q#QiB+b kNU8V- RRykĜRRRj UkyRjV
GB- >X- w?M;- sX- uQ- _X- amM- CX- SM- :X- w?m- qX, PTiBKH 6Q`+2 .Bbi`B#miBQM "b2/ QM
aH+F _QT2 JQ/2H BM i?2 AM+QKTH2i2Hv *QMbi`BM2/ *#H2@.`Bp2M S`HH2H J2+?MBbK Q7
6ah h2H2b+QT2- TTX 3dĜRykX aT`BM;2` "2`HBM >2B/2H#2`; UkyRjV
J2`H2i- CXSX, S`HH2H `Q#QibX kX 2/MX aT`BM;2`- .Q`/`2+?i UkyyeV
JB2`K2Bbi2`- SX- SQii- X, .2bB;M Q7 +#H2@/`Bp2M T`HH2H `Q#Qib rBi? KmHiBTH2 THi7Q`Kb
M/ 2M/H2bb `QiiBM; t2bX AM, E2+bF2Kûi?v- X- :2m 6HQ`2b- 6X U2/bXV AMi2`/Bb+BTHBM`v
TTHB+iBQMb Q7 EBM2KiB+b- TTX kRĜkNX aT`BM;2` AMi2`MiBQMH Sm#HBb?BM;- *?K UkyR8V
SQii- X, *#H2@.`Bp2M S`HH2H _Q#Qib- pQHX RkyX aT`BM;2` AMi2`MiBQMH Sm#HBb?BM;- *?K
UkyR3V
SQii- X- "`m+FKMM- hX- JBF2HbQMb- GX, *HQb2/@7Q`K 7Q`+2 /Bbi`B#miBQM 7Q` T`HH2H rB`2
`Q#QibX AM, E2+bF2Kûi?v- X- JɃHH2`- X U2/bXV *QKTmiiBQMH EBM2KiB+b- TTX k8Ĝj9X
aT`BM;2` "2`HBM >2B/2H#2`; UkyyNV
SQii- X- JB2`K2Bbi2`- SX, qQ`FbT+2 M/ AMi2`72`2M+2 MHvbBb Q7 *#H2@.`Bp2M S`HH2H
_Q#Qib rBi? M lMHBKBi2/ _QiiBQM tBb- TTX j9RĜj8yX aT`BM;2` AMi2`MiBQMH Sm#HBb?@
BM;- *?K UkyR3V
a?BM;- qXCX- *MMQM- .X- :Q`KM- CX, PTiBKH 7Q`+2 /Bbi`B#miBQM TTHB2/ iQ `Q#QiB+ +`M2
rBi? ~2tB#H2 +#H2bX AM, A111 AMi2`MiBQMH *QM72`2M+2 QM _Q#QiB+b M/ miQKiBQMX-
pQHX k- TTX RN93ĜRN89 pQHXk UkyyyV
o2`?Q2p2M- _X, MHvbBb Q7 i?2 rQ`FbT+2 Q7 i2M/QM@#b2/ ai2rì THi7Q`KbX S?/ i?2bBb-
lMBp2`bBi i .mBb#m`;@1bb2M- .mBb#m`;- :2`KMv Ukyy9V
Part III
Workspace
Calculation of the cable-platform collision-free
total orientation workspace of cable-driven
parallel robots
Marc Fabritius1 , Christoph Martin1 , and Andreas Pott2

1
Fraunhofer Institute for Manufacturing
Engineering and Automation (IPA)
70569 Stuttgart, Germany
[email protected]
2
University of Stuttgart
70049 Stuttgart, Germany
Abstract. The large workspace of cable-driven parallel robots is one

of their main benefits over conventional parallel robots with rigid links.
Therefore, it is crucial to measure its size and analyze its constraints.
One of the limits are collisions between the cables and the platform of
the robot. They can damage the robot and cause malfunctioning of its
control algorithms. In the literature, methods for the detection of this
collision type only consider the constant orientation workspace and are
ill-suited for platform geometry data supplied from a CAD model.
This paper presents a new approach for the approximation of the cable-
platform collision-free total orientation workspace with various platform
orientation sets. The collision detection is based on a convex collision
cone data structure that precisely extracts the relevant information for
collision detection from the platform geometry data. This method is com-
patible with various workspace approximation algorithms to facilitate its
integration into the design process of cable-driven parallel robots. It is
tested on the IPAnema 3 cable robot geometry and its performance is
evaluated in terms of computation time.
Keywords: cable-driven parallel robots, workspace analysis, total ori-

entation workspace, cable-platform collisions
1 Introduction
While collisions between the cables and the platform of cable-driven parallel
robots have already been studied [3, 5, 9, 4], the approaches found in the
literature only consider the resulting constraints on the constant orientation
workspace of the robot. Furthermore, they all have other draw-backs such as
high computational costs or restrictive assumptions on the platform geometry.
Merlet [3] describes the platform geometry as a polyhedral object and calculates
bounding surfaces for the collision-free constant orientation workspace.
Perreault et al. [5] use a similar approach to analyze both cable-platform and
cable-cable collisions with the same symbolic method.

138 M. Fabritius et al.
The platform’s geometry data is usually available from CAD as a triangulation,

e.g. in .STL format [2]. Such data sets can contain thousands of triangles, of
which only a small subset may be relevant for collisions between the cables and
the platform. Both approaches mentioned above [3, 5], perform collision detection
for all edges in the platform geometry triangulation, which could result in high
computational costs. To make the detection of collisions more efficient, Nguyen et
al. [4] propose simplifying the platform geometry by only considering its convex
hull. Another approach, taken by Yue [9], is to describe the platform with a
distribution function, which is then approximated by splines.
The approach presented in this paper overcomes these draw-backs and efficiently
approximates the collision-free total orientation workspace for various platform
orientation sets without any restrictive assumptions on the geometry of the plat-
form.
2 Kinematic model of cable-driven parallel robots

This paper uses the standard kinematic and geometric model for cable-driven
parallel robots from [6, see chapter 3.2]. A cable-driven parallel robot consists
of m cables, which connect a mobile platform to fixed winches.
Each cable i ∈ {1, ..., m} leaves its winch at the proximal anchor point, whose
coordinates are denoted as ai ∈ R3 in the fixed coordinate system K0 . Its distal
anchor point on the platform is denoted as bi ∈ R3 in the platform coordinate
system KP . Depending on the length of all cables, the tuple of position r ∈ R3
and orientation R ∈ SO3 defines the platform’s pose (r, R) ∈ SE3 in the space of
spatial rigid body transformations SE3 . The cable vectors li between the anchor
points ai and bi , which provide an estimate of the cable lengths, can be expressed
in K0 for i ∈ {1, ..., m} as
li = ai − (r + Rbi ) . (1)
These relationships between the vectors associated with the i-th cable and the
coordinate systems K0 and KP are illustrated in Figure 1. Hereby, all elements
in the platform coordinate system KP are colored blue.
3 Assumptions for cable-platform collision detection

The analysis of cable-platform collisions presented in this paper is based on the
following assumptions:
– The mechanisms through which the cables leave their winches and are at-
tached to the platform are modeled as points ai and bi for i ∈ {1, ..., m},
from which the cables can extend into any direction.
– All cables i ∈ {1, ..., m} are modeled as straight line segments li between
their anchor points ai and bi .
– A triangulation of the platform geometry in KP is available, e.g. in .STL
format.
Calculation of the cable-platform collision-free total orientation… 139
bi
KP
bi
li
r+R
r, R
ai
K0
Fig. 1: Illustration of the vectors associated with cable i
– The distal anchor points bi for i ∈ {1, ..., m} are located on the surface or
outside of the platform geometry.
– At each distal anchor point bi in KP , a normal vector nbi is given, which
points to the outside of the platform geometry.
4 Workspace definitions
For a given workspace criterion, the workspace of a cable-driven parallel robot
is defined as the set of platform poses (r, R) ∈ SE3 for which the criterion
is satisfied. Visualizing and understanding the topology of such a set in the
six dimensional space of platform poses SE3 can be difficult. To circumvent this
problem, various subsets of the workspace can be studied, which lend themselves
to be visualized in 3D.
This paper uses the total orientation workspace definition from [6, see chapter
5.1.3], since it offers an intuitive interpretation. For a fixed set of rotation ma-
trices R• ⊂ SO3 , the total orientation workspace is the set of positions r ∈ R3
where all poses (r, R) with R ∈ R• satisfy the workspace criterion.
In the following sections, two sets of rotation matrices, i.e. platform orientations,
are defined, which are relevant in the context of cable-platform collisions.
4.1 A homogeneous set of rotations

The set Rη ⊂ SO3 contains all rotations up to a maximal angle of η ∈ [0, π]
about any axis d ∈ R3

Rη = R (d, θ) ∈ SO3 d ∈ R3 , θ ∈ [0, η] ⊂ SO3 . (2)

Hereby, R (d, θ) denotes the matrix describing a rotation about the axis d with
the angle θ. The total orientation workspace with this set of platform rotations
is a conservative approximation of the workspace based on the single parameter
η. Within this workspace, the platform can perform arbitrary rotations up to a
maximal deviation angle of η from the initial orientation R = I.
4.2 Rotations about a single axis

The set Rd,η ⊂ SO3 contains all rotations about the fixed axis d ∈ R3 in the
angle interval [−η, η] with η ∈ [0, π]

Rd,η = R (d, θ) ∈ SO3 θ ∈ [−η, η] ⊂ SO3 . (3)

The total orientation workspace with this set of platform rotations is relevant
for applications where it is known that the robot will only perform rotations
about a single axis. This is often the case in pick and place applications where
only one rotational degree of freedom is required.
5 Calculation of the collision cones

For each cable i ∈ {1, ..., m}, the set of outgoing cable directions from bi , in
which the cable would intersect with the platform, forms a cone. These cones
are referred to as collision cones since they contain the platform and therefore all
cable directions where collisions occur. They can be calculated with the method
of polar sorting, which was introduced by Pott [7] in the context of cable-driven
parallel robots.
For each anchor point bi , the following steps are performed to generate the
collision cone associated with cable i in KP :
1. Generate a local coordinate system using the Gram-Schmidt orthonormal-
ization scheme [1, see page 544], starting from the normal vector nbi and
two unit vectors of the KP coordinate system.
2. Use polar sorting on all vertices of the platform geometry to group them into
s ∈ N equally spaced counter-clockwise ordered angular segments around the
axis nbi through the point bi in KP .
3. Within each segment k ∈ {1, ..., s}, compare the direction vectors from bi
towards the vertices in the segment. The direction vector dk , which has the
minimal angle to nbi , is added to collision cone i to represent its segment.
4. Discretize each edge in the platform geometry, whose endpoints are in dis-
tinct and non neighboring angular segments, into linearly spaced points along
the edge, which no longer satisfy this condition.
Repeat steps 2 and 3 for each of these points.
The collision cones calculated by this procedure contain precisely the necessary
information about the platform geometry to detect cable-platform collisions.
Figure 2 shows the IPAnema 3 platform from [8], whose .STL file contains 12564
vertices. It can be encapsulated by m = 8 collision cones that each only consist
of s = 36 vectors. This resembles a reduction of the geometry data by 97.7%.
Therefore, the following collision detection algorithms, which operate on this
reduced data structure, have a significant performance advantage over any algo-
rithm that directly uses the triangulation data of the platform geometry.
Before these cones can be used to detect collisions, they must be subdivided into
convex cones, since the collision detection relies on this property.
Each collision cone i ∈ {1, ..., m} is partitioned through the following steps:
Fig. 2: IPAnema 3 platform (left) and its collision cones (right)
1. Iterate over the directions dk and calculate the normal vectors nk = dk−1 ×
dk for k ∈ {1, ..., s} with d0 = ds , which point to the inside of the cone.
2. Group consecutive directions dk , dk+1 into the same convex collision cone if
they satisfy the relationship nk , dk+1 ≥ 0.
3. To each group, add the vector −nbi to form a convex collision cone.
The result of this procedure for a collision cone of the IPAnema 3 platform is
visualized in Figure 3.
dk
dk−1
−nbi
bi
Fig. 3: Bundle of convex collision cones illustrated alone and with the platform
From now on, the direction vectors, that make up each convex collision cone
i ∈ {1, ..., m} centered at bi , are denoted as dk and are sorted in counter-
clockwise direction such that the normal vectors nk = dk−1 × dk point to the
inside of the convex cone.
5.1 Collision detection for a platform pose
A platform pose (r, R) ∈ SE3 is free of collisions between the cables and the
platform if and only if, all cables are outside of all associated convex collision
cones. Cable i is outside of a convex collision cone for the platform pose (r, R) ∈
SE3 , if and only if, the cone has a normal vector nk such that
0 > li , Rnk = ai − (r + Rbi ) , Rnk . (4)
Note that this condition relies on the convexity of the collision cones.
Figure 4 illustrates this condition by showing a planar cut through the convex
collision cone and cable i.
convex
collision Rnk
cone
li
Rnk+1
ai
Rbi
r
O
Fig. 4: Illustration of a convex collision cone
5.2 Collision detection for a set of platform orientations
To simultaneously check the condition from Equation (4) for a platform position
r ∈ R3 and a set of orientations R• from Section 4, it is not sufficient to solely
consider the normal vectors nk of the convex cone. Figure 5 illustrates this by
showing the setting from Figure 4 for all platform orientations R ∈ R• . To better
understand the relative positions of the cable directions towards the convex cone,
its position is fixed by rotating all involved vectors {ai , r + Rbi , nk , nk+1 } with
R−1 for each possible orientation R ∈ R• . With this visualization, it becomes
apparent that for both of the normal vectors nk and nk+1 , the condition from
Equation (4) is not simultaneously satisfied for all R ∈ R• . To validate that
the cable is indeed outside of the collision cone for all orientations R ∈ R• , the
condition must be checked for various convex linear combinations of the normal
vectors
λnk+1 + (1 − λ) nk
nλk = with λ ∈ [0, 1] . (5)
λnk+1 + (1 − λ) nk
convex
collision
cone nk
nλk
nk+1

•
R
∈
i
R
b
−
r)
−
−
1 (a i

R
Fig. 5: Illustration of a convex collision cone for all R ∈ R•
If such a vector nλk can be found to satisfy

0 > li , Rnλk = ai − (r + Rbi ) , Rnλk ∀ R ∈ R• ,

(6)
then cable i is outside of the convex collision cone for all platform poses (r, R)
with R ∈ R• . In the following sections, explicit checks of this condition are
developed for the orientation sets defined in Section 4.
5.3 Collision detection for Rη

For all R in the set Rη from Section 4.1, Equation (6) can be reformulated to
0 > li , Rnλk = ai − r, Rnλk − bi , nλk

(7)
= cos(α + β) ai − r − bi , nλk ,

where α = ai − r, nλk denotes the angle between (ai − r) and nλk and β ∈

[−η, η] is the deviation angle resulting from the rotation R ∈ Rη . The inequality
in Equation (7) can be checked for all β by simply considering the maximum of
the expression cos(α + β) for all β ∈ [−η, η].
The i-th cable is below the plane defined by the normal vector nλk for all poses
(r, R) with R ∈ Rη if and only if
max cos(α + β) ai − r < bi , nλk .

(8)
β∈[−η,η]
Since cos(·) is a continuous periodic function, with a maximum at zero, it suffices

to check the condition in Equation (7) at this value when it is contained in the
interval [α − η, α + η]. If zero is not in the interval, the maximum has to occur
at the interval boundary {α − η, α + η}.
A similar check for collisions with respect to the orientation set Rd,η is developed
in the following section.
5.4 Collision detection for Rd,η

For all R in the set Rd,η from Section 4.2, the inner normal
vector
nλk can be
λ
split up into the part that is not affected by the rotation d, nk d and the part
n⊥d that is orthogonal to the rotation axis d
Rnλk = d, nλk d + R nλk − d, nλk d .

(9)
n⊥d
With this notation, Equation (6) is reformulated to
0 > li , Rnλk = ai − r, Rnλk − bi , nλk

(10)
= ai − r, Rn⊥d + ai − r, d d, nλk − bi , nλk .

According to Figure 6, a planar coordinate system {x, y} is created to derive a

suitable expression for Rn⊥d .
ai − r
α β
x Rn⊥d
n⊥d
Fig. 6: Construction of a planar coordinate system to calculate ai − r, Rn⊥d

The axes of the planar coordinate system are determined by

d × (ai − r) π
x= , y = R d, x. (11)
d × (ai − r) 2
The signed angle of n⊥d to the x-axis of the planar coordinate system is calcu-
lated as
α = atan2 n⊥d , y , n⊥d , x .

(12)
With this notation, the scalar product ai − r, Rn⊥d from Equation (10) can

be expressed as
ai − r, Rn⊥d = sin (α + β) ai − r, y ,

(13)
where β is the angle of the rotation R ∈ Rd,η . Substitution into the condition
from Equation (10) leads to
0 > sin (α + β) ai − r, y + ai − r, d d, nλk − bi , nλk .

(14)
Cable i is outside of the convex collision cone if this condition is satisfied for
all β ∈ [−η, η]. This can be easily checked using the extremal values of sin(·)
in the interval [α − η, α + η]. It suffices to check Equation (14) at the interval
boundary {α − η, α + η} and the points α + β = ± 12 π, ± 32 π if they are contained
in the interval.
6 Implementation and evaluation

The cable-platform collision detection method developed in this paper, is inte-
grated into WireCenter, a cable robot design tool developed by Pott et al. [8].
It is compatible with any point-based workspace approximation method.
To create a visualization for the IPAnema 3 robot [8], the numerical method from
[6, see chapter 5.5.1] is chosen. It provides an efficient approximation for compact
workspaces of redundantly constrained cable robots, such as the IPAnema 3.
This method is based on a bundle of rays originating from a common center
point, that uniformly discretize the unit sphere around that point. Along each
of these rays, a line search is performed with a boolean workspace criterion to find
the furthest points on the rays that satisfy this criterion. These points are then
connected by triangles to form the boundary of the approximated workspace.
The workspace criterion, used in the following visualizations, is whether a plat-
form position r ∈ R is in the cable-platform collision-free total orientation
workspace with the orientation set R• . This can be translated into a condi-
tion for all convex collision cones associated with the m cables of the robot. The
position r is in the workspace if, for each of them, there exists a convex lin-
ear combination of neighboring inner normal vectors nλk such that the condition
from Equation (6) is satisfied. Explicit methods to check this condition for the
sets Rη and Rd,η are described in Sections 5.3 and 5.4.
In Figure 7, the cable-platform collision-free total orientation workspace of the
IPAnema 3 robot is visualized from two perspectives for the orientation set Rη
with η = 0◦ , 15◦ , 30◦ . Figures 8 and 9 show equivalent visualizations for the
T T
orientation set Rd,η with d = [0, 0, 1] and d = [0, 1, 0] .
In all three visualizations it can be seen that the workspace significantly shrinks
when η is increased. Furthermore, one can observe that the workspace for η = 30◦
in Figure 7 is significantly smaller than in Figures 8 and 9. This is no surprise
since Rd,η ⊂ Rη , but it underscores the importance of considering specialized
orientation sets to analyze the capabilities of cable-driven parallel robots for
specific applications.
The computation times of the calculations for Figures 7 to 9 are measured on
a laptop with an Intel R
CoreTM i5-7440HQ CPU running at 2.80 GHz. For the
orientation set R , the average computation time is 0.559 seconds, while for Rd,η
η
it is 1.078 seconds. This difference in computation time is due to the fact that
η = 0◦ η = 15◦ η = 30◦
Fig. 7: Total orientation workspace for Rη
η = 0◦ η = 15◦ η = 30◦
T
Fig. 8: Total orientation workspace for Rd,η with d = [0, 0, 1]
η = 0◦ η = 15◦ η = 30◦
T
Fig. 9: Total orientation workspace for Rd,η with d = [0, 1, 0]
the collision detection method for Rd,η in Section 5.4 requires more calculations
than the one for Rη in Section 5.3.
7 Conclusion and outlook
A new approach for detecting cable-platform collisions based on an efficient data

structure for the platform geometry is presented. For each cable of the robot, the
relevant information regarding collisions with the platform is extracted from the
geometry data of the platform and distilled into a convex collision cone bundle.
In contrast to other methods, no assumptions or restrictions have to be imposed
on the platform geometry.
The cable-platform collision-free total orientation workspace can be approxi-
mated for various sets of platform orientations Rη , Rd,η . Other orientation sets
can also be incorporated into the method to analyze a robot’s capabilities for
specific applications. Thus, the presented method provides a powerful tool for
the analysis of cable robot geometries.
In practice, the locations of the distal anchor points bi can often easily be
changed on the surface of the platform, while the platform geometry itself is
fixed through application specific design constraints. For such settings, future
research could investigate the constrained optimization problem of choosing lo-
cations for the anchor points bi that maximize the cable-platform collision-free
workspace.
Additionally, one could also proceed by integrating the method from this pa-
per with other relevant workspace criteria to work towards a universal design
framework for cable-driven parallel robots.
Acknowledgements
The results presented in this paper are originated in the research project HIND-
CON (Hybrid INDustrial CONstruction) funded by the European Commission
(Grant Agreement No 723611).
References
[1] Ward Cheney and David Kincaid. Linear algebra: Theory and applications.
Vol. 110. 2009.
[2] Jonathan D. Hiller and Hod Lipson. “STL 2.0: a proposal for a universal
multi-material additive manufacturing file format”. In: Proceedings of the
Solid Freeform Fabrication Symposium. 1. 2009, pp. 266–278.
[3] Jean-Pierre Merlet. “Analysis of the influence of wires interference on the
workspace of wire robots”. In: On Advances in Robot Kinematics. Springer,
2004, pp. 211–218.
[4] Dinh Quan Nguyen and Marc Gouttefarde. “On the Improvement of Ca-
ble Collision Detection Algorithms”. In: Cable-Driven Parallel Robots: Pro-
ceedings of the Second International Conference on Cable-Driven Parallel
Robots. Ed. by Andreas Pott and Tobias Bruckmann. Cham: Springer In-
ternational Publishing, 2015, pp. 29–40.
[5] Simon Perreault et al. “Geometric determination of the interference-free
constant-orientation workspace of parallel cable-driven mechanisms”. In:
Journal of Mechanisms and Robotics 2.3 (2010), p. 031016.
[6] Andreas Pott. Cable-Driven Parallel Robots: Theory and Application. Springer,
2018.
[7] Andreas Pott. “Determination of the Cable Span and Cable Deflection of
Cable-Driven Parallel Robots”. In: Cable-Driven Parallel Robots. Springer,
2018, pp. 106–116.
[8] Andreas Pott et al. “IPAnema: a family of cable-driven parallel robots for
industrial applications”. In: Cable-Driven Parallel Robots. Springer, 2013,
pp. 119–134.
[9] Suilu Yue and Qi Lin. “A Solution for interference between mobile platform
and wires in wire-driven parallel manipulators”. In: 2010 IEEE Interna-
tional Conference on Intelligent Computing and Intelligent Systems (ICIS).
Vol. 3. IEEE. 2010, pp. 693–696.
Workspace Analysis of Cable Parallel Manipulator for
Side Net Cleaning of Deep Sea Fishing Ground
澻澼
Liping Wang1,2,3, Haisheng Li3, Zhufeng Shao1,2 , Zhaokun Zhang1,2
and Fazhong Peng2,3

1
Key Laboratory of Tribology, Tsinghua University, Beijing 100084
2 Beijing Key Lab of Precision/Ultra-precision Manufacturing Equipment and Control,
Tsinghua University, Beijing 100084
3
School of Mechanical and Electrical Engineering, University of Electronic Science and
Technology of China, Chengdu 611731
[email protected]
Abstract. Deep sea fishing ground (DSFG) is a large-scale fish farming

equipment that allows the full use of marine environment to meet human needs
for food quantity and quality. To maintain the good aquaculture environment of
the DSFG, regular cleaning of its side nets is necessary. This paper discusses the
application of the four-cable parallel manipulator in performing the cleaning task.
The impact of buoyancy on the workspace of the cable parallel manipulator is
studied, which reveals that reasonable use of buoyancy can improve the reachable
workspace. The stiffness anisotropy and stiffness average indices are adopted to
analyze the stiffness performance. Then, the preliminary workspace of the four-
cable parallel manipulator is determined, and covers 88.8% of the entire side net
of the DSFG and meet the coverage requirement of 85%. Optimization of the
scale and end effector rotation will be carried out in the future, which can further
improve the coverage.
Keywords: Cable-driven parallel manipulator, Workspace, Stiffness, Buoyancy
1 Introduction
With the continuous exacerbation of global population and economic unbalance,

problems on hunger and obesity are on the rise. At the same time, the ocean, which
accounts for 70% of Earth’s surface, provides only 2% of human food and is recognized
as the future granary [1]. Deep sea fishing ground (DSFG) is a large-scale fish farming
equipment that provides efficient and healthy food supply and possesses great potential.
The main structure of DSFG is a regular prism with a circumscribed circle diameter of
100 meters and a height between 30 and 40 meters (see Fig. 1). Considering that the
DSFG is immersed in seawater every day, algae will cover the net, deteriorating the
internal environment. Therefore, regular cleaning of side nets is strictly required. Due
to the large side area of the DSFG, the rigid manipulator is bulky and expensive to
finish this work, while the underwater robot experiences difficulty in meeting the

150 L. Wang et al.
automation and efficiency requirement. The cable-driven parallel robot features the
advantages of simple structure, small mass, low cost, and large workspace, and thus
appears as an ideal candidate.
The cable-driven parallel manipulator (CDPM), which adopts ropes instead of rigid
branches, has been successfully applied in many fields, such as wind tunnel test [2],
glass curtain wall cleaning [3], and large radio telescope [4]. According to the
configuration, the CDPM can be divided into mutual-pulled and suspended types. In
the first category, cables are distributed on both sides of the end effector, and complete
or over constraint is realized with advantages of high stiffness and speed [5]. For the
suspended one, cables are kept in tension usually by the weight with huge workspace
and load capacity. Considering the disturbance of water flow, stiffness is required. The
mutual-pulled and over-constrained CDPM, which is the planar four-cable parallel
manipulator with fixed posture, is adopted to carry out the DSFG cleaning operation.,
Workspace coverage is required for the side net cleaning of the DSFG. The
workspace of the mutual-pulled CDPM can be divided into four categories. Static
workspace [6] is the position and posture set of the end effector that satisfies the static
equilibrium condition without considering external force and torque. On the basis,
wrench-closure workspace [7] pertains to poses balancing any given wrench on the end
effector. Wrench-feasible workspace [8] is defined as a set of resulting poses where the
tensions can counteract a specific set of wrenches applied on the end effector. Force-
closure workspace [9] is a set of poses where resultant cable tensions can sustain an
arbitrary external wrench acting on the moving platform. As no concentrated external
force or torque exists during澳 the cleaning and the speed is slow, static workspace is
adopted in this research to perform follow-up analysis.澳
For the cleaning condition of the DSFG, the stiffness of the CDPM needs to be
studied. Many scholars completed considerable research on this topic. Verhoeven [10]
proposed the definition of the stiffness matrix. Liu [11] analyzed the 6激DOF cable
parallel mechanism of the wind tunnel test, and obtained the stiffness matrix by
differential transformation. Cui [12] solve the problem of variable stiffness of cable-
driven parallel robots (CDPR), a new static stiffness analysis and cable tension
distribution method are proposed for studying the CDPRs’ controllable stiffness. Li
[13] proposed that the stiffness of CRPM and RRPM is only related to the elastic
properties of the cable and the Jacobian matrix of the cable force, regardless of the
magnitude of the cable tension. Subsequently澿澳the stiffness condition number is defined
as the square root of the ratio of the minimum eigenvalue to the maximum eigenvalue
of the stiffness matrix [14]. Wang [15] used the geometric mean of the eigenvalues of
the stiffness matrix to define the average stiffness. In the present study, the stiffness
condition number and average stiffness are used to refine the workspace of the CDPM.澳
Considering the impact of the buoyancy force on the workspace, this paper discusses
the application of the CDPM in cleaning the DSFG side nets. In the first section, the
kinematic model of the four-cable parallel manipulator is illustrated, and the cable
tension is derived. The second section studies the workspace of the CDPM with fixed
and passively changed buoyancy. The stiffness of the CDPM is evaluated, and the final
workspace is determined. Finally, conclusions are provided.
Workspace Analysis of Cable Parallel Manipulator for Side Net Cleaning… 151
2 Model construction
The structure of the DSFG is a regular polygonal prism (see Fig. 1). Each side facet is
rectangular, and the size is 28.6 m ×32.9 m. Fig濁 2 presents the cable layout. The layout
of single sided net and the cable parallel manipulator are shown in Fig. 3. The Cartesian
coordinate system XOY is established at the lower left corner of the rectangular
frame of the sided net (the static platform). The points corresponding to the cable
outputs of the static platform are denoted as A1 ,A 2 ,A3 ,A4 . The side lengths are
recorded as l3 and l4 . The vector connecting the origin of the fixed reference frame
to point Ai is denoted as a1 , a2 , a3 , a4 . The end effector (moving platform) of the
cable parallel manipulator is also a rectangular structure with the size of 0.5 m ×0.8 m.
The four vertices are recorded as B1 ,B2 ,B3 ,B4 , and the side lengths are recorded as l1
and l2 . The local coordinate system X 'O'Y ' is attached to the geometric center of the
end effector. The position of the end effector with respect to the origin of the fixed
reference frame is denoted as p , and p ri represents the vector of each vertex of the
end effector with respect to the local frame.
Fig. 1. Structure of deep sea fishing ground Fig. 2. Layout of the cable manipulator
A4 A3
t4 YĆ t3
Fb
B4 B3
l1 XĆ
l4 OĆ
B1 G B2
l2
Y t1 t2
A1 A2
O X l3
Fig. 3. Kinematic model of the 4-cable CDPM
152 L. Wang et al.
The unit vector of the cable can be expressed as

ai p O R p p ri
ui i 1, 2,3, 4 , (1)
ai p O R p p ri
where
ªcos I sin I 0º
O
Rp « sin I cos I 0 »» .
«
«¬ 0 0 1 »¼
O
Rp is the rotation matrix, and I is the rotation angle of the moving platform, which
remains zero in subsequent studies.
According to the force and torque balances, equations can be derived as
m
¦ ti f 0, (2)
i 1
m
¦ ri u ti m 0. (3)
i 1
where ti is the cable force, f is the resultant force, which is the vector sum of
gravity G , buoyancy Fb and disturbing force, and m is the resultant torque.
Rearranging in matrix form yields the following equation:
JT W 0 . (4)
where J is called the structure matrix, T is the cable force matrix, and W is the
external force matrix.
For the redundant cable manipulator, the cable force is hard to derive with the matrix
equation. Instead, it is transformed into the optimization problem to solve the cable
force, and can be expressed as follows:
Objective function z min (t12 t22 t32 t 42 ) ,
subject to
JT W 0
® .
¯ti t tmin˄i=1ˈˈ
2 3ˈ˅
4
3 Workspace analysis
According to the size of the moving platform and considering the ergodic ability of the
workspace, step lengths of 0.5 m and 0.8 m are adopted to search the entire range to
determine workspace of the CDPM. The maximum cable force is set to be five times
the load weight. In addition, taking into account the disturbance of water flow to the
rope, the minimum cable force is set to be 0.2 times the load weight. To clarify the
effects of buoyancy, the workspace is first analyzed without the buoyancy. The
workspace of the four-cable manipulator is determined and illustrated in Fig. 4. The
CDPM workspace can cover 89.8% of the side net. Unreachable areas lies in the upper
middle as well as the left and right.
Fig. 4. Workspace without considering buoyancy

Y/m
Y/m
0 0
0 0
X/m X/m
(a) The force map of cable 4 (b) The force map of cable 3
Y/m
Y/m
0 0
0 0
X/m X/m
(c) The force map of cable 1 (d) The force map of cable 2
Fig. 5. Cable forces of the four-cable manipulator
154 L. Wang et al.
The corresponding cable force distributions are illustrated in Fig. 5, which reveals that
the tension distribution of the left and right cables is symmetrical. Generally, tensions
of upper cables are greater than the lower cables due to weight. Tension of the upper
cables increase when the end effector moves toward the up, left and right borders, while
tension of lower cables sharply rise when the end effector approach left or right borders.
The CDPM workspace can be considered determined by the maximum tension limit of
the upper cables.
The coverage of workspace
The ratio of the buoyancy to the weight of the end effector
Fig. 6. Workspace coverage under different buoyancies
As DSFG is immersed in the water, the moving platform will be subjected to 瀇濻濸
buoyancy. Therefore, the influence of buoyancy on the workspace should be
considered. Since the buoyancy can be adjusted flexibly with the volume change,
optimal buoyancy is discussed to improve the workspace. Under the same tension
range, the buoyancy is changed from 0 to 2 times the weight of the end effector, and
the workspace coverage of the CDPM is calculated, and shown in Fig.6.
Fig. 7. Workspace when the buoyancy is 0.6 times the weight of the end effector
When different buoyancies are selected, the workspace coverage of the four-cable
manipulator fluctuates changes between 89.5% and 92.0%. The workspace coverage
curve of the four-cable CDPM is symmetrical at the line that the buoyancy equals the
weight. The weight on the left side is larger than the buoyancy, while the weight on the
right side is less than the buoyancy. The workspace coverage of the four-cable CDPM
reaches a maximum value of 91.8% when the ratio of buoyancy to weight is about 0.6
and 1.4. Fig. 7 shows the maximum workspace when the ratio is 0.6. The unreachable
area on the upper, left and right borders of Fig. 4 shrinks. When the ratio 1.4 is adopted,
the workspace is the situation after Fig. 7 is upside down. Since the lower area is harder
to reachable, the ratio of 0.6 is more practical.
Considering that the object in different water depths experiences varying pressures,
if a simple tank is installed on the moving platform which can passively adjust the
volume according to the water pressure, and the buoyancy can changed accordingly.
With the relationship of pressure and volume at different positions, it can be expressed
as:
PV
0 0 PV
1 1. (5)
Setting the initial volume V0 , P0 is the atmospheric pressure, initial buoyancy is
F0 U gV0 P mg , and P is a coefficient to be further discussed. The pressure at
§h ·
different depths (h) can be expressed as P1 ¨ 1¸ P0 .
© 10 ¹
Subsequently, the buoyancy changes with the depth can be expressed as
PV 1
F1 U gV1 0 0
Ug P mg . (6)
P1 §h ·
¨ 1¸
© 10 ¹
The nonlinear function of buoyancy change with the change of height contains
parameters related to initial buoyancy. The initial buoyancy changes between 0.5 and
5 times of the load weight. A curve of the change of the workspace coverage with the
change of initial buoyancy is depicted in Fig. 8. The abscissa represents the ratio of
initial buoyancy to load weight, and the ordinate represents the workspace coverage of
the cable manipulator.
The coverage of workspace
The ratio of the initial buoyancy to the weight of the end effector
Fig. 8. Workspace coverage when buoyancy is a nonlinear function

156 L. Wang et al.
When the initial buoyancy differs, the workspace coverage of the four-cable
manipulator fluctuates, ranging between 92% and 96%. Compared with the workspace
of the cable manipulator with constant buoyancy, the working space of the four-cable
parallel manipulator is further expanded. The workspace coverage of the four-cable
manipulator reaches a maximum value of 95.7% near abscissa 2.5.
4 Stiffness analysis
In the mechanical design process, stiffness describes the manipulator’s ability to resist
external torque (external force and acting torque). The higher the stiffness, the smaller
the deformation of the manipulator under the same acting torque. Usually, stiffness is
guaranteed within a specific range to meet the motion accuracy of the manipulator濁
When the generalized external force acting on the terminal wW slightly changes, it
will inevitably cause the moving platform to produce wX corresponding micro motion
vector. Thus, the stiffness of the system can be expressed as:
wW wJ wT
K T J , (7)
wX wX wX
wJ wT
where T is the active stiffness, J is the passive stiffness, X is the position
wX wX
and posture of the moving platform, and X R n .
Cable manipulator stiffness [16] can be expressed as:
ª kai k 'li1 1 k ' 1ti º
K H ª¬ J W I J J O º¼ Jdiag « » JT , (8)
«¬ kai k 'li1 1 k ' 1ti »¼
where the stiffness of the driving unit is kai , the linear elastic stiffness of the rope is
EA / l0i , E , A are the elastic modulus and cross-sectional area of the rope, respectively,
and l0i is the length of the i th rope before deformation.
The overall stiffness of the system is divided into two parts. One is related to the
rope pulling force, whereas the other depends on the geometric arrangement of the rope,
position and posture of the moving platform, and physical parameters of the driving
branch. Active stiffness usually exerts little effect on the overall stiffness of the system,
so this study only analyzes passive stiffness.
Tang [17] mentioned the simplified formula of passive stiffness of cable parallel
manipulator
K k ' JL1J T (9)
'
where k is the rope stiffness of unit length. This study assumes that all the driving
cables have same type of rope with the same parameters. For the ultra-high-strength
composite fiber with a rope diameter of 8 mm, the unit stiffness of the rope is 125000
N/m.
ªl11 0 0 0 º
« »
0 l21 0 0 »
L1 « , l1 , l2 , l3 , l4 is the length of the four cables.
« 0 0 l3 1
0»
« 1
»
«¬ 0 0 0 l4 »¼
The eigenvalues of the passive stiffness matrix of the cable parallel manipulator are
calculated, and they are arranged in ascending order: 0 d O1 d d On , where n is the
number of translational degrees of freedom of the cable parallel manipulator. The
stiffness condition number S t of the static workspace is the square root of the ratio of
the minimum eigenvalue to the maximum eigenvalue of the stiffness matrix.
min Oi
St i 1, 2, , n (10)
max Oi
According to the definition of the performance parameter, the closer the value of the
stiffness condition number S t to 1, the closer the stiffness along each direction of the
parallel manipulator at this position and posture and the better the isotropy of stiffness.
If the value of the stiffness condition number is closer to 0, then the greater the
difference in stiffness along each direction of cable parallel manipulator at this position
and posture, the worse the isotropy of stiffness.
Fig. 9 depicts the distribution map of the stiffness condition number in the static
workspace of the four-cable parallel manipulator at different positions.
Fig濁 9 shows that the moving platform has better isotropic stiffness at the central area
of the workspace. Furthermore, near the positions of (15,12) and (15,22), the workspace
obtains the best isotropy of stiffness, which can reach above 0.9, indicating that the
closer the workspace boundary, the worse the isotropy. In this study, the stiffness
condition number is greater than 0.4, which is the critical value of the normal operation
of the external system.
0.3
0.4
0.5
0.6 0.8
0.7 0.9
0
0
Fig. 9. Location distribution of stiffness condition number

158 L. Wang et al.
The stiffness condition number provides the difference distribution of the cable parallel
manipulator stiffness in varying directions. However, the stiffness condition number is
the normalized dimensionless parameter, which cannot reflect the difference
distribution of the absolute value of the stiffness of varying positions and postures. This
research only focuses on the translation of the moving platform, and the rotation angle
of the moving platform remains zero in subsequent studies, so considering the problem
of dimension non-uniformity is unnecessary. To study the distribution of the overall
stiffness of the workspace, Wang [15] proposed the concept of average stiffness, which
is defined as
KA n O1 O2 On . (11)
The average stiffness KA is the geometric mean of the stiffness matrix eigenvalues,
which can reflect the stiffness value of different position and posture according to the
average stiffness value. Considering the stability condition of the system affected by
the external environment, Fig. 10 illustrates the average stiffness distribution map of
the stiffness condition number larger than 0.4.
Fig.10 reveals that the average stiffness is distributed symmetrically from top to
bottom and from left to right, and the stiffness decreases gradually along the direction
from top to bottom and left to right and increases gradually along the direction of four
angles. As the element in the stiffness matrix contains the reciprocal length of the cable,
a large stiffness is generated at the position and posture where the cable length is small,
and a sudden change in stiffness occurs even in the four corners. The average stiffness
of the workspace is above 1u 107 N/m , which meets the operational requirement of the
DSFG.
Fig. 10. Average stiffness distribution
Combining the above discussion that various factors affecting the workspace of the
1
four-cable manipulator and selecting the nonlinear function F2 2.5mg of
§ h ·
¨ 1 ¸
© 10 ¹
the change of buoyancy with height, Fig.11 shows the workspace of the four-cable
manipulator having a workspace coverage of 88.8%.
Ύ Ύ
Fig. 11. Workspace diagram
Fig.11 indicates that the workshop is distributed symmetrically from top to bottom and
from left to right, due to cable force and isotropic constraint of stiffness, the workspace
of the four corners, and the surrounding boundary being reduced inward. The
workspace can reach the upper and lower boundaries because of the optimization
algorithm of the change of buoyancy with height.
5 Conclusion
The workspace and stiffness of the four-cable planar CDPM are analyzed to evaluate
the application feasibility of cleaning the side nets of the DSFG. The influence of
buoyancy on the static workspace of the CDPM is discussed. Simulations reveal that
the workspace coverage of the CDPM is 89.7% without buoyancy and that buoyancy
can improve the workspace. When the buoyancy force changes passively with depth,
workspace can cover 95.7% of the side nets, which indicates that buoyancy should be
used to improve the performance of the CDPM for underwater applications. Stiffness
isotropic and average stiffness are also studied. After refining the stiffness, the four-
cable CDPM can still cover 88.8% of the side nets, meeting the cleaning requirements.
In the future, optimization of the geometry and posture will be performed to further
improve the workspace and stiffness performance of the CDPM.
References
1. Kang Sun, Lidan Li. Analysis on the Effect and Time-Space Differences of China's Marine
Fisheries Transformation. Economic review, 2018, 7(4):72-83.
160 L. Wang et al.
2. Yaqing Zheng, Qi Lin, Xiongwei Liu, et al. On Wire-driven Parallel Suspension Systems
for Static and Dynamic Derivatives of Aircraft in Low-speed Wind Tunnels. Acta
Aeronautica et Astronautica Sinica, 2009, 30(8):1549-1554.
3. Yongde Zhang, Jingang Jiang, Shu Zhang, et al. Development and experimental study on
glass-curtain wall cleaning robot driven by flexible rope. Chinese Journal of Scientific
Instrument, 2013, 34(3):494-501.
4. Zhiyuan Liu, Lian Chen, Zhufeng Shao, et al. Terminal Accuracy of the Feed Support
System in FAST. Journal of Mechanical Engineering, 2017, 53(17): 50-59.
5. MING A, HIGUCHI T. Study on multiple degree-of-freedom positioning manipulator using
wires (part 1) -concept, design and control. International Journal of Japan Social
Engineering, 1994, 28(2):131-138.
6. Albus J S, Bostelman R V, Dagalakis N. The NIST ROBOCRANE. Journal of
RoboticsSystem, 1992, 10(5): 709-724.
7. Gouttefarde M, Gosselin C M. Analysis of the wrench-closure workspace of planar parallel
cable-driven manipulators. IEEE Transactions on Robotics, 2006, 22(3): 434-445.
8. Bosscher P, Riechel A T, Ebert-Uphoff I. Wrench-feasible workspace generation for cable-
driven robots. IEEE Transactions on Robotics, 2006, 22(5): 890-902.
9. Pham C B, Yeo S H, Yang G, et al. Force-closure workspace analysis of cable-driven parallel
manipulators. Manipulator and Machine Theory, 2006, 41(1): 53-69.
10. Verhoeven, R. Analysis of the workspace of Tendon-based Stewart platforms. Doctoral
dissertation. University Duisburg-Essen, 2004.
11. Xin Liu, Yuanying Qiu, Ying Sheng. Stiffness Enhancement and Motion Control of a 6-
DOF Wire-driven Parallel Manipulator with Redundant Actuations for Wind Tunnels. Acta
Aeronautica et Astronautica Sinica, 2009, 30(6):1156-1164.
12. Zhiwei Cui, Xiaoqiang Tang, Senhao Hou, etc. Research on Controllable Stiffness of
Redundant Cable-Driven Parallel Robots. IEEE/ASME Transactions on Mechatronics,
2018, 23(5):2390-2401.
13. Hui Li, Wenbai Zhu. Static Stiffness Analysis of Flexible-cable-driven Parallel Mechanism.
Journal of Mechanical Engineering, 2010, 46(3):8-16.
14. Lewei Tang. Research and Experiments on Dynamical Trajectory Planning of Cable-driven
Parallel Manipulators濁澳Doctoral dissertation. Tsinghua University, 2015.
15. Weifang Wang. Research on Redundantly Restrained Cable-driven Parallel Manipulator for
Simulating Force. Doctoral dissertation. Tsinghua University, 2016.
16. Xin Liu, Yuanying Qiu, Ying Sheng. Analysis on the Static Stiffness of Wire-driven Parallel
Manipulators. Journal of Mechanical Engineering, 2011, 47(13):35-43.
17. Xiaoqiang Tang, Lewei Tang, Jinsong Wang, et al. Workspace quality analysis and
application for a completely restrained 3-Dof planar cable-driven parallel manipulator.
Journal of Mechanical Science and Technology, 2013, 27(8):2391-2399.
18. Yaqing Zheng, Xiaoling Jiang. Static Stiffness Analysis and Optimization of Four-cable-
driven Under-restrained Parallel Manipulator with 6 DOFs Based on Least Square-support
Vector Machine. Journal of Mechanical Engineering, 2012, 48(13): 49-55.
19. Xiaoqiang Tang, Weifang Wang, Lewei Tang. A geometrical workspace calculation method
for cable-driven parallel manipulators on minimum tension condition. Advanced Robotics,
2016, 30(16):1061-1071.
20. Trevisani A. Planning of dynamically feasible trajectories for translational, planar, and
underconstrained cable-driven robots. Journal of Systems Science and Complexity, 2013,
26(5): 695-717.
Identifying the largest sphere inscribed in the
constant orientation wrench-closure workspace
of a spatial parallel manipulator driven by seven
cables
Ambuj Shahi1 and Sandipan Bandyopadhyay2 ( )
Department of Engineering Design,

Indian Institute of Technology Madras, Chennai 600 036, India
1
[email protected]
2
[email protected],
WWW home page: http://ed.iitm.ac.in/~sandipan
Abstract. This paper presents a methodology to find the largest sphere

inside the constant orientation wrench-closure workspace of spatial cable-
driven parallel robots driven by seven cables. The sphere is centred at
a prescribed point of interest and is obtained for a given orientation of
the moving platform. The method builds upon the analytical description
of the boundary of the constant orientation wrench-closure workspace
to obtain the desired spheres. The problem has been reduced to solv-
ing seven systems, each consisting of three cubic polynomial equations
in three unknowns. A computer algebra system (CAS) has been used to
solve these systems of equations, using a formulation based on Sylvester’s
dialytic elimination and generalised eigenproblem, which has been illus-
trated through an example.
Keywords: Cable-driven parallel robots, wrench-closure workspace,

singularity-free sphere, parallel robots, cable robots
1 Introduction
Cable-driven parallel robots (CDPRs) form a particular class of parallel manip-
ulators, in which cables are used to connect the moving platform (MP) to the
base-frame, instead of rigid links. Cables with substantial total length can easily
be stored on winches fixed to the base-frame, providing CDPRs with the possi-
bility of working inside a considerably larger space as compared to the rigid-link
parallel mechanisms. One inherent drawback of CDPRs is that the cables can
only pull the platform and not push, thus denying the CDPRs the ability to
operate at all the points inside this workspace. In this context, the concept of
wrench-closure workspace (WCW) [1] has been introduced, as the set of poses
for which all the applied forces and torques on the MP, represented together as
wrenches, can be balanced by a set of positive cable tensions. In [1, 2], an ana-
lytical method has been presented to obtain the boundary of the constant ori-
entation WCW (denoted by Wco ) of CDPRs, using the null-space of the wrench

162 A. Shahi and S. Bandyopadhyay
transformation matrix. In the present work, building upon this description of

the boundary of the Wco , the largest sphere is found, which is contained inside
the Wco of a spatial parallel robot driven by 7 cables (7-SCDPR). The centre of
this sphere is at a chosen point inside the Wco of the robot. The identification
of such a sphere is vital from the aspect of design and path-planning of such
robots, as the sphere forms a convex region inside R3 . Thus, any line-segment
joining two points inside this maximal sphere would also lie inside that space
entirely. The above sphere is found for a particular Wco . If it is to be found for
a range of orientations of the MP, then this range can be scanned, up to some
desired resolution. For each of these Wco , a similar sphere, as described above,
can be obtained. The smallest sphere among all such spheres would then form
the maximal sphere which would lie entirely inside the WCW of a 7-SCDPR, for
the specified range of orientations. Thus, it would be possible to formulate the
problem of finding an algorithm for the kinematic design of CDPRs, where the
requirement would be to identify the geometric parameters which would allow
the robot to have a desired spherical region inside its WCW, over a given range of
orientations. It should be noted here that the above-mentioned maximal sphere
is obtained approximately, which is accurate only up to the resolution used while
scanning the range of orientations.
The problem of finding the largest gain-type singularity-free spheres for the case
of conventional parallel manipulators has been studied in many previous works,
such as in [3, 4]. Bayani et al. [5] have studied the problem of identifying the
maximal ellipsoid inside the wrench-feasible workspace (WFW) [6] of a CDPR.
In their work, the non-linear expressions representing the WFW boundary have
first been approximated through polynomial expressions, that together form a
convex region, and then the maximal ellipsoid is found inside this region using
convex optimisation. However, the present work tries to cover a gap in the ex-
isting literature on finding a maximal convex shape, like a sphere, inside the
Wco of a spatial CDPR, harnessing the exact closed-form definition of the Wco .
The proposed method of obtaining the maximal spheres also promises to be
computationally efficient as the closed-form expressions are being used for its
computation. The maximal sphere being found is tangent to the boundary of
the Wco of a 7-SCDPR which leads to seven different systems, each consisting
of three cubic equations in three unknowns. The unknowns in these systems are
the coordinates of the points of tangency of the sphere with the seven cubic
surfaces. These seven surfaces form parts of the boundary of the Wco of the
7-SCDPR. Each of the seven systems of equations mentioned above is solved
through Sylvester’s dialytic elimination method (see, e.g., [7], pp. 86-87), which
is converted to a generalised eigenvalue problem (see, e.g., [8], pp. 414-416). The
solution of this eigenvalue problem provides the points of tangency of the sphere
with the boundary surface of Wco , which further leads to the desired sphere.
The details of this formulation, as described above, have been presented in Sec-
tion 2, which is then followed by a numerical example in Section 3 and finally,
the paper is concluded in Section 4.
Identifying the largest sphere inscribed in the constant orientation… 163
2 Mathematical Formulation
This section presents the description of the geometry of an SCDPR and the
formulation of the equations describing the boundary of its Wco .
2.1 Geometry of the robot
Fig. 1: Geometry of a general cable-driven parallel robot (data adapted from [9]).
In this section, the geometric model of the robot has been described and the
relevant notations have been introduced. The schematic of an SCDPR is shown in
Fig. 1, which has been adapted from [9]. The robot consists of a moving platform
(MP) connected by m cables to m fixed points Bi , i = 1, . . . , m. The ith cable,
Ci , is attached to the MP at Pi , and winds at Bi around an actuated winch. To
analyse the motion of the MP, two frames of reference are considered: the global
coordinate frame B, which is fixed to the base, and the moving frame P, which
is attached to the reference point p on the MP. The following notations have
been used in the rest of the paper:
– Vector bi = [bix , biy , biz ] ∈ R3 represents the position of the actuated
winch Bi in the global frame B.
– Vector pi = [pix , piy , piz ] ∈ R3 is a constant vector and represents the
position of the attachment point Pi of the ith cable in frame P.
– Vector p = [x, y, z] ∈ R3 is expressed in B and represents the position of
the point P w.r.t. point O.
– Vector li points from Pi to Bi , and its norm represents the length of Ci . In

this work, all the cables have been assumed to be taut and thus Ci can be
represented geometrically as a straight line.
– The orientation of the MP w.r.t. B is represented by the rotation matrix
R ∈ SO(3), and has been parametrised by the X-Y-Z body-fixed Euler-
angles, (α, β, γ). Thus, the pose of the MP is described by the pair (p, R) ∈ SE(3).
In the aforementioned notation, vector li along Ci is obtained as:
li = bi − p − Rpi , for i = 1, . . . , m. (1)
li
The unit vector defining the axis of the wrench along Ci is given by ui = ,
li 2
where · 2 represents the Euclidean norm. The wrench applied at P by Ci is
ti ŵi , where:
ŵi = [u
i , (Rpi × ui ) ] . (2)
In Eq. (2), ŵi is the ith screw which carries the wrench ti ŵi , where ti is the
tension in Ci . It is to be noted that ti > 0 ∀ i, since the cables have been
assumed to be taut. Therefore, for a 7-SCDPR, the force and torque equilibrium
of the MP leads to the following equation:

Ŵ (p, R)t + wp = 0, with Ŵ = ŵ1 | . . . | ŵ7 , t = t1 , . . . , t7 . (3)
In Eq. (3), wp is the external wrench acting on the MP at P , t is the vector

of cable tensions and Ŵ is the wrench transformation matrix. The matrix Ŵ
is a pose-dependent 6 × 7 matrix for a 7-SCDPR. Henceforth, for any vector v,
v 0, v 0, v ≺ 0 and v 0 would mean that all the components of v are
greater than zero, greater than or equal to zero, smaller than zero and smaller
than or equal to zero, respectively.
2.2 Wrench-closure workspace (WCW) and the derivation of the

singularity-free sphere (SFS)
In this section, the largest sphere in R3 which is completely inside the Wco of a
7-SCDPR, is being found. There are different criteria [1, 6] for deciding whether
a pose of a CDPR belongs to the workspace, the most important one being the
ability of the robot to be statically balanced with positive tensions in the cables.
This condition forms the basis for the following definition of the WCW of a
CDPR:
A pose (p, R) belongs to the WCW of a 7-SCDPR, if for every wrench1 wp ∈ R6 ,
there exists at least one vector of cable-tensions, 0 ≺ t ∈ R7 such that Ŵ t + wp = 0.
1
Strictly speaking, wp ∈ se(3)∗ which is locally approximated by R6 (see,
e.g., [10], pp. 420).
Theorems 1, 2 and 3, introduced in [1], help in obtaining a geometric description

of the boundary of the Wco , for the 7-SCDPR. The boundary of Wco can be
described by cubic polynomials in x, y and z, obtained from the null-space of
Ŵ [2] and is represented here by the following set of equations:
η0i (x, y, z) = 0, i = 1, . . . , 7. (4)
Equations (4) are of the following form:
a1 x3 + a2 x2 y + a3 x2 z + a4 x2 + a5 xy 2 + a6 xyz + a7 xy + a8 xz 2
+ a9 xz + a10 x + a11 y 3 + a12 y 2 z + a13 y 2 + a14 yz 2 + a15 yz + a16 y (5)
3 2
+ a17 z + a18 z + a19 z + a20 = 0.
The coefficients ak ∈ R, k = 1, . . . , 20 are functions of the X-Y-Z Euler-angles

α, β, γ, and the architecture parameters bi and pi , i = 1, . . . , 7 [2]. A pose
(p, R) of a CDPR is considered to be singular, when the wrench transformation
matrix, Ŵ , becomes rank deficient [11], and thus by the definition of WCW
in [1], that particular pose does not belong to the WCW of the robot. As men-
tioned earlier, through this work, the description of a sphere centred around a
prescribed point, say, s0 = [x0 , y0 , z0 ] , is being sought. This sphere is re-
quired to lie completely inside the Wco of a 7- SCDPR and is, therefore, free
of the aforementioned singularity. In the rest of the work, this sphere is being
referred to as the singularity-free sphere (SFS), following [3], where it has been
used in the same sense in the context of semi-regular Stewart platform manipu-
lator (SRSPM). The equation of the SFS is given by:
g(x, y, z) = (x − x0 )2 + (y − y0 )2 + (z − z0 )2 − r2 = 0, (6)
where r is the radius of the sphere. Let, s = [x, y, z] be the point of tangency
between the sphere and the boundary surface. Therefore, at s, the normals to
these two surfaces should align, giving rise to the following equations:
∇η0i × ∇g = 0 ⇒ hij (x, y, z) = 0, j = 1, 2, 3 for i = 1, . . . , 7. (7)
As only two of the three equations hij = 0 are linearly independent for any i, any
two along with the equation of the boundary surface, namely one of Eqs. (4) with
which the tangency is being sought, form a system of three linearly independent
equations in x, y, and z. One needs to solve seven such systems of three cubic
equations in three unknowns. In each case, every set of real solutions represents
a sphere that is tangent to the boundary of the Wco of the 7-SCDPR. The one
with the smallest value of the radius, r, among these is the desired SFS.
2.3 Solution methodology
The degrees of hi1 , hi2 and hi3 in (x, y, z) are (2, 3, 3), (3, 2, 3) and (3, 3, 2),
respectively, while the total degree in x, y, and z equals 3 for each polynomial.
Fig. 2: Tangency of the sphere with the boundary surface at point s
Considering the first two equations hi1 = 0 and hi2 = 0 along with the equation
η0i = 0, the ith system of three cubic equations in x, y, and z, is obtained. Such
cubic systems can be solved by using Sylvester’s dialytic elimination method,
which involves the formulation of the Sylvester matrix and then the construction
of an eigenvalue problem involving the said Sylvester matrix. The solution pro-
cedure for the first system of three cubic equations, given below in Eqs. (8a-8c),
is explained in the following.
η01 (x, y, z) = 0, (8a)

h11 (x, y, z) = 0, (8b)
h12 (x, y, z) = 0. (8c)
Forming the Sylvester matrix: Following [7], pp. 86-87, first, one can rewrite
Eqs. (8a-8c) by suppressing the unknown x, by absorbing it into the coefficients,
leading to the equations:
m=3−k
k=3
dkm,i y k z m = 0,

i = 1, 2, 3, (9)
k=0
m=0
where dkm,i is a polynomial in x of degree (3 − k − m). There are 10 monomi-

als in y and z in Eqs. (9). If the individual monomials are treated as a single
variable, then the system of Eqs. (9) represents a homogeneous linear system
of equations. In order to solve Eqs. (9) uniquely, new equations must be intro-
duced, so as to have as many linearly independent equations as the number of
distinct monomials. Multiplying each of Eqs. (9) by y and z, respectively, one
obtains six new linearly independent equations, which together contain five new
monomials. Hence, the total number of distinct monomials rises to fifteen and
thus, one needs six additional equations introducing no new monomials. The six
new equations are obtained in the following manner. The system of Eqs. (9) can
be cast in the form:
Ei y 3 + Fi z + Gi = 0, i = 1, 2, 3, (10)
where Ei is a constant, Fi is a quadratic polynomial in y and z and Gi is a
quadratic polynomial in y. The system of Eqs. (10) can be organised in the
matrix form as:
⎡ ⎤⎡ ⎤ ⎡ ⎤
E1 F1 G1 y3 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎦ ⎣ z ⎦ = ⎣0⎦ . (11)
⎢E2 F2 G2 ⎥ ⎢ ⎥ ⎢ ⎥
⎣
E3 F3 G3 1 0
Equation (11) admits a non-trivial solution iff :

⎡ ⎤
E1 F1 G1
⎢ ⎥
⎣E2 F2 G2 ⎦ = 0.
det ⎢ ⎥ (12)
E 3 F3 G 3
The determinant in the LHS of Eq. (12) expands into a quartic polynomial in y
and z, which is linearly independent of Eqs. (9). Five more equations can be ob-
tained in an analogous manner by collecting in Eqs. (9) each of the variable-pairs
from the set, S = {(y 2 , z 2 ), (y, z 3 ), (y 2 , z), (y, z 2 ), (y, z)} and writing them in
a manner similar to that in Eq. (10). Thus, a total of six additional quartic
polynomials in y and z are obtained. These six equations, along with the nine
equations obtained earlier, i.e., Eqs. (9) and the equations obtained by multiply-
ing each of Eqs. (9) by y and z, respectively, form a homogeneous linear system
of equations which can be written as:
S(x)v = 0, (13)
where S(x) ∈ R15×15 , whose elements are polynomials in x, with maximum de-
gree of seven, and v is the following vector containing 15 monomials in y, z:
v = [y 4 , y 3 z, y 3 , y 2 z 2 , y 2 z, y 2 , yz 3 , yz 2 , yz, y, z 4 , z 3 , z 2 , z, 1] . (14)
Reduction to a generalised eigenvalue problem: The matrix S can be

expressed as:
S = H 1 x7 + H 2 x6 + H 3 x5 + H 4 x4 + H 5 x3 + H 6 x2 + H 7 x + H 8 , (15)
where H i ∈ R15×15 , i = 1, . . . , 8. The matrix equation (13) can be solved
through the following generalised eigenvalue problem (see, e.g., [8], pp. 414-416):
M 1 z = xM 2 z, (16)
where M 1 , M 2 ∈ R105×105 , and z = [x6 v , x5 v , . . . , v ] is the eigenvector

corresponding to the eigenvalue x. The matrices M 1 , M 2 are obtained as:
⎡
−H 2 − H 3 . . . −H 7 − H 8
⎤ ⎡
H1
⎤
0 ... 0
⎢ I I ...
⎢ ⎥ ⎢ ⎥
0 ... 0 0 ⎥ ⎢ 0 0⎥
M1 = ⎢ ⎥ , M2 = ⎢ ⎥.
⎢ ⎥ ⎢ ⎥
.. ..
⎢
⎣ . ⎥
⎦
⎢
⎣ . ⎥
⎦
0 0 ... I 0 0 0 ... I
In the above, I, 0 ∈ R15×15 are the identity and null matrices, respectively. The
eigenvalues of the generalised problem in Eq. (16) are also the roots of the equa-
tion, det(S(x)) = 0. However, as the problem of obtaining roots of such polyno-
mials obtained after the determinant expansion is prone to ill-conditioning (see,
e.g., [12]), Eq. (13) is posed as an eigenvalue problem. Once the eigenvalues and
eigenvectors of Eq. (16) have been computed, for a particular eigenvalue x, the
corresponding values of y and z are then computed from the known form of the
corresponding eigenvector z. In this way, all the points of tangency between the
sphere and a part of the boundary of Wco formed by the cubic surface, η01 = 0,
are obtained and the real solutions are identified. Equation (6) is then used to
compute the radii of the spheres, corresponding to each real point of tangency.
The smallest among all such radii provides the radius of the largest sphere tan-
gent to η01 = 0, which does not intersect it at any other point. Similar tangency
problems are then obtained for each of the remaining six cubic surfaces, which
form the other parts of the boundary of Wco , and the real points of tangency
are found. Then the radii of the largest spheres, centred at s0 , which are tan-
gent to each of these surfaces and do not intersect them at any other point, are
computed. As the largest spheres obtained for each of the seven surfaces are con-
centric, the smallest among all such spheres is guaranteed to be non-intersecting
with the other six surfaces. This smallest sphere, thus, forms the desired SFS.
3 Results
A representative problem was solved in Mathematica [13] operating with its de-
fault numerical precision. The values of the architecture parameters used for this
problem have been adopted from [14]. The fixed centre of the SFS is chosen to be
at s0 = [0.5 m, 2.0 m, 0.5 m] and the orientation parameters in terms of X-Y-Z
Euler-angles are considered to be α = 0.5 rad, β = 0.5 rad, and γ = 0.5 rad. The
equations of the seven cubic surfaces forming parts of the boundary of the Wco ,
for these input values are of the form as described by Eq. (5).
While solving the tangency problem of the sphere in Eq. (6) with all the afore-
mentioned cubic surfaces, there can be possible inclusions of extraneous roots
because of the non-homogeneous nature of Eqs. (4-5) (see, e.g., [7], pp. 87-90).
Therefore, the solutions obtained for each system have been substituted back
into the three original equations (i.e., Eqs. (4, 7)) and the residue of the vector
η i = [η0i , hi1 , hi2 ] for i = 1, . . . , 7, is computed in terms of its Euclidean norm:
ei = η i 2 for i = 1, . . . , 7. (17)
Fig. 3: Sphere with r = 0.376 m, Fig. 4: Sphere with parts lying outside
tangent to the surface η07 = 0 at s the Wco , r = 0.451 m
Based on the obtained value of ei for i = 1, . . . , 7, all the extraneous roots have
been filtered out. For the example at hand, the value of ei for i = 1, . . . , 7, was ob-
tained to be less than 10−14 for all the admissible solutions. The minimum value
of the radius, r, obtained corresponding to all the real points of tangency, is found
to be 0.376 m. For the given inputs, this is the radius of the desired SFS, that is
tangential to the boundary surface of Wco at s = [0.449 m, 1.989 m, 0.128 m] , as
shown in Fig. 3. For spheres with radii greater than the aforementioned critical
radius, tangency with the boundary of Wco may occur at some point, but there
will always be other points where the spheres will intersect the boundary of Wco .
Such a case has been depicted in Fig. 4. The validation of the obtained SFS has
also been done by generating 100,000 random points inside a sphere of radius 1.2
times that of the SFS. As anticipated, some of these points lie outside the Wco
of the 7-SCDPR but none of these points, outside the Wco , lie inside the SFS.
This validates the fact that the obtained SFS lies completely inside the Wco .
This has been depicted in Fig. 6, where the points in red are outside the Wco .
In Fig. 5, the Wco has been shown along with the above obtained SFS 2 , which
can be seen to lie completely inside the boundary of the Wco . This Wco has been
approximated numerically utilising its analytical description presented in [1, 2]
and has been plotted using the built-in function RegionPlot3D of Mathematica.
4 Conclusions
A method for computing the maximal sphere inside the constant orientation
wrench-closure workspace (Wco ) of a 7-SCDPR has been presented in this pa-
per. The formulation leads to seven systems consisting of three cubic equations
each, in terms of the coordinates of the points of tangency between the sphere
and the boundary surface of the Wco . These systems are then solved succes-
sively to obtain the desired singularity-free sphere (SFS). As per the best of the
knowledge of the authors, no such analytical formulation appears in the existing
literature for finding such a maximal sphere inside the Wco of a spatial CDPR.
2
For this example, it took about 1.3 seconds to obtain the radius of the desired SFS
using Mathematica, on a computer with Intel Core i7-4790 CPU @ 3.60 GHz
and 16 GB of RAM.
Fig. 5: Maximal sphere inside the Fig. 6: Points in red lie outside the
Wco of a 7-SCDPR, r = 0.376 m Wco , and none are inside the SFS
The need for such a method, with the intention of using it further in the design
of CDPRs, motivated this work. For the geometric design of CDPRs, one of the
objectives can be to determine the architecture parameters of a CDPR for which
the resulting SFS would completely enclose a prescribed region in R3 . The pro-
posed method also promises to be computationally efficient as the closed-form
expressions are being used in the computation of the maximal sphere. For the
aforementioned example, it took about 1.3 seconds to obtain the radius of the
desired SFS using Mathematica. However, the computation time may be reduced
further by implementing the proposed methodology in the C programming lan-
guage.
This method also assumes that the desired centre of the sphere being sought,
has been specified a priori. One might think that it would be preferable to calcu-
late the largest sphere regardless of its centre and then shift the location of the
robot’s base to suit the purposes of the desired application. However, this idea
works when the CDPR is to operate in a workspace with the orientation of its
MP held fixed. When the orientation of the MP also varies, different orientation
slices are going to have maximal spheres with different centre-point locations.
This would make the task of finding the maximal sphere for the total orientation
workspace comparatively difficult.
The above-mentioned choice of sphere as the convex bounding surface for each
orientation is not unique, and can be modified to fit the specific application
at hand. In this work, the spherical geometry has been adopted for the Wco ,
motivated by the practical considerations listed below.
– The sphere has the highest order of symmetry among all the simple closed
surfaces and it encloses a convex region. The intersection of any number of
concentric spheres is still a convex region, i.e., the smallest of the spheres,
which can be identified trivially. While such convex regions of intersection
can be computed easily in the case of other convex regions as well, the com-
putational cost is higher in all of them. This point has been illustrated in
Figs. 7 and 8, where the region represented by Fig. 8 is computed as the in-
Fig. 7: Three concentric ellipsoids Fig. 8: The convex region of intersec-

of the same volume tion
tersection of three concentric ellipsoids, which have been depicted in Fig. 7.

Given this form of the workspace of the robot, to determine if a given point
is inside it, or outside, one needs to perform more computations, than to
simply compare two real numbers, namely, the distance of the said point
from the centre of the SFS, O, and the radius of the SFS, had spheres been
used instead. In real-life design problems, as in [15], the number of such
sections used is of the order of tens, and hence the situation is a lot more
computationally intensive than what Figs. 7 and 8 suggest.
– Following the same methodology for obtaining the maximal sphere, a maxi-
mal ellipsoid inside the Wco could have been obtained as well, by replacing
the equation of the sphere by that of an ellipsoid in Eq. (6), while keeping
the rest of the formulation, essentially, the same. If one were to choose these
ellipsoids instead of the spheres for the above-mentioned design problem, the
difference in the CPU time required for one section would be insignificant.
However, one needs to keep in mind that the actual utility of such compu-
tations is to enable the design algorithms and any additional computation
makes a significant impact in the total time needed in the design computa-
tions.
– In light of the above considerations, the spherical geometry has been chosen,
in the present work, out of necessity and not to simplify the problem.
References
1. Gouttefarde, M., Gosselin, C. M.: On the Properties and the Determination of

the Wrench-Closure Workspace of Planar Parallel Cable-Driven Mechanisms. In:
Proceedings of DETC’04, ASME 2004 International Design Engineering Technical
Conferences and Computers and Information in Engineering Conference, Salt Lake
City, Utah, USA, pp. 337–346. American Society of Mechanical Engineers (2004)
2. Sheng, Z., Park, J.-H., Stegall, P., Agrawal, S. K.: Analytic Determination of
Wrench Closure Workspace of Spatial Cable Driven Parallel Mechanisms. In: Pro-
ceedings of the ASME 2015 International Design Engineering Technical Confer-
ences and Computers and Information in Engineering Conference, Boston, Mas-
sachusetts, USA, p. V05CT08A048; 10 pages. American Society of Mechanical
Engineers (2015)
3. Nag, A., Reddy, V., Agarwal, S., Bandyopadhyay, S.: Identifying Singularity-Free
Spheres in the Position Workspace of Semi-regular Stewart Platform Manipulators.
In: J. Lenarčič, J.P. Merlet (eds.) Advances in Robot Kinematics 2016, pp. 421–
430. Springer International Publishing, Cham, Switzerland (2018). DOI 10.1007/
978-3-319-56802-7
4. Jiang, Q., Gosselin, C. M.: Determination of the maximal singularity-free orienta-
tion workspace for the Gough-Stewart platform. Mechanism and Machine Theory
44(6), 1281–1293 (2009)
5. Bayani, H., Masouleh, M. T., Karimi, A., Cardou, P., Ebrahimi, M.: On the deter-
mination of the maximal inscribed ellipsoid in the Wrench-Feasible Workspace of
the cable-driven parallel robots. In: Proceedings of the 2nd RSI/ISM International
Conference on Robotics and Mechatronics (ICRoM), Tehran, Iran, pp. 422–427.
IEEE (2014)
6. Gouttefarde, M., Merlet, J.-P., Daney, D.: Wrench-Feasible Workspace of Parallel
Cable-Driven Mechanisms. In: 2007 IEEE International Conference on Robotics
and Automation, Roma, Italy, pp. 1492–1497. IEEE (2007)
7. Salmon, G.: Lessons Introductory to the Modern Higher Algebra (4th Ed.). Hodges,
Figgis, and Co., Grafton Street, Dublin, Ireland (1885)
8. Golub, G. H., Van Loan, C. F.: Matrix Computations. Johns Hopkins Studies in
the Mathematical Sciences. Johns Hopkins University Press, Baltimore, Maryland,
USA (2013)
9. Azizian, K., Cardou, P.: The Dimensional Synthesis of Spatial Cable-Driven Par-
allel Mechanisms. Journal of Mechanisms and Robotics 5(4), 044,502; 8 pages
(2013)
10. Murray, R. M., Li, Z., Sastry, S. S.: A Mathematical Introduction to Robotic
Manipulation. CRC Press, Boca Raton, Florida, USA (1994)
11. Verhoeven, R.: Analysis of the Workspace of Tendon-based Stewart Platforms.
PhD thesis, Germany: University of Duisburg-Essen (2004)
12. Wilkinson, J. H.: The Evaluation of the Zeros of Ill-conditioned Polynomials. Part
I and II. Numerische Mathematik 1(1), 150–166 and 167–180 (1959). DOI 10.
1007/BF01386382
13. Wolfram Research, Inc.: Mathematica, Version 11.1.1. Champaign, IL (2018)
14. Pott, A., Kraus, W.: Determination of the Wrench-Closure Translational
Workspace in Closed-Form for Cable-Driven Parallel Robots. In: 2016 IEEE In-
ternational Conference on Robotics and Automation (ICRA), Stockholm, Sweden,
pp. 882–887 (2016). DOI 10.1109/ICRA.2016.7487218
15. Kilaru, J., Karnam, M.K., Agarwal, S., Bandyopadhyay, S.: Optimal design of
parallel manipulators based on their dynamic performance. In: Proceedings of the
14th IFToMM World Congress, pp. 406–412. Tapei, Taiwan (25-30, October, 2015)
A Bounding Volume of the Cable Span for Fast
Collision Avoidance Verification
M. Lesellier, M. Gouttefarde
LIRMM, Université de Montpellier, CNRS, Montpellier France

ICUBE, Université de Strasbourg
Abstract. The problem of verifying the absence of collision between a

cable and the mobile part(s) of a device located on-board the mobile
platform of a Cable-Driven Parallel Robot (CDPR) is addressed. The
set of all positions taken by one cable of the CDPR for all the poses
of the mobile platform in a prescribed workspace is called the cable
span. A simple bounding volume approximation of the cable span is
proposed in this paper. This bounding volume is a polyhedron and the
characterization of the faces of this polyhedron is discussed. Using this
polyhedron as a bounding volume of the cable span allows to accelerate
computations related to collision avoidance checking.
Keywords: Cable-Driven Parallel Robot, Cable, Span, Collision
1 Introduction
Cable-Driven Parallel Robots (CDPRs) consist of a mobile platform driven by
cables. Using cables instead of rigid links gives CDPRs interesting characteristics
such as a potentially very large workspace, large payload-to-weight ratio, and
high dynamics capabilities.
However, one of the major drawbacks of CDPRs is their low stiffness due to
the use of cables. This low stiffness may be the cause of vibrations affecting the
platform positioning accuracy. Several techniques can be used to damp these
vibrations: input shaping [1], modal space control [2], or active damping by
creating a transient wrench to compensate the vibrations. This wrench can be
created by additional actuated stabilizing mechanical devices placed on board
the mobile platform [3–6]. In reference [5], a stabilizer consisting of rotating arms
is located on-board the CDPR mobile platform to actively damp vibrations, as
illustrated in Figure 1. The work reported in the present paper is part of a project
aiming to embed a similar stabilizer on the platform of the 6 Degree-of-Freedom
(DOF) CDPR CoGiRo [7], shown in Figure 2.
The design of CoGiRo ensures that there is no cable-cable collision nor cable-
platform collision within the workspace of this CDPR. However, when active
mechanical devices are placed on-board the CDPR mobile platform, the moving
parts of these on-board devices may collide with the cables or platform. This
eventuality of collisions is an issue that must be taken into account in the design

174 M. Lesellier and M. Gouttefarde
−
→
z0
• • − •
A3 A1 →
y0 A2
k 3 , l 03
k2 , l02
k1 , l01
m1
λ1
m2 B3 B2 •
B1 • θ1
→
−
zp
• r 1 O1
λ2 θ2 →
−yp
•
O2 r 2 Gp r 3
2π − θ3
O3
−
→
z0 λ3
−
→
zp −
→
yp
α m3
Gp −
→
y0
Fig. 1: Planar 3-DOF CDPR with 3 stabilization arms
Fig. 2: 6-DOF CDPR CoGiRo
optimization of on-board active devices. In order to prevent such collisions, a

first step is to determine the volume of space occupied by each cable when
the CDPR platform moves throughout a given workspace. Indeed, during the
platform motions, the positions of the cables change. The set of all positions
taken by one cable for all the poses of the platform in the CDPR workspace is
called cable span [8]. The latter reference is the only previous work dealing with
the issue of determining the cable span. It proposes a method to represent the
cable span by a volume object but it does not deal with the issue of using this
volume to check the absence of collisions.
While the problem of cable-cable interferences as well as cable-platform and
cable-object collisions has been addressed several times in the literature, e.g.
[9–13], to the best of our knowledge, the problem of avoiding collisions between
a cable and the mobile part(s) of a device located on-board the CDPR platform
A Bounding Volume of the Cable Span for Fast Collision Avoidance Verification 175
has never been treated yet. Hence, the contribution of this paper is a method to
check the absence of such collisions within a box-shaped workspace of a CDPR.
In order to accelerate computations related to collision avoidance checking, this
method is based on an original bounding volume approximation of the cable
span which is simpler than the one discussed in [8]. Moreover, this cable span
approximation is calculated with respect to the mobile platform frame to enable
simple testing of collisions with on-board devices.
This paper is organized as follows. The cable span of a CDPR is defined
in Section 2. Based on the approximation of the cable span introduced in [8],
two algorithms are discussed in Section 3 to test the eventuality of a collision
between a cable and an object located on-board the mobile platform. A new
and simpler bounding volume approximation of the cable span is proposed in
Section 4. The characterization of the faces of this bounding volume and the
corresponding collision avoidance testing are finally presented in Section 5.
2 Definition of the cable span

The frame attached to the CDPR mobile platform is denoted Fp and the fixed
base frame is denoted F0 . A vector v expressed in frame F0 is denoted by 0v
and in Fp by pv . The pose of the platform is given by the vector x , composed
of its position vector p in F0 and of a vector θ p of three Euler angles defining
the orientation of Fp in F0 . Let R (θθ p ) be the rotation matrix describing the
orientation of the platform in F0 . Each cable is considered to be a straight line
segment between its base drawing point Ai and its platform fixing point Bi .
These notations are shown in Figure 3.
The vector along the straight line segment of cable i, denoted u i , is given in
frame F0 by:
0
u i = 0p + R (θθ p )pb i − 0a i (1)
The interval that contains all values of a scalar variable e is denoted [e, e],
with e the minimal value of e (lower bound) and e its maximal value (upper
bound). The concatenation operator is written ||. In this paper, the prescribed
workspace W of a 6-DOF CDPR is defined as follows:
W = { x = {0p , θ p } | { 0p ∈ [x, x] × [y, y] × [z, z]}

(2)
|| {θθ p ∈ [αp , αp ] × [βp , βp ] × [γp , γp ] } }
It is a box-shaped workspace in the sense that, for a given orientation θ p , the

set of positions lying inside W is a box. In the remainder of this paper, it is
assumed that W is fully included in the wrench-feasible workspace [14, 15] and,
hence, that all poses in W can be reached with feasible cable tensions.
The cable span of cable i with respect to F0 is defined as CS 0i = {0u i | x ∈
W}. Vector 0u i is defined in (1) and according to the right-hand side of this
equation and to (2), CS 0i is the Minkowski sum of three sets:
– The box [x, x] × [y, y] × [z, z] (position vectors 0p )
Ai• ••
•
ui
•• ••
•
•Bi
•
•Fp •
p •
•
F0
•
Fig. 3: Schematic of the CoGiRo CDPR
R (θθ p )pb i | θ p ∈ [αp , αp ] × [βp , βp ] × [γp , γp ]}

– The portion of a sphere {R
– The point Ai having as position vector 0a i
Note that pb i and 0a i are constant vectors. Hence, when the Euler angle set
[αp , αp ] × [βp , βp ] × [γp , γp ] is not reduced to a single value, the cable span is a
relatively complex geometric object (not necessarily convex) even for the simple
definition of the workspace W given in (2). In [8], based on a discretization of the
CDPR workspace and for a fixed orientation of the mobile platform, the cable
span is approximated as a so-called generalized cone (a kind of cone whose cross
section is a polygon not necessarily convex).
With the aim of checking the absence of collisions between a cable and the
mobile part(s) of a device located on-board the mobile platform of a CDPR, the
present paper deals with the cable span calculated with respect to the mobile
platform frame Fp . For cable i, this cable span is defined as CS pi = {−pu i | x ∈
W} where:
−pu i = R (θθ p ) ( 0ai −0 p ) − pb i (3)
Equation (3) is directly obtained from (1) by multiplying both sides by R (θθ p ) .
CS pi is thus the translation of the set {R R (θθ p ) ( 0ai −0 p ) | x ∈ W} by vector
p
b i . This set corresponds to the volume swept by the box { 0ai −0 p | 0p ∈
[x, x] × [y, y] × [z, z]} when this box is rotated by R (θθ p ) for all θ p in [αp , αp ] ×
[βp , βp ]×[γp , γp ]. Similarly to the case of the cable span CS 0i calculated in F0 , the
cable span CS pi is thus a relatively complex three-dimensional geometric object
which is not necessarily convex.
Q
×Q
Bi •
Fig. 4: Convex polyhedral cone containing the cable span
3 Collision avoidance checking
In [8], based on a discretization of the CDPR workspace W, the cable span

CS 0i is approximated as a so-called generalized cone. The same method could
be applied to the cable span CS pi . This generalized cone approximation is not
necessarily convex (see Figure 3 in [8]) and may contain a large number of vertices
depending on the number of points in the workspace discretization (and on the
workspace definition). Therefore, it is not well suited for checking the absence
of collision between a cable and mechanical parts on-board the mobile platform
since this check consists in verifying that the volumes swept by the mechanical
parts during their motions are fully outside of the cable span.
As an example, let us consider the convex polyhedral cone shown in Figure 4.
Note that this cone is a simpler geometric object than the generalized cone
considered in [8]. Let us also consider the simple case of testing whether or not a
point Q is located inside this convex polyhedral cone. Several algorithms can be
used to determine if Q is inside the cone, two of them are now briefly described.
First, a point-in-polygon approach, whose principle is illustrated in Figure 4,
performs the following steps:
1. Obtain Q the projection of Q on the axis of the cone. This defines a ratio k
of, on the one hand, the distance between the base of the cone and Q and,
on the other hand, the length of the cone axis.
2. Apply the ratio k to the distances between the vertices of the base of the
cone, which is a polygon, and the center of this polygon. The new polygon
hereby obtained called P is the section of the cone containing the point Q .
3. Test whether or not Q belongs to the polygon P .
The third step can be implemented with a ray crossing algorithm [16] (which
can also be used for non-convex polygons) whose complexity is O(n), n being
the number of vertices of the base polygon.
Another approach consists first in determining the faces of the convex poly-
hedral cone. These faces may be given by the convex hull procedure or they can
be determined within the polar sorting phase of the generalized cone determina-
tion as detailed in [8]. Then, point Q is located inside the polyhedral cone if and
only if it is located on the same side of each of the planes containing the cone
faces. This second approach complexity is also O(n) where n is the number of
vertices (or edges) of the base polygon.
The number of vertices n depends on the number of points used to discretize
the CDPR prescribed workspace. Hence, the computational time needed to test
if a given set is not interfering with the cable span, i.e. to check the absence of
collision, depends on n. Moreover, the check must be done N times, N being the
number of cables. Hence, using one of the two algorithms described above within
an iterative optimization process slows down the whole procedure when n is
(relatively) large. Using a simple, yet not overly conservative, bounding volume
approximation of the cable span is thus important to enhance computational
efficiency.
4 A simple bounding volume approximation of the cable

span
Based on a discretization of the CDPR workspace W, a simple bounding volume

of the cable span CS pi (cable span in Fp ) can be obtained as follows.
Cable i is considered as a straight line segment between points Ai and Bi .
In the mobile platform frame Fp , the position of Bi is constant whereas the
position vector of point Ai is R (θθ p ) ( 0ai −0 p ) and thus it depends on the mobile
platform pose x = {0p , θ p } ∈ W. In Fp , the discretization of W generates a set
of positions of point Ai and taking the convex hull of all these points produces a
convex polyhedron. If the discretization is fine enough, this convex polyhedron is
a good approximation of all possible positions of Ai in Fp when x ∈ W. However,
as discussed in Section 3, the number of vertices of the bounding volume of the
cable span should be kept small. Hence, instead of the convex hull, the smallest
box B (aligned with the axes of Fp ) containing all positions of point Ai for x in
the discretized workspace W is calculated. To this end, it suffices to determine,
along each of the three coordinate axes of Fp , the minimum and maximum
coordinates of this set of positions of Ai .
The box B encloses all positions of points Ai in Fp for x in the discretized
workspace W, as illustrated in Figure 5. Hence, the polyhedron P obtained by
connecting the fixed point Bi to B is a bounding volume approximation of the
cable span CS pi , i.e., a bounding volume of all the positions of the cable segment
Ai Bi for x in the discretized workspace W. In the remainder of this paper, it is
assumed that the discretization of the workspace is such that this polyhedron P
encloses the cable span CS pi .
An example of such a polyhedron P forming a bounding volume approxima-
tion of the cable span is shown in Figure 6. As can be seen in this figure, P
consists of the union of the box B and of a pyramid (convex polyhedral cone)
having the fixed point Bi as apex. The base of this pyramid is a convex polygon
whose vertices are some of the vertices of the box B.
Fig. 5: Convex hull of all possible po- Fig. 6: Bounding volume approxima-
sitions of Ai (and of the fixed point tion of the cable span: The union of
Bi ) and box containing the convex a box and a pyramid
hull
5 Collision avoidance checking with the new cable span

bounding volume approximation
In Section 4, a polyhedron P forming a bounding volume approximation of the

cable span CS pi has been introduced. A method to ensure collision avoidance
between a cable and a moving device on-board the CDPR mobile platform con-
sists in testing if the set of all possible positions of this device does not intersect
P. Indeed, P being a bounding volume of the cable span CS pi , no intersection
with P means no intersection with the cable span and thus no possible colli-
sion with the cable in the workspace W. In order to test if a point or a set is
outside polyhedron P, the determination of the polyhedron faces is useful. In-
deed, the polyhedron can be defined as the intersection of halfspaces bounded by
the planes containing the faces (mathematically, a system of linear inequalities).
Then, a point or a set is outside the polyhedron if and only if it is outside of
at least one of these halfspaces (it violates at least one of the inequalities). The
method to determine the faces of the polyhedron P forming a bounding volume
approximation of the cable span is described below.
As mentioned at the end of Section 4, the polyhedron P is the the union of
the box B and of a pyramid, the apex of the pyramid being the fixed point Bi
(Fig. 6). The number of faces of this polyhedron depends on the number of faces
of the box B that are visible from point Bi . Three cases must be distinguished.
– One face of B is visible from Bi : As shown in Figure 7, P consists of nine

faces. Four of these faces are triangles which form the pyramid. Point Bi ,
the apex of the pyramid, is a vertex common to these four faces. The five
other faces of P are the faces of the box B that are not visible from Bi .
– Two faces of B are visible from Bi : As shown in Figure 8, P consists of ten
faces. Six of these faces form the pyramid with point Bi as a common vertex.
The four other faces of P are the faces of B that are not visible from Bi .
Bi
• •
• •
Bi •
•
•
• •
B •
B
Fig. 7: Faces of the polyhedron P
if one face of B is visible from Bi Fig. 8: Faces of the polyhedron P
if two faces of B are visible from Bi
Bi
•
• •
• •
Fig. 9: Faces of the polyhedron P

if three faces of B are visible from Bi
– Three faces of B are visible from Bi : As shown in Figure 9, P consists of nine

faces. Six of these faces form the pyramid with point Bi as common vertex.
The three other faces of P are the faces of B that are not visible from Bi .
In addition to these three cases, particular cases must be dealt with. These
particular cases appear when point Bi lies in one or several planes containing
the faces of the box B. For simplicity, these particular cases are not detailed in
this paper. They do not cause any particular difficulty.
Based on this characterization of the faces of the polyhedron P, the mathe-
matical description of each of these faces is straightforward. Let us consider face
k of P. Its mathematical description consists in a point Pk lying on the face and
a vector n k normal to the face and pointing to the outside of P. Point Pk and
vector n k define the plane containing face k. A point Q is then outside of the
polyhedron if and only if, for at least one face k, it is not on the same side of
−−→
this plane than P, i.e., if and only if there exists k such that Pk Q · n k > 0.
In the case of a face k of P which is also a face of the box B (a rectangle),
point Pk can for instance be one vertex of the rectangle or its center and vector
n k is taken along the normal to this rectangle pointing to the outside of B. In the
case of a face of the pyramid having point Bi as apex, the face is a triangle whose
three vertices are known so that point Pk and vector n k are also straightforward
to define. Note that the determination of the faces of P and the calculation
of their mathematical descriptions (Pk and nk ) can be done offline, i.e., before
any use of P to test the absence of collision between a cable and a device on-
board the CDPR mobile platform. Typically, within an iterative optimization
(a) 3D view (b) Top view (projection along z0 )
Fig. 10: All the bounding volumes of the cable spans of the CDPR CoGiRo as
seen in the platform frame Fp
−−→
process, only dot products (Pk Q · n k ) need to be calculated. Consequently, the
computation cost of testing collision avoidance with the cables across the CDPR
workspace is small.
Figure 10 show the polyhedra P which are bounding volume of the cable
spans of the eight cables of the CDPR CoGiRo for a given workspace W. As
can be seen, the cable span bounding volumes represent a significant part of the
space around the platform and, in fact, surround the platform. This is mostly
due to the cable arrangement of CoGiRo [7]. This CDPR being suspended, it
can also be observed that the space below the platform is free.
6 Conclusion
This paper dealt with the problem of verifying the absence of collisions between
the cables of a CDPR and mobile devices located on-board the mobile platform,
across a prescribed workspace. The cable span is defined as the set of all positions
taken by one cable of the CDPR for all the poses of the mobile platform in the
prescribed workspace. In this paper, the cable span as seen in the platform
coordinate frame is considered. If the set of all possible motions of a device on-
board the mobile platform is fully outside of the cable span, then no collision
between the device and the cable can occur within the prescribed workspace.
However, the cable span being a relatively complex geometric object, a bounding
volume approximation of the cable span has been proposed in this paper. This
bounding volume is a polyhedron and the characterization of the faces of this
polyhedron has been given. Using this polyhedron as a bounding volume of
the cable span allows to accelerate computations related to collision avoidance
checking.
7 Acknowledgement
This work was supported by the ANR under grant ANR-15-CE10-0006, project
DexterWide.
References
1. Forrest Montgomery and Joshua Vaughan. Suppression of cable suspended parallel

manipulator vibration utilizing input shaping. In Conference on Control Technol-
ogy and Applications, 2017.
2. Xavier Weber, Loı̈c Cuvillon, and Jacques Gangloff. Active vibration canceling of
a cable-driven parallel robot in modal space. In IEEE International Conference
on Robotics and Automation, 2015.
3. Rushton, Mitchell. Vibration control in cable robots using a multi-axis reaction
system. Master’s thesis, Univ. Waterloo, 2016.
4. Xavier Weber, Loc Cuvillon, and Jacques Gangloff. Active vibration canceling of
a cable-driven parallel robot using reaction wheels. In IEEE/RSJ International
Conference on Intelligent Robots and Systems, 2014.
5. Maximilian Lesellier, Loı̈c Cuvillon, Jacques Gangloff, and Marc Gouttefarde. An
active stabilizer for cable-driven parallel robot vibration damping. In IEEE/RSJ
International Conference on Intelligent Robots and Systems, 2018.
6. R. de Rijk, M. Rushton, and A. Khajepour. Out-of-plane vibration control of a
planar cable-driven parallel robot. IEEE/ASME Transactions on Mechatronics,
23(4):1684–1692, 2018.
7. Marc Gouttefarde, Jean-François Collard, Nicolas Riehl, and Cédric Baradat. Ge-
ometry selection of a redundantly actuated cable-suspended parallel robot. IEEE
Transactions on Robotics, 31(2), 2015.
8. Andreas Pott. Determination of the cable span and cable deflection of cable-driven
parallel robots. In C. Gosselin, P. Cardou, T. Bruckmann, and A. Pott, editors,
Cable-Driven Parallel Robots. Springer, 2017.
9. Jean-Pierre Merlet. Analysis of the influence of wires interference on the workspace
of wire robots. In Advances in Robot Kinematics (ARK), 2004.
10. Martin J-D Otis, Simon Perreault, Thien-Ly Nguyen-Dang, Patrice Lambert, Marc
Gouttefarde, Denis Laurendeau, and Clément Gosselin. Determination and man-
agement of cable interferences between two 6-dof foot platforms in a cable-driven
locomotion interface. IEEE Transactions on Systems, Man, and Cybernetics-Part
A: Systems and Humans, 39(3), 2009.
11. Simon Perreault, Philippe Cardou, Clément M Gosselin, and Martin J-D Otis.
Geometric determination of the interference-free constant-orientation workspace
of parallel cable-driven mechanisms. Journal of Mechanisms and Robotics, 2(3),
2010.
12. D. Q. Nguyen and M. Gouttefarde. On the improvement of cable collision detection
algorithms. In T. Bruckmann and A. Pott, editors, Cable-Driven Parallel Robots,
pages 29–40. Springer, 2014.
13. A. Martin, S. Caro, and P. Cardon. Geometric determination of the cable-cylinder
interference regions in the workspace of a cable-driven parallel robot. In C. Gos-
selin, P. Cardou, T. Bruckmann, and A. Pott, editors, Cable-Driven Parallel Robots,
pages 117–127. Springer, 2017.
14. Paul Bosscher, Andrew T Riechel, and Imme Ebert-Uphoff. Wrench-feasible

workspace generation for cable-driven robots. IEEE Transactions on Robotics,
22(5):890–902, 2006.
15. Marc Gouttefarde, David Daney, and Jean-Pierre Merlet. Interval-analysis-based
determination of the wrench-feasible workspace of parallel cable-driven robots.
IEEE Transactions on Robotics, 27(1), 2011.
16. J. O’Rourke. Computational Geometry in C. Cambridge University Press, 2nd
edition edition, 1998.
Computation of the interference-free wrench
feasible workspace of a 3-DoF translational
tensegrity robot
Marc Arsenault
Laurentian University, Sudbury, Ontario, Canada P3E 2C6

[email protected]
Abstract. This paper proposes a method to compute the workspace

of a three-degree-of-freedom translational tensegrity robot. Equivalent
compression spring legs incorporating variable radius drums are used to
emulate linear compression springs that replace the struts of the tenseg-
rity system from which the robot is inspired. The workspace is com-
puted based on the interval analysis evaluation of constraints related to
the kinematics of the equivalent spring legs, the avoidance of interfer-
ences between the robot’s components and the need to generate required
wrenches on the robot’s mobile platform while ensuring acceptable ca-
ble tensions. Sufficient conditions for the satisfaction of these constraints
that may be suitably evaluated using interval analysis are developed.
Results suggest that the size of the workspace may be increased by in-
troducing pre-load in the robot’s components.
Keywords: tensegrity, workspace computation, interval analysis, wrench

feasible workspace, cable interference
1 Introduction
Inspired by Snelson’s [19] sculptures, the word tensegrity was originally intro-
duced by Fuller [7] to describe assemblies of axially-loaded members, where each
member is subjected either to tension or compression forces. Since cables may be
used for the members in tension, tensegrity systems have the potential to achieve
high strength-to-weight ratios, which make them attractive for the development
of robots where high acceleration motions are sought (e.g., pick and place ap-
plications). Moreover, in comparison to parallel cable-driven robots, tensegrity
robots avoid the need to surround their mobile platform (MP) with cables in or-
der to achieve wrench closure without relying on gravitational forces. Combined
with a self-stress capability [13], these attributes provide the required motivation
to explore the development of robots based on tensegrity systems.
Most often, tensegrity robots are developed by adding actuation to previously
documented systems [14]. For instance, Snelson’s X-shape tensegrity system [19]
has inspired many planar tensegrity robots (e.g., [2, 21]) as well as some spatial
robots [3]. The tensegrity simplex (Fig. 1(a)) has also frequently been used to de-
velop robots (e.g., [12, 16, 20]). However, it suffers from an inherent infinitesimal

186 M. Arsenault
Compressive strut or leg link Spherical joint Universal joint

Cable Revolute joint
(a) (b)
A2
A1 O X
Y A3
Z
C2 C3
Q
g B3 C1
P B1
B2 (c)
Fig. 1. Development of tensegrity robot: (a) tensegrity simplex, (b) reinforced tenseg-
rity simplex and (c) proposed robot architecture
mechanism [6] which severally affects its stiffness along a direction corresponding
to a screw-like motion between its opposing triangular faces. This issue is ad-
dressed in the reinforced tensegrity simplex [14] by adding three additional tensile
members (Fig. 1(b)). It may be shown, using the methods described in [13], that
the reinforced tensegrity simplex has three states of self-stress requiring forces
to be applied to three of its members in order to stress the remaining members.
The three-degree-of-freedom (3-DoF) translational tensegrity robot that is
the focus of this work, developed from the reinforced tensegrity simplex, has
already been proposed in [10] and a proof-of-concept prototype has been de-
veloped [4]. In [10], the boundaries of the robot’s reachable workspace were
Computation of the interference-free wrench feasible workspace… 187
determined based on kinematic constraints on the lengths of its members along

with the requirement to maintain tension in its cables with no external wrench
applied to its end-effector. However, potential interferences between the robot’s
members were not considered. In this work a method to compute the robot’s
interference-free wrench feasible workspace (WFW) is presented. The method is
based on the use of interval analysis [11] techniques within a branch-and-prune
algorithm to yield sets of poses that are guaranteed to be located within the
robot’s workspace. In comparison, the approach described in [10] represented
the robot’s workspace as a finite set of discrete poses.
2 Description of the tensegrity robot

The proposed 3-DoF translational tensegrity robot is illustrated schematically in
Fig. 1(c). The cables in the opposing triangular faces of the reinforced tensegrity
simplex (Fig. 1(b)) have been replaced by rigid plates corresponding to the
robot’s base (A1 A2 A3 ) and MP (B1 B2 B3 ). A fixed reference frame XY Z
is chosen to have its origin located at the centroid of A1 A2 A3 , its Y axis
directed toward node A3 and its Z axis perpendicular to the base plate and
directed toward the MP. In this way, the position of node Ai (i = 1, 2, 3) in
frame XY Z is ai = rb [− sin ψi , cos ψi , 0]T with ψi = i · (2π/3) and where rb is
the radius of the circle that circumscribes A1 A2 A3 .
In order to exploit the self-stress capability of the tensegrity system and
maintain tension in the cables, it is chosen to have the compressive struts be-
have as compression springs of stiffness k and rest length 0 . However, it has
been shown [10] (and may be observed in Fig. 1(b)) that the lines joining nodes
Ai and Bi intersect at a common point Q regardless of the robot’s configuration.
To avoid this situation, equivalent compression spring legs (ECSLs) consisting of
proximal and distal links of length L, connected together at nodes Ci by revolute
joints, are used. Universal joints connect the proximal links to the base at nodes
Ai while the distal links connect to the MP through spherical joints at nodes Bi .
The universal joints’ axes are selected so as to minimize the occurrence of inter-
ferences between the ECSLs and the cables. With this in mind, the unit vector
e1i parallel to the ith universal joint’s first axis is chosen to be perpendicular to
OAi and the Z axis such that
e1i = [− cos ψi , − sin ψi , 0]T (1)
while e2i is parallel to the universal joint’s second axis (and the axis of the
revolute joint at Ci ).
In [10], torsion springs were mounted in parallel with the revolute joints
at nodes Ci such that the ECSLs generated a nonlinear compression spring
behaviour. In this work, based on previous research [1], a variable radius drum
(VRD) is mounted on the distal link as illustrated in Fig. 2(a) such that its axis
coincides with the revolute joint axis at node Ci . A cable wrapped around the
VRD is routed through an eyelet and is attached to a linear extension spring
located in the robot’s base (the cable itself is considered to be inextensible). It
188 M. Arsenault
d
Bi Bi
L
Variable
radius Di Q
PQ ∩ Ai Bi Ci
drum θ
Ci γ PV
Ci i
γ
L e2i (out of plane)

Ai
e 1i Xi
Ai
k
(a) (b)
Fig. 2. Schematic representation of the ith ECSL: (a) VRD mechanism used to emulate
the linear compression spring behaviour and (b) variables used in the kinematic and
static modeling.
has been found that, by properly designing the profile of the VRD, it is possible
to achieve the desired linear compression spring behaviour with the ECSLs.
The tensegrity robot is actuated by varying the lengths of the cables con-
necting the base to the MP. Using motor-driven winches as described in [4], the
cable lengths become ρ1 = a2 − p − b3 = a3 − p − b2 , ρ2 = a1 −p−b3 =
a3 − p − b1 and ρ3 = a1 − p − b2 = a2 − p − b1 , where ai and bi are
the positions of nodes Ai and Bi in the XY Z frame. In this way, so long as the
cables remain taut, parallelograms A1 A2 B1 B2 , A2 A3 B2 B3 and A3 A1 B3 B1 are
formed between the base and MP, the latter’s motion is purely translational 1 ,
and bi = −ai . It follows that the position p = [x, y, z]T of node P , which is
located at the centroid of B1 B2 B3 , represents the pose of the robot. However,
it may be observed from Fig. 1(c) that the robot has a third-order rotational
symmetry about the Z axis. To exploit this fact for the purpose of workspace
computation, cylindrical coordinates r, φ and z are instead used such that
p = [r cos φ, r sin φ, z]T (2)
and the robot’s pose may alternatively be described by x = [r, φ, z]T .
3 Description of the workspace constraints

3.1 Constraints due to limits on member lengths (C1 )
The robot’s workspace is normally constrained by the admissible lengths of the
cables and ECSLs. However, since large lengths of cable may be stored on the
1
Note that this assumes the cables to be inextensible.
motor-driven winches, it is assumed that they do not apply any constraint on

the workspace. Meanwhile, the ECSLs lengths, which may be obtained as
i = p + bi − ai = p − 2ai (3)
must satisfy min ≤ i ≤ max . Geometrically, these bounds on i represent

spheres centred at 2ai of radii min and max , respectively. While min and max
could be determined from the mechanical limits of the revolute joints at nodes
Ci , min = 0 and max = 2L will be used here.
3.2 Constraints due to interferences between members (C2 )
Referring to Fig. 1(c), it is observed that the robot’s cables cannot interfere
with each other. Similarly, interferences between the ECSLs are not possible
given their geometrical arrangement and the design of the universal joints at
nodes Ai [4]. However, as the length of the ith ECSL increases and its proximal
and distal links unfold, one of the links will eventually interfere with the cable
joining nodes Aj and Bk (j = i − 1 if i = 2, 3, j = 3 if i = 1, k = i + 1 if i = 1, 2
and k = 1 if i = 3) that is otherwise piercing through Ai Bi Ci . For the purpose
of checking for interferences, the cables as well as the proximal and distal links
of the ECSL are modeled as line segments. The position of the midpoint Ri of
the cable joining Aj and Bk in the XY Z frame is ri = (p + aj + bk )/2. Defining
a vector directed from Bi to Ai as ui = ai − p − bi it may subsequently be
shown that (ri − ai )T (e1i × ui ) = 0. In other words, regardless of the tensegrity
robot’s configuration, Ri is always the point on the cable that is located within
the plane defined by Ai Bi Ci . As a result, the verification of whether the cable
is in contact with the links of the ECSL simplifies to determining whether its
midpoint is located on one of the links of the ECSL. It may also be observed
from Fig. 1(c) that the cable’s midpoint will always be located in a horizontal
plane PQ passing through the robot’s centre Q. Defining Di as the point on
either the proximal or distal link of the ith ECSL (depending on the robot
configuration) that passes through PQ , interferences between the ith ECSL’s
links and the cable joining Aj and Bk are avoided if the distance d(Q, Di ) from
Q to Di is greater than d(Q, √ Ri ). In addition, it may easily be shown that
d(Q, Ri ) = ri − p/2 = 3rb /2 remains constant regardless of the robot’s
configuration. In order to calculate d(Q, Di ), angle γ is first defined as being
measured about e2i from a vertical plane PV passing through OAi to Ai Bi
(refer to Fig. 2(b)). Defining a Xi Yi Zi reference frame such that Xi e1i and
Zi Z, one has γi = − sin−1 (xi /i ) where xi is the position of P measured along
the Xi axis. The condition for avoiding interferences between the ith ECSL and
the cable joining nodes Aj and Bk is then expressed as
√
i cos βi 3
d(Q, Di ) = > rb (4)
2 sin (βi + γi ) 2
where βi = θi /2.
190 M. Arsenault
3.3 Constraints related to wrench feasibility (C3 )

The wrench feasible workspace (WFW) is defined as the set of poses where the
robot may generate a required set of wrenches with admissible cable tensions.
The cables must remain taut at all times to ensure the tensegrity robot’s motion
is translational. The cable tensions also should not exceed their breaking strength
nor the capabilities of the actuators used to change their length. Admissible cable
tensions are thus defined as tmin ≤ tjk ≤ tmax where tjk is the tension in the
cable joining nodes Aj and Bk (j, k = 1, 2, 3 and j = k). The feasibility at a
given pose x of a wrench w, applied by the MP to its environment, requires that
W c t∗ = Ws f ∗ + w (5)
be satisfied with admissible cable tensions where t∗ = [t∗32 , t∗23 , t∗31 , t∗13 , t∗12 , t∗21 ]T
is a vector of cable tensions normalized with respect to their lengths (i.e., t∗jk =
tjk /ρi ). Moreover, f ∗ = [f1∗ , f2∗ , f3∗ ]T is a vector of forces (positive if in compres-
sion) in the ECSLs normalized with respect to their lengths, i.e.,

0
fi∗ = k −1 (6)
i
Meanwhile,

n1 n1 n2 n2 n3 n3
Wc = (7)
b2 × n1 b3 × n1 b1 × n2 b3 × n2 b2 × n3 b1 × n3

u1 u2 u3
Ws = (8)
b1 × u1 b2 × u2 b3 × u3
are pose dependent wrench matrices where
ni = aj − p − bk with i, j, k = 1, 2, 3, j, k = i and j = k (9)
is a vector directed along the cable joining nodes Aj and Bk .
4 Workspace computation using interval analysis

4.1 Some concepts of interval analysis
Interval analysis tools allow for the computation of rigorous enclosures on math-
ematical expressions and have regularly been used for the analysis and design
of robots, e.g., [5, 8, 18]. While some basic concepts of interval analysis are in-
troduced here, further details may be found in [9, 11]. A Matlab-based imple-
mentation of interval analysis tools [17] is used in this work. An interval variable
[x] = {x ∈ R : x ≤ x ≤ x} is a bounded set of real numbers where x and
x are the interval’s lower and upper bounds, respectively. An interval vector
T
[x] = [x1 ], [x2 ], . . . , [xn ] is the set of real vectors x = [x1 , x2 , . . . , xn ]T such

that xi ∈ [xi ] (i = 1, 2, . . . , n). Similarly, the element located on the ith line and
jth column of an interval matrix [A] is an interval [Aij ] such that [A] contains
the set of real matrices A whose elements satisfy Aij ∈ [Aij ]. The interval eval-
uation of a real-valued function f (x) over an interval vector [x] will produce an
interval [f ] such that f ([x]) = {f (x) : x ∈ [x]} ⊆ [f ] where x ∈ [x] is interpreted
component-wise (i.e., xi ∈ [xi ] ∀ i) and f ([x]) is the image of [x] under f (x).
The fact that [f ] generally overestimates f ([x]) is due to the interval dependency
problem [11] that occurs when the expression of f (x) includes multiple instances
of a given interval variable. This presents a challenge with the use of interval
analysis tools that must be addressed for each workspace constraint.
4.2 Branch-and-prune workspace computation algorithm

T
Starting from an annulus sector [x] = [r], [φ], [z] corresponding to a set of
poses known to encompass the entire workspace, sectors are processed as follows:
– If all workspace constraints are rigorously satisfied: [x] → Lin
– If at least one workspace constraint is rigorously violated: [x] → Lout
– If decisions on constraints are not rigorous and N < Nb : [x] → bisection
– If decision on constraints are not rigorous and N ≥ Nb : [x] → Lbnd
where Lin is a list of sectors whose union defines the workspace, Lout contains
sectors known to be completely outside the workspace and the workspace bound-
aries are known to be located within the sectors in Lbnd . This process continues
until the total number of sectors N has reached a specified maximum Nb . The
bisection of a sector is performed by splitting one interval among [r], [φ] or [z].
The sectors resulting from the bisection are then added at the end of the list
of those needing to be evaluated. While other (perhaps more efficient) bisection
methods could be considered, in this work sectors are split along their largest
dimension (i.e., max(r̆, r̂ · φ̆, z̆)) where, given an interval variable [x], the interval
width is defined as x̆ = x − x while its midpoint is x̂ = (x + x)/2. For example,
if for a given sector [x] it is found that r̂ · φ̆> max (r̆, z̆), the sectors resulting
from its bisection will be [r], [φ, φ̂], [z] and [r], [φ̂, φ], [z] .
4.3 Interval evaluation of the member length constraints (C1 )

In order to verify the ECSL length constraint described in Section 3.1, the
squares of the ECSL lengths (i.e., 2i ) are considered for convenience. By substi-
tuting Eq. (2) in Eq. (3), one may find
[2i ] = [2i , 2i ] = 4rb2 + [r]{[r] − 4 sin ([φ] − ψi )} + [z]2 (10)
With the above expression having [r] as the only repeated interval variable, the
interval evaluation was found to yield sufficiently tight bounds for the purposes
of constraint verification. In order for the sector described by [x] to rigorously
satisfy the ECSL length constraints and thus be located completely inside the
robot’s workspace, min 2i ≥ 2min and max 2i ≤ 2max must both be satisfied.

If max 2i > 2max or min 2i < 2min then the sector rigorously violates the
ECSL length constraints and is completely outside the workspace.
192 M. Arsenault
4.4 Interval evaluation of the member interference constraints (C2 )

The verification of the constraints due to interferences between members outlined
in Section 3.2 requires the interval evaluation [d(Q, Di )] (from Eq. (4)). Referring
to Fig. 2(b), one may observe that

[i ]
[βi ] = sin−1 (11)
2L
Furthermore, referring to the definition of γi (Section 3.2) and noting that xi =
rb cos (φ − ψi ), one has
rb cos ([φ] − ψi )
[γi ] = sin−1 (12)
[i ]
[i ] cos [βi ]
[d(Q, Di )] = (13)
2 sin ([βi ] + [γi ])
It may quickly be ascertained that interval dependency will lead to challenges
during the interval evaluation [d(Q, Di )]. In order to overcome this issue, the
Skelboe-Moore algorithm [11] is used. Through an iterative process, the algo-
rithm proceeds to find the lower bound of the interval evaluation [f ] of a function
f (x) up to a specified tolerance. Afterwards, considering the negative function
−f (x), an upper bound on the interval evaluation [f ] is also found. Further
details on the algorithm’s operation are available in [11]. With respect to the
constraints related to interferences between the ECSLs and the cables, a sec-
tor described by [x] will be considered to be in the robot’s workspace√if none
of the ECSLs interfere with the actuated cables, i.e., min(d(Q, Di )) > 3rb /2.
Meanwhile, if at least √
one ECSL is found to interfere with an actuated cable,
i.e., min(d(Q, Di )) < 3rb /2 then [x] rigorously violates the constraints and
will be outside the workspace.
4.5 Interval evaluation of the wrench feasibility constraints (C3 )

Henceforth, the required set of wrenches used to define the WFW is assumed to
have the shape of a box in R6 such that it may be represented by a n-dimensional
interval vector [w]. Moreover, the set of admissible normalized cable tensions is
represented as an interval vector [t∗ ] with t∗jk = tmin /ρi and t∗jk = tmax /ρi .2 The
use of normalized cable tensions to compute interval evaluations of the wrench
feasibility constraints was first proposed in [8] in order to manage the interval
dependency issue. The WFW is then defined as the set of sectors [x] for which
∀ x ∈ [x], ∀ w ∈ [w], ∃ t∗ ∈ [t∗ ] : Wc t∗ = Ws f ∗ + w (14)
In order to determine whether a sector [x] is located entirely inside or outside
the WFW, Eq. (5) is first decomposed into
Wc t∗f = Ws f ∗ (15)
Wc t∗w = w (16)
2
This is only valid if tmin /ρi ≤ tmax /ρi .
where t∗f and t∗w are the normalized cable tensions due to the ECSL forces
and external wrench, respectively, such that t∗ = t∗f + t∗w . It may be shown
by solving Eq. (15) for t∗f that the elements of the latter which are associated
with pairs of parallel cables are equal, i.e., t∗fjk = t∗fkj . Given that the num-
ber of unknown normalized cable tensions in Eq. (15) is thus reduced by half,
these may be found by considering only the sum of forces acting on the mobile
platform. In other words, Eq. (15) may be replaced by Wc t̃∗f = Ws f ∗ where
t̃∗f = [2t∗f23 , 2t∗f13 , 2t∗f12 ]T , Wc = n1 n2 n3 and Ws = u1 u2 u3 . Interestingly,

while Wc and Ws are pose dependent, Ws−1 Wc is not. An interval enclosure of

t̃∗f over the sector [x] may thus be obtained as
⎡ ⎤
1 1 1
⎢0 − [1 ] + [2 ] + [3 ] − 1⎥
⎢ ⎥
1 1 1
[t̃∗f ] = Wc−1 Ws [f ∗ ] = k ⎢ − −
⎢ ⎥
+
⎢ 0 [1 ] [2 ] [3 ] 1 ⎥
⎥ (17)
⎢ ⎥
1 1 1
− −1
⎣ ⎦
0 +
[1 ] [2 ] [3 ]
where [f ∗ ] is the interval evaluation of f ∗ over [x] and where the determination
of [t∗f ] is straightforward once [t̃∗f ] is known. Meanwhile, based on an algorithm
described in [15], the interval hull [t∗w ] of the solution set of Eq. (16) evaluated
over a sector [x], i.e.,
∗
tw : ∀ Wc ∈ [Wc ], ∀ w ∈ [w], Wc t∗w = w ∈ [t∗w ]

(18)
may be computed. Given [t∗f ] and [t∗w ], a sufficient condition for a sector [x] to
be located entirely within the WFW may be defined as [t∗f ] + [t∗w ] ⊆ [t∗ ] or,
equivalently, as
t∗fjk + t∗wjk ≥ t∗jk and t∗fjk + t∗wjk ≤ t∗jk with j, k = 1, 2, 3 and j = k (19)
Meanwhile, a sufficient condition for asector [x] to be located entirely outside

the WFW requires that [t∗f ] + [t∗w ] [t∗ ] = ∅ or, alternatively, that

t∗fjk + t∗wjk ≤ t∗jk or t∗fjk + t∗wjk ≥ t∗jk with j, k = 1, 2, 3 and j = k (20)
The above conditions are not necessary since the interval matrix [Wc ] used in
the determination of [t∗w ] in Eq. (18) is an enclosure (not a precise description)
of the set Wc (x) ∀ x ∈ [x] (i.e., there are matrices within [Wc ] which are not
proper normalized cable wrench matrices of the robot).
5 Examples
As noted in Section 2, the robot has a rotational symmetry about the Z axis. Its
workspace is thus computed for a 120 degree slice (e.g., φ ∈ [−150, −30] degrees)
and is then rotated twice (at 120-degree intervals) to obtain the total workspace.
194 M. Arsenault
3 3 3
2 2 2
1 1 1
Y 0 Y 0 Y 0
-1 -1 -1
-2 -2 -2
-3 -3 -3
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
X X X
(a) (b) (c)
Fig. 3. 2D slices of the robot’s workspace for z = 0.75 m and 0 = 2L while considering
the (a) member length constraints, (b) member interference constraints and (c) wrench
feasibility-based constraints (Lin ≡ green, Lbnd ≡ blue, Lout ≡ red).
Parallel computing is also used to reduce the computation time. Parameter values
that apply to all examples presented herein are rb = 1 m, L = 2 m, tmin = 0
N, tmax = 2000 N and k = 1 kN/m. The required wrench set is based on the
requirement of supporting the MP’s weight as [w] = [0, 0, −mg, 0, 0, 0]T with
m = 2 kg and g = 9.81 m/s2 .3
In Fig. 3, two-dimensional (2D) horizontal slices of the robot’s workspace,
considering each of the constraints separately, are plotted for z = 0.75 m, k =
1 kN/m and Nb = 2000. The interval evaluation of constraints C1 and C3 is
efficient, with only a thin band of sectors belonging to Lbnd . In comparison,
constraint C2 is more challenging due to increased interval dependency. As is
suggested by Fig. 3 and confirmed by observing Fig. 1(c), constraint C1 is always
less constraining than C2 when max = 2L as the ECSLs will collide with the
cables prior to their proximal and distal links being fully aligned. The robot’s
workspace is thus constrained by some combination of C2 and C3 . In fact, it may
be observed that C2 is generally responsible for limiting the workspace in the
regions surrounding φ = −90, 30 and 150 degrees while C3 is more constraining
elsewhere.
It may be verified that the boundaries in Fig. 3(c) correspond to one of the
cables losing tension (rather than a cable exceeding its maximum permissible
tension). Moreover, testing has shown that modifying either k or m will not lead
to increases in the size of the workspace. In fact, to the contrary, increases to k
and m will eventually yield cable tensions that exceed tmax in some configurations
(thus decreasing the workspace size). It is interesting then to study the effect
on the workspace of the ECSL rest length 0 . Selecting 0 > 2L is akin to
introducing pre-load in the ESCLs. While this has no effect on constraints C1
and C2 , it does have the initial tendency to push the workspace boundaries due
to C3 outwards. This leads to corresponding increases in the workspace size as
3
Note that gravity is considered to act in the positive Z axis direction so that it is
“pulling” the MP upward and away from the robot’s base when the MP is located
above the XY plane.
3 3 3 3
2 2 2 2
Z Z Z Z
1 1 1 1
0 0 0 0
1 1 1 1
0 1 0 1 0 1 0 1
Y 0 Y 0 Y 0 Y 0
-1 -1 -1 -1 -1 -1 -1 -1
X X X X
(a) (b) (c) (d)
Fig. 4. Workspace with (a) 0 = 2L, (b) 0 = 2.5L, (c) 0 = 3L, (d) 0 = 3.5L.
seen in Figs. 4(a-c). Once the pre-load exceeds a certain threshold, however,
the maximum cable tension limits are breached and any subsequent increase
in 0 would decrease the workspace’s volume (e.g., Fig. 4(d)). It may also be
observed that this effect is generally more significant in regions of the workspace
located further away from the XY plane (i.e., in regions where, in the absence
of pre-load, the ECSL forces are smaller).
6 Conclusion
In this work, a method is proposed to compute the workspace of a 3-DoF trans-

lational tensegrity robot based on the reinforced tensegrity simplex. In order
to avoid the inherent intersection of the latter’s compressive members (refer to
Fig. 1(b)), these are replaced with ECSLs. However, interferences between the
cables and the ECSL links remain a possibility. Moreover, the translational mo-
tion of the robot depends on the cables remaining in tension. The workspace
is thus computed based on constraints due to the kinematics of the ECSLs,
the need to avoid member interferences and the ability of the robot to support
a desired wrench set while maintaining acceptable tensions in its cables. The
methodology proposed herein is based on the use of interval analysis such that
all poses found to be within the workspace have been verified to be reachable.
While the stiffness of the ECSL springs was not found to significantly influence
the workspace boundaries, it was shown that introducing an appropriate level of
pre-load in the ECSLs leads to a larger workspace.
The proposed approach assumes the cables to be rigid in tension. Since pairs
of parallel cables will not support equal tensions, the effect of cable elongations
on the translational motion of the robot could be investigated. Moreover, the
effect of the weights of the ECSL links, considered negligible with respect to the
weight of the MP in this work, could also be quantified. Given the additional
kinematic equations required to locate the centers of mass of the links, however,
it is expected that the use of interval analysis in such a case would be very chal-
lenging. Finally, future work will focus on the design of the robot to improve its
performance (e.g., workspace size, stiffness, etc.) while also investigating differ-
ence ECSL designs.
196 M. Arsenault
References
1. Arsenault, M.: Design of convex variable radius drum mechanisms. Mechanism
and Machine Theory 129, 175–190 (2018)
2. Arsenault, M., Gosselin, C.M.: Kinematic and static analysis of a planar modular
2-DoF tensegrity mechanism. In: Proceedings - IEEE International Conference on
Robotics and Automation, vol. 2006, pp. 4193–4198 (2006)
3. Arsenault, M., Gosselin, C.M.: Kinematic and static analysis of a three-degree-of-
freedom spatial modular tensegrity mechanism. International Journal of Robotics
Research 27(8), 951–966 (2008)
4. Arsenault, M., Mohr, C.: Design and fabrication of a functional prototype for a
3-DoF translational tensegrity robot. In: Proceedings of the 2016 CSME Interna-
tional Congress (2016)
5. Berti, A., Merlet, J.P., Carricato, M.: Solving the direct geometrico-static problem
of underconstrained cable-driven parallel robots by interval analysis. International
Journal of Robotics Research 35(6), 723–739 (2016)
6. Calladine, C., Pellegrino, S.: First-order infinitesimal mechanisms. International
Journal of Solids and Structures 27(4), 505–515 (1991)
7. Fuller, B.: Tensile-integrity structures. United States Patent No. 3,063,521 (1962)
8. Gouttefarde, M., Daney, D., Merlet, J.P.: Interval-analysis-based determination of
the wrench-feasible workspace of parallel cable-driven robots. IEEE Transactions
on Robotics 27(1), 1–13 (2011)
9. Hansen, E., Walster, G.W.: Global Optimization Using Interval Analysis, second
edn. Marcel Dekker Inc., New York (2004)
10. Mohr, C.A., Arsenault, M.: Kinematic analysis of a translational 3-DoF tensegrity
mechanism. Transactions of the CSME 35(4), 573–584 (2011)
11. Moore, R.E., Kearfott, R.B., Cloud, M.J.: Introduction to Interval Analysis. Soci-
ety for Industrial and Applied Mathematics, Philadelphia (2009)
12. Paul, C., Lipson, H., Valero-Cuevas, F.: Design and control of tensegrity robots for
locomotion. IEEE Transactions on Robotics 22(5), 944–957 (2006)
13. Pellegrino, S.: Analysis of prestressed mechanisms. International Journal of Solids
and Structures 26(12), 1329–1350 (1990)
14. Pugh, A.: An introduction to tensegrity. University of California Press (1976)
15. Rohn, J.: An algorithm for computing the hull of the solution set of interval linear
equations. Linear Algebra and Its Applications 435(2), 193–201 (2011)
16. Rovira, A.G., Tur, J.M.: Control and simulation of a tensegrity-based mobile robot.
Robotics and Autonomous Systems 57, 526–535 (2009)
17. Rump, S.: INTLAB – INTerval LABoratory. In: T. Csendes (ed.) Developments in
Reliable Computing, pp. 77–104. Kluwer Academic Publishers, Dordrecht (1999)
18. Schmidt, V., Muller, B., Pott, A.: Solving the forward kinematics of cable-driven
parallel robots with neural networks and interval arithmetic. pp. 103–110. Kluwer
Academic Publishers (2014)
19. Snelson, K.: Continuous tension, discontinuous compression structures. United
States Patent No. 3,169,611 (1965)
20. Sultan, C., Corless, M., Skelton, R.: Tensegrity flight simulator. Journal of Guid-
ance, Control and Dynamics 23(6), 1055–1064 (2000)
21. Wenger, P., Chablat, D.: Kinetostatic analysis and solution classification of a class
of planar tensegrity mechanisms. Robotica (2018)
Antipodal Criteria for Workspace Characterization of
Spatial Cable-Driven Robots
Leila Notash
Department of Mechanical and Materials Engineering

Queen’s University, Kingston, Ontario, Canada
[email protected]
Abstract. In this paper, the implementation of the antipodal principles on the

spatial cable-driven parallel robots is investigated. The spatial form of the antip-
odal method is presented and the criteria for examining the wrench-closure work-
space of cable robots are developed. The generalization of the 3D conditions on
the 6 DOF fully constrained spatial robots, with one or two cables at each plat-
form attachment points, is proposed. Comprehensive mathematical models are
developed and implemented in simulation for a variety of 5 and 6 degrees of
freedom cable robots.
Keywords: Cable-Driven Parallel Robots, Antipodal Method
1 Introduction
In parallel robots, the moving platform is connected to the base by several legs/cables,
which act in parallel to share the payload and control the platform motion. Considering
the actuation, parallel robots could be categorized as solid-link (kinematic chains of
links with actuated and passive joints, referred to as legs/branches) and wire/cable-
driven/actuated. The actuators of parallel robots could be mounted at or close to the
base to reduce the moving mass. Thus, these robot manipulators could be designed to
have high accuracy, stiffness, load capacity, and dynamic characteristics (e.g., high ac-
celerations). In addition, cable robots could have lower mass, cost and power consump-
tion, as well as larger workspace to mass ratio and easier assembly/disassembly and
reconfiguration, compared to the equivalent solid-link parallel robots.
Two example cable robots are depicted in Fig. 1. Cables apply tension force (can
pull but not push). Fully constrained cable robots have more cables/actuators than their
degrees of freedom (DOF). Thus, there are many solutions for the cable tension vector
for given platform forces and moments (wrench). Generally, the minimum 2-norm non-
negative solution is sought, which is a convex set [1]. The generalized inverse (GI) of
the transposed Jacobian matrix results in the minimum 2-norm solution. However, un-
like the solid-link manipulators in which the minimum 2-norm solution could include
both positive and negative inputs from actuators, negative cable tension is not allowed.
On the other hand, the minimum 2-norm solution for the vector of cable forces could
result in negative tension for cables while the platform is within the wrench-closure
workspace, i.e., even if it is possible to have positive tension.

198 L. Notash
Cable 5 Cable 7
A
Cable 4
Cable 4 Cable 3 Cable 6
Platform
Platform
Cable 2 Cable 1 Cable 3
Cable 1
Cable 2
A
(a) (b)
Fig. 1. Cable robots (a) 3 DOF planar, four cables; (b) 3 DOF spatial, seven cables.
The non-negative cable tension and the workspace of the fully constrained cable
parallel robots have been investigated extensively using a variety of methods and the-
orems, e.g., [2-9]. The conditions for static equilibrium and fully constrained poses
were proposed in [2]. The production of a set of specified wrenches at a pose in the
workspace, with bounded cable tension, was studied in [6]. In [4,5], analytical functions
based on Cramer’s rule were applied to generate the boundary of the wrench-closure
workspace of fully constrained and redundant cable robots. In [3], the application of
antipodal theorem on two cable-driven robots was examined. A generalized form of the
antipodal method was presented in [7] for the workspace generation of non-redundant
and redundant planar cable manipulators, acting under external force/gravity.
In this article, the implementation of the antipodal criteria for the workspace charac-
terization of fully constrained spatial cable robots is presented. Two techniques, includ-
ing the antipodal theorem, for identifying the controllable poses are reviewed in Sec. 2.
The formulations are implemented for the workspace generation of numerous 5 DOF
and 6 DOF manipulators; results for some cases are presented in Sec. 3.
2 Fully Constrained Poses
For the cable-driven robot manipulators, the platform wrench F is linearly related to the
cable forces by the transposed Jacobian matrix ۸ ் as
௡
۴ ൌ ۸ ் ࣎ ൌ ሾ۸ଵ் ۸ଶ் ǥ ்
۸௡ିଵ ۸௡் ሿ࣎ ൌ ෍ ۸௝் ߬௝ (1)
௝ୀଵ
where ۸௝் is the Plücker line coordinates of the force axis of cable j, i.e., the unit vector
of cable force and its moment about the operation point of platform. Because of the
unidirectional force property of cables, the cables could only apply pulling force on the
platform, i.e., they provide unidirectional and irreversible inputs.
The non-negative solution for the cable tension vector ૌ could be identified utilizing
the following closed-form method. Applying ૌ ൌ ۸ ்͓ F, if the Moore-Penrose GI ۸ ்͓
Antipodal Criteria for Workspace Characterization of Spatial Cable-Driven Robots 199
results in negative tension for k cables in the particular (minimum 2-norm) solution,
then the reduced Jacobian matrix ۸௥் is formed by removing the columns of ۸ ் that
correspond to cables with negative tension. After adjusting the negative tension of these
k cables to non-negative values (calculated or assigned depending on the existence of
dominant cables as discussed in [8,9]), and using
۸௥் ࣎௧௢௧ೝ ൌ ۴ െ ۴ఛ್ೌ೗ ൌ ۴ െ ෍ ۸௝் ߬௖௝ (2)

௞
the minimum 2-norm non-negative solution τ tot r for the cables with positive tension
in particular solution is identified. This closed-form method could be used to identify
the wrench-closure workspace of cable-driven manipulators.
Alternatively, antipodal method from multi-finger grasping could be used to charac-
terize the workspace of cable-driven robots. In grasping, using the antipodal theorem,
a point-contact with friction is modeled as a “friction cone”, with the contact point be-
ing the vertex (apex) of the cone. Then, the magnitudes of the normal and friction forces
are respectively the height (altitude) and radius of the base circle of cone. The planar
grasp is wrench closed if and only if the line connecting the contact points of two fingers
lies within the two corresponding “friction cones” [10,11]. With frictionless contacts,
each friction cone is formed by the forces of two fingers. The intersection point of the
axes of two forces represents the vertex of cone, and these two axes form the sides of
cone [7]. Thus, four frictionless contacts are needed to form the required two friction
cones of planar grasping.
2.1 Planar Cable Robots

For planar robots, cables and platform move on the same or parallel planes. Planar cable
robots could have up to three degrees of freedom; two translations on the plane of mo-
tion and a rotation about the normal of plane. Thus, at least four cables are required for
the fully constrained 3 DOF planar cable robots.
Considering two adjacent cables of planar cable robots, the “friction cones” of grasp-
ing are signified by triangles, each with a vertex at the intersection point of the two
cable axes and the other vertices at the pertinent cables’ anchors (base attachment points
of cables). That is, the axes of cable forces represent the sides of the “cone” and their
intersection point corresponds to the apex of the “cone” as depicted in Fig. 2.
The adjacent cables of the four-cable planar robot in Fig. 2 connect to the same point
on the platform. That is, cables 1 and 4 connect to B1 and cables 2 and 3 connect to B2,
forming the 4-2 cable attachment layout. In Fig. 2(a) the line connecting the two verti-
ces (line passing through B1 and B2) is inside the two “cones”. Thus, the pose is within
the wrench-closure workspace (referred to as workspace for the remainder of paper),
i.e., the four cables can constrain the translation of platform on the plane of motion and
its rotation about the normal of plane. The platform pose in Fig. 2(b) is not in the work-
space because line B1B2 is not within the “cone” formed by the axes of cables 2 and 3.
Therefore, the four cables cannot apply a moment in the clockwise direction.
200 L. Notash
Cable 4 Cable 3
Cable 4
fB4 Cable 3
Platform
Pl tfform fB3
fB4 Platform
fB3
B1
fB1 B2 fB2
fB1 B1
fB2
B2
Cable 1 ³)ULFWLRQ&RQHV´ Cable 2 Cable 1 Cable 2
³)ULFWLRQ&RQHV´
L & ´
(a) (b)
Fig. 2. Planar cable robot poses (a) within and (b) outside workspace.
2.2 Spatial Cable Robots

Spatial cable robots could have up to six degrees of freedom. For the fully constrained
6 DOF robots, at least seven cables are required.
A1 A4
Platform
A3
A2 B1 B2
Fig. 3. Geometric shape formed by four non-coplanar intersecting cable axes.
Three or more non-coplanar cables at each platform attachment point. For spatial
cable robots, the set of three (or more) non-coplanar cables attached to one of B1 or B2
form a 3D geometric shape, e.g., a rectangular pyramid as illustrated in Figs. 3 and 4.
The cable forces act along the pertinent cable axes and each pair of the intersecting
cable axes form a triangular shape. When at least three non-coplanar cable axes meet
at one point, the pairs of adjacent cable axes form the triangular faces of a convex pol-
yhedron. That is, the set of three intersecting cables form a tetrahedron (triangular pyr-
amid) while four intersecting cables form a rectangular pyramid, Fig. 4. The summation
of cable forces lies inside the pyramid formed by the cable axes. The potential poses
are certified to be within the workspace if the line that connects the vertices is fully
constrained, i.e., by checking if the line is inside the pertinent pyramids.
For example, the pyramid in Fig. 3, formed by four cables, has its apex at the inter-
section point B1, and includes four triangular faces B1A1A2, B1A2A3, B1A3A4, B1A1A4. As
well, the anchors of these cables form the rectangular base A1A2A3A4. When the line
connecting the two apexes B1 and B2 is within the two pyramids, then the cables con-
strain line B1B2. Unlike the 3D grasping, if line B1B2 is inside the cone but outside the
pyramid defined by the intersecting cable axes, then it is not fully constrained.
It is noteworthy that with the antipodal method, the largest possible workspace is
identified as this method does not rely on the magnitude of cable forces. The workspace
would be smaller for the bounded cable forces.
(a) Eight-cable (b) Seven-cable, 4 cables up
(c) Seven-cable, 3 cables up (d) Six-cable, 4 cables up
(e) Six-cable, 2 cables up (f) Six-cable, 3 cables up
Fig. 4. Eight-, seven- and six-cable 5 DOF robot manipulators.
One or two cables at each platform attachment point. For spatial cable robots, the
set of three or four non-coplanar cables attached to different points on the platform form
a 3D geometric shape.
For the fully constrained spatial cable robot of Fig. 5(a), seven cables are attached
to three points on the platform; e.g., three cables at B1, two at B2, two at B3. The pyramid
formed by three cables at B1 includes four triangular faces. The 3D shape formed by
the set of cable pairs attached to B2 and B3 and their anchors includes five faces.
Namely, two triangular faces (each formed by one of B2 or B3, and the pertinent two
cables), one rectangular face (formed by the four anchors), and two faces formed by
four points (B2, B3 and two anchors). Each face generated by B2, B3 and two anchors is
a ruled surface with the line segment B2B3 and the line segment connecting the pertinent
202 L. Notash
two anchors as its directrix lines. At zero platform rotation, each set of these four points
become coplanar and the 3D shapes transform to polyhedral.
(a) Three attachments on platform (b) Four attachments on platform

Fig. 5. Seven- and eight-cable 6 DOF robot manipulators.
In Fig. 5(b) the eight cables are attached in pairs at four points, e.g., B1, B2, B3, B4.
Each set of four adjacent cables (connected to two of these points) form a 3D shape
with five faces as discussed in the previous paragraph.
For the 6 DOF fully constrained ݊-cable robots (݊ ൒ ͹), the cables could be con-
nected to the platform at a minimum of three non-collinear points. When four adjacent
cables connect to four distinct points on the platform (with coplanar or non-coplanar
anchors), they form a 3D geometric shape, e.g., a twisted pyramidal frustum. In general,
the faces are ruled surfaces formed by four non-coplanar lines (passing through four
non-coplanar points).
3 Workspace of Cable Robots
3.1 Fully Constrained 5 DOF Robots

To investigate the spatial form of the antipodal theorem, the 5 DOF fully constrained
eight-cable, seven-cable and six-cable robots of Fig. 4 are considered here. For the
eight-, seven- and six-cable robots, four, four and three cables are respectively con-
nected to the platform at B1 and four, three and three others at B2, hence, forming the 8-
2, 7-2 and 6-2 cable attachment layouts.
The cable anchors ‫ܣ‬௝ , ݆ ൌ ͳǡ ǥ ǡ ݊, are set on a cube centered at the origin with a side
length of 2.2 meters. The anchors of the eight-cable (݊ ൌ ͺ) robot are at the vertices of
the cube, e.g., the coordinates of ‫ܣ‬ଵ and ‫ܣ‬ଶ are respectively (–1.1, –1.1, 1.1) and (–1.1,
–1.1, –1.1) meters in Fig. 4(a). For the seven-cable robots in Figs. 4(b)-(c), six cables
are attached to the vertices and one cable is attached to the midpoint of opposite edge.
For the six-cable robots, four cables are connected to the vertices and two cables are
attached to the midpoint of opposite edges as illustrated in Figs. 4(d)-(f). The cables are
connected to the platform at two locations ‫݅( ࢐࢏ܤ‬௝ ൌ ͳʹ), with the platform radius of
0.09 meters. Thus, column ݆ of the Jacobian matrix ۸ ் ൌ ሾ۸ଵ் ‫ ڮ‬۸௡் ሿ is formulated
in terms of the unit vector ‫ܝ‬௝ of the pertinent cable force vector ܎஻௝ ൌ ݂஻௝ ‫ܝ‬௝ acting on
the platform, and its moment with respect to the platform operation point ܲ as
் ࢀ
۸௝் ൌ ቂ‫ܝ‬௝
்
‫ܚ‬௉ǡ஻ ೔
ൈ ‫ܝ‬௝் ቃ (3)
ೕ
The platform pose is examined by forming the line segment B1B2. The cable axes are
utilized to form the surfaces of the pyramids. When the line connecting the two apexes
B1 and B2 is within the two pyramids, then the line is on the same side of each lateral
face (e.g., planes B1A1A2, B1A2A3, B1A3A4, B1A1A4 of Fig. 3) as the geometric center of
the pertinent pyramid. It is noteworthy that pyramid B1A1A2A3A4 collapses to rectangle
A1A2A3A4 when apex B1 is on the base A1A2A3A4. Thus, in the absence of a horizontal
force, the workspace is bounded by coplanar anchors A1, A2, A3, A4. The proposed pro-
cedure is verified by checking the existence of the non-negative null space basis of the
Jacobian matrix, e.g., [8].
(a) Eight-cable (b) Six-cable, 4 cables up; Seven-cable, 4 up
(c) Six-cable, 2 cables up; Seven-cable, 3 up (d) Six-cable, 3 cables up

Fig. 6 Workspaces for zero platform orientation of cable robots in Fig. 4.
For zero platform orientation, the symmetric workspaces of the eight-, seven- and
six-cable manipulators of Fig. 4 are illustrated in Fig. 6. The workspace is generated
with increments of 0.1 meters for െͳ ൑ ‫ ݔ‬൑ ͳ , െͳ ൑ ‫ ݕ‬൑ ͳand െͳ ൑ ‫ ݖ‬൑ ͳ meters.
For the eight-cable robot the workspace is a cuboid, and the largest among the six ro-
bots, with the size of 9261 units, Fig. 6(a). For the six-cable robots with four cable
anchors at ‫ ݖ‬ൌ ͳǤͳ, Fig. 4(d), or at ‫ ݖ‬ൌ െͳǤͳ, Fig. 4(e), the workspace is a triangular
prism (flipped shapes). Whereas when three cable anchors are at ‫ ݖ‬ൌ ͳǤͳ and three at
‫ ݖ‬ൌ െͳǤͳ, Fig. 4(f), the workspace is a rhombic prism and the smallest among the four
204 L. Notash
robots, with the size of 2457 units. Within these workspaces, the unidirectional cable
forces can fully constrain the platform. For zero platform rotation, the workspaces of
seven-cable robots in Figs. 4(b)-(c) are identical to the related six-cable ones in Figs.
6(b)-(c).
The workspace plots of these cable robots when the platform orientation is repre-
sented using the roll-pitch-yaw angles as ߶௭ ൌ ʹͲ‫ ל‬, ߰௬ ൌ െͳͲ‫ ל‬and ߠ௫ ൌ Ͳ are shown
in Fig. 7. The size of the workspaces depicted in Figs. 7(a)-(f) are respectively 5233,
3539, 2617, 1494, 1494 and 1617 units.
(a) Eight-cable (b) Seven-cable, 4 cables up
(c) Seven-cable, 3 cables up (d) Six-cable, 4 cables up
(e) Six-cable, 2 cables up (f) Six-cable, 3 cables up

Fig. 7 Workspaces for non-zero platform orientation of 5 DOF cable robots in Fig. 4.
3.2 Fully Constrained 6 DOF Robots

For three (or more) cable connections on the platform, the ruled surfaces of the 3D
geometric shapes are formed by a pair of adjacent cables (generators). The line segment
passing through the pertinent platform connections, e.g., B1B2, forms directrix ‫ܚ‬஻ , and
the line segment passing through the corresponding anchors, e.g., A1A2, forms directrix
‫ܚ‬஺ . Hence, the ruled surface is generated using directrices as
‫ܚ‬ሺ‫ݑ‬ǡ ‫ݒ‬ሻ ൌ ሺͳ െ ‫ݒ‬ሻ‫ ܣܚ‬ሺ‫ݑ‬ሻ ൅ ‫ ܤܚݒ‬ሺ‫ݑ‬ሻ for Ͳ ൑ ‫ݑ‬ǡ ‫ ݒ‬൑ ͳ (4)
To check if the pertinent line segment is on the same side as the geometric center of the
3D shape, the ruled surface is divided into planar sections (using generators).
(a) Zero orientation (b) ߶௭ ൌ Ͳ‫ ל‬, ߰௬ ൌ െ͵Ͳ‫ל‬, ߠ௫ ൌ Ͳ

Fig. 8 Workspaces of 6 DOF seven-cable robot in Fig. 5(a).
For the 6 DOF seven- and eight-cable robots of Figs. 5(a)-(b), the angular positions
of the cable connections on the platform are െͳͺͲ‫ ל‬ǡ െͳͺͲ‫ ל‬ǡ െͳͺͲ‫ ל‬,െ͵Ͳ‫ ל‬ǡ െ͵Ͳ‫ ל‬ǡ ͵Ͳ‫ ל‬ǡ
͵Ͳ‫ ל‬and െͳͷͲ‫ ל‬, െͳͷͲ‫ ל‬ǡ ͳͷͲ‫ ל‬ǡ ͳͷͲ‫ ל‬ǡ െ͵Ͳ‫ ל‬ǡ െ͵Ͳ‫ ל‬ǡ ͵Ͳ‫ ל‬ǡ ͵Ͳ‫ ל‬, respectively, with the plat-
form radius of 0.09 meters. The workspace plots with zero platform orientation and for
the noted roll-pitch-yaw angles are respectivelyshown in Figs. 8 and 9(a)-(b), with the
size of 2583, 236, 1617 and 383 units.
When the angular positions of the eight platform connection points are
െͳ͹ͻ‫ ל‬ǡ െͳ͹ͻ‫ ל‬ǡ ͳ͹ͻ‫ ל‬ǡ ͳ͹ͻ‫ ל‬ǡ െͳ‫ ל‬ǡ െͳ‫ ל‬ǡ ͳ‫ ל‬ǡ ͳ‫( ל‬very close to the 5 DOF case), for the zero
platform orientation, the workspace size (and shape) is the same as the one in Fig. 6(a)
for the 5 DOF eight-cable robot, i.e., 9261 units, using increments of 0.1 meters. The
workspace size is 2920 units for ߶௭ ൌ ʹͲ‫ל‬, ߰௬ ൌ െͳͲ‫ל‬, ߠ௫ ൌ Ͳ in Fig. 9(c), compared to
5233 in Fig. 7(a) for the 5 DOF robot. The workspace increases to 4095 units for ߰௬ ൌ
െ͵Ͳ‫ ל‬, Fig. 9(d), compared to the one in Fig. 9(b).
4 Conclusion
The wrench-closure workspaces of the fully constrained spatial cable-driven robots

were investigated in this paper. Implementation of the antipodal criteria for the work-
space characterization of cable robots was presented. The criteria were applied for the
5 DOF eight-, seven- and six-cable, as well as 6 DOF seven- and eight-cable spatial
manipulators. The workspace envelope of spatial cable robots, using the presented an-
tipodal criteria, will be investigated in future.
206 L. Notash
(a) Zero orientation (b) ߶௭ ൌ Ͳ‫ ל‬, ߰௬ ൌ െ͵Ͳ‫ל‬, ߠ௫ ൌ Ͳ
(c) ߶௭ ൌ ʹͲ‫ל‬, ߰௬ ൌ െͳͲ , ߠ௫ ൌ Ͳ (d) ߶௭ ൌ Ͳ‫ ל‬, ߰௬ ൌ െ͵Ͳ‫ל‬, ߠ௫ ൌ Ͳ

Fig. 9 Workspaces of 6 DOF eight-cable robot in Fig. 5(b).
References
1. Notash, L.: Manipulator Deflection for Optimum Tension of Cable-Driven Robots with Pa-
rameter Variations, In Cable-Driven Parallel Robots, Springer, Cham, pp. 26–36 (2018).
2. Roberts, R.G., Graham, T., Lippit, T.: On the Inverse Kinematics, Statics, and Fault Toler-
ance of Cable-Suspended Robots, J. Robot. Syst. 15(10), 581–597 (1998).
3. Voglewede, P.A., and Ebert-Uphoff, I.: Application of the Antipodal Grasp Theorem to Ca-
ble-Driven Robots, IEEE Trans. Robot., 21(4), 713–718 (2005).
4. Stump, E., Kumar, V.: Workspace of Cable-Actuated Parallel Manipulators, ASME J. Mech.
Des., 128, 159–167 (2006).
5. McColl, D., Notash, L.: Workspace Envelope Formulation of Planar Wire-Actuated Parallel
Manipulators, Trans. Can. Soc. Mech. Eng., 33(4), 547–560 (2009).
6. Bouchard, S., Gosselin, C., and Moore, B.: On the Ability of a Cable-Driven Robot to Gen-
erate a Prescribed Set of Wrenches. ASME J. Mech. Rob., 2(1), p. 011010 (2010).
7. McColl, D., Notash, L.: Workspace Formulation of Planar Wire-Actuated Parallel Manipu-
lators, Robotica, 29(4), 607–617 (2011).
8. Notash, L.: Designing Positive Tension for Wire-Actuated Parallel Manipulators. In: Kumar,
V., Schmiedeler, J., Sreenivasan, S. V., Su, H.-J. Editors, Advances in Mechanisms, Robot-
ics and Design Education and Research, pp. 251-263, Springer, Cham, Switzerland (2013).
9. Notash, L.: Failure Recovery for Wrench Capability of Wire-Actuated Parallel Manipula-
tors, Robotica, 30(6), 941–950 (2012).
10. Nguyen, V. D.: Constructing force-closure grasps, Int. J. Robot. Res. 7(3), 3–16 (1988).
11. Murray, R., Li, Z., Sastry, S.: A Mathematical Introduction to Robotic Manipulation, CRC
Press (1994).
Part IV
Control
Robust Adaptive Control of Over-Constrained Actuated
Alireza Izadbakhsh1 and Hamed Jabbari Asl2 and Tatsuo Narikiyo2

ϭ
Department of Electrical Engineering, Garmsar branch, Islamic Azad University, Garmsar, Iran
2
Control System Laboratory, Toyota Technological Institute, Japan
[email protected]
Abstract. This paper proposes a robust adaptive controller for cable-driven pa-
rallel robots subject to dynamic uncertainties. The main objective is to study the
performance of the proposed controller on these systems designed based on
function approximation technique (FAT). Compared to the previous adaptive
control strategies, the proposed controller is simpler, and does not require prior
knowledge of the uncertainties upper bound, linear parameterization of the ki-
nematic and dynamic models. The reason is that FAT estimators consider the
uncertainty as a time-varying function rather than a function of different state
variables. To ensure that all the cables are in tension, internal force concept is
used in the proposed adaptive control algorithm. Stability of the whole system
is analyzed through a Lyapunov-based method, and the uniformly ultimately
bounded stability is guaranteed. Simulation studies on a planer cable-driven pa-
rallel robot indicate the effectiveness of proposed method.
Keywords:Cable-driven parallel robot, function approximation technique, ro-

bust adaptive control
1 Introduction
Cable-driven parallel robots (CDPRs) have become popular due to their widespread
applications in various fields such as gait rehabilitation [1], handling heavy compo-
nents [2], high-speed manipulation [3], etc. The challenges in the control of these
systems have attracted the attention of many researchers, and several control schemes
are reported. A task-space controller based on the back-stepping approach has been
presented in [4]. A feed-forward nonlinear controller in cable length coordinates has
been developed in [5]. A Lyapunov-based nonlinear controller for CDPRs has also
been presented in [6], in which the nonlinearities related to manipulator dynamics are
successfully eliminated using an inverse dynamic controller. In addition, many other
approaches such as adaptive robust control [7], passivity-based control [8], adaptive
fuzzy approaches [9], and robust sliding mode control [10] have been studied. It is
worth noting that most of them have overlooked the actuator dynamics in their design
procedure. In other words, their control laws calculate the desired tension that should
be applied to the manipulator dynamics.
Since most robotic systems use electrical motors as actuators, recently, some torque-

210 A. Izadbakhsh et al.
based controllers have been presented for CDPRs [11-12]. It should be emphasized
that considering the dynamics of actuators increases the system order and consequent-
ly the complexity of the design. Moreover, actuator nonlinearities are another chal-
lenging issue, since they reduce the bandwidth of the inner-loop dynamics [12]. Thus,
one of the objectives of this paper is to design a suitable controller for CDPRs consi-
dering the dynamics of whole system including the manipulator and actuators, and
hence to provide a satisfactory transient performance for large initial errors.
Studying the recent literature in the field of adaptive and robust control, remains no
doubt that uncertainty estimation and compensation play a key role in improving the
controller performance. At the heart of this area is the Stone-Weierstrass theorem
[13]. Many advances and successes in neuro-fuzzy control of complicated uncertain
nonlinear multivariable systems owe to this fundamental theorem and during the last
few decades numerous neuro-fuzzy control structures for various systems have been
developed [14-16]; since with the help of Stone-Weierstrass theorem, determination
of the regressor matrices can be avoided. It is worthy to note that the Stone-
Weierstrass theorem is a general theorem that covers many other estimators such as
the Fourier series (FS) expansion, Legendre polynomials, Chebyshev polynomials,
Hermite polynomials, Laguerre polynomials, and Bessel functions. It has been also
shown that the response of linear differential equations satisfies the Stone-Weierstrass
theorems’ conditions and it can be used as an uncertainty estimator in robust and
adaptive control [17-19]. These approaches are referred as FAT. In comparison with
neuro-fuzzy estimators, FAT-based estimators have simpler structures and fewer tun-
ing parameters [20]. Therefore, in this paper, a FAT-based adaptive control law using
the Fourier series expansion is developed for actuated CDPRs. To the best knowledge
of authors, this control scheme has not been previously investigated for CDPRs. Uti-
lizing the FAT-based estimator, the proposed scheme can adaptively learn the uncer-
tain terms in dynamics of both manipulator and actuators. The overall closed-loop
system is proven to be robust, and uniformly ultimately bounded (UUB) based on the
Lyapunov's stability concept. The performance of the controller is evaluated through
simulations, where the results are compared to those of the torque-based robust adap-
tive controller reported in [12].
This paper is organized into six sections. Following the introduction, in section 2, the
dynamic model of the CDPR manipulators is presented. Section 3 presents the con-
troller design scenario. Stability analysis and performance evaluation are discussed in
section 4. Section 5 presents the simulation results, and finally, conclusions are given
in section 6.
2 Robot kinematics and dynamics

The dynamic equations of an n-cable parallel robot actuated by geared rotary motors,
assuming no elasticity in the cables and neglecting frictions in the pulleys can be de-
scribed by [12]
M ( q ) q C ( q, q ) q G ( q ) Fd q Fs ( q ) Td ( t ) JW (1)
Robust Adaptive Control of Over-Constrained Actuated Cable-Driven Parallel Robots 211
rpW W m ( Fdg Z Fsg (Z ))

(2)
Wg
in which q > x y z I T \ @T 6 represents the pose vector of the end-effector,

M ( q) 6u6 represents a symmetric positive definite inertia matrix, C ( q, q )q 6
represents the vector of centripetal and Coriolis forces, G ( q) 6 represents the gra-
vitational force vector, Fd 6u6 denotes the constant viscous friction coefficient
matrix, Fs 6 represents the Coulomb friction, and Td (t ) 6 is a time-varying
disturbance term. In addition, the matrix J 6un is the Jacobian of the robot which
is defined as in [12], and W n represents the vector of tension forces. Furthermore,
rp represents the radius of pulley, W m n represents the motor torque vector,
W g n represents the viscous and Coulomb friction torques of gearboxes which
assumed to be differentiable and bounded, Z n denotes the vector of shaft speed,
Fdg nun is a diagonal constant matrix, and Fsg (Z ) n denotes the Coulomb fric-
tion. Dynamic Equation (1) has the following properties, which are useful in the sta-
bility analysis [12].
Property 1: The inertia matrix M ( q ) is symmetric, positive definite, and both M ( q )
and M 1(q) are uniformly bounded as a function of q 6 .
Property 2: The matrix M ( q ) 2C ( q , q ) is skew symmetric, which satisfies
y T M ( q ) 2C ( q, q ) y 0, y 6 .
3 Control design
In this section, a back-stepping-like control scheme will be used for the actuated
CDPRs. First, the desired tension W d n is designed to ensure the convergence of q
to the desired trajectory qd . Next, the control input W m is constructed to ensure the
convergence of tension tracking error, eW W W d n , around zero. Before going to
the details of controller derivation, we present the following assumption.
Assumption 1: The desired trajectory of the end-effector, and its first two time deriv-
atives are continuous and uniformly bounded.
3.1 Outer-loop controller design

Let us define an error vector as
S q /q (3)
where q q qd 6 , and / diag( /1 ,..., / 6 ) 6u6 with / i ! 0 for all i 1,..., 6 .
Adding and subtracting the same term W d to the right hand side of Equation (1), and
introducing Q qd /q n , we have
(4)
M (q)S C(q, q )S M ( q)Q C(q, q )Q G( q) Fd q Fs ( q ) Td JeW JW d
The objective of the controller design is to regulate the dynamics in Equation (4) by
designing an appropriate tension vector W d , keeping in mind that the cables should
always be in tension. To satisfy the later, the following general solution is adopted,
which follows the internal forces concept [3]:
Wd Jˆ †W 0 N n (5)
where Jˆ 6un denotes an estimate of Jacobian matrix, N is an arbitrary value

chosen in a way to yield a positive tension for the cables, n n is the null space of
ˆ
Jˆ satisfying Jn 0 , and Ĵ †W 0 is the particular solution for the tension, in which
Jˆ † nu6 denotes the pseudo-inverse of Jˆ , and W 0 6 is the applied wrench on the
end-effector, which is proposed as
W0 Mˆ (q)v Cˆ ( q, q )v Gˆ (q) Hˆ Kd S (6)
In the last equation, Mˆ ( q ) , Cˆ ( q, q ) , Gˆ ( q ) , and Ĥ denote the estimates of M ( q ) ,

C ( q, q ) , G ( q ) , and H N ( Jˆ J )n ( I JJˆ † )W 0 Fd q Fs ( q ) Td 6 , respectively;
in which I 6u6 represents the identity matrix, and K d 6u6 is a constant diagon-
al gain matrix. Now, plugging (5) and (6) into (4), and introducing ( < ) ( <) ( ˆ<) , the
closed-loop dynamics of the error signal can be written as
MS CS K D S G H
Cv
JeW Mv (7)
If some proper update laws can be found so that M̂ o M , Ĉ o C , Ĝ o G and

Ĥ o H , we will show that S will be drive toward zero. By definition of (3), conver-
gence of S also implies the convergence of the output error q . With this in mind, FS
will be used to describe M , C , G and H as linear combinations of sinusoidal func-
tions as
T
M WM Z M H M , C WCT ZC H C , G WGT ZG H G , H WHT Z H H H (8)
where WM 36E M u6 , WC 36 EC u6 , WG 6 EG u6 , and WH 6 E H u6 are constant

weighting matrices; Z M 36E M u6 , and ZC 36EC u6 are matrices, and ZG 6 EG

and Z H 6 E H are vectors of sinusoidal terms, respectively. H M , H C , H G , and H H
are truncation errors of M , C , G , and H respectively, caused by considering a
finite number of sinusoidal functions in the proposed FS; E M , E C , E G , and E H
represent the number of sinusoidal functions fixed by the designer. The corresponding
estimates of M , C , G , and H are introduced as
T
Mˆ Wˆ M ZM , Cˆ WˆCT ZC , Gˆ WˆGT Z G , Hˆ Wˆ HT Z H (9)
Therefore, Equation (7) can be rewritten as

T
MS CS K D S JeW W M Z M Q WCT ZCQ WGT ZG W HT Z H H1 (10)
in which W(x) W(x) Wˆ(x) is the error vector of FS coefficients, and

H1 H1 (H M , H C , H G , H H , v, v) denotes the vector of lumped approximation error.
3.2 Inner-loop controller design

The control objective is to design the motor torque W m , guaranteeing the convergence
of output tension W to the desired tension W d , given by Equation (5). With this in mind
and inspired from the dynamic extension method applied in [12], the following dy-
namic can be considered for Equation (2):
rpW Wm W g (11)
where Wm can be considered as the new control input. The actual input can then be
t
recovered through the time integration of Wm as W m ³0Wm (X )dX . The considerable
point is that the uncertainty term Wg cannot be evaluated directly, since the actual
values of the motor dynamic parameters and the external disturbance are unknown.
Under these circumstances, we propose the control law in the form of
Wm rpWd Wˆg K c eW (12)
where Wˆg n is an estimate of Wg , called residual uncertain vector, and Kc nun is
a constant diagonal matrix. Inserting Equation (12) into (11), and some mathematical
simplification, we obtain
rp eW K c eW Wˆg Wg (13)
If an adaptation rule is designed based on the stability analysis so that Wˆg o Wg , then
eW o 0 can be concluded from (13). With this in mind, the FS is used to represent Wg
as linear combinations of basis functions as
Wg WWT ZW HW (14)
where ZW n EW is the vector of sinusoidal function, WW n EW un is its unknown
coefficients matrix to be estimated, İW n is the bounded approximation error due
to finite-term approximation, and EW ! 0 is the number of terms of basis functions
used in approximation. Supposing that the same set of basis functions are used, the
corresponding estimates can also be represented as
Wˆg WˆWT ZW (15)
where WˆW nEW un is the estimate of WW . Now, substituting expression (14) and (15)
into (13), we have
rp eW Kc eW WWT ZW HW (16)
in which WW WW WˆW is the error vector of the FS coefficients.
4 Stability analysis and performance evaluation

In order to assure the stability of this control system and to find the update laws for
matrices Wˆ M , WˆC , WˆG , Wˆ H and ŴW , a Lyapunov candidate function is devised as
V (S , eW ,WM ,WC ,WG ,W H ) 0.5S T MS 0.5eWT rpeW 0.5Tr[WM

T
QM W M
(17)
WCT QCWC WGT QGWG W HT QH W H WWT QWWW ]
where QM 36 E M u36 E M , QC 36EC u36EC , QG 6 EG u6 EG , QH 6 E H u6 E H and

QW n EW un EW are the convergence rates (the positive definite constant matrices).
Using property 2, the derivation of V (t ) along (10) and (16) is
ªSº ªH º
V ªST eWT º Q « » ª S T eWT º « 1 » Tr[W M T
(QM Wˆ M Z M Q S T )]
¬ ¼ ¬ eW ¼ ¬ ¼ ¬HW ¼
(18)
Tr[WCT (QCWˆC ZCQ S T )] Tr[WGT (QGWˆG Z G S T )]

Tr[W HT (QH Wˆ H Z H S T )] Tr[WWT (QWWˆW ZW eWT )]
ª K 0.5 J º
where Q « D . In order to have positive definite Q , according to the
¬ 0.5 J K c »¼
Shur complement we must have K D ! 0 , and K c ! 0.25J T K D1J . Let us select the up-
date laws with V -modification terms as:
1
Wˆ M QM ( Z M Q S T +V M Wˆ M ) , WˆC QC1 ( Z CQ S T V CWˆC )

WˆG QG1 ( Z G S T +V GWˆG ) , Wˆ H 1
Q H ( Z H S T +V H Wˆ H ) (19)

WˆW QW 1 ( ZW eWT +V WWˆW )
where V ( < ) 's are positive scalars. Upon the substitution of the update laws from (19),
we obtain
ªSº ªH º
V ªST eWT º Q « » ª S T eWT º « 1 » V M Tr (W M T ˆ
WM )
¬ ¼ ¬ eW ¼ ¬ ¼ ¬HW ¼
(20)
V C Tr (WCT WˆC ) V G Tr (WGT WˆG ) V H Tr (W HT Wˆ H ) VWTr (WWT WˆW )
Result 1: Consider the situation in which the required number of sinusoidal functions
for proper performance of the robotic system is selected and the truncation error is
negligible. Then, we can exclude the V -modification terms in (19). Hence, asymptot-
T
ic convergence of ª S T eWT º can be concluded using the Barbalat's Lemma.
¬ ¼
Result 2: Assume that the truncation error cannot be ignored. It is easy to show that
2 2
ªS º ªH º 1 ªS º 1 ª H1 º
ªST eWT º Q « » ª S T eWT º « 1 » d Omin (Q) «e » « » (21)
¬ ¼ ¬eW ¼ ¬ ¼ ¬HW ¼ 2 ¬ W¼ 2Omin (Q) ¬HW ¼
1 1
Tr (W(T< )Wˆ( <) ) d Tr (W(T<)W( <) ) Tr (W(T< )W(< ) ) (22)
2 2
and
2
1 ªSº T T
V d [Omax ( A) «e » Omax (QM )Tr (WM WM ) Omax (QC )Tr (WC WC )
2 ¬ W¼ (23)
Omax (QG )Tr (WGT WG ) Omax (QH )Tr (W HT W H ) Omax (QW )Tr (WWT WW )]
where A diag( M , rp ) ; O max ( < ) denotes the maximum eigenvalue of (<) , and <
represents the Euclidean norm. Using these inequalities, one may rewrite Equation
(20) as
2 2
ªS º 1 ª H1 º
V d G V 0.5 >GOmax ( A) Omin (Q ) @ « » « »
¬ eW ¼ 2Omin (Q ) ¬HW ¼
T
0.5 >GOmax (QM ) V M @Tr (W M WM ) 0.5 >GOmax (QC ) V C @Tr (WCT WC )
0.5 >GOmax (QG ) V G @Tr (WGT WG ) 0.5 >GOmax (QH ) V H @Tr (W HT W H ) (24)
0.5[GOmax (QW ) V W ]Tr (WWT WW ) 0.5 ªV M Tr (WM

T
WM ) V C Tr (WCT WC )
¬
V GTr (WGT WG ) V H Tr (WHT WH ) V WTr (WWT WW ) º
¼
with Omin ( < ) denoting the minimum eigenvalue of (<) . Let us define a constant para-
° Omin (Q) VM VC VG VH
meter G satisfying G d min ® , , , , ,
°̄ Omax ( A) Omax (QM ) Omax (QC ) Omax (QG ) Omax (QH )
V W ½°
¾ . Then (24) becomes
Omax (QW ) °¿
2
1 ª H1 º 1ª T
V d G V V M Tr (WM WM ) V C Tr (WCT WC )
2Omin (Q ) «¬HW »¼ 2¬ (25)
V G Tr (WGT WG ) V H Tr (WHT WH ) VWTr (WWT WW ) º
¼
This implies that V 0 whenever
2
1 ªH (t ) º 1 ª
V ! sup 1 T
V M Tr (WM WM ) V C Tr (WCT WC )
2GOmin (Q ) t tt0 «¬HW ( t )»¼ 2G ¬ (26)
V G Tr(WGT WG ) V H Tr (WHT WH ) VWTr (WWT WW ) º
¼
As a result, we have guaranteed that S , eW , W M , WC , WG , W H , and WW are UUB.
In order to make the set in which V t 0 as small as possible, the parameters in the
right side of (26) play important roles. Boundedness of S implies that q and q are
also bounded, that means boundedness of q and q , respectively, since qd and qd are
bounded. Furthermore, solving inequality (25) and using the lower bound of (17), it is
concluded that
G (t t0 )
ªS º 2V (t0 ) 2 1 ª H1 ( t ) º
«e » d e sup «H ( t ) »
¬ W¼ Omin ( A) GOmin ( A)Omin (Q ) t0 t t ¬ W ¼
1 T
ªV M Tr (WM WM ) V C Tr (WCT WC ) (27)
GOmin ( A) ¬
1
V GTr (WGT WG ) V H Tr (WHT WH ) V WTr (WWT WW ) º 2
¼
ªS º
that means « » is bounded. This completes the transient performance analysis.Ŷ
¬eW ¼
5 Simulation results
In this section, simulations of a CDPR, which has four cables with three degrees of
freedom, are conducted to illustrate the performance of the proposed controller. The
geometrical structure and configuration of this manipulator is shown in Fig.1. Subject
to planar motion of the robot, the generalized coordinates vector is as
q [ x , y , I ]T 3 . Thus, Equation (1) can be simplified to
Mq G Td JW (28)
T
where M diag (m, m, I z ) , G >0 mg 0@ ; m and I z represent the mass and inertia of
the end-effector, respectively, and g represents the gravity acceleration. The parame-
ters of these dynamics and the actuator-gearbox model are set as m 8kg ,
Iz 0.3kgm2 , rp 0.04m , Fdg 0.5 , and Fsg (Z ) 0.5tanh(10Z ) .
Fig 1. Configuration of the studied 4-cable planar parallel robot
In the simulations, it has been assumed that matrices of M ( q ) , C ( q, q ) ; and vectors of

G ( q ) , Fd , Fs ( q ) , and W g are unknown. The disturbance term Td (t ) is also modeled
T
as Td (t ) sin(t ) >0.7 0.7 0.05@ , which is bounded with finite energy. Furthermore,
we assume %20 uncertainty in the Jacobian matrix. Comparisons between the pro-
posed controller and the robust trajectory control presented in [12] have also been
done. Assume that a circle with the radius of 0.25m is defined as the desired trajectory
for the manipulator end-effector. The orientation was also commanded to stay con-
stant ( I 0 ) throughout the circular path. The initial value of the end-effector posi-
tion is set at q(0) [0.1 0.55 0]T , that is far from the desired one. The controller
parameters were selected as Kd diag (100,100,100) , Kc diag (20,20,20,20) , and
/ diag (10,10,10) . Assume that the first term of FS are used for approximation of
9u3 3u3 3u3
Wˆ M , WˆG , Wˆ H , and ŴW , respectively. Therefore, WˆM , WˆG , WˆH
, and ŴW 4u4 . The initial conditions for the Fourier series coefficients have been set
to zero and the convergence rate in the adaptation rule (19) are defined as
QM 1000 u I 9 , QG QH 0.033 u I3 , and QW 0.033 u I 4 , where I ( x ) indicates identity
matrix.
Under the aforementioned settings, Fig. 2a shows the actual and desired motions in
the x-y plane tracked by the proposed adaptive controller, and the robust control
scheme [12]. As it can be seen, the proposed approach has better performance com-
pared with the control system presented in [12]. Also, Fig. 2b represents norm of posi-
tion error. Finally, Fig. 3 shows the applied positive tensions to all cables. According
to this figure, the control signals are smooth during the robot manoeuvres. The large
initial value of the control signal is due to difference between the initial positions of
the end-effector and desired one.
1RUPRISRVLWLRQHUURU
<>P@

Fig 2a.Tracking performance, Fig 2b.Norm of position error

3URSRVHGFRQWUROOHU
5REXVWFRQWURO>@

>1@
>1@

7LPH>VHF@ 7LPH>VHF@

>1@
>1@

7LPH>VHF@ 7LPH>VHF@
Fig 3. Cable forces
6 Conclusion
A FAT-based robust adaptive controller using the FS has been developed for CDPRs.
The controller has been designed so that the information of the system parameters is
not required. The system nonlinearities including inertia matrix, Coriolis/centrifugal
matrix, and gravitational torques vector, have been estimated using the FS. Satisfacto-
ry performance in the transient state has also been investigated. Simulation results on
CDPRs with three DOF verify the efficiency of the proposed controller.
References
1. Abbasnejad, G., Yoon, J., Lee, H.: Optimum kinematic design of a planar cable-driven pa-
rallel robot with wrench-closure gait trajectory. Mech. Mach. Theory 99, 1–18 (2016)
2. Oh, S.-R., Ryu, J.-C., Agrawal, S.K.: Dynamics and control of a helicopter carrying a
payload using a cable-suspended robot. J. Mech. Des. 128(5), 1113–1121 (2006)
3. Kawamura, S., Kino, H., Won, C.: High-speed manipulation by using parallel wire-driven
robots. Robotica 18(01), 13–21 (2000)
4. You, X., Chen, W., Yu, S., Wu, X.: Dynamic control of a 3-DOF cable-driven robot based
on backstepping technique. 6th IEEE Conference on Industrial Electronics and Applica-
tions, pp. 1302–1307 (2011)
5. Fang, S., Franitza, D., Torlo, M., Bekes, F., Hiller, M.: Motion control of a tendon-based
parallel manipulator using optimal tension distribution. IEEE/ASME Trans. Mecha-
tron. 9(3), 561–568 (2004)
6. Oh, S.-R., Agrawal, S.K (2005) Cable suspended planar robots with redundant cables: con-
trollers with positive tensions. IEEE Trans. Robot. 21(3), 457–465.
7. Babaghasabha, R., Khosravi, M.A., Taghirad, H.D (2015) Adaptive robust control of fully-
constrained cable driven parallel robots. Mechatronics 25, 27–36
8. R. J. Caverly, J. R. Forbes, and D. Mohammad shahi.: (2015) Dynam-
ic Modeling and Passivity-Based Control of a Single Degree of Freedom Cable-
Actuated System. IEEE Transactions on Control Systems Technology, Vol: 23, pp. 898 -
909
9. Tiechao Wang, Shaocheng Tong, Jianqiang Yi, and Hongyi Li.: 2015. Adap-
tive Inverse Control of CableDriven Parallel System Based on Type2 Fuzzy Logic System
s. IEEE Transactions on Fuzzy Systems. Vol: 23. Pp. 1803 – 1816.
10. Oh, S.-R., Albus, J.S., Mankala, K., Agrawal, S.K (2005) A dual-stage planar cable robot:
dynamic modeling and design of a robust controller with positive inputs. J. Mech.
Des. 127, 612
11. Jabbari Asl, H, and F. Janabi-Sharifi (2017) Adaptive neural network control of cable-
driven parallel robots with input saturation. Engineering applications of artificial intelli-
gence. Vol: 65, pp. 252-260.
12. Jabbari Asl, H, and J. Yoon (2017) Robust trajectory tracking control of cable-driven pa-
rallel robots. Nonlinear Dynamics, Vol. 89, pp. 2769-2784.
13. A. Izadbakhsh, and P. Kheirkhahan, Adaptive Fractional-Order Control of electrical Flexi-
ble-Joint Robots: Theory and Experiment, Proceedings of the Institution of Mechanical
Engineers. Part I: Journal of Systems and Control Engineering, DOI:
10.1177/0959651818815384.
14. A. Izadbakhsh, and P. Kheirkhahan (2019) An alternative stability proof for "Adaptive
Type-2 fuzzy estimation of uncertainties in the control of electrically flexible-joint robots",
Journal of Vibration and Control, Vol: 25. pp. 977–983.
15. B. Chen, C. Lin, X. Liu, and K. Liu (2016) Observer-based adaptive fuzzy control for a
class of nonlinear delayed systems, IEEE Transactions on Systems, Man, and Cybernetics:
Systems, Vol: 46. pp. 27-36.
16. Q. Zhou, H. Li, C. Wu, L. Wang, and C. K. Ahn (2017) Adaptive fuzzy control of nonli-
near systems with unmodeled dynamics and input saturation using small-gain approach,
IEEE Transactions on Systems, Man, and Cybernetics: Systems.
17. A. Izadbakhsh, S. Khorashadizadeh, and P. Kheirkhahan (2018) Tracking control of electr-
ically driven robots using a model free observer, Robotica,
DOI:10.1017/S0263574718001303.
18. A. Izadbakhsh (2018) Robust adaptive control of voltage saturated flexible joint robots
with experimental evaluations,AUT Journal of Modeling, and simulation, Vol: 50. pp. 31-
38.
19. A. Izadbakhsh, and S. M. R. Rafiei (2009) Endpoint perfect tracking control of robots – A
robust non inversion-based approach, International Journal of Control, Automation, and
systems, Springer, vol.7, no.6, pp. 888-898.
20. A. Izadbakhsh (2017) FAT-based robust adaptive control of electrically driven robots
without velocity measurements, Nonlinear Dynamics, vol. 89, pp. 289-304.
Model Predictive Control of Large-Dimension
João Cavalcanti Santos, Ahmed Chemori, and Marc Gouttefarde

{joao.cavalcanti-santos, ahmed.chemori, marc.gouttefarde}@lirmm.fr
Abstract. A Model Predictive Control (MPC) strategy is proposed in

this paper for large-dimension cable-driven parallel robots working at
low speeds. The latter characteristic reduces the non-linearity of the
system within the MPC prediction horizon. Therefore, linear MPC is
applied and compared with two commonly used strategies: Sliding mode
control and PID+ control. The simulations aim at comparing disturbance
rejection performances and the results indicate a superior performance of
the proposed controller. Indeed, MPC takes into account control limits
(cable tension limits) directly in the control design which allows the
controller to better exploit the robot capabilities. In addition, actuation
redundancy is resolved as an integral part of the control strategy, instead
of calculating the desired wrench and then applying a tension distribution
method.
Keywords: Cable-driven parallel robots, model predictive control, dis-

turbance rejection
1 Introduction
A rather typical control strategy for Cable-Driven Parallel Robots (CDPRs) is

the combination of a linear feedback control with computed torque applied as
a feedforward term [1]. Indeed, the computed torque (also known as feedback
exact linearization) enables the application of usual linear control methods [2, 3].
Besides, Sliding Mode Control (SMC) is an advanced nonlinear feedback control
that has been implemented successfully in CDPRs [4–7]. The main advantages of
SMC are the possibility to attain finite time convergence, simple implementation
and robustness to uncertainties. However, recent and advanced SMC methods
still present chattering issues in experimental setups [6] even if some previous
studies presented methods in order to reduce it [7].
Since a fully-constrained CDPR has more cables than degrees of freedom
(DOF), there are infinitely many possible combinations of cable tensions gen-
erating a desired wrench. The choice of one of these combinations is an actua-
tion redundancy resolution problem where cable tension lower and upper limits
should be taken into account. The lower limit is a positive tension to avoid
cable slackness. The upper limit is set in order to account for the mechanical

222 J. C. Santos et al.
limitations of the cables, motors, etc. Several previous works deal with CDPR
actuation redundancy resolution, e.g. [8, 9].
Usually, the control strategy and the resolution of actuation redundancy are
addressed separately. This paper proposes the use of Model Predictive Control
(MPC) as the control strategy, which has the advantage of solving the tension
distribution problem as an integral part of the control strategy. MPC is a feed-
back control design which, at each decision instant, computes the sequence of
future controls that optimizes a cost function satisfying the system constraints.
The cost function is formed by a weighted sum of individual costs (tracking er-
rors, control input and other performance measures). MPC is considered as one
of the most general way of posing a control problem in the time domain [10,
11]. It defines an optimized admissible control sequence if the considered model
is sufficiently close to reality. Moreover, control limits can be directly handled,
which is an important advantage since the optimized performance is often ob-
tained with active constraints. If the MPC is applied to a linear optimization
problem without constraints, the solution is analytic, e.g. in [12], where Gener-
alized Predictive Control (GPC) is presented. Nevertheless, constraints [13] and
nonlinear systems [14] have been successfully addressed as well.
This paper presents the design, implementation and simulation of an MPC
scheme for motion control of large-dimension CDPRs. To the best of our knowl-
edge, MPC has never been used to control CDPRs. Several applications of CD-
PRs involve a large workspace and relatively low velocities of the mobile platform
[15–18]. Hence, within a reasonable prediction horizon, the CDPR dynamics may
be approximated as being a linear time invariant system. Based on this assump-
tion, linear MPC is applied in this paper. Simulation results compare the per-
formances obtained with the proposed MPC and two other control strategies,
namely SMC and PID+ Controller [19]. The performance evaluation focuses
mainly on external disturbance rejection.
The paper is organized as follows. The modeling of a spatial CDPR is in-
troduced in Section 2. Two state-of-the-art control schemes, SMC and PID+
Control, and the proposed MPC strategy are introduced in Section 3. These
three methods are compared through numerical simulations in Section 4. Con-
clusions and future works are drawn in Section 5.
2 Dynamic Modeling of CDPRs
The dynamic modeling of a spatial CDPR is presented in this section. The

control schemes introduced later in Section 3 are based on this model. Figure 1
illustrates a CDPR with the notations used in its kinematic modeling. The robot
consists of a n-DOF mobile platform driven by m cables. The spatial case where
n = 6 is considered and the m cables (m ≥ 6) are considered massless and
inextensible. Each cable has one end attached to the platform and the other
end wound on a winch drum. The cables are responsible for transmitting to
the platform the efforts applied by the winches. Cable tensions τ applied on
Model Predictive Control of Large-Dimension Cable-Driven Parallel Robots 223
the platform generates the wrench f . These variables are related linearly by the
wrench matrix W, so that f = W τ [9].
The length li is defined as the distance between the drawing point Ai and and
the attachment point Bi . Point Ai is the drawing point defined by the pulley
attached to the base frame. This point is considered as being fixed. Point Bi
is the attachment point of cable i on the platform. The cable length vector is
l = [l1 , ... , lm ]T .
Ai
li
Pulleys
Bi
Platform
z
p
Ob y bles Winches
x Ca
Fig. 1. Illustration of a CDPR with notations for kinematic modeling
Let x, ẋ, ẍ be the platform pose, velocity and acceleration, respectively1 . The
pose (position and orientation) of the platform is then defined by x = [pT , ψ T ]T .
p is the position vector of the reference point of the platform. Vector ψ defines
the orientation of the platform. Typically, it is composed of three Euler angles.
Using Newton-Euler formalism, the dynamics of the platform can be written
as [20]
M(x) ẍ + C(x, ẋ) ẋ = g(x) + f (1)
where matrices M and C are given by
mp I3 −mp ĉ mp ω̂ ω̂ c

M(x) = , C(x, ẋ) ẋ = (2)
mp ĉ H ω̂ H ω
Scalar mp is the platform mass and I3 is the identity matrix of dimension 3.

The present model considers that the platform geometric center and its center
T
of mass may not be coincident and, accordingly, c = cx cy cz is the vector
going from the platform geometric center to its center of mass. Matrices ω̂ and
ĉ are the skew-symmetric matrices associated to ω and c, respectively, with
ω the angular velocity of the platform. The matrix H is defined as H = I +
mp ĉ ĉT where I is the platform inertia matrix relative to its center of mass.
1
We highlight that ẋ = dx
dt
, since angular velocity is not equal to the derivative of the
2
vector of Euler angles. Similarly, ẍ = ddtx .
T
The vector of gravitational forces is g(x) = mp g 0 0 −1 −cy cx 0 . The

wrench applied by the cables on the platform is f = W τ .
3 Proposed Control Schemes
The goal of this paper is to compare MPC performance to those obtained with
strategies commonly used for CDPRs. Namely, MPC is compared to: (i) A lin-
ear PID+ controller [19], based on the controller proposed in [20], and (ii) SMC.
These two strategies were recently validated experimentally, demonstrating ap-
plicability and good performances. A brief description of these control methods
is presented in the following section.
3.1 Background on the State of the Art Controllers
For a given reference trajectory in time ti t tf , the desired poses, velocities

and accelerations are denoted as xd (t), ẋd (t) and ẍd (t), respectively. At a given
instant t, the error in the Cartesian space is expressed as ex (t) = xd (t) − x(t).
Similarly, the error in joint space is ej (t) = ld (t) − l(t), where ld (t) is the vector
of desired cable lengths obtained with the inverse kinematics.
PID+: The PID+ control strategy applies the following wrench

t
f = M(x)ẍd + C(x, ẋ) ẋd − g(x) + W Kp ej + Ki ej (τ ) dτ + Kd ėj (3)
ti
where Kp , Ki and Kd are diagonal matrices containing the linear feedback PID
gains.
Sliding Mode Control: The SMC strategy defines a sliding surface s = ex + Ce ėx ,
with Ce = diag(c1 , ..., cn ). The wrench to be applied on the platform is
f = M ẍd + Cd (ẋd − ẋ) + K sat(s) + Q s + C(x, ẋ) ẋd − g(x) (4)
where K, Q and Cd are diagonal gain matrices. The function sat(s) is a contin-
uous approximation of the sign function. Each component of this vector-valued
function is calculated as follows
⎧
⎨ 1, if si > Δ
sat(si ) = si
, if |si | Δ (5)
⎩ Δ
−1, if si < Δ
The resulting function presents the same output than the sign(s) function
except for the interval −Δ s Δ in which a linear interpolation eliminates
the discontinuity.
Redundancy Resolution: The two above control strategies define wrench f .

The final control output is the vector of cable tensions τ (or motor torques). For
fully-constrained CDPRs, m > n and actuation redundancy shall be resolved to
determine τ being given f , i.e., the equation system W τ = f is underdetermined
and an appropriate vector of cable tensions τ shall be determined among the
infinitely many possible ones (assuming a non-singular pose). In this work, the
following common optimization problem is used to resolve actuation redundancy
minτ 2 (6)
τ
s.t. W τ = f (7)
τmin τ τmax (8)
Using (6), the 2-norm of τ is minimized. As constraints, the tension distri-

bution shall generate the desired wrench (7) and the cable tensions shall be in
an admissible interval (8). The constraint (8) is necessary mainly for two rea-
sons: avoiding cable slackness (0 τmin τ ) and not violating mechanical
limitations of the cables, motors, etc (τ τmax ).
In some cases, the wrench demanded by the controller is not feasible. In
the space of cable tensions, the intersection of the subspaces defined by the
constraints (7) and (8) is empty. Another strategy should then be defined and
the following optimization problem is proposed
minW τ − f P
τ
(9)
s.t. τmin τ τmax
where the subscript P indicates that the 2-norm is calculated with a weighting
positive definite diagonal matrix P, which is necessary since f has components
with inconsistent units (forces and moments).
3.2 Proposed Model Predictive Controller (MPC)

A model predictive control strategy is proposed in the present section. The
continuous model (1) may be approximated as a discrete-time system

x(t + Δt)
y(t + Δt) = = A(t) y(t) + B(t) τ (t) + v(t). (10)
ẋ(t + Δt)
where the vector v and the matrices A and B are given by

0nx1
v= (11)
Δt M−1 g

In Δt In
A= (12)
0nxn In − Δt M−1 C

0nxm
B= , (13)
Δt M−1 W
where time t has been omitted, being given that v, A and B are calculated at
time t.
MPC predicts the states over a given horizon t + N Δt. Equation (10) is
applied N times in order to obtain y(t + i Δt), with i = 1, ..., N . As mentioned
earlier, the proposed MPC strategy is applied to large-dimension CDPRs moving
at low velocities. Therefore, for a small horizon N Δt, the linear time varying
system (10) may be approximated by a linear time invariant system. Matrix A is
considered as A(t + i Δt) = A(t), i = 1, ..., N . Similarly, B and v are considered
constant over the MPC optimization horizon.
With this approximation, linear MPC can be applied. A vector Y(t) contain-
ing the states over the prediction horizon is calculated as follows
⎡ ⎤
⎡ ⎤ B 0 ... 0 ⎡ ⎤
⎡ ⎤ A τ (t)
y(t + Δt) ⎢ AB B ... 0⎥
⎢ 2⎥ ⎢ 0 ⎥⎢ ⎥
⎢ ⎥ ⎢A ⎥ ⎢ ⎥⎢ τ (t + Δt) ⎥
⎢ .. ⎥ ⎢ ⎥ ⎢ 2 . . .⎥ ⎢ ⎥
Y(t) = ⎢ . ⎥ = ⎢ . ⎥ y(t) + ⎢ A B AB B 0 ⎥⎢ .. ⎥ +...
⎣ ⎦ ⎢ . ⎥ ⎢ .. ⎥⎢ ⎥
⎣ . ⎦ ⎢ ⎥⎣ . ⎦
y(t + N Δt) ⎣ . ⎦
AN τ (t + (N − 1) Δt)
AN −1 B AN −2 B . . . B
D U(t)
E
⎡ ⎤
v
⎢ ⎥
⎢ Av + v ⎥
⎢ ⎥
... + ⎢ .. ⎥
⎢ ⎥
⎣ . ⎦
AN −1 v + AN −2 v + ... + v

F
Similarly, the reference trajectory yd = [xTd , ẋTd ]T should be defined over the
prediction horizon
⎡ ⎤
yd (t + Δt)
⎢ yd (t + 2 Δt) ⎥
w(t) = ⎢ (14)
⎢ ⎥
.. ⎥
⎣ . ⎦
yd (t + N Δt)
For a given current state y(t), the control is responsible for finding a trade-off
between minimizing the predicted error w(t) − Y(t) and minimizing the control
effort U(t). To this end, the following cost function can be considered
J y, U = (w − Y)T KY (w − Y) + UT KU U (15)
where Ky and KU are diagonal weight matrices.

The minimization of J for a given current state is equivalent to
min UT (ET KY E + KU ) U + 2 (D y + F − w)T KY E U

U (16)
Hc dT
This minimization is a case of quadratic programming (QP). Constraints on

cable tension limits and rates of change can be easily added

τmin τ τmax
⇒ Aineq U bineq (17)
|τ˙i | τ̇max
Accordingly, at each time step, a QP problem is solved in order to determine

the optimal control output τ (t). This problem is stated as
1
min UT Hc U + dT U
U 2 (18)
s.t. Aineq U b
The QP problem (18) can be efficiently solved with the Interior Point method
implemented in the MATLAB function quadprog.
The optimal argument U(t) = [τ T (t), τ T (t + Δt), ..., τ T (t + (N − 1)Δt)]T
contains all vectors of future control outputs over the control prediction horizon.
The controller applies the first sample of the obtained sequence and maintains
this action until the next time step. After that, the algorithm is repeated. The
new state y(t+Δt) is measured and the states Y(t+Δt) = [y(t+2 Δt)T ... y t+
T
(N + 1) Δt ]T are estimated and optimized taking U(t + Δt) as argument. The
procedure is repeated until t tf .
4 Simulation Results
This section presents simulation results that compare the performances obtained
with the three control strategies presented in Section 3. The context of these
simulations is the project Hephaestus [21] where a large-dimension CDPR is
intended to automatize several tasks in the construction and maintenance of
building facades. The main task of the CDPR is the installation of curtain wall
modules. The robot workspace is a rectangular region in front of the building
facade. Thereby, the CDPR mobile platform can get curtain wall modules on
the ground and position them where needed on the building facade. Since the
CDPR will operate in an outdoor environment, it will be subjected to exter-
nal disturbances. One of the main concerns is the incidence of wind gusts. For
this reason, the simulations presented in this section focus on external distur-
bance rejection performances of the proposed control strategies. An impulsive
disturbance is applied and the response of the CDPR is analyzed. Note that
the simulated trajectory is relatively short. However, the simulation of a longer
trajectory would not affect the results which highlight the disturbance rejection
capabilities of each control strategy.
The initial and final positions are depicted in Fig. 2. The path between these
two positions is a straight line segment. The trajectory is the fastest possible re-
specting upper bounds on linear velocities, accelerations and jerks. These bounds
are 0.3m/s, 0.3m/s2 and 1.0m/s3 , respectively. The resulting trajectory has con-
tinuous derivatives up to the acceleration level. The desired orientation of the
platform is constant along the trajectory. An impulsive disturbance fd is applied

T
at the instant t = 2 s. This impulsive wrench is fd = 55 55 550 0 0 0
(N) and it is applied at the reference point of the platform.
The CDPR configuration (cable drawing points, cable-platform attachments,
and cable arrangement) can also be seen in Fig. 2. Moreover, the parameters
of the CDPR dynamic model are the following: τmin = 100N, τmax = 14kN,
mp = 1000kg, [c]Fp = [0 5 0]T m, I = diag([400 100 400])kg.m2 .
Fig. 2. Initial and final positions of the simulated trajectory
In the following, the obtained results with the three proposed motion control
strategies of Section 3 are presented and discussed. Control parameters used for
the simulations are the following: Kp = 71400 I8 , Ki = 71400 I8 , Kd = 71400 I8 ,
Ce = 2 × 10−3 I6 , Cd = 36 I6 , Q = 40 I6 , K = 0.2 I6 , N = 20, Δt = 6 × 10−4 ,
Δr = 5 × 10−1 , KY p = 6 × 109 , KY v = 1 × 10−2 , KU = 2 × 10−5 I8N .
Scalar KY p are the components of KY related to pose errors, whereas KY v is
related to velocity errors. Value of Δt is used in (5) as Δ for translational inputs
of sat(s), and Δr is used for rotational inputs.
Figure 3 shows the evolution versus time of the norms of the translational
and rotational errors. All the control strategies are able to cancel the tracking
error caused by the applied external disturbance but PID+ presents an oscil-
latory behavior and increased settling time. SMC responds faster and without
oscillations. MPC presents the fastest response, resulting in the smallest tracking
errors along the whole trajectory.
The histograms of Fig. 4 present a performance comparison on different as-
pects. Let et be the 2-norm of the translational error. Histogram (a) quantifies
the maximum value et over the trajectory. Histogram (b) compares the RMS
(a) (b)
20 0.8
SMC
MPC
15 PID 0.6
mm
rad
10 0.4
5 0.2
0 0
0 2 4 6 8 0 2 4 6 8
s s
Fig. 3. Norm of the (a) translational and (b) rotational errors
value of et along the trajectory. Taking these two performance measures, MPC
leads to the smallest error.
Fig. 4. Comparative results: (a) maximum error, (b) error RMS, (c) maximum cable
tensions, (d) maximum cable tension derivatives, (e) RMS of cable tension derivatives,
and (f) consumed energy.
Regarding cable tension values, as shown in Fig. 4-(c), MPC demands the
maximum allowed value τmax = 14 kN. This is also shown in Fig. 5, which depicts
the cable tensions near the instant of application of the impulsive disturbance.
Indeed, as discussed earlier, the main advantage of MPC is that the controller
takes into account the constraints of the system and optimizes the control actions
in order to reduce the tracking errors. Here, the maximum allowed cable tension
is an active constraint just after the impulsive disturbance is applied. In the
case of SMC, the maximum tension value is 13.7 kN along the trajectory which
indicates that this controller response is close to the largest admissible value
τmax = 14 kN. If higher gains were used, the wrench f would then be unfeasible.
In order to not exceed τmax , the strategy described in Eq. (9) would be necessary
and the resulting wrench would not be equal to f . However, the MPC strategy
does not lead to this risk. The PID+ controller has |τmax | = 9.1 kN. This
relatively low tension is a consequence of the use of small gains Kp . Indeed, small
gains were used because larger gains lead to high frequency oscillations of cable
tensions. For instance, increasing the gains of less than 1% with this strategy,
RMS(τ̇ ) is equal to 86 kN (more details about this measure are discussed in the
following paragraph).
The time derivative of the cable tensions is a measure of the aggressiveness
of the control action. Large values of this variable may excite high frequency dy-
namics which are difficult to control. Figure 4-(d) presents the maximum deriva-
tive of cable tensions over the trajectory considering all cables. Mathematically,
values in Fig. 4-(d) are equal to maxt maxi |τ̇i (t)| . The smallest maximum ca-
ble tension derivative is obtained for the PID+, which is an advantage of this
method. SMC is the most aggressive controller considering this performance mea-
sure. The proposed MPC strategy constraints this variable (Eq. (17)). Therefore,
any value can be imposed independently of the rest of the controller parame-
ters. In simulations, the value used is 800 kN/s and Fig. 4-(d) shows that this
value is reached. Fig. 4-(e) presents the values of the maximum RMS value of
τ̇i considering all cables i = 1, ...8. More precisely, Fig. 4 (e) presents values
of maxi RMS(τ̇i ) . MPC presents the largest RMS of the cable tension deriva-
tives. Note that the MPC optimization cannot take as constraint variables that
depends on the system states beyond the prediction horizon. The prediction
horizon covers only a small part of the whole trajectory over which the RMS
values are calculated.
(a) (b) (c)

14 14 14
Actual Cable Tensions
12 t max 12 12
10 10 10
8 8 8
kN
kN
kN
6 6 6
4 4 4
2 2 2
0 0 0
2 2.1 2.2 2.3 2.4 2.5 2.6 2 2.1 2.2 2.3 2.4 2.5 2.6 2 2.1 2.2 2.3 2.4 2.5 2.6
s s s
Fig. 5. Simulated cable tensions for (a) PID+ Controller, (b) Sliding Mode Controller
and (c) Model Predictive Control.
The consumed energy over the trajectory (Fig.4-(f)) is roughly the same for
all the control schemes. Nevertheless, MPC leads to the highest consumption.
t
The consumed energy is calculated as tif |l̇(t)T τ (t)|dt.
5 Conclusions and Future Work

An MPC strategy was proposed for large-dimension CDPRs. Considering that
the robot works at low velocities, a linear MPC approach was selected. A linear
time invariant system is considered along the MPC optimization horizon. MPC
allows to minimize the error between the desired trajectory and the predicted
positions while taking into account constraints on cable tensions. Since cable
tensions are the argument of this optimization, actuation redundancy resolution
is integrated within the MPC optimization. Simulation results validate these
advantages by comparing the disturbance rejection performances of the proposed
MPC strategy with two other commonly used controllers (PID+ and SMC).
MPC yields the smallest errors when compared to these two other controllers.
In future works, the modeling errors due to the assumption of a linear time
invariant system should be quantified. Indeed, the non-linear dynamics of the
CDPR will differ from the proposed linear time invariant system according to the
mobile platform dynamics and the MPC optimization horizon. These modeling
errors may be predicted and the relevance of the proposed linear MPC may be
evaluated in a general case. Additionally, experimental tests should be conducted
to validate the simulations presented in this paper. Experimental validation may
also clarify the feasibility regarding computational burden, which is a critical
aspect of using MPC for real-time control of CDPRs.
ACKNOWLEDGMENT
The research leading to these results has received funding from the European
Union’s H2020 Programme (H2020/2014-2020) under grant agreement No. 732513
(Hephaestus project).
References
1. R. L. Williams, P. Gallina, and J. Vadia, “Planar translational cable-direct-driven
robots,” Journal of Robotic Systems, vol. 20, no. 3, pp. 107–120, 2003.
2. M. H. Korayem, H. Tourajizadeh, and M. Bamdad, “Dynamic load carrying ca-
pacity of flexible cable suspended robot: Robust feedback linearization control
approach,” Journal of Intelligent and Robotic Systems: Theory and Applications,
vol. 60, no. 3-4, pp. 341–363, 2010.
3. M. A. Khosravi and H. D. Taghirad, “Robust PID control of fully-constrained cable
driven parallel robots,” Mechatronics, vol. 24, no. 2, pp. 87–97, 2014.
4. A. Alikhani and M. Vali, “Sliding Mode Control of a Cable-driven Robot
via Double-Integrator Sliding Surface,” in International Conference on Control,
Robotics, and Cybernetics, 2012.
5. G. El-Ghazaly, M. Gouttefarde, and V. Creuze, “Adaptive terminal sliding mode

control of a redundantly-actuated cable-driven parallel manipulator: CoGiRo,” in
Cable-Driven Parallel Robots. Springer, 2015, pp. 179–200.
6. C. Schenk, C. Masone, A. Pott, and H. H. Bülthoff, “Application of a Differentiator-
Based Adaptive Super-Twisting Controller for a Redundant Cable-Driven Parallel
Robot,” in Cable-Driven Parallel Robots. Springer, 2018, pp. 254–267.
7. M. Zeinali and A. Khajepour, “Design and Application of Chattering-Free Sliding
Mode Controller to Cable-Driven Parallel Robot Manipulator: Theory and Exper-
iment,” in Volume 2: 34th Annual Mechanisms and Robotics Conference, Parts A
and B. ASME, jan 2010, pp. 319–327.
8. C. Gosselin and M. Grenier, “On the determination of the force distribution in
overconstrained cable-driven parallel mechanisms,” Meccanica, vol. 46, no. 1, pp.
3–15, 2011.
9. M. Gouttefarde, J. Lamaury, C. Reichert, and T. Bruckmann, “A Versatile Tension
Distribution Algorithm for n - DOF Parallel Robots Driven by n+2 Cables,” IEEE
Transactions on Robotics, vol. 31, no. 6, pp. 1444–1457, 2015.
10. E. F. Camacho and C. B. Alba, Model predictive control. Springer Science &
Business Media, 2013.
11. J. M. Maciejowski, Predictive control: with constraints. Pearson education, 2002.
12. L. Cuvillon, E. Laroche, J. Gangloff, and M. De Mathelin, “GPC versus H∞ con-
trol for fast visual servoing of a medical manipulator including flexibilities,” in
International Conference on Robotics and Automation - ICRA. IEEE, 2005, pp.
4044–4049.
13. G. Buondonno, F. Patota, H. Wang, A. De Luca, and K. Kosuge, “A model pre-
dictive control approach for the Partner Ballroom Dance Robot,” in Robotics and
Automation (ICRA), 2015 IEEE International Conference on. IEEE, may 2015,
pp. 774–780.
14. A. Chemori and M. Alamir, “Multi-step limit cycle generation for Rabbit’s walking
based on a nonlinear low dimensional predictive control scheme,” Mechatronics,
vol. 16, no. 5, pp. 259–277, 2006.
15. J. Albus, R. Bostelman, and N. Dagalakis, “The NIST robocrane,” Journal of Field
Robotics, vol. 10, no. 5, pp. 709–724, 1993.
16. M. Gouttefarde, J.-F. Collard, N. Riehl, and C. Baradat, “Geometry selection of
a redundantly actuated cable-suspended parallel robot,” IEEE Transactions on
Robotics, vol. 31, no. 2, pp. 501–510, 2015.
17. G. Meunier, B. Boulet, and M. Nahon, “Control of an overactuated cable-driven
parallel mechanism for a radio telescope application,” IEEE transactions on control
systems technology, vol. 17, no. 5, pp. 1043–1054, 2009.
18. X. Tang and R. Yao, “Dimensional Design on the Six-Cable Driven Parallel Ma-
nipulator of FAST,” Journal of Mechanical Design, vol. 133, no. 11, 2011.
19. B. Paden and R. Panja, “Globally asymptotically stable pd+’ controller for robot
manipulators,” International Journal of Control, vol. 47, no. 6, pp. 1697–1712,
1988.
20. J. Lamaury and M. Gouttefarde, “Control of a Large Redundantly Actuated Cable-
Suspended Parallel Robot,” in International Conference on Robotics and Automa-
tion. IEEE, 2013, pp. 4644–4649.
21. [Online]. Available: www.hephaestus-project.eu
Linearised Feedforward Control
of a Four-Chain Crane Manipulator
Michael Stoltmann1 , Pascal Froitzheim2 , Normen Fuchs2 , Wilko Flügge2 , and

Christoph Woernle1
1 University of Rostock, 18059 Rostock, Germany,
{michael.stoltmann,woernle}@uni-rostock.de,
WWW home page: http://www.ltmd.uni-rostock.de
2 Fraunhofer Research Institution for Large Structures in Production Engineering IGP,
18059 Rostock, Germany
WWW home page: https://www.igp.fraunhofer.de
Abstract. For a crane manipulator suspending a flexible metal plate by four

chains feedforward control is derived that moves the payload along desired spa-
tial trajectories. Due to the statically indeterminate suspension, the stiffness of
the payload has to be taken into account. The actuator coordinates can be al-
gebraically calculated from the desired trajectory of the plate at the position,
velocity and acceleration levels exploiting the flatness property of the system.
As the system is kinematically redundant, a technically meaningful solution like
minimal inclination angles of the chains are determined by optimisation. As the
iterative calculation of the nonlinear constrained optimisation problem is compu-
tationally expensive and not suitable for implementation on the crane controller,
a linearised inverse dynamics model is derived that describes small sway motions
of the flexible payload around a static equilibrium state. Linearised and nonlinear
feedforward control are compared in a numerical simulation of the crane system.
Keywords: crane manipulator, elastic payload, underactuated system, linearisa-

tion, inverse dynamics, quadratic programming, feedforward control
1 Introduction
Metal plates can be formed by means of a so-called ship building press in a cold plastic
forming process. Hereby the workpiece is positioned above the forming tool by four
chain hoists that are mounted on trolleys of an overhead crane with two bridges each
with two trolleys as schematically shown in Fig. 1. Bridges, trolleys and chain hoists
are individually controllable to achieve the desired workpiece position. In addition to
the flexibilty of the workpiece itself the four chain attachment points at the metal plate
are compliant in order to isolate the crane gear from process forces.
This crane manipulator can be considered as a 4-4 cable-driven parallel manipula-
tor (CDPM) with movable winch positions that is underconstrained (incompletely re-
strained) or underactuated [12, 13]. Compared to completely constrained and overcon-
strained CDPMs the pose of the payload is not fully defined by the actuator positions,
here the bridge and trolley positions and the chain lengths, but by the static equilibrium

234 M. Stoltmann et al.
u1 x0
y0 z0
u12
u2 u11
u22
u21
w11 w22 w12
a
b η
w21
z1 x1
y1
Fig. 1. Cable manipulator with four load chains suspending a flexible rectangular metal plate [15].
of the system under gravity. Accordingly the definitions of direct and inverse kinematics
of completely constrained manipulators are extended to direct and inverse statics, also
designated as geometrico-static analysis in [6]. The task of direct statics of a CDPM
is to find stable equilibrium poses of the payload for given actuator coordinates, while
inverse statics comprises the determination of the actuator positions for a desired pay-
load pose. Solution of direct statics with finding all possible stable equilibrium poses
of CDPMs was solved with different methods e. g. in [1, 2, 9]. Inverse statics for a four-
cable CDPM was considered in [5].
The payload trajectory of an underactuated CDPM cannot be kinematically con-
trolled but requires dynamic control in order to avoid undesired residual sway motions.
Feedforward and feedback approaches for anti-sway control of cranes are mostly based
on linearised equations of motion that enable linear controller design, see e.g. [14, 11],
or the development of input shaping strategies [10]. Dynamic models of cranes and
CRPMs with idealised massless cables exhibit the property of flatness [7] that can be
exploited to derive a nonlinear feedforward control model [8]. The actuator coordinates
and actuator forces are algebraically calculated from the desired trajectory of the pay-
load and their time derivatives up to the second and fourth order, respectively. The twice
time differentiation of the equation of motion here required can be avoided by applying
an implicit integration procedure [4, 18].
For the crane manipulator shown in Fig. 1 direct and inverse statics and flatness-
based feedforward trajectory control were formulated in [15] using the nonlinear equa-
tions of motion. However, this requires the iterative solution of a nonlinear constrained
optimisation problem that is computationally expensive and not directly suitable for im-
plementation into a plant controller. Therefore a simplified feedforward control scheme
is studied in the present contribution. It is derived from the linearised equations of mo-
Linearised Feedforward Control of a Four-Chain Crane Manipulator 235
tion under the assumption that the motion takes place in the neighbourhood of a static
equilibrium position of the system.
The contribution is organised as follows. In section 2 the nonlinear dynamic model
of the crane manipulator formulated in [15] is summarised. The linearised equations of
motion are analytically derived by Taylor series expansion in section 3. Feedforward
control based on the linearised equations of motion is described in section 4. A simula-
tion example comparing linear and nonlinear feedforward control is shown in section 5.
2 Nonlinear Dynamics of the Crane Manipulator

The equations of motion of the crane manipulator are formulated in terms of the abso-
lute pose coordinates of the flexible plate. The masses of the chains are neglected. The
bridge and trolley displacements and the chain lengths are treated as kinematic inputs
of the dynamic model under the assumption of individual drive controllers.
2.1 Coordinates
According to Fig. 1 the independently controllable actuator coordinates are, with i = 1, 2
and j = 1, 2, the positions of the two bridges ui , the positions of the four trolleys ui j and
the lenghts of the four chains wi j including the lengths of the unloaded attachment
springs. They are summarised in the vectors
u = [ u1 u2 u11 u12 u21 u22 ]T , w = [ w11 w12 w21 w22 ]T . (1)
The spatial pose of the flexible plate relative to K0 is described by the six po-
sition coordinates of the body frame K1 , comprising the three Cartesian coordinates
r1 = [ r1x r1y r1z ]T of O1 , three Cardan angles ϕ = [ ϕ1 ϕ2 ϕ3 ]T , defined in the rotation
sequence around the follower axes z1 –y1 –x1 , and a coordinate to describe the elastic
deformation of the plate η defined in the following subsection. The pose coordinates
are summarised in the 7-vector
T
r̂ = rT1 ϕ T η .

(2)
To describe the spatial velocity of the flexible plate relative to K0 , the velocity v1 = ṙ1
of O1 , the angular velocity vector ω of K1 and the time derivative vη of η are used,
summarised in the 7-vector
T
v̂ = vT1 ω T vη .

(3)
The relation between r̂ and v̂ is given by the kinematic differential equation [17]
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ṙ1 I 0 0 v1
ϕ ⎦ = ⎣ 0 Hϕ (ϕ
⎣ ϕ̇ ϕ) 0 ⎦ ⎣ ω ⎦
η̇ 0 0 1 vη (4)
r̂˙ = H(r̂) v̂
ϕ ) relating the time derivatives
with the (3, 3) identity matrix I and the (3, 3) matrix Hϕ (ϕ
of the Cardan angles ϕ to the angular velocity ω .
2.2 Nonlinear Equations of Motion
The workpiece is modelled by a rectangular Kirchhoff plate (edge lengths a, b, thick-

ness h and density ρ being constant over the area A, body coordinate system K1 located
in the mass center O1 ) that is suspended at its corners (Fig. 2). The displacement of
the plate in z1 -direction is only modelled by the shape function φ (x, y) and the elastic
coordinate η(t) describing the torsion [3],
x y
W (x, y,t) = φ (x, y) η(t) with φ (x, y) = 4 . (5)
ab
Compared to other flexible modes of the workpiece, the torsion most strongly influences
the force distribution among the four load chains. The vectors from O1 to the corners
Qi j then are
di j = di j,0 + φi j ez1 η (6)
with vectors di j,0 of the planar plate and φi j = 1 for i = j and φi j = −1 for i = j.
The nonlinear equations of motion of the crane manipulator are [15]
M v̂˙ = kc (v̂) + ke (r̂, u, w) + kel (r̂) (7)
with the mass matrix M and the generalised centrifugal and Coriolis forces kc according
to ⎡ ⎤ ⎡ ⎤
mI 0 0 0
M = ⎣ 0 Θ 1 0 ⎦, ω Θ 1ω ⎦.
kc = ⎣ −ω̃ (8)
⎢ ⎥ ⎢ ⎥
0 0 91 m 0
Neglecting quadratic terms in the thickness h and the flexible coordinate η, the inertia
tensor of the undeformed plate Θ 1 = diag 12 1 1
mb2 , 12 1
ma2 , 12 m(a2 + b2 ) , represented
in K1 , can be used.
The vector of generalised external forces on the plate ke in (7) is governed by the
gravity force fG = m g ez0 at the mass center O1 and the four suspension forces fi j at the
−f11
−f12
η Q12
−f21 d11 fG η
Q11
d12
Q21 b
η d 21
z1
O1 2
a e z1 d22 x1
b
2 2
y1 a η
2
−f22
Q22
Fig. 2. Model of the flexible metal plate [15].

O0
y0 x0
z0
ui
Pij trolley ij
uij bridge i
eij
r1
chain ij
O1
wij dij x1
c z1
lij Qij
y1
Fig. 3. Geometric relations for calculation of the suspension forces [15].
attachment points Qi j for i = 1, 2 and j = 1, 2,

fG − f11 − f12 − f21 − f22
⎡ e⎤ ⎡ ⎤
kf
ke = ⎣ keτ ⎦ = ⎣ −ττ 11 −ττ 12 −ττ 21 −ττ 22 ⎦ with τ i j = d̃i j fi j . (9)
⎢ ⎥ ⎢ ⎥
kηe
−ez1 (−f11 + f12 + f21 − f22 )
T
Here kef is the resulting force, keτ the resulting moment of the suspension forces with
respect to O1 , and kηe is the generalised force on the elastic coordinate η due to the
suspension forces.
With the position vectors rPi j = ey0 ui + ex0 ui j of points Pi j and rQi j = r1 + di j of
points Qi j according to Fig. 3 and under consideration of linear attachment springs
(stiffness c) the chain forces are
li j
fi j = c (li j − wi j ) ei j with ei j = , li j = rQi j − rPi j , li j = |li j | . (10)
li j
The generalised elastic force of the plate kel in (7) is related to the bending stiffness
of the plate K under consideration of Young’s modulus E and Poisson’s ratio ν [3],
⎡ ⎤
0
32 K (1 − ν) E h3
kel = ⎣ 0 ⎦ with kηel = −cf η , cf = , K= . (11)
kel ab 12 (1 − ν 2 )
η
2.3 Direct and Inverse Dynamics and Statics

Direct dynamics of the crane manipulator yields the spatial motion of the plate r̂(t)
for given time trajectories of the actuator coordinates u(t), w(t) by means of numeri-
cal integration of the equations of motion (7) together with the kinematic differential
equations (4).
Inverse dynamics enables feedforward control by calculating the actuator posi-
tions u(t), w(t) for a desired spatial trajectory of the plate given in terms of posi-
tion r̂d (t), velocity v̂d (t) and acceleration v̂˙ d (t). This calculation exploits the flatness
property of the system [7]. The equations of motion (7) represent a system of seven
nonlinear equations for the ten actuator positions u, w. Thus the problem is kinemat-
ically triple redundant. Appropriate solutions can be found by optimisation [15]. A
technically meaningful objective criterion are minimal chain inclinations.
Direct and inverse statics is based on the static equilibrium conditions
ke (r̂, u, w) + kel (η) = 0 (12)
included in (7). Direct statics comprises calculation of the static equilibrium pose of the
plate
T
ϕ T η̄
r̂¯ = r̄T1 ϕ̄

(13)
and of the forces of the chains for given stationary actuator coordinates ū, w̄. Here the
seven equilibrium equations (12) have to be iteratively solved with respect to the seven
unknowns r̂¯ under the condition of positive chain forces. Inverse statics yields the ten
actuator coordinates ū, w̄ to achieve a desired equilibrium pose of the plate r̂.¯ As for
inverse dynamics redundancy resolution by optimisation is necessary.
3 Linearised Dynamics of the Crane Manipulator

The linearised equations of motion are formulated to describe small oscillations of the
platform around a static equilibrium pose r̂¯ according to (13). The small deflections of
the platform from its equlibrium posen are described by
T
Δz = ΔrT1 Δψ ψ T Δη

(14)
comprising the deflection Δr1 of point O1 , the rotation vector Δψψ with ω = ψ̇
ψ describing
small rotations from the equilibrium orientation, and the increment of the elastic coordi-
nate Δη. According to the kinematic differential equation (4) the rotation increment Δψ ψ
is linked to the corresponding increments of the Cardan angles Δϕ ϕ by Δϕ
ϕ = Hϕ (ϕ̄ ϕ ) Δψ
ψ.
With the increments of the actuator coordinates Δu and Δw from the respective station-
ary values ū and w̄ a Taylor series expansion of the nonlinear equations of motion (7)
up to first-orders in the Δ-variables yields the linearised equations of motion in the form
M Δz̈ + K Δz = Bu Δu + Bw Δw (15)
with the mass matrix M ∈ the stiffness matrix K ∈
R7,7 , R7,7
and the input matrices
Bu ∈ R7,6 , Bw ∈ R7,4 .
The linearised equations of motion (15) are obtained by successively linearising all
calculations during formulation of the nonlinear equations of motion (7). The incre-
ments of the chain forces fi j from (10) become (magnitudes related to the equilibrium
values denoted by an upper bar)
Δfi j = c (Δli j − Δwi j ) ēi j + c (l¯i j − w̄i j ) Δei j . (16)
The increments of the suspension lengths Δli j and of the chain unit vectors Δei j are
obtained by calculating firstly from (10) the increments of the suspension vectors
Δli j = ΔrQi j − ΔrPi j with ΔrPi j = ey0 Δui + ex0 Δui j , ΔrQi j = Δr1 + Δdi j . (17)
Using the tensor representation of a vector product a × b ≡ ã b ≡ −b̃ a with the skew-
symmetric tensor ã = −ãT , the vector increments Δdi j are expressed by the rotation
increment Δψψ according to
Δd = Δψ
ij ψ d̄ + φ e Δη = −d̄˜ Δψ
ij i j z1ψ + φ e Δη , ij (18)
i j z1
again with φi j = 1 for i = j and φi j = −1 for i = j. The increments of the suspension

vectors then are
⎡ ⎤
⎤ Δu1
I |−d̄˜ 11 | ez1 ⎡ ⎤ −ey0 0 | −ex0 0
⎡ ⎤ ⎡ ⎤ ⎡
Δl11 0 0 ⎢ Δu2 ⎥
˜ ⎥ Δr1 ⎢−e | −e
⎢ 12 ⎥ ⎢ I |−d̄12 |−ez1 ⎥⎣ ⎦ ⎢ y0
⎢Δl ⎥ ⎢ 0 0 0 0 ⎥⎢ ⎥
ψ
x0 ⎢Δu11 ⎥. (19)
⎥⎢ ⎥
⎥ =⎢ ˜ Δψ +
⎣Δl21 ⎦ ⎣ I |−d̄21 |−ez1 ⎦ ⎣ 0 −ey0 | 0 0 −ex0 0 ⎦⎢Δu12 ⎥
⎢ ⎥ ⎢ ⎥⎢
Δη ⎥
Δl22 ˜
I |−d̄22 | ez1 0 −ey0 | 0 0 0 −ex0 Δu21
⎣ ⎦
Δz Δu22
Jlz Jlu
Δu
The increments of the suspension lengths Δli j are obtained by
Δli j = ēTi j Δli j = ēTi j Jlz,i j Δz + ēTi j Jlu,i j Δu (20)
whereby Jlz,i j and Jlu,i j are the row entries of Jlz and Jlu , respectively, in (19). The
increments of the chain unit vectors Δei j are obtained from the taylor series expansion
of Δli j = li j ei j yielding
Δli j ēi j
Δei j = ¯ − ¯ Δli j (21)
li j li j
and with (19) and (20)
I − ēi j ēTi j I − ēi j ēTi j
Δei j = J lz,i j Δz + Jlu,i j Δu . (22)
l¯i j l¯i j

Jez,i j Jeu,i j
Inserting (20) and (21) into (16) yields the increments of the chain forces
c (s̄11 Jez,11 + ē11 ēT11 Jlz,11 ) c (s̄11 Jeu,11 + ē11 ēT11 Jlu,11 )
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
Δf11
⎢Δf12 ⎥ ⎢c (s̄12 Jez,12 + ē12 ēT12 Jlz,12 )⎥ ⎢c (s̄12 Jeu,12 + ē12 ēT12 Jlu,12 )⎥
⎣Δf21 ⎦ = ⎣c (s̄21 Jez,21 + ē21 ēT Jlz,21 )⎦ Δz + ⎣c (s̄21 Jeu,21 + ē21 ēT Jlu,21 )⎦ Δu
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
21 21
Δf22 c (s̄22 Jez,22 + ē22 ēT22 Jlz,22 ) c (s̄22 Jeu,22 + ē22 ēT22 Jlu,22 )

Jfz Jlu
(23)
−c ē11 0
⎡ ⎤⎡ ⎤
0 0 Δw11
⎢ 0 −c ē12 0 0 ⎥ ⎥ ⎢Δw12 ⎥ .
⎢ ⎥
+⎢⎣ 0 0 −c ē21 0 ⎦ ⎣ Δw21 ⎦
0 0 0 −c ē22 Δw22

Jfw Δw
The increments of the generalised forces kef in (9) become with the row entries
J f z,i j , Jlu,i j and J f w,i j of J f z , Jlu and J f w , respectively,
2 2 2 2 2 2
Δkef = − ∑ ∑ J f z,i j Δz − ∑ ∑ J f u,i j Δu − ∑ ∑ J f w,i j Δw . (24)
i=1 j=1 i=1 j=1 i=1 j=1

Jk f ,z Jk f ,u Jk f ,w
The increments of the moments τ i j = d̃i j fi j of the chain forces fi j with respect to O1
are
Δττ i j = Δd̃i j fi j + d̃i j Δfi j = −f̃i j Δdi j + d̃i j Δfi j (25)
and after inserting (18) and (23)

Δττ i j = 0 | ˜f̄i j d̄˜ i j | φi j ez1 + d̄˜ i j JFz,i j Δz + d̄˜ i j J f u,i j Δu + d̄˜ i j J f w,i j Δw . (26)

Jτz,i j Jτu,i j Jτw,i j
In analogy to (24) the increment of the resulting torque keτ in (9) reads
2 2 2 2 2 2
Δkeτ = − ∑ ∑ Jτz,i j Δz − ∑ ∑ Jτu,i j Δu − ∑ ∑ Jτw,i j Δw . (27)
i=1 j=1 i=1 j=1 i=1 j=1

Jkτ,z Jkτ,u Jkτ,w
The increment of the generalised force kηe in (9) becomes
Δkηe = − (−f̄11 + f̄12 + f̄21 − f̄22 )T Δez1 − ēTz1 (−Δf11 + Δf12 + Δf21 − Δf22 ) (28)

f̄Tη
with
ψ ēz1 = −ē˜ z1 Δψ
ψ = 0 | −ē˜ z1 | 0 Δz

Δez1 = Δψ (29)

Jψ z
and together with Δfi j from (23)
Δkηe = − f̄Tη Jψ z + ēTz1 (−JFz,11 + JFz,12 + JFz,21 − JFz,22 ) Δz

Jkη,z
− ēTz1 (−JFu,11 + JFu,12 + JFu,21 − JFu,22 ) Δu
(30)
Jkη,u
− ēTz1 (−JFw,11 + JFw,12 + JFw,21 − JFw,22 ) Δw .

Jkη,w
The increment of the elastic force kηel from (11) is directly obtained by
Δkηel = −cη Δη = − 0 | 0 | cη Δz .

(31)

Jηz
With (24), (27), (30) and (31) the linearised equations of motion (15) become
−Jk f ,u −Jk f ,w
⎡ ⎤⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤
mI 0 0 Δr̈ Jk f ,z Δr
⎣ 0 Θ 1 0 ⎦⎣Δψ̈ ψ ⎦ + ⎣ Jkτ,z ⎦⎣Δψ ψ ⎦ = ⎣ −Jkτ,u ⎦Δu + ⎣ −Jkτ,w ⎦Δw
0 0 91 m Δη̈ Jkη,z + Jηz Δη −Jkη,u −Jkη,w (32)
M Δz̈ + K Δz = Bu Δu + Bw Δw .
4 Feedforward trajectory control

The objective of feedforward control is to move the workpiece along a desired trajec-
tory between two rest positions with pose r̂d (t), velocity v̂d (t) and acceleration v̂˙ d (t).
The desired trajectory is discretised in N intervals with a time increment Δt leading to
pose increments between subsequent time steps tk and tk+1 = tk + Δt, k = 0, 1, . . . , N,
according to Δr̂dk = r̂d (tk+1 ) − r̂d (tk ) and
Δzdk = H−1 (r̂dk ) Δr̂dk (33)
with the kinematic matrix H from (4). The desired acceleration at time tk is Δz̈d = v̂˙ dk .
The linearised equations of motion (32) represent conditions for increments of the
actuator positions Δuk and Δwk that are consistent with the desired trajectory,

Δuk
= M Δz̈dk + K Δzdk .

Bu Bw (34)
Δwk
This is an underdetermined system of seven linear equations for the increments of the
ten actuator coordinates Δuk and Δwk , corresponding to the triple kinematic redun-
dancy. Solutions can be found by quadratic programming using an objective function
being quadratic in terms of the actuator coordinate increments Δuk and Δwk and (34) as
linear constraint equations. The actuator coordinates are then updated according to
uk+1 = uk + Δuk , wk+1 = wk + Δwk , k = 0, 1, . . . , N, (35)
with the actuator coordinates of the initial rest position u0 , w0 . The position-dependent
system matrices in (34) are calaculated with the values of the initial rest position and
kept constant during execution of the trajectory.
The actuator coordinates and the increments are used as inputs for individual axis
controllers that compensate disturbing effects like inertia, friction and varying loads.
Within the following numerical example it is assumed that the axis controllers make the
actuators track the desired trajectories ideally. In a next step simplified dynamic transfer
behaviours of the controlled drives will be incorporated into the control scheme [11].
5 Numerical Example
As an example a rectangular plate with a = 2 m, b = 3 m, h = 0.01 m, ρ = 7800 kg m−3 ,
E = 2.1 · 1011 N m−2 , ν = 0.3 and c = 1.5 · 104 N m−1 is considered. The numerical cal-
culations were done in Matlab. According to Fig. 4a the plate is moved from the initial
interpolated plate position bridge positions

1.5 1.5
1 r1dy u1 u2
1
r1dz
[m]
[m]
0.5
O0 0.5
y0 x0 0
r1dx
0
z0 0 5 10 0 5 10
interpolated plate Cardan angles trolley positions

6 0
4
ϕ d3 u11
ϕ 1d -0.2
2 u12
[deg]
[m]
0 -0.4 u22
ϕ d2 u21
-2 -0.6
0 5 10 0 5 10
O1 interpolated plate torsion chain lengths
initial position 0.02
t0 = 0s 1 w22 w21
0.015 w12
ηd
[m]
[m]
0.01
0.5 w11
final position
0.005
t N = 10s
0 0
0 5 10 0 5 10
time [s] time [s]
a b c
Fig. 4. Motion between prescribed rest positions of the plate. a Initial and final rest posi-
tions. b Time trajectories of the interpolated plate position r̆d1 (t) = rd1 (t) − rd1 (t0 ), plate Car-
dan angles ϕ̆ ϕ (t) = ϕ d (t) − ϕ d (t0 ), and plate torsion η d (t). c Time trajectories of the bridge
coordinates ŭi (t) = ui (t) − ui (t0 ), trolley coordinates ŭi j (t) = ui j (t) − ui j (t0 ), and chain lengths
w̆i j (t) = wi j (t) − wi j (t0 ).
rest position at t0 = 0 s into the final rest position at tN = 10 s with the pose coordinates
rd1 (t0 ) = [ 2.0 3.0 3.0 ]T m , ϕ d (t0 ) = [ 0 0 0 ]T deg , η d (t0 ) = 0.02 m , (36)
T T
rd1 (tN ) = [ 1.6 4.5 4.0 ] m , ϕ (tN ) = [ 5 − 3 6 ] deg ,
d
η (tN ) = 0 m .
d
(37)
The actuator coordinates belonging to the initial plate position are calculated by the
nonlinear optimisation procedure with minimised chain lengths described in [15] that
leads here to almost vertical chains.
The time trajectories of the seven position variables of the plate in Fig. 4b are in-
terpolated by functions being steady up to the fourth-order time derivatives. The time
trajectories of the ten actuator coordinates in Fig. 4c are obtained as described in sec-
tion 4 by calculating the minimum norm solution of the actuator coordinates in (34).
For feedforward control simulation the trajectories of the actuator coordinates from
Fig. 4c were applied as rheonomic constraints to a Simpack multibody model of the
crane manipulator. The time history of point O1 in the horizontal x,y-plane is shown
in Fig. 5. The comparison with the results from interpolations with simple trapezoidal
velocity profiles in Fig. 5a shows the effect of the feedforward control. There are only
small differences between the nonlinear and linear inverse dynamic models that are
seen in detail in Fig. 5b. The residual sway motions can be traced back to the stiff
plate representation used in the Simpack forward dynamics model. Thus the linearised
inverse dynamics model well approximates the nonlinear one and can be used for further
investigations.
plate position r1x(t) residual sway motions r1x(t)

2
1.6002
1.9 1.6
trapezoidal velocity profile
nonlinear
[m]
1.8
[m]
1.5998 feedforward
linear control
nonlinear/linear feedforward 1.5996
1.7 feedforward
control
control
1.6 1.5994
0 5 10 15 20 10 15 20
plate position r1y(t) residual sway motions r1y(t)
4.5
4.5004 nonlinear
feedforward
nonlinear/linear feedforward 4.5002
4 control
control
[m]
4.5
[m]
4.4998
3.5 trapezoidal velocity profile linear
4.4996
feedforward
4.4994 control
3
0 5 10 15 20 10 15 20
a time [s] b time [s]
Fig. 5. Feedforward control trajectories applied to a Simpack forward dynamic model of the crane
manipulator. a Time trajectories r1x (t), r1y (t) of point O1 between the prescribed initial and final
rest positions from (36), (37) with interpolated actuator coordinates by linear feedforward control
according to Fig. 4c, nonlinear feedforward control calculated from (12) according to [15] (not
distinguishible from linear feedforward results in this representation) and, for comparison, with
trapezoidal velocity profiles. b Zooms on residual sway motions of r1x (t), r1y (t) under linear and
nonlinear feedforward control.
6 Conclusion and Outlook
The nonlinear equations of motion of a CDPM suspending a flexible workpiece by four

load chains are analytically linearised around a static equilibrium state of the system.
From the linearised equations of motion a flatness-based linear feedforward control
scheme is derived that calculates the actuator coordinates for given trajectories of the
payload between rest positions. The linearised model is parametrised by the positions
of the actuator coordinates and the equilibrium position of the workpiece that are to be
measured during operation of the system. A numerical simulation of the crane system
shows negligeable differences between nonlinear and linearised feedforward control.
In a next step simplified transfer behaviours of the controlled axes will be incor-
porated into the control scheme. Model incertainties and external disturbances in the
industrial environment will have to be compensated by an additional feedback con-
troller that actively dampens sway motions of the payload. For this purpose a control
design model will be used that is reduced to the three dominant coordinates of the sway
motion Δzsway = [ Δrx Δry Δψz ]T that have to be estimated from measurements [16].
The control design model is obtained by static (Guyan) condensation of the linearised
equations of motion (32). A prototype industrial crane manipulator is being prepared
for experimental validations.
Acknowledgement This research and development project was supported by the Eu-
ropean Regional Development Fund (EFRE). Support was also provided by the lead
partner Technologie-Beratungsinstitut (TBI) according to the directive for support, de-
velopment and innovation of the Ministry of Economics, Construction and Tourism of
Mecklenburg-Vorpommern.
References
1. Abbasnejad, G., Carricato, M.: Direct geometrico-static problem of underconstrained cable-
driven parallel robots with n cables. IEEE Transactions on Robotics 31, 468–478 (2015)
2. Berti, A., Merlet, J.P., Carricato, M.: Solving the direct geometrico-static problem of under-
constrained cable-driven parallel robots by interval analysis. The International Journal of
Robotics Research 35, 723–739 (2016)
3. Bhaskar, K., Varadan, T.: Plates. John Wiley Sons (2014)
4. Blajer, W., Kolodziejczyk, K.: Improved DAE formulation for inverse dynamics simulation
of cranes. Multibody System Dynamics 25, 131–143 (2011)
5. Carricato, M., Abbasnejad, G., Walter, D.: Inverse geometrico-static analysis of under-
constrained cable-driven parallel robots with four cables. In: J. Lenarc̆ic̆, M. Husty (eds.)
Latest Advances in Robot Kinematics, pp. 365–372. Springer, Dordrecht (2012)
6. Carricato, M., Merlet, J.P.: Geometrico-static analysis of under-constrained cable-driven par-
allel robots. In: J. Lenarc̆ic̆, M. Stanis̆ić (eds.) Advances in Robot Kinematics, pp. 309–319.
Springer, Berlin (2010)
7. Fliess, M., Lévine, J., Martin, P., Rouchon, P.: Flatness and defect of nonlinear systems:
Introductory theory and examples. International Journal of Control 51, 1327–1361 (1995)
8. Heyden, T., Woernle, C.: Dynamics and flatness-based control of a kinematically undeter-
mined cable suspension manipulator. Multibody System Dynamics 16, 155–177 (2006)
9. Husty, M., Schadlbauer, J., Zsombor-Murray, P.: A new approach to the direct geometrico-
static problem of cable suspended robots using kinematic mapping. In: C. Gosselin, P. Car-
dou, T. Bruckmann, A. Pott (eds.) Cable-Driven Parallel Robots, pp. 97–105. Springer, Cham
(2018)
10. Hwang, S.W., Bak, J.H., Yoon, J., Park, J.H., Park, J.O.: Trajectory generation to suppress
oscillations in under-constrained cable-driven parallel robots. Journal of Mechanical Science
and Technology 30, 5689–5697 (2016)
11. Knierim, K.L., Krieger, K., Sawodny, O.: Flatness based control of a 3-dof overhead crane
with velocity controlled drives. IFAC Proceedings Volumes 43, 363 – 368 (2010)
12. Ming, A., Higuchi, T.: Study on multiple degree-of-freedom positioning mechanisms using
wires (parts 1 and 2). Int. Journal of the Japanese Society for Precision Engineering 28,
131–138 and 235–242 (1994)
13. Pott, A.: Cable-Driven Parallel Robots - Theory and Application. Springer, Berlin (2018)
14. Sawodny, O., Aschemann, H., Lahres, S.: An automated gantry crane as a large workspace
robot. Control Engineering Practice 10, 1323 – 1338 (2002)
15. Stoltmann, M., Froitzheim, P., Fuchs, N., Woernle, C.: Flatness-based feedforward control
of a crane manipulator with four load chains. In: B. Corves, P. Wenger, M. Hüsing (eds.)
EuCoMes 2018, pp. 61–68. Springer, Berlin (2019)
16. Woernle, C.: Trajectory tracking for a three-cable suspension manipulator. In: T. Bruckmann,
A. Pott (eds.) Cable-Driven Parallel Robots, pp. 371–386. Springer, Berlin (2013)
17. Woernle, C.: Mehrkörpersysteme. Springer, Berlin (2016)
18. Yang, Y., Betsch, P.: Computer simulation of the inverse dynamics of underactuated mechan-
ical systems. In: Proc. ECCOMAS Thematic Conference on Multibody Dynamics, Prague
(2017)
$QH[SHULPHQWDOVWXG\RQFRQWURODFFXUDF\RI)$67FDEOHURERW
澥
IROORZLQJ]LJ]DJDVWURQRPLFDOWUDMHFWRU\
+XL/,0LQJ]KH/,
1DWLRQDO$VWURQRPLFDO2EVHUYDWRULHV&KLQHVH$FDGHP\RI6FLHQFHV%HLMLQJ
&$6.H\/DERUDWRU\RI)$671$2&&KLQHVH$FDGHP\RI6FLHQFHV%HLMLQJ
$%675$&7
7KHKXJHFDEOHURERWRI)$67LVXVHGLQIROORZLQJDVWURQRPLFDOWUDMHFWRULHVFRQWLQXRXVO\DQGWUDFNLQJ
VWDULQWKHVN\7KHFDEOHURERWLVDFRPSOLFDWHFRXSOHGVHULHVFRQWUROV\VWHP,WVFRQWURODUFKLWHFWXUHLV
LQWURGXFHGLQFOXGLQJDIRUPXODWLRQRIDVWURQRPLFDOWUDMHFWRU\DVZHOO6RPHFRQWUROWHVWVDUHGRQHDQG
FRPSDUHGDVWKHURERWIROORZVDVSHFLDO]LJ]DJ WUDMHFWRU\$FFRUGLQJWRWKHSUREOHPVRIFRQWUROWHVWV
VRPHWHQWDWLYHLPSURYHPHQWVDUHWULHG)LQDOO\DVKRUWVXPPDU\LVJLYHQRQFRQWUROSHUIRUPDQFHRIWKH
FDEOHURERWDQGIXUWKHUZRUNLQWKHIXWXUH
,QWURGXFWLRQ
7KH)LYHKXQGUHGPHWHU$SHUWXUH6SKHULFDOUDGLR7HOHVFRSH )$67 KDVWKHODUJHVWFDEOHGULYHQSDUDOOHO

URERWDOORYHUWKHZRUOGVSDQQLQJPZKLFKGLVWLQJXLVKHVLWREYLRXVO\IURPPDQ\RWKHUWHOHVFRSHV
>a@
,WLVGULYHQE\VL[VXVSHQVLRQFDEOHVDQGGHVLJQHGWRWUDFNSUHGHILQHGWUDMHFWRU\DQGDORQJSODQQHG
RULHQWDWLRQLQXVXDODVWURQRPLFDOREVHUYDWLRQDVZHOO7KHGHVLUHGXSSHUOLPLWVRIFRQWUROHUURUDUHPP
LQURRWPHDQVTXDUH UPV IRUVSDWLDOSRVLWLRQDQGGHJUHHIRURULHQWDWLRQUHVSHFWLYHO\6RDIHZSRVH
FRPSHQVDWRUVIRUILQHWXQLQJKDYHWREHLQVWDOOHGLQWKHHQGHIIHFWRUWKHIHHGFDELQLQFOXGLQJDD[LV
URWDWRUDQGDKH[DSRG>@7KHZKROHFRQWURORIWKHFDEOHURERWDUHPDGHXSRIDFRXSOHGVHULHVV\VWHP
7KH\KDYHLQWHUORFNLQJUHODWLRQV>a@
7KH FRQVWUXFWLRQ RI WKH FDEOH URERW ZDV FRPSOHWHG LQ 6HSWHPEHU RI GHFODULQJ WKH SHULRG RI LWV
FRPPLVVLRQLQJVWDJH6LQFHWKHQDORWRIH[SHULPHQWVKDGEHHQGRQHWRWHVWLI\LWVIXQFWLRQDQGFRQWURO
SHUIRUPDQFH>@,QFRPSOHWHO\FRQVWUDLQHGIOH[LEOHDQGORZGDPSLQJ>@WKHKXJHFDEOHURERWKDVDOZD\V
D PDLQ IRFXV RQ SRVLWLRQLQJ DFFXUDF\ %HVLGHV ZLQG PLVFHOODQHRXV WUDMHFWRULHV DQG UHODWHG PRWLRQ
SODQQLQJDOVRKDYHDJUHDWDIIHFWLRQRQLWVFRQWURODFFXUDF\$IWHU\HDUV¶WHVWVQRZLWFDQJUDFHIXOO\
GULIWVLQWKHDLUDORQJDVLPSOHDUFVKDSHGWUDMHFWRU\ZLWKVPDOOYLEUDWLRQ7KHUHDUHKRZHYHUVWLOODIHZ
FKDOOHQJHVLQWKHFRQWUROSHUIRUPDQFHDORQJPDQ\LUUHJXODUWUDMHFWRULHV,QWKHQH[WFKDSWHUVRIWKHSDSHU
WKHDXWKRUVZRXOGOLNHWRLQWURGXFHDVSHFLDOW\SHRI]LJ]DJDVWURQRPLFDOWUDMHFWRU\VRFDOOHGRQWKHIO\
PDSSLQJ 27) VFDQQLQJ)XUWKHUPRUHLWVUHODWHGWUDFNLQJHUURUVDUHVWXGLHGH[SHULPHQWDOO\$IHZVWHSV
DUHDOVRWULHGWRLPSURYHWKHDFFXUDF\%HIRUHWKDWLWLVQHFHVVDU\WRLQWURGXFHWKHFRXSOHGVHULHVFRQWURO
V\VWHPRIWKHURERW
&RXSOHGVHULHVFRQWUROV\VWHP
7KH FRQWURO RIWKHFDEOH URERW DUH GLYLGHG LQWR DSUHOLPLQDU\ SRVHFRQWURO FDEOHV IXUWKHU URWDWLRQ
FRPSHQVDWLRQ URWDWRU DQGWKHVHFRQGDU\ILQHWXQLQJRISRVH KH[DSRG DVVKRZQLQ)LJ7KHFDEOHV
GULYHWKHFDELQFRRUGLQDWHO\WRIROORZSODQQHGWUDMHFWRULHVLQDODUJHPRYHPHQWUDQJH'XULQJWKDWSURFHVV
FDELQRULHQWDWLRQPD\YDU\WRDGDSWLWVHOIWRDUHODWLYHO\XQLIRUPO\GLVWULEXWHGWHQVLRQVDPRQJWKHFDEOHV
7KHSHUIRUPDQFHLVHQVXUHGE\DVHULHVRIVHQVRUVWKDWVHQGIHHGEDFNVDQGDOD\HUHGQHWZRUNFRQWURO
澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳澳
(PDLORIWKHFRUUHVSRQGLQJDXWKRUOLKXL#QDRFDVFQ

246 H. Li and M. Li
V\VWHP>@7KHURWDWRUFRQQHFWVWKHFDELQWUXVVWKHPDLQVXSSRUWLQJVWUXFWXUHZLWKDKH[DSRGYLDULQJV
ZKRVHURWDWLRQD[HVDUHKRUL]RQWDOO\RUWKRJRQDO5LQJ$LVFRQQHFWHGZLWKWKHFDELQWUXVVDQGVXSSRUWV
5LQJ % ZKLFK LV DOVR WKH KH[DSRG EDVH$V WKH ILQDO HQGHIIHFWRU RI WKH URERW WKH UHFHLYHU DQG LWV
VXSSRUWLQJ SODWIRUP ZHLJKLQJ DERXW WRQV DUH GLUHFWO\ GULYHQ DQG FRQWUROOHG E\ WKH OHJV RI WKH
KH[DSRGIRUILQHSRVLWLRQLQJDQGSRLQWLQJLQWKHDFFXUDF\DVDERYHGHVLUHG

)LJXUH 3RVLWLRQLQJ SRLQWLQJPHFKDQLVPVRIWKH)$67FDEOHURERW
7KHWRWDOFRQWURORIWKHFDEOHURERWFDQEHGLYLGHGLQWROD\HUVDVVKRZQLQ)LJ7KHORZHUPRVWOD\HUV
DOODUHFORVHORRSFRQWUROLQFOXGLQJPRWRUVDQGFDSVWDQV,QWKHXSSHUOHYHODOORIWKHFRQWUROORRSV
PXVW UHFHLYH IHHGIRUZDUG LQIRUPDWLRQ SODQQHG WUDMHFWRU\ DQG RULHQWDWLRQ %XW WKHUH DUH GLIIHUHQW
DSSOLFDWLRQVDPRQJWKHP6LQFHWKHWUDFNLQJVSHHGRIWKHFDELQLVDVVORZDVPPVRUOHVVWKHFDELQ
RULHQWDWLRQ FKDQJHV VORZO\ DV ZHOO +HQFH WKH URWDWRU LV GHVLJQHG WR EH D TXDVLVWDWLF RSHQORRS
FRPSHQVDWRUIRUDQDQJXODUGLIIHUHQFHEHWZHHQWKHGHVLUHGSRLQWLQJDQGQDWXUDOWLOWRIWKHFDELQWUXVV
6XFKDQJXODUGLIIHUHQFHFDQEHFDOFXODWHGWKHRUHWLFDOO\GHFLGHGRQO\E\WKHFDELQSRVLWLRQ>a@0RVWRI
RULHQWDWLRQHUURUVFDQEHFRPSHQVDWHGLQWKLVZD\>@2QWKHRWKHUKDQGWKHFDEOHFRQWURODQGWKHKH[DSRG
FRQWURO KDYH WR EH FORVH ORRS VHW ZLWK SURSRUWLRQDO DQG LQWHJUDO 3, SDUDPHWHUV WR UHVLVW GLVWXUEDQFHV
LQGXFHGE\ZLQGRUXQHYHQO\LQWHUDFWLYHFDELQFDEOHG\QDPLFV,WLVZRUWKQRWLQJWKDWWKHUHDUHDWOHDVW
WDUJHWV RI ODVHU PHDVXUHPHQW aPP UPV LQVWDOOHG RQ WKH FDELQ WUXVV DQG WKH UHFHLYHU SODWIRUP
UHVSHFWLYHO\VRWKDWSRVHIHHGEDFNFDQEHFDOFXODWHGIRUWKHFDELQDQGWKHUHFHLYHU$VDUHVXOWRIHUURU
EXGJHWLWLVDQWLFLSDWHGWKDWWKHFDELQFDQEHORFDWHGLQVSDWLDODFFXUDF\RIDERXWPPUPVDQGPD[LPDO
URWDWLRQDOHUURURIGHJUHH FRPSDUHGZLWKQDWXUDOWLOW YLDWKHSUHOLPLQDU\FDEOHFRQWURO7KHUHVLGXDO
HUURUVDUHFRPSHQVDWHGE\WKHKH[DSRGZKLFKLVGHVLJQHGWRKDYHDVSKHULFDOWUDQVODWLRQDOZRUNVSDFHRI
PPLQGLDPHWHUDQGDURWDWLRQDOZRUNVSDFHRIGHJUHHV

)LJXUH &RQWUROORRSVRI)$67FDEOHURERW
,WLVKRZHYHUQRWDOZD\VWKHFDVHWKDWELJJHUZRUNVSDFHEULQJVEHWWHUILQHWXQLQJSHUIRUPDQFH)OH[LEOH
An experimental study on control accuracy of FAST cable robot… 247
VXVSHQVLRQ FDEOHV DQG ORZ GDPSLQJ RI WKH G\QDPLF V\VWHP PD\ OLPLW ERWK WXQLQJ IUHTXHQF\ DQG
DGMXVWPHQW UDQJH RI WKH KH[DSRG>@ ,Q RWKHU ZRUGV WKH VXVSHQGHG KH[DSRG PD\ RQO\ EH DEOH WR
FRPSHQVDWHHIIHFWLYHO\WKHUHODWLYHO\VORZDQGVPDOOUHVLGXDOHUURUVOHIWE\FDEOHFRQWURORULQGXFHGE\
FDELQFDEOHLQWHUDFWLYHYLEUDWLRQV,QWKHFDVHRIQRQGLIIHUHQWLDOWUDMHFWRU\RUYHU\VPDOOFXUYDWXUHUDGLXV
YLEUDWLRQV RI WKH FDELQ DQG FDEOHV PD\ VLJQLILFDQWO\ UHGXFH WKH FRQWURO DFFXUDF\7KHUHIRUH LW LV YHU\
LPSRUWDQWLQGHYHORSLQJDVWDEOHFRQWURODOJRULWKPRIWKHKH[DSRG
$VWURQRPLFDOWUDMHFWRU\
,I)$67ZDQWVWRREVHUYHDVWDUWKDWLVVHQGLQJUDGLRZDYHVWRWKHHDUWKLWKDVWREHDOLJQHGWRWKHVWDU
ZLWKWKHUHFHLYHULQWKHFDELQWRIROORZLWVWUDMHFWRU\IRUDIHZKRXUVLQWKHVN\7KHVWDU¶VSRVLWLRQFDQEH
GHVFULEHGPDWKHPDWLFDOO\YLD&DUWHVLDQFRRUGLQDWHDVIROORZV>@
‫ ݔ‬ൌ ሺͳ െ ݂ሻ ή ܴ ή ߜ ή ‫ܪ‬
ቐ‫ ݕ‬ൌ ሺͳ െ ݂ሻ ή ܴ ή ሺ߮ ή ߜ ή ‫ ܪ‬െ ߮ ή ߜሻ
‫ ݖ‬ൌ ሺͳ െ ݂ሻ ή ܴ ή ሺ ߮ ή ߜ ή ‫ ܪ‬൅ ߮ ή ߜሻ
+HUHWKHRULJLQRIWKH&DUWHVLDQIUDPHLVVHWRQWKHFHQWHURIVSKHULFDOIRFDOVXUIDFHLQ)LJZLWKWKH
KRUL]RQWDO[D[LVSRLQWLQJWRWKHHDVWDQG\D[LVWRWKHQRUWK7KHV\PEROVfRDQGĳDUHWKHFRQVWDQWVWKDW
UHSUHVHQWUHVSHFWLYHO\WKHIRFDOUDWLRWKHVSKHULFDOUDGLXVDQGWKHORFDOJHRJUDSKLFODWLWXGHRIWKH)$67
WHOHVFRSH7KHV\PEROVHDQGįDUHDSDLURILQGHSHQGHQWFRRUGLQDWHVLQWKHFHOHVWLDOIUDPHUHSUHVHQWLQJ
DVFHQVLRQ DVWURQRPLFDO ORQJLWXGH DOVR FDOOHG KRXU DQJOH DQG GHFOLQDWLRQ DVWURQRPLFDO ODWLWXGH
UHVSHFWLYHO\%DVLFDOO\GLIIHUHQWUHODWLRQEHWZHHQHDQGį SURGXFHVGLIIHUHQWWUDFNLQJPRGHV,QWKHPRGH
RIVWDUWUDFNLQJWKH\FDQEHVLPSO\H[SUHVVHGDV>@
‫ܪ‬ሺ‫ݐ‬ሻ ൌ ‫ܪ‬଴ ൅ ߱଴ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ

൜
ߜሺ‫ݐ‬ሻ ൌ ߜ଴
+HUHWKHV\PERO ߱଴ UHSUHVHQWVWKHVHOIURWDWLRQRIWKHHDUWKDFRQVWDQWRI(UDGVHDQGį
DUHWKHVWDUWLQJKRXUDQJOHDQGGHFOLQDWLRQ6RWKHWUDMHFWRU\LVDKRUL]RQWDOVWUDLJKWOLQHLQWKHFHOHVWLDO
IUDPHDVVKRZQLQWKHWRSOHIWRI)LJ,WFKDQJHVWRDUHJXODUDUFLQ&DUWHVLDQFRRUGLQDWHDFFRUGLQJWR
(T ,QWKHFDVHRI27)PRGHWUDMHFWRU\LVKRZHYHUDUHJXODUEDFNDQGIRUWKVFDQQLQJSDWKLQWKH
FHOHVWLDO IUDPH VXUURXQGLQJ D VPDOO DUHD QHDU WKH REMHFW VWDU DV VKRZQ LQ WKH ERWWRP OHIW RI )LJ
6FDQQLQJSDWKVZLWFKHVDOWHUQDWHO\EHWZHHQYHUWLFDODQGKRUL]RQWDOVWUDLJKWOLQHVZLWKWKHVDPHLQWHUYDO
į RU ș$ORQJWKHkWKYHUWLFDOVFDQQLQJWKHHTXDWLRQVRIHDQGį DUHH[SUHVVHGDV>@
‫ܪ‬ሺ‫ݐ‬ሻ ൌ ‫ܪ‬௞ ൅ ߱଴ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ ఋ ିఋ

൜ ǡ ሺ‫ ݐ‬൏ ‫ݐ‬଴ ൅ ౣ౗౮ ౣ౟౤ሻ
ߜሺ‫ݐ‬ሻ ൌ ߜ୫୧୬ ൅ ߱ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ ఠ
RU
‫ܪ‬ሺ‫ݐ‬ሻ ൌ ‫ܪ‬௞ ൅ ߱଴ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ ఋ ିఋ

൜ ǡ ሺ‫ ݐ‬൏ ‫ݐ‬଴ ൅ ౣ౗౮ ౣ౟౤ሻ
ߜሺ‫ݐ‬ሻ ൌ ߜ୫ୟ୶ െ ߱ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ ఠ
+HUHWKHV\PERO ߱ LVWKHSODQQHGFRQVWDQWDQJXODUVSHHGtWKHVWDUWLQJWLPHRIHDFKVLQJOHVFDQQLQJ
6LPLODUO\DORQJWKHkWKYHUWLFDOVFDQQLQJWKHHTXDWLRQVRIHDQGį DUHH[SUHVVHGDV>@
‫ܪ‬ሺ‫ݐ‬ሻ ൌ ‫ܪ‬௞ ൅ ʹ ή ߱଴ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ ఏ

൜ ǡ ሺ‫ ݐ‬൏ ‫ݐ‬଴ ൅ ሻ
ߜሺ‫ݐ‬ሻ ൌ ߜ୫ୟ୶ ଶήఠబ
RU
248 H. Li and M. Li
‫ܪ‬ሺ‫ݐ‬ሻ ൌ ‫ܪ‬௞ ൅ ʹ ή ߱଴ ή ሺ‫ ݐ‬െ ‫ݐ‬଴ ሻ ఏ

൜ ǡ ሺ‫ ݐ‬൏ ‫ݐ‬଴ ൅ ሻ
ߜሺ‫ݐ‬ሻ ൌ ߜ୫୧୬ ଶήఠబ
+HUHWKHV\PEROșLV WKH KRUL]RQWDOLQWHUYDORIH7KHDERYHHTXDWLRQVGHVFULEHWKHFDVHRIVFDQQLQJ

YHUWLFDOO\$QRWKHUFDVHRIVFDQQLQJKRUL]RQWDOO\FDQDOVREHGHVFULEHGVLPLODUO\+HUHYHUWLFDOO\VFDQQLQJ
LVWKHPDLQIRFXV6XEVWLWXWLQJWKHUHODWLRQLQWR(T SURGXFHVD]LJ]DJW\SHRIWUDMHFWRU\DVVKRZQLQ
WKHULJKW)LJ%\WKHZD\ERWKW\SHVRIWUDMHFWRULHVDUHMXVWORFDWHGLQWKHVSKHULFDOIRFDOVXUIDFHRIWKH
WHOHVFRSHZLWKLWVFXUYDWXUHUDGLXVHTXDOWR f R,QHDFKREVHUYDWLRQPRGHRULHQWDWLRQLVDOVRSODQQHG
VRWKDWWKHUHFHLYHUKRUQVKRXOGNHHSSRLQWLQJWRWKHVSKHULFDOFHQWHURIWKHIRFDOVXUIDFH*HQHUDOO\LWLV
TXLWHKDUGWRPDNHFDELQWLOWWRGLUHFWO\DSSURDFKWKDWSRLQWLQJRQO\YLDWKHSUHOLPLQDU\FDEOHFRQWURO6R
WKHURWDWRULVXVHGWRFRPSHQVDWHPRVWRIWKHGLIIHUHQFHLQDTXDVLVWDWLFPRGH
)LJXUH 7UDMHFWRULHVLQWUDFNLQJ UHG DQG27)VFDQQLQJ EODFN
'LIIHUHQWLDWLQJ(T ZLWKUHJDUGWRWLPHtWKHSODQQHGVSHHGVFDQEHH[SUHVVHGLQ&DUWHVLDQ
FRRUGLQDWHDV
‫ݔ‬ሶ ൌ ሺͳ െ ݂ሻ ή ܴ ή ൫ ߜ ή ‫ ܪ‬ή ‫ܪ‬ሶ െ ߜ ή ‫ ܪ‬ή ߜሶ ൯
൞‫ݕ‬ሶ ൌ ሺͳ െ ݂ሻ ή ܴ ή ൫߮ ή ൫െߜ ή ‫ ܪ‬ή ߜሶ െ ߜ ή ‫ ܪ‬ή ‫ܪ‬ሶ൯ െ ߮ ή ߜ ή ߜሶ ൯
‫ݖ‬ሶ ൌ ሺͳ െ ݂ሻ ή ܴ ή ൫ ߮ ή ൫െߜ ή ‫ ܪ‬ή ߜሶ െ ߜ ή ‫ ܪ‬ή ‫ܪ‬ሶ൯ ൅ ߮ ή ߜ ή ߜሶ ൯
+HUH WKH DQJXODU VSHHGV ‫ܪ‬ሶ DQG Ɂሶ DUH REWDLQHG DFFRUGLQJ WR (T RU (T ,Q WKH PRGH RI 27)
VFDQQLQJWKH\DUHSODQQHGDV>@
‫ܪ‬ሶ ൌ ߱଴ Ǣߜሶ ൌ ሺെͳሻ௞ ߱Ǣሺ݇ ሻ

ቊ
‫ܪ‬ሶ ൌ ʹ߱଴ Ǣߜሶ ൌ ͲǢሺሻ
+HUHWKHV\PERO k LV DQHYHQ QXPEHU LI į LV LQFUHDVLQJ RU RWKHUZLVH RGG QXPEHU 6LPLODUO\ D IXUWKHU
GLIIHUHQWLDWLRQRI(T ZLWKUHJDUGWRWLPHtFDQJLYHWKHVFDQQLQJDFFHOHUDWLRQV
&RQWUROWHVWVDQGDQDO\VLV
6HYHUDOWHVWVDUHWULHGRQVLWHWRWHVWLI\FRQWURODFFXUDF\RIWKHFDEOHURERWDORQJNLQGVRIWUDMHFWRULHV
RQ6HSWHPEHU$VVKRZQLQ)LJWKHFDEOHURERWIROORZHGDWUDFNLQJWUDMHFWRU\ UHGDUF IRU
PLQXWHV7KHFDELQVSHHGLVVHWDVFRQVWDQWPPV7KHQWKHURERWIROORZHGD]LJ]DJ27)WUDMHFWRU\
EODFN IRUPLQXWHVZLWKHDFKVLQJOHVWUDLJKWOLQHODVWLQJIRUPLQXWHVDQGVFDQQLQJVSHHGDSSURDFKLQJ
FRQVWDQWPPV7KHFRRUGLQDWHVRIWDUJHWVZKLFKDUHLQVWDOOHGRQWKHFDELQWUXVVDQGWKHUHFHLYHU
SODWIRUPUHVSHFWLYHO\DUHUHFRUGHGLQUHDOWLPHDQGFKDQJHGLQWRSRVHV&RQWUROHUURUVDUHFDOFXODWHGDIWHU
FRPSDULVRQZLWKWKHSODQQHGSRVH,WLVZRUWKQRWLQJWKDWWKHTXDVLVWDWLFRULHQWDWLRQDOHUURUVOHIWE\WKH
URWDWRUDUHDOVRDGGHGXSWRWKRVHE\FDEOHFRQWURO
)LJXUH 7UDMHFWRULHVDQGVSHHGSODQQLQJLQFRQWUROWHVWV
$VVKRZQLQ)LJDJRRGFRQWURODFFXUDF\LVREWDLQHGDORQJWKHWUDFNLQJWUDMHFWRU\,WVKRZVWKDWWKH
WRWDOFRQWUROHUURUVOHIWE\WKHKH[DSRGDUHZHOOORZHUWKDQWKHGHVLUHGPPDQGGHJUHHVGXULQJWKH
PRVWRIWUDFNLQJWLPH7KHIDFWVLQGLFDWHVWKDWLWLVDFFHSWDEOHIRUWKHIOH[LEOHFDEOHURERWWRIROORZDUF
VKDSHG WUDFNLQJ WUDMHFWRU\ %HVLGHVLW LV ZRUWK QRWLQJ WKDW GXULQJWKHWHVWDEQRUPDOODUJH ZLQG VSHHG
DYHUDJHaPVPD[LPDOPV ZDVUHFRUGHGDWDSSURDFKRIDVXSHU W\SKRRQ 0DQJNKXW 7LPHHUURU
FXUYH DQG IUHTXHQF\HUURU FXUYH DUH JLYHQ IURP WKH OHIW WR WKH ULJKW UHVSHFWLYHO\ ,Q WKH FDVH RI
WUDQVODWLRQDOHUURUVLWUHYHDOVWKDWWKHHUURUVOHIWE\FDEOHFRQWURODUHPDGHXSRIWZRPDLQSDUWV2QHLV
D VWDWLF RIIVHW HUURU LQGXFHG E\ VWHDG\ VWDWH ZLQG7KH RWKHU LV YLEUDWLRQV RI WKH FDELQFDEOH G\QDPLF
V\VWHPDURXQGLWVQDWXUDOIUHTXHQF\ a +] 7KHIRUPHULVFRPSHQVDWHGZHOOE\WKHKH[DSRG
EXW ERWK DUH DOPRVW WKH VDPH IRU WKH ODWWHU7KH IDFW LQGLFDWHV WKDW WKH KH[DSRG ZRUNV ZHOO RQO\ LQ D
IUHTXHQF\UDQJH +] PXFKORZHUWKDQWKHILUVWQDWXUDOIUHTXHQF\7KHSKHQRPHQRQLVLQDJRRG
DFFRUGDQFHZLWKSUHYLRXVVLPXODWLRQZRUN>@,WDOVRH[SODLQVZK\EHWWHUFRQWURODFFXUDF\LVREWDLQHGLQ
WKHFDVHRIRULHQWDWLRQDOHUURUV

D 7UDQVODWLRQDOHUURUV
澳
250 H. Li and M. Li

E 2ULHQWDWLRQDOHUURUV
)LJXUH &RQWUROHUURUVLQWUDFNLQJ OHIWWLPHGRPDLQULJKWIUHTXHQF\GRPDLQ

D 7UDQVODWLRQDOHUURUV

E 2ULHQWDWLRQDOHUURUV
)LJXUH &RQWUROHUURUVLQ27)VFDQQLQJ PLQXWHIRUDVLQJOHVFDQQLQJ
$IWHUVZLWFKLQJWRWKHFDVHRI27)WUDMHFWRU\LWLVKRZHYHUQRWVRPXFKRSWLPLVWLF$WHVWRQWKHWUDMHFWRU\
RI27)VFDQQLQJZDVGRQH$VVKRZQLQ)LJ D REYLRXVYLEUDWLRQVFDQEHREVHUYHGIRUWKHERWKWZR
FRQWUROOHYHOV7KHQXPEHURISHDNVDQGLQWHUYDOVEHWZHHQQHLJKERUSHDNV aV DUHMXVWFRUUHVSRQGLQJ
WR WKH VKDUS WXUQV RI WUDMHFWRU\ ,W WDNHV UHODWLYHO\ ORQJ WLPH IRU WKH KH[DSRG WR VXSSUHVV VXFK ODUJH
YLEUDWLRQVLQHDFKLQWHUYDO,WLVDOVRREVHUYHGLQWKHIUHTXHQF\GRPDLQWKDWEHVLGHVWKHQDWXUDOIUHTXHQF\
PDQ\VXEKDUPRQLFVKDSSHQDVZHOO(YHQLQYHU\ORZIUHTXHQF\ a+] WKHKH[DSRGFRQWUROFDQQRW
ZRUN DV ZHOO DV LW GRHV LQ WUDFNLQJ 1HYHUWKHOHVV RULHQWDWLRQDO HUURU LV PXFK EHWWHU FRPSDUHG ZLWK
WUDQVODWLRQHUURULQWKDWWKHPRVWSDUWRIHUURUVOHIWE\FDEOHFRQWUROLVVWLOOVWHDG\VWDWHRULQYHU\ORZ
IUHTXHQF\UDQJH6RLQWKHFDVHRI27)VFDQQLQJWKHPDLQIRFXVLVKRZWRLPSURYHWUDQVODWLRQDODFFXUDF\
6KDUSWXUQDUHWKHPDLQFDXVHRIVXFKODUJHYLEUDWLRQVZKLFKPD\EHSURSRUWLRQDOWRWKHFDELQVSHHG
*RRGDFFXUDF\PD\EHREWDLQHGLIVSHHGSODQQLQJLVFKDQJHGVRWKDWWKHFDELQVSHHGFDQEHGHFHOHUDWHG
澳
QDWXUDOO\WRDVLWLVFORVHWRVKDUSWXUQDQGYLFHYHUVD,WLVKRZHYHUQRWSRVVLEOHDWWKDWWLPHGXULQJWKH
WHVW EHFDXVH LW PHDQV WR FKDQJH WKH FRQWURO VRIWZDUH JUHDWO\ $V WKH VHFRQG FKRLFH D SRVVLEOH
LPSURYHPHQWPD\EHWRUHGXFHWKHVFDQQLQJVSHHGRUWRLQFUHDVHWKHVFDQQLQJWLPHRIDVLQJOHVWUDLJKW
OLQHIURPPLQXWHWRPLQXWHV7KHUHIRUHVORZVSHHGUHGXFHVFDELQRVFLOODWLRQVQHDUVKDUSWXUQVDQGLW
LQWXUQDFFRPPRGDWHVPRUHWLPHIRUWKHFRPSHQVDWLRQRIKH[DSRGFRQWURO6RWKHVDPH27)WUDMHFWRU\
ZDVWULHGDJDLQRQHGD\ODWHU7KHQDQHZFRQWUROUHVXOWVZHUHREWDLQHGLQ)LJ
&RPSDUHGZLWKWKHUHVXOWVLQ)LJLPSURYHPHQWLVREYLRXVLQWKHWLPHGRPDLQ3HDNVVHHPPRUHQDUURZ
DQGWDNHOHVVWLPHUDWLRRIDQLQWHUYDO7KHWRWDOHUURUVHHPWRKDYHDVOLJKWGHFUHDVHDVZHOO1HYHUWKHOHVV
LQWKHIUHTXHQF\GRPDLQLPSURYHPHQWORRNVOLWWOH([FHSWIRUWKHVWDWLFRIIVHWWKHKH[DSRGVHHPVWRQRW
ZRUN DW RWKHU KDUPRQLFV ,W VHHPV ORZHU VSHHG GRHV LPSURYH WKH DFFXUDF\ DQG LWV HIIHFW LV OLPLWHG
)XUWKHUPRUHWKHUHLVQRPRUHVSDFHRIWKLVLPSURYHPHQWEHFDXVHRIWKHQHHGWRNHHSWKHORZHVWVFDQQLQJ
VSHHG

)LJXUH 7UDQVODWLRQDOHUURUVLQ27)VFDQQLQJ PLQXWHVIRUDVLQJOHVFDQQLQJ
$QRWKHUSRVVLEOHLPSURYHPHQWPD\EHWRRSWLPL]HWKHUHODWHGSDUDPHWHUVRIKH[DSRGFRQWURO$FFRUGLQJ
WRWUDFNLQJUHVXOWVLQ)LJWKHKH[DSRGFRQWUROVKRXOGZRUNDWDZLGHUIUHTXHQF\UDQJHWKDQLWGRHVLQ
)LJDQG)LJ,QWKHSUHYLRXVWHVWRI27)VFDQQLQJLWHYHQGLGQRWFRPSHQVDWHYLEUDWLRQVRIDYHU\
ORZKDUPRQLFV2QHSRVVLEOHUHDVRQH[SODLQVWKDWWKHKH[DSRGFRPSHQVDWLRQPD\EHWRRODUJHWRVWRS
UHVRQDQFHRIWKHWRWDOFDELQFDEOHFRXSOHGG\QDPLFV\VWHP$QRWKHUUHDVRQDVVXPHVWKDWWKHKH[DSRG
FRPSHQVDWLRQPLJKWEHWRRVPDOOWRFRPSHQVDWHHQRXJKHUURUV1RPDWWHUZKLFKRQHLVWKHWUXWKWKHIDFW
LQGLFDWHVWKDWWKHFXUUHQWFRQWUROSDUDPHWHUVPD\VWLOOQHHGWREHRSWLPL]HG1HYHUWKHOHVVRSWLPL]DWLRQLV
ERULQJORQJWHUPZRUNDQGQHHGVDORWRIWHVWGDWD7KHDXWKRUVWULHGWHQWDWLYHWHVWVRQHGD\ODWHUZLWK
FRQWUDU\RSWLRQV2QHLVWRIXUWKHUUHGXFHFRPSHQVDWLRQRIWKHKH[DSRGE\FORVLQJWKH3,SDUDPHWHUVRI
WKHKH[DSRGFRQWURO7KHRWKHULVWRHQODUJHFRPSHQVDWLRQVSHHGRIWKHKH[DSRGE\%HFDXVHRI
VRPHRWKHUUHDVRQVWKHWHQWDWLYHWHVWVVHOHFWHGDQHZWUDMHFWRU\RI27)VFDQQLQJEXWNHSWWKHVDPH
VHWWLQJDVPLQXWHVIRUDVLQJOHVFDQQLQJODVWLQJIRUPLQXWHLQWRWDO7KHFRQWUROHUURUVDUHFRPSDUHG
DQGVKRZQLQ)LJ

252 H. Li and M. Li
D +H[DSRGFRQWUROZLWKRXW3,IHHGEDFN

E 1RUPDOKH[DSRGFRQWURO

F )DVWKH[DSRGFRQWURO
)LJXUH 7UDQVODWLRQDOHUURUVLQ27)VFDQQLQJ PLQXWHVIRUDVLQJOHVFDQQLQJ
,WLVFOHDULQ)LJ D WKDWWKHUHVXOWLVHYHQZRUVHIRUWKHFDVHRIQR3,IHHGEDFNSURYLQJWKDWWKH3,
SDUDPHWHUV PD\ EH TXLWH LPSRUWDQW RWKHU WKDQ LJQRUDQFH 2Q WKH RWKHU KDQG WKHUH LV LPSURYHPHQW RI
DFFXUDF\LQWKHFDVHRIIDVWKH[DSRGFRQWURO%XWWKHUHLVQRVLJQLILFDQWGLIIHUHQFHEHWZHHQ)LJ E DQG
F 0RUHLPSRUWDQWLVWKDWYLEUDWLRQVDURXQGVXEKDUPRQLFVWKDWDUHPXFKORZHUWKDQQDWXUDOIUHTXHQF\
DUHQRW\HWVXSSUHVVHGDVH[SHFWHG
6XPPDU\
7KHKXJHFDEOHURERWRIWKH)$67WHOHVFRSHLVDSSOLHGLQSRLQWLQJWUDFNLQJRUVFDQQLQJVWDUVLQWKHVN\
ZLWKWKHUHTXLUHGDFFXUDF\,WLVKRZHYHUDVLJQLILFDQWO\IOH[LEOHDQGORZGDPSLQJG\QDPLFV\VWHP7KH
IDFWEULQJVJUHDWGLIILFXOW\LQIROORZLQJLUUHJXODUWUDMHFWRU\ZLWKVKDUSWXUQVOLNH]LJ]DJ27)VFDQQLQJ
6HYHUDO FRPSDUDWLYH FRQWURO WHVWV DUH GRQH IRUDVVHVVPHQW DQG ODWHULPSURYHPHQW,QWRWDOWKHFRQWURO
DFFXUDF\RIWKHFDEOHURERWUHGXFHVJUHDWO\LQ27)VFDQQLQJWKDQLQWUDFNLQJHVSHFLDOO\QHDUHDFKVKDUS
WXUQ6RPHSDUDPHWHUVDUHWULHGWRLPSURYHWKHFRQWURODFFXUDF\VXFKDVUHGXFLQJVFDQQLQJVSHHGDQG
FKDQJLQJWKHFRPSHQVDWLRQVSHHGRIWKHKH[DSRG7KHIRUPHUKDVVRPHHIIHFWVEXWWKHLPSURYHPHQWLV
OLPLWHG7KHODWWHULVDFWXDOO\UHODWHGZLWKRSWLPL]DWLRQRIFRQWURODOJRULWKPDFRPSUHKHQVLYHDQGORQJ
WHUPZRUNXQGHUWKHVXSSRUWRIDORWRIH[SHULPHQWV6LPSOHDGMXVWPHQWRIWKHKH[DSRGSDUDPHWHUVVHHPV
WRKDYHOHVVHIIHFWWRLPSURYHWKHDFFXUDF\7KHUHVKRXOGEHVRPHSODQQHGZRUNWRDQDO\]HWKHFRXSOLQJ
G\QDPLFV DQG LQWHUDFWLRQ DPRQJ WKH KH[DSRG WKH FDELQ DQG FDEOHV LQ WKH IXWXUH %DVHG RQ WKRVH
2SWLPL]DWLRQRIFRQWURODOJRULWKPZLOOEHEXLOWRQDGHHSXQGHUVWDQGLQJRIG\QDPLFV%HVLGHVPRWLRQ
SODQQLQJQHDUVKDUSWXUQLVDOVRDUHVHDUFKIRFXV
$FNQRZOHGJPHQWV
7KLVZRUNLVVXSSRUWHGE\WKH1DWLRQDO1DWXUDO6FLHQFH)RXQGDWLRQ 116) XQGHU*UDQW1R

DQGWKH2SHQ3URMHFW3URJUDPRIWKH.H\/DERUDWRU\RI)$671$2&&KLQHVH$FDGHP\RI6FLHQFHV
5HIHUHQFH
5' 1DQ )LYH KXQGUHG PHWHU DSHUWXUH VSKHULFDO UDGLR WHOHVFRSH )$67 6FLHQFH LQ &KLQD 6HULHV * 3K\VLFV
0HFKDQLFV $VWURQRP\ SS
1DQ5/L'DQGHWDO7KH)LYH+XQGUHG0HWHU$SHUWXUH6SKHULFDO5DGLR7HOHVFRSH )$67 3URMHFW,QWHUQDWLRQDO
-RXUQDORI0RGHUQ3K\VLFV' SS
+-.lUFKHU+/L-6XQHWDO3URSRVHG'HVLJQ&RQFHSWVIRUWKH)RFXV&DELQ6XVSHQVLRQRIWKHP)$67
7HOHVFRSH3URFHHGLQJVRI63,(0DUVHOOLH)UDQFH-XQHSS
+-.lUFKHU)$67IRFXVFDELQVXVSHQVLRQVLPXODWLRQVWXG\,QWHUQDOWHFKQLFDOUHVHDUFKUHSRUWRI)$67
+/L-6XQDQG;=KDQJ([SHULPHQWDOVWXG\RQWKHGDPSLQJRI)$67FDELQVXVSHQVLRQV\VWHP3URFRI63,(
*URXQG%DVHGDQG$LUERUQH7HOHVFRSHV,9$PVWHUGDP1HWKHUODQGVSS
+/L-6XQHWDO3UHOLPLQDU\UXQQLQJDQGSHUIRUPDQFHWHVWRIWKHKXJHFDEOHURERWRI)$67WHOHVFRSH&DEOH
GULYHQ3DUDOOHO5RERWV0HFKDQLVPVDQG0DFKLQH6FLHQFH9ROSS
+/L;=KDQJ5<DRHWDO2SWLPDOIRUFHGLVWULEXWLRQEDVHGRQVODFNURSHPRGHOLQWKHLQFRPSOHWHO\FRQVWUDLQHG
FDEOHGULYHQSDUDOOHOPHFKDQLVPRI)$67WHOHVFRSH&DEOHGULYHQ3DUDOOHO5RERWV0HFKDQLVPVDQG0DFKLQH6FLHQFH
9RO
+/L$JLDQW6DJJLQJ&DEOH'ULYHQ3DUDOOHO5RERWRI)$677HOHVFRSH漣LWVWHQVLRQIHDVLEOHZRUNVSDFHRIRULHQWDWLRQ
DQGRULHQWDWLRQSODQQLQJ3URFRIWK:RUOG&RQJUHVVLQ0HFKDQLVPDQG0DFKLQH6FLHQFH7DLSHLSS

+/LDQG5<DR2SWLPDORULHQWDWLRQSODQQLQJDQGFRQWUROGHYLDWLRQHVWLPDWLRQRQ)$67FDEOHGULYHQSDUDOOHOURERW
$GYDQFHVLQ0HFKDQLFDO(QJLQHHULQJ SS
6'HQJ)-LQJ=/LDQJHWDW5HVHDUFKRQSRVHGLVWULEXWLRQDOJRULWKPRI)$67IHHGVXSSRUWV\VWHP LQ&KLQHVH
2SWLFVDQG3UHFLVLRQ(QJLQHHULQJ
'<X$OJRULWKPRQWUDMHFWRU\SODQQLQJRIWKHIHHGVXSSRUWRI)$67WHOHVFRSH 9 ,QWHUQDOWHFKQLFDOUHVHDUFK
UHSRUWRI)$67 )$67-6&.
Part V
Motion Planning
Path Planning of a Mobile Cable-Driven Parallel Robot
in a Constrained Environment
Tahir Rasheed1 , Philip Long2 , David Marquez-Gamez3, and Stéphane Caro4

1 École Centrale de Nantes, Laboratoire des Sciences du Numérique de Nantes, UMR
CNRS 6004, 1, rue de la Noë, 44321 Nantes, France,
[email protected],
2 RIVeR Lab, Department of electrical and computing engineering, Northeastern University,
USA [email protected]
3 IRT Jules Verne, Chemin du Chaffault, 44340, Bouguenais, France,
[email protected]
4 CNRS, Laboratoire des Sciences du Numérique de Nantes, UMR CNRS 6004, 1, rue de la
Noë, 44321 Nantes, France, [email protected]
Abstract. A Mobile Cable-Driven Parallel Robot (MCDPR) is a special type of

Reconfigurable Cable-Driven Parallel Robot (RCDPR) composed of a classical
Cable-Driven Parallel Robot (CDPR) mounted on multiple mobile bases. The
additional mobility of the mobile bases allows such systems to autonomously
modify their geometric architecture, and thus make them suitable for multiple
manipulative tasks in constrained environments. Moreover, these additional mo-
bilities make MCDPRs kinematically redundant. Therefore, the subject of this
paper is to introduce a two stage path planning algorithm for MCDPRs. The first
stage searches for a feasible and collision free path of mobile bases. The second
stage deals with generating an optimal path of the moving-platform to displace
it from an initial to a desired pose. The proposed algorithm is validated through
simulation on a three degree-of-freedom (DoF) point mass moving-platform dis-
placed by four cables with each cable carried by an independent mobile base.
Keywords: Mobile Cable-Driven Parallel Robot, Reconfigurability, Kinematic

Redundancy, Path Planning, Wrench Analysis
1 Introduction
Cable-Driven Parallel Robots (CDPRs) are a particular class of parallel robots whose
actuated limbs are cables, connecting the moving-platform with a fixed base frame. The
platform is operated by changing the cable lengths between the moving-platform and
the base. CDPRs are interesting for various applications, such as construction [1], high
speed tasks [2], operations over large workspace [3] and as haptic devices [4].
In spite of numerous applications of CDPRs, several challenges are still unresolved.
For example, CDPRs require free circulation of the cables without any interference
with the environment [19]. Furthermore, CDPRs have fixed cable layout, i.e., fixed exit
points and cable configuration, which must be chosen carefully in order to maximize
its workspace. Therefore, it is reasonable to alter robot’s geometric structure based on

258 T. Rasheed et al.
its environment and the required task. CDPRs with the ability to change their geomet-
ric architecture are known as Reconfigurable Cable-Driven Parallel Robots (RCDPRs).
A recent study on RCDPRs [10–12] proposes different approaches for maximizing the
robot workspace and increasing platform stiffness. However, for most existing RCD-
PRs, reconfiguration is usually discrete and performed manually.
To achieve autonomous reconfigurability of RCDPRs, a novel concept of Mobile
Cable-Driven Parallel Robots (MCDPRs) has been introduced in [5] which uses a
unique combination of CDPR and mobile bases. The first MCDPR prototype, FASTKIT,
was designed and built in the framework of ECHORD++ project5 for logistic applica-
tions. Recently, another MCDPR prototype named MoPICK6 has been developed com-
posed of a three DoF point mass moving-platform pulled by four cables mounted on
four mobile bases, as shown in Fig.1(a). The targeted applications of MoPICK are mo-
bile tasks in a constrained environment, for example, a workshop or logistic operations
in a warehouse.
Generally MCDPRs are kinematically redundant due to the additional mobility of
the mobile bases [7, 8]. As a consequence, there exist multiple paths for the MCDPR
mobile bases to displace the moving-platform from one pose to another and to perform
a desired task. [7] addresses the problem of determining an optimal path also referred to
as a redundancy planning of FASTKIT with one degree of kinematic redundancy. On the
contrary, MoPICK has eight degrees of kinematic redundancy. In order to perform the
desired tasks in a potentially cluttered environments, safe and collision free trajectories
are required for the mobile bases and the moving-platform of MoPICK.
A* and Dijkstra are search algorithms which can find globally optimal paths in a
discretized workspace. Yet, such systematic algorithms scale poorly to higher dimen-
sions [20]. In contrast by exploiting randomization, a feasible path can be obtained
quickly, for instance, using sampling based motion planning algorithms [21]. However,
the resulting paths are often of poor quality and thus the methods require larger compu-
tation time to find reasonable solutions in complex environments. A common method
to reduce the complexity is by obstacle inflation [18] with the objective of reducing the
planning problem to a moving point in a 2D environment. Finally, an additional prob-
lem with these algorithms is the difficulty of incorporating the complex kinodynamic
constraints associated with CDPRs.
In this paper, we present a sampling based iterative path planning algorithm for
MoPICK by optimizing the wrench capabilities of its moving-platform. The proposed
algorithm searches for a feasible and continuous path of its mobile bases between the
initial and desired pose of the moving-platform by making a locally optimal choice at
each step. Thus the system’s kinodynamic constraints can be enforced at each instant of
the trajectory. The proposed algorithm decomposes the problem in two parts. The first
part aims to find a feasible, continuous and collision free path for the mobile bases. A
path is generated using an iterative procedure by generating a sequence of straight line
paths. The paths are then smoothed using B-Splines. The second part of the algorithm
takes as input the smoothed B-Splines and locally optimizes the moving-platform’s
wrench capability.
5 FASTKIT Videos: https://www.youtube.com/channel/UCJ8QRs818MBc8YSbn-bZVjA

6 Demonstration of MoPICK: https://youtu.be/ zfqtNsrpHI
Path Planning of a Mobile Cable-Driven Parallel Robot… 259
Aj Aj
Cj
z0 [m]
Cj uj
P
Mj
MP xb j
zb j Ob j
Fb j
y0
yb j
[m
Co j eo j
]
Mj O0 x0 [m]
F0
(a) (b)
Fig. 1. (a) MoPICK prototype and its (b) Parameterization
The paper is organized as follows. Section 2 deals with the description and pa-
rameterization of MoPICK. Section 3 details the Static Equilibrium (SE) conditions
of MCDPR under study that is used to determine its wrench feasible poses. Section 4
deals with the formulation of the task which is to be performed by MoPICK. Section 5
proposes a path planning algorithm for MoPICK to displace its moving-platform from
the initial pose to a desired pose while respecting its SE conditions. Section 6 discusses
the results acquired from the path planning algorithm for the required task. Finally,
conclusions are drawn and future work is presented in Section 7.
2 Manipulator Description and Parameterization

The MoPICK prototype consists of four mobile bases (p = 4) that carry the exit points
of the CDPR itself composed of four cables (q = 4) and a three degree-of-freedom
(DoF) point mass moving-platform (n = 3) as shown in Fig. 1(a). The jth mobile base,
denoted as M j , j = 1, . . . , 4, is carrying a single cable named as C j . Let u j be the
directional vector of C j pointing from the the point-mass P to the cable exit point A j .
Let t j be the C j ’s cable tension vector expressed as,
t j = t ju j, (1)
where t j denotes the tension in the cable C j .

Each mobile base has a cylindrical shape of radius rmb equal to 0.25 m. The support
structure for the cables is composed of a 1.2 m high aluminum frame mounted on a
cylindrical base. Let F0 be the base frame of origin O0 and axes x0 , y0 and z0 . Let
Fb j be the frame attached to M j of origin Ob j and axes xb j , yb j and zb j , respectively.
Each mobile base is composed of four wheels, with two motorized wheels and two
supporting wheels at the front and rear. It uses a differential drive mechanism through
the motorized wheels to move, thus is subjected to non-holonomic constraints [17],
i.e., it can only generate a transnational motion along xb j and a rotational motion about
zb j . Straight line motion is achieved by turning the motorized wheels at the same rate
in the same direction, however, pure rotational motion is accomplished by turning the
motorized wheels at the same rate in the opposite direction. The exit point A j lies on
the axis zb j , thus, the rotational motion of M j does not alter u j . As a consequence, for
a given position of P, u j can be determined by localizing the reference point Ob j of M j
in x0 y0 plane (see Fig. 1(b)).
3 Wrench Feasibility
In this section the Wrench Feasible Workspace (WFW) for MoPICK is studied. It is de-
fined as the set of platform poses for which the required set of wrenches can be balanced
with wrenches generated by the cables while keeping the MCDPR in Static Equilibrium
(SE) [6]. A MCDPR is in the state of equilibrium if and only if its mobile bases and
moving-platform are all in SE. Therefore, first the SE conditions of the moving-platform
and the mobile bases of MoPICK are formulated. From this, the wrench capabilities of
the moving-platform can be studied.
3.1 Static Equilibrium of the MCDPR
The SE of the moving-platform is expressed as [9]
Wt = f, (2)
where W is a (n × m) wrench matrix mapping the cable tension vector t ∈ Rm onto the
wrenches f ∈ Rn applied by the cables onto the moving-platform.
⎡ ⎤
⎡ x⎤ t1
f
t2 ⎥
W = u1 u2 u3 u4 , f = ⎣ f y ⎦ , t = ⎢ ⎥.
⎢
(3)
t3 ⎦
fz
⎣
t4
The cable tensions are all bounded between a minimum tension t j and a maximum
tension t j
t j ≤ t j ≤ t j , j = {1, . . . , 4}, (4)
The SE of a wheeled mobile base is characterized by its tipping conditions, which
depend on the moments generated at the boundaries of the mobile base footprint. The
footprint of M j is formed by joining the wheel contact points, denoted by Co j , o =
1, . . . , 4 in anticlockwise direction. The directional vector of the footprint boundary be-
tween the two consecutive contact points (Co j ,Co+1 j ) is denoted as eCo j as shown in
Fig. 1(b). Let mCo j be the moment generated about the footprint boundary between Co j
z 0[m] MP way-points
Pl = Pf Obstacles
P1
y 0[m
safe region
]
around obstacles
FO
x 0[m]
Fig. 2. Case Scenario
and Co+1 j at the instant when M j loses contact with the ground at the contact points
other than Co j and Co+1 j , expressed as
mCo j = eCTo j ((g j − co j ) × wg j ) + eCTo j ((co j − p) × u j )t j , (5)
where wg j represents the weight of M j . g j = [gxj gyj gzj ]T and co j = [cxo j cyo j czo j ]T
denote the Cartesian coordinate vector of the center of gravity and the contact point
Co j , respectively. p denotes the Cartesian coordinate vector P. For M j to be in SE,
mCo j , o = {1, . . . , 4} should be negative, namely,
mCo j ≤ 0, o = {1, . . . , 4} (6)
3.2 Available Wrench Set
For MCDPRs, the Available Wrench Set (AWS), is defined as the set of wrenches a
mechanism can generate while respecting the cable tension limits and the SE conditions
of the mobile bases [6], i.e.,
A = f ∈ R3 | f = Wtt , t j ≤ t j ≤ t j , mCo j ≤ 0, o = {1, . . ., 4}, j = {1, . . . , 4} .
(7)
The AWS defined in Eq. (7) corresponds to a n-dimensional convex polytope. The facets
of the polytope depend on the MCDPR configuration, the constraints associated with
the cable tension limits and the SE conditions of the mobile bases. Recent work on
MCDPRs [6, 7] discusses different strategies to determine the AWS for MCDPRs. The
latter proposes mapping the SE equations defined by Eqs. (4) and (5) into the wrench
space of the moving-platform using Eq. (2). The SE conditions in the wrench space are
exploited to determine the facets of the AWS.
An index called Capacity Margin [15, 16] determines if a given pose is wrench fea-
sible using the facets of the AWS and the vertices of the Required Wrench Set (RWS).
It is a measure of the robustness of the equilibrium of the robot, denoted as μ , namely,
μ = min (min sd,l ), (8)
where sd,l is the signed distance from dth vertex of the RWS to the lth facet of the AWS.
μ is positive as long as all the vertices of RWS are inscribed by A , i.e. RWS can be
generated by the cables while respecting all the SE conditions of a MCDPR.
4 Task Formulation
The task is defined as displacing the moving-platform from an initial position P1 to a
desired position Pf while ensuring that the moving-platform passing through a set of
way-points. The task is to be performed in a constrained environment having numer-
ous tables and obstacles in it. The way-points on the tables require a task action, for
example grasping and/or releasing an object. Some intermediate way-points are placed
in order to guide the system to navigate between the two consecutive task locations.
These intermediate way-points are interpolated between the two task locations while
maximizing the distance from the nearest obstacle.
Let l be the total number of way-points. Each way-point is denoted as Pi , i = 1, . . . , l.
The Cartesian coordinate vector of the ith way point is denoted as pi . The obstacles
are defined as cylinders. As discussed earlier, the mobile bases are also modeled as
cylinders with radius rmb . A common technique [18] used to ensure no collision between
the obstacles and the mobile bases is to inflate the obstacles by at least rmb in x0 y0 plane,
denoted as ‘safe region around obstacles’, as depicted in Fig. 2. As a consequence, the
mobile bases are treated as a point.
In order to perform the desired task, a feasible and collision free path of the mobile
bases is required. Accordingly, the path of the moving-platform is also required to dis-
place it from P1 to Pf by sequentially following the intermediate way-points. Therefore,
Section 5 addresses the aforementioned problem and presents a path planning algorithm
for MCDPR under study.
5 Path planning Algorithm

This section presents a Path Planning algorithm for MoPICK. In order to find a path be-
tween ith and (i + 1)th way-points, the following steps are taken. First, let’s assume that
a feasible solution has been obtained for the mobile bases with moving-platform located
at Pi . The wrench capability of the system is calculated when the moving-platform is
located at Pi+1 while the mobile bases are still located at the solution of previous way-
point (Pi ). The mobile bases are then iteratively displaced from Pi to Pi+1 such that each
displacement maximizes the wrench capability while avoiding collisions. This results
in a feasible path for the mobile bases. Given this feasible path for the mobile bases,
the path of the moving-platform is optimized to further increase the wrench capability
during the transition.
Cables , , and denote the previous steps of M1 , M2 , M3 and M4

Way-Points , , and denote possible steps for M1 , M2 , M3 and M4
y0 [m]
y0 [m]
safe region s4 =7
round obstacles s
4 =7
Pi+1
Pi+1
Circular grid s3 =7
Pi Pi
s1 =6 s3 =7
s2 =7
s2 =7
s1 =7 change of d1
x0 [m] x0 [m]
(a) First iteration (k = 2) (b) Change in step size of M1 (k = 6)

Final MBs
y0 [m]
s4 =7
y0 [m]
Position
s3 =7
Pi+1 Pi+1
s1 =7
Pi Pi
s2 =5
Avoiding Obstacle
x0 [m] x0 [m]
(c) Avoiding Obstacle for M2 (k = 13) (d) Iterative process stops at ki (k = 21)
P
y0 [m]
20
15 Pi4
Pi3
10 Pi+1
5 Pi,MP
Pi1
0
Pi
-5
-10
-15 Pi2
1 6 11 16 21 k x0 [m]
(e) μ as a function of iteration number (f) Paths of the mobile bases and the
moving-platform
Fig. 3. Illustration of MCDPR path planning

5.1 Generation of feasible path for mobile bases
The first phase of the algorithm iteratively searches for a collision free continuous path
for the mobile bases between any two way-points of the moving-platform. As an illus-
trative example, we will show how the algorithm evolves and calculates a continuous
path for M j , j = 1, . . . , 4, between ith and (i + 1)th way-points shown in Fig. 3(a). Let
k represents an iteration. Let Mi j be a matrix whose kth column contains the Cartesian
coordinate vector of M j in x0 y0 plane at the kth iteration of the algorithm.
From the initial configuration of the mobile bases at k = 1, the iterative process
starts at k = 2 by displacing the moving-platform at Pi+1 . This results in the drastic
decrease of the moving-platform wrench capabilities (μ ). At each kth iteration, the al-
gorithm searches for the best step of M j on a circular grid with a step size d j . On the
corresponding search grid, M j can either retain its current position or due to the non-
holonomic constraints, can move in the forward, backward, or diagonal directions, all
denoted by ‘+’ in Figs. 3(a), 3(b) and 3(c). Let M j have s j possibilities of collision free
steps. There exists s1 × s2 × s3 × s4 combinations for four mobile bases. The algorithm
chooses the combination that results in the maximum increase in μ . The matrix Mi j is
updated with the Cartesian coordinates of the new step taken by M j and a line segment
is created between its locations at kth and (k − 1)th iteration. If there are no feasible
steps for M j due the blockage around any obstacle, the step size (d j ) is reduced and
the search is repeated until a feasible step is obtained as illustrated in Fig. 3(b). The
procedure stops at k = ki when μ does not increase any further (see Fig. 3(d)). For the
illustrative example, the evaluation of μ as a function of iteration number is shown in
Fig. 3(e).
The next phase is to smooth the sequence of straight line segments generated be-
tween the Cartesian coordinates of M j during the iterative process using B-Splines [13].
The following function in MATLAB bspline f oot point [14] is used which requires the
two parameters knot sequence and control points to be tuned. It takes ki Cartesian coor-
dinates of M j in Mi j as input, and generates ten times the smooth sequence of Cartesian
coordinates, denoted as Pi j , as depicted in Fig. 3(f).
The above procedure is repeated to find a collision free continuous path for the mo-
bile bases between all the way-points of the moving-platform. It is noteworthy to men-
tion that location of M j at ki th iteration is used as an initial configuration to compute
P(i+1) j i.e. path of M j between (i + 1)th and (i + 2)th way-points. The Pseudo-code for
generating the path of mobile bases is presented in Algorithm. 1.
5.2 Generation of the moving-platform optimal path
The second phase of the path planning algorithm generates an optimal path of the
moving-platform, denoted as Pi,MP between ith and (i + 1)th way-points. Similar to
the first phase, the second phase of the algorithm is also an iterative process which
computes an optimal moving-platform pose for each location of the mobile bases in
Pi j , j = 1, . . . , 4. Let r represents an iteration. Total number of iterations are fixed for
the second phase i.e. equal to the number of Cartesian coordinates of M j in Pi j .
Given the initial (pi ) moving-platform pose at r = 1, the iterative process begins
at r = 2. It builds Pi,MP by searching for the best pose of the moving-platform on a
Algorithm 1: Generation of MB feasible paths

Input : A matrix with initial Cartesian coordinates of mobile bases Mi j (1)
Maximum number of iterations kmax
Cartesian coordinates of the moving-platform at (i + 1)th way-point pi+1
Output : Feasible Path of mobile bases Pi j
Notations : j represents the MB number j = 1, . . . , 4
Cartesian Coordinates of M j at the kth iteration Mi j (k)
A vector with Wrench Capability (μ ) of the moving-platform at the kth
iteration μ (k)
Invoked functions : Determine smooth path of M j BSpline(Mi j )
1 k = 1;
2 μ (1) = Determine μ with M j at Mi j (1) and moving-platform at pi+1
// Section 3.2
3 repeat
4 k = k + 1;
5 Determine the new step of M j // Section 5.1
6 Mi j (k) = Cartesian coordinates for the new location of M j
7 μ (k) = Determine μ with M j at Mi j (k) and MP at pi+1
8 until (μ (k) = μ (k − 1) or k > kmax );
9 Pi j = BSpline(Mi j )
10 return (Pi j )
circular grid with a step size dMP . In contrast to mobile bases, the moving-platform can
either retain its previous pose or take a step in all the neighboring eight directions i.e.,
forward, back, left, right and diagonals. The algorithm chooses a step with a maximum
μ among all the possible steps. For the illustrative example, Pi,MP is shown in Fig. 3(f).
The above process is repeated to find an optimal moving-platform path between all the
way-points. The Pseudo-code used to generate the moving-platform path is presented
in Algorithm. 2.
6 Results and Discussion
As discussed in Section 4, the required task is to displace the MoPICK moving-platform

from point P1 to Pl by sequentially following the intermediate way-points as depicted
in Fig. 2. In order to perform the desired task, the output of the proposed path plan-
ning algorithm is illustrated in Fig. 4. The simulation showing the complete process of
searching for a feasible path of mobile bases and the generation of the optimal path of
the moving-platform between all the way-points can be seen at7 . The video also shows
the resultant motion of the complete system along with the pick-and-place operations
performed at the task locations. The first step of the algorithm took 109.9 minutes to
generate the path of the mobile bases. The second step of the algorithm took 15.92
minutes to determine the path of the moving platform.
7 https://youtu.be/0wrdLBvM9-s
Algorithm 2: Generation of the moving-platform optimal path

Input : Path of mobile bases Pi j
Cartesian coordinates of the moving-platform at the ith way-point pi
Output : Optimal path of moving-platform Pi,MP
Notations : j denotes number of mobile bases j = 1, . . . , 4
Cartesian Coordinates of M j at rth iteration Pi j (r)
Optimal moving-platform pose at rth iteration Pi,MP (r)
Invoked functions : Determine the number of Cartesian coordinates of M j in Pi j
size(Pi j )
1 Pi,MP (1) = pi ;
2 for r = 2 : size (Pi1 ) do
3 Pi,MP (r) = Compute optimal moving-platform pose with M j at Pi j (r) // see
Section 5.2
4 end
The cable tension lower bound is null. The cable tension upper bound depends on
the actuation system used to actuate the cables of the MCDPR, i.e., motors, winches etc.
In MoPICK prototype, Dynamixel MX-64AT actuators and winches whose drum diam-
eter is equal to 0.2 m, are used, to pull the cables. Based on the hardware specification
and safety issues, the cable tension upper bound is set to 15 N.
The tunning parameters for the proposed MCDPR algorithm are the step sizes of
the search grids, d j , j = 1 . . . , 4 and dMP . It is important to tune these parameters in
order to obtain a feasible path of the MCDPR mobile bases and the moving-platform.
For example, during the first phase of the algorithm, the step size d j must stay smaller
than the diameter of the smallest obstacle in the environment in order to detect its col-
lision with the mobile bases. On the contrary, making d j smaller also increases the
computation time to calculate the feasible paths of the mobile bases. During the sec-
ond phase of the algorithm, a very small step size dMP can result in not achieving the
desired moving-platform position while a big dMP can generate large discontinuities in
the moving-platform path.
7 Conclusion and Future Work

The aim of this study is to propose a path planning algorithm for kinematically redun-
dant Mobile Cable-Driven Parallel Robots (MCDPRs). The proposed algorithm plans
the MCDPR path in two subsequent stages. In the first stage, the algorithm searches for
a feasible and collision free path of mobile bases. The second stage generates an optimal
path of the moving-platform to reach at the desired pose. Although the obtained path
between the initial and final poses may not be the shortest one, it leads to a feasible path,
when it exists. The proposed algorithm is validated by simulations on MoPICK, which
is a MCDPR with a three degree-of-freedom point mass end-effector displaced by four
cables mounted on four mobile bases. Future work will deal with the experimental val-
idation and the extension of the proposed methodology for the trajectory planning of
MCDPRs while taking into account the cable and mobile base velocity limits.
Path of M1 Path of M2 Path of M3

Path of M4 Path of the MP Way-Points
y [m]
0
7
Pl
5
Pi3
4 Pi4 Pi,MP
Pi+1 Pi
3 Pi1
Pi2
1
P1
0
0 1 2 3 4 5 6 7 x 0[m]
Fig. 4. Path of MoPICK for the required task
References
1. J. Albus, R. Bostelman, and N. Dagalakis. “The nist spider, a robot crane”. Journal of research
of the National Institute of Standards and Technology, 97(3):373–385, 1992.
2. S. Kawamura, H. Kino, and C. Won. “High-speed manipulation by using parallel wire-driven
robots”. Robotica, 18(1):13–21, 2000.
3. Lambert, C., Nahon, M., and Chalmers, D., 2007. “Implementation of an aerostat positioning
system with cable control”. IEEE/ASME Transactions on Mechatronics, 12(1), pp. 32-40.
4. P. Gallina, G. Rosati, and A. Rossi. “3-dof wire driven planar haptic interface”. In Journal of
Intelligent and Robotic Systems, 32(1):23–36, 2001.
5. Rasheed, T., Long, P., Marquez-Gamez, D. and Caro, S., “Tension Distribution Algorithm for
Planar Mobile Cable-Driven Parallel Robots”, The Third International Conference on Cable-
Driven Parallel Robots (CableCon 2017), Quebec City, Canada, August 24 2017.
6. Rasheed, T., Long, P., Gamez, D. M., and Caro, S. “Available wrench set for planar mobile
cable-driven parallel robots”. In 2018 IEEE International Conference on Robotics and Au-
tomation (ICRA).
7. Rasheed, T., Long, P., Marquez-Gamez, D. and Caro, S., 2018, August.“Optimal Kinematic
Redundancy Planning for Planar Mobile Cable-Driven Parallel Robots”. In ASME 2018 Inter-
national Design Engineering Technical Conferences and Computers and Information in Engi-
neering Conference (pp. V05BT07A025-V05BT07A025). American Society of Mechanical
Engineers.
8. Rasheed, T., Long, P., Gamez, D. M., and Caro, S.: “Kinematic modeling and twist feasibility
of mobile cable-driven parallel robots”. In Advances in Robot Kinematics 2018.
9. Hiller, M., Fang, S., Mielczarek, S., Verhoeven, R., and Franitza, D. “Design, analysis and
realization of tendonbased parallel manipulators”. In Mechanism and Machine Theory, 40(4),
pp. 429–445, 2005.
10. L. Gagliardini, S. Caro, M. Gouttefarde, and A. Girin. “A reconfiguration strategy for recon-
figurable cable-driven parallel robots”. In IEEE International Conference on Robotics and
Automation (ICRA), pages 1613–1620, IEEE, 2015.
11. L. Gagliardini, S. Caro, M. Gouttefarde, and A. Girin. “Discrete reconfiguration planning for
cable-driven parallel robots”. In Mechanism and Machine Theory, pages 313-337, 2016.
12. G. Rosati, D. Zanotto, and S. K. Agrawal. “On the design of adaptive cable-driven systems”.
In Journal of mechanisms and robotics, 3(2):021004, 2011.
13. Pottmann, H., Leopoldseder, S. and Hofer, M., 2002. “Approximation with active B-spline
curves and surfaces”. In Computer Graphics and Applications, 2002. Proceedings. 10th Pa-
cific Conference on (pp. 8-25). IEEE.
14. https://www.mathworks.com/matlabcentral/fileexchange/27374-b-splines
15. Guay, F., Cardou, P., Cruz-Ruiz, A.L. and Caro, S., 2014. “Measuring how well a structure
supports varying external wrenches”. In New Advances in Mechanisms, Transmissions and
Applications (pp. 385-392). Springer, Dordrecht
16. Ruiz, A.L.C., Caro, S., Cardou, P. and Guay, F., 2015. Arachnis: “Analysis of robots actuated
by cables with handy and neat interface software”. In Cable-Driven Parallel Robots (pp. 293-
305). Springer International Publishing.
17. Barraquand, J. and Latombe, J.C., “On nonholonomic mobile robots and optimal maneu-
vering”. In Proceedings. IEEE International Symposium on Intelligent Control, pp. 340-347,
September 1989, IEEE.
18. Connors, J. and Elkaim, G., “Analysis of a spline based, obstacle avoiding path planning
algorithm”. In Vehicular Technology Conference, VTC 2007-Spring, pp. 2565-2569, April
2007, IEEE.
19. Y. Wischnitzer, N. Shvalb, and M. Shoham, “Wire-driven parallel robot: Permitting collisions
between wires,” Int. J. Rob. Res., vol. 27, no. 9, pp. 1007–1026, Sep. 2008
20. S. M. LaValle, IPlanning algorithms. Cambridge university press, 2006.
21. S. Karaman and E. Frazzoli, “Sampling-based algorithms for optimal motion planning,” The
international journal of robotics research, vol. 30, no. 7, pp. 846–894, 2011.
Development of Emergency Strategies for
Cable-Driven Parallel Robots after a Cable
Break
Roland Boumann and Tobias Bruckmann
Universität Duisburg-Essen, Forsthausweg 2, 47057 Duisburg, Germany,

{roland.boumann,tobias.bruckmann}@uni-due.de
Abstract. This paper investigates cable breaks of a cable-driven par-

allel robot and proposes emergency strategies to recover the payload. A
simulation model of a simplified two dimensional robot is set up and the
workspace of the robot is analyzed before and after a cable break. Two
methods are proposed for dealing with the issue of guiding the end effec-
tor into the remaining workspace and to stop the system: An approach
to minimize the kinetic energy of the system is made as well as the use of
potential fields in combination with a method for calculating reasonable
cable force distributions outside of the wrench feasible workspace. Both
methods are tested in simulation and the results are presented.
Keywords: cable-driven parallel robot, cable break, cable failure, force

distribution, emergency strategies, model prediction, potential fields
1 Introduction
Cable-driven parallel robots are currently making their way into application.
This covers multiple fields and industrial branches, including large-scale 3D
printing [1], warehousing and other applications [2]. As this typically includes
high end effector masses, from a safety perspective, system failures have to be
investigated to avoid damage to the robot, the payload, the environment and
especially humans. Intuitively, one of the most expected failures for cable robots
is a failure of a cable due to cable break. On the other hand, for the construction
of machines employing cables, extensive compilations of guidelines and norms
exist which ensure a durable and reliable operation of cable-driven systems. Still,
and especially under prototyping conditions, the failure of mechanical compo-
nents, sensors, motors and control algorithms may be an issue that leads to
loss of cable tension or a breaking cable, respectively. Clearly, this is a sudden
change in the dynamics of the system and there is the immediate danger of an
end effector crash, which includes the platform and the payload. On the other
hand, cable robots typically use a large number of cables which opens the po-
tential to recover and to safely guide the platform into a safe position with the
remaining cables. This issue has already been adressed in some publications, e.g.
in [3, 4]. Both publications focus mainly on defining dynamic trajectories back

270 R. Boumann and T. Bruckmann
into the post-failure workspace. This work instead proposes two new strategies
to recover the payload, in which the trajectory does not need to be pre-defined.
At first a brief introduction about cable-robots is given, afterwards the basics
of a kinematic and dynamic robot model are explained. This is followed by a
short discussion about the problems to be faced after a cable breakage at the
system and why it is important to develop special strategies for guidance of
the robot platform afterwards. Two strategies are introduced, which consist of
known methods, combined in new ways. A simulation model for evaluation is set
up and analyzed regarding its workspace before and after a cable failure. Both
strategies are tested using the model. In the end a conclusion and outlook for
future work is given.
2 Modeling and cable-robot basics
mP , I
P
6
-
P
pi
fi τE
fE
B
RP , Φ
B
rP
B
li
B
bi
6
B
-
Fig. 1. Cable-robot model parameters
The robot model equations and notation used here base on [2]. The robot is
defined by the inertial coordinate-system 6 B and the fixed coordinate-system
-
on the platform 6 P
- , using a number of m cables and n degrees-of-freedom. The
redundancy of the platform is given by r = m − n. The posture of the platform
in coordinates of the inertial frame is B xP = [B r T P Φ ] , which contains the
T T
position vector r P and the orientation Φ of the end effector. The orientation
B
with respect to the inertial coordinate-system is described by the rotation matrix

B
RP in the means of roll-pitch-yaw angles. Assuming a point-shaped guidance
of the cable, the modeling of pulleys is neglected in the following. The cable
vectors B li as shown in figure 1, can be obtained by
B
l i = B bi − (B r P + B R P P pi ), 1 ≤ i ≤ m. (1)

B
p Bi
Development of Emergency Strategies for Cable-Driven… 271
Given the cable vectors, the length of each cable and hence the joint coordinates
can be determined as
qi = B li 2 , 1 ≤ i ≤ m. (2)
The cable force vector f ∈ Rm×1 contains all tensions fi where each fi is mea-
sured along cable direction ν i . The direction of the ith cable force f i is obtained
from inverse kinematics by
li
f i = fi · = fi · ν i , 1 ≤ i ≤ m. (3)
li 2
Based on the force equilibrium, the force and torque acting on the end effector
wp and a structure matrix AT can be obtained as
⎡ ⎤
f1
f ν1 . . . νm

⎢ .. ⎥
wp = p = ⎣ . ⎦ = A f.
T
(4)
τp p 1 × ν 1 . . . pm × ν m
fm
Setting up Newton-Euler equations and assuming that the geometrical center

point of the end effector is also the center of gravity it holds that
B
mp E 0 r̈ P 0 f

+ − E = AT f . (5)

0 IH Φ̈ I(Ḣ Φ̇) + (H Φ̇) × I(H Φ̇) τE

wE

M p (xp ) ẍp wC (xp ,ẋp )

−w
The vectors ẋp and ẍp are the first and second time derivative of the end effector
pose, respectively. The matrix H and its time derivative can be obtained from
the kinematic Kardan equations [5]. M p is the mass matrix of the platform,
dependent on the mass of the platform mp and the inertia tensor I. E is a
3 × 3 identity matrix. Note that the inertia tensor needs to be expressed in
the inertial system 6 B and depends on the orientation of the platform. The
-
Coriolis and centrifugal forces and torques are described by the cartesian space
vector wC while vector wE contains the acting external forces and torques. To
avoid cable sagging on one hand and to limit the maximum motor torque on
the other hand, the cable forces are limited by a minimum force vector f min
and a maximum force vector f max . Solving Eq. 5 with respect to the cable force
limits is a known problem in the field of cable-robotics. Approaches include e.g.
calculating the forces as an optimization-problem [6] or applying the barycentric
method [7].
Now assume the case of a cable break. Typically, the robot is suddenly located
outside the post-failure workspace of its remaining set of cables. Thus, a solution
of Eq. 5 cannot be found in that case and the given methods fail to operate. A
method to deal with this circumstance is proposed in [8]: The constraints of
minimum and maximum cable forces cannot be changed, therefore Eq. 5 needs
to be relaxed. Hence a so called slack variable s is added to the equation. In [8], a
way to solve the resulting problem via quadratic programming is also proposed.
The implementation takes the form
minimize sT D 1 s + (f − f
)T D 2 (f − f
)
with AT f + w + s = 0 (6)
f min ≤ f ≤ f max .
The matrices D 1 and D 2 are in general diagonal weighting matrices, f indi-

cates the desired tension levels. By choosing the magnitude of the terms of D 1
greater than those in D 2 , the solution will more likely follow the desired wrench.
Otherwise, the resulting forces will tend to follow f . By making a proper choice
of the desired tension, the solution can be adjusted e.g. for minimum power or
maximum stiffness.
3 Development of emergency-strategies
The occurrence of a cable breakage can have fatal consequences towards loss of
payload, machine damage or harm to humans and the environment, caused by
an uncontrolled movement of the robot. As the control system often depends on
model-based control [9], it cannot react to the changed system condition. For
simplicity, assume the break of one cable. Afterwards, a cable robot structure
with m − 1 cables still exists. The structure matrix of the remaining system is
reduced by the column of the broken cable AT∗ ∈ Rn×m−1 . The remaining cable
force distribution is f ∗ ∈ R(m−1)×1 . Note, the following considerations might
also work for the failure of multiple cables as long as a single cable remains.
Accordingly, the control system can still influence the platform motion and thus
stop the motion in a controlled manner, but has to deal with the fact that the
platform may be outside the workspace of the remaining cable robot. Thus, two
strategies are presented in the following.
3.1 Minimization of kinetic energy

The first approach aims at minimizing the kinetic energy of the system after
cable breakage. Since the kinetic energy is directly linked to the velocity of
the end effector, its minimization leads to a stop of the robot. If the robot
is able to stop, it will automatically be within in the post-failure workspace.
As the final position does not need to be predefined, no knowledge or analysis
of the workspace is needed, which is a major advantage of this method. To
generate appropriate cable forces, a model-based predictive approach is chosen
as explained in [10]. The method should predict the system state based on the
robot’s nonlinear dynamic model and find cable forces to minimize the kinetic
energy (and thus the end effector’s velocity). Model predictive control on parallel
manipulators is explained e.g. in [11]. Assuming a rigid body, the kinetic energy
of the system is given by
1 B TB 1
EKin = mp ṙ P ṙ P + Ω T IΩ. (7)
2 2
Since the angular velocity Ω = H Φ̇ is directly dependent on Φ̇ [5], the magnitude

T
of ẋP = [r˙P T Φ̇ ]T must be minimized in order to minimize the kinetic energy.
The system is described by the nonlinear discrete state equations
x(k + 1) = f x(k), u(k)

(8)
y(k + 1) = C x(k + 1), (9)
with the state vector of velocity and position
x(k) = [ ṙ p (k), Φ̇(k), r p (k), Φ(k) ]T , (10)
the output vector

ṙ p (k + 1)

y(k + 1) = v(k + 1) = (11)
Φ̇(k + 1)
and therefore the output matrix is
J 0 00

C= . (12)
0 K00
J and K are 3 × 3 identity matrices. The input vector consists of the vector of
the remaining cable forces
u(k) = f ∗ (k). (13)
The nonlinear system function f x(k), u(k) returns the state vector in the next

time step x(k+1) under consideration of Eq. 5 and a numerical integration based
on the Euler-Cromer method:
ṙ p (k + 1) ṙ p (k)

T∗ ∗
= + M −1 A f + w E − w C (k)Δtc (14)
Φ̇(k + 1) Φ̇(k) p
r p (k + 1) r (k) ṙ p (k + 1)

= p + Δtc . (15)
Φ(k + 1) Φ(k) Φ̇(k + 1)
The prediction step size is Δtc . The quality function over prediction horizon np
and control horizon nc can be formulated as
np
T
y(k + i) − y r (k + i) Q y(k + i) − y r (k + i) + . . .

J=
i=1
nc
∗
f (k + i − 1) − f ∗ (k + i − 2) r f ∗ (k + i − 1) − f ∗ (k + i − 2) .
T
i=i
(16)
This quality function is minimized every time step under consideration of the
cable force limits, to determine the set-point cable forces f ∗ . If the reference
output y r is chosen as zero, the function holds the minimum value when the
robot comes to a full stop and the cable forces remain at a constant value. The
parameters Q and r can be used to weight the optimization problem.
3.2 Potential fields

As a second approach, potential fields are used to guide the end effector into
a safe position. A reference for applying the potential field method to parallel
cable-based manipulators can be found in [12]. The used potential fields, as they
are described by [13], will be shortly explained in the following section. The
robot is set under the influence of a potential field U which is the sum of all
attractive fields U att and repulsive fields U rep . The virtual forces acting on the
end effector are considered as negative gradient of U , where
F (r p ) = −∇U (r p ) = −∇U att (r p ) − ∇U rep (r p ). (17)
To guide the end effector, a goal needs to be defined, which is chosen inside
the post-failure workspace of the system. The strategy to determine a goal pose
quickly and depending on the broken cable is subject to current work. The robot
will be attracted to this position by the attractive virtual forces. To avoid colli-
sion between the end effector and the frame or obstacles inside the workspace,
repulsive fields can be used, which is a strong advantage of this method.
Attractive fields The attractive field should be monotonically increasing with

the distance to its origin xf and continuously differentiable to avoid stability
issues. Therefore a field that grows squarely with the distance is chosen. Be
ρf (r p ) the Euclidean distance between actual and final posture. The quadratic
field with respect to ρf (r p ) and the goal position xf is then defined as
1 2 1
U att (r p ) = ζρ (r p ) = ζr p − xf 2 . (18)
2 f 2
The parameter ζ can be used to adjust the field strength. With rising distance to
the fields origin, the attractive forces are growing strongly. With larger ρf (r p ),
the force might be too high. Therefore the quadratic potential is combined with
a conic one, which will be switched at the distance d, resulting in lower virtual
forces on higher distances from the origin. Note that the gradient is equivalent
at the boundary of both fields ρf (r p ) = d to ensure continuity. The resulting
attractive force equals
⎧
⎪
⎪ −ζ(r p − xf ) : ρf (r p ) ≤ d
⎨
F att (r p ) = −∇U att (r p ) = (19)
dζ(r p − xf )
⎩− : ρf (r p ) > d.
⎪
⎪
ρf (r p )
Repulsive fields While the repulsive fields should avoid collision between the
end effector and obstacles, they should not influence the platform when its far
enough away. Thus, a field is chosen which equals zero at a certain distance from
the obstacle ρ0 and rises to infinity in direction of the obstacle boundary. In case
of ρ(r p ) ≤ ρ0 , the potential of such field yields
2
1 1 1
U rep (r p ) = η − . (20)
2 ρ(r p ) ρ0
Herein, ρ(r p ) is the shortest distance between the end effector-position r p and the
nearest boundary of an obstructive object and η is a scalar amplification of the
field. If the object boundary consists of a single convex region, the corresponding
repulsive force is given by the negative gradient of the repulsive field:
⎧
1 1 1
⎨η
⎪ − ∇ρ(r p ) : ρ(r p ) ≤ ρ0
F rep (r p ) = −∇U rep (r p ) = ρ(r p ) ρ0 ρ2 (r p )
⎪
0 : ρ(r p ) > ρ0 .
⎩
(21)
The gradient of the distance to the nearest obstacle is given by
rp − b
∇ρ(r p ) = , (22)
r p − b
where b is the point on the obstacle boundary closest to the actual platform-
position. The virtual force acting onto the end effector, caused by all attracting
and repelling fields and based on the current position, is included in wE with
reverse sign.
4 Simulation and results
Parameter Value

0011
Position of pulleys B= m
0110
Cable force limits fmin = 5N, fmax = 150N

Mass of end effector mp = 1kg
Simulation time step Δt = 1ms
Table 1. Parameter of the two-dimensional point-mass model
To test the emergency strategies, a simplified dynamic simulation model is set

up, see Tab. 1. The chosen planar two-dimensional model consists of four cables
and a point-mass describing the robot platform (m = 4, n = 2) and is subject
to gravity. Modeling of winches and drives is neglected as well as friction and
disturbances. The cables are assumed to be massless and cannot sag. According
to that the, dynamic effect of a snapping cable is also neglected, which is a
limitation of the simulation. The numbering of the cables is clockwise, beginning
with cable one at the bottom left. By numerical integration using the Euler-
Cromer method, speed and position in the next time step k + 1 are determined
with a fixed simulation time step Δt. Testing a dynamic model with a higher
number of degrees of freedom will be part of future work.
1 1
0.8 0.8
y-axis [m]
y-axis [m]
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x-axis [m] x-axis [m]
(a) Workspace of simulation model (b) Failure of upper cable
Fig. 2. Static workspace before and after cable failure, black dots indicate the winch
positions
4.1 Examination of the Workspace
Especially for cable robots the wrench-feasible workspace is of interest as the set
of all postures that can be reached under a certain wrench with respect to Eq. 5
and the cable force limits [9]. Neglecting dynamic effects, the static equilibrium
workspace can be obtained. Using a point grid and solving Eq. 5 for each point,
an approximation of the workspace is generated. Fig. 2 compares the workspace
before and after a break of one upper cable. It can be observed that the static
workspace after a cable failure is nearly halved. This also occurs for the break of
a lower cable. Thus the probability is high, that the robot is no longer inside the
workspace after the cable break. Now the robot must be guided into a position
inside of the remaining workspace to stop the motion and to prevent the end
effector from hitting the ground or the robot frame. This calls out for emergency
strategies as they are introduced in Sec. 3.
4.2 Application of emergency methods
The emergency strategies described in Chap. 3 are implemented in the simu-

lation model to calculate desired cable forces at each time step. Note, due to
lack of space, only the results for a breakage of one upper cable are presented.
At the start of the simulation, the kinematic state of the robot is according to
Tab. 2. The platform is placed in the top left area of the frame and is therefore
outside and far away from the post-failure workspace. In the scope of the paper
the velocity has been set to zero but first investigations indicate also applica-
bility for non-zero velocities. At first, the system behavior under usage of the
Parameter Value
T
Start position platform xstart = 0.1 0.9 m
T
Start velocity platform ẋstart = 0 0 m/s
Control and prediction horizon np = n c = 1

0.7 0
Weighting Paramters for prediction Q= , r = 2.5e − 5
0 1
System reference output yr = 0

T
Selected target position xf = 0.65 0.35 m
Border distance d = 0.2m

Potential field scaling parameters ξ = 100,
η1 = η2 = 1000

10 30 0
Weighting matrices slack optimization D1 = , D2 =
01 0 30
Damping factor of virtual damping DP = 500 Ns/m
Table 2. Parameter of simulation and both methods
minimization of kinetic energy is investigated. To solve the nonlinear and con-

strained optimization problem and to minimize the quality function in Eq. 16, an
interior-point-algorithm is used. The model prediction will be carried out only
for one time step. The goal is to maximize the reduction of kinetic energy in
each time step. To increase the efficiency of the algorithm and to reduce com-
putation time, the last timestep cable forces are always taken as a warm start.
The weighting parameters Q and r are set as described in the Tab. 2 and the
results of the simulation are shown in Fig. 3. In this example, no collision with
the geometric boundaries of the robot can be detected. The maximum occurring
speed per axis is about 4m/s. After a short time of 0.4s, the velocity is nearly
zero. After 0.5s the cable forces reach steady values and the robot comes to a
full stop, which shows the successful rescue of the platform in the workspace.
The cable force limits are reached but not exceeded. By adjusting the weighting
parameters of the quality function, the desired behavior can of course be shifted,
either to slower force progressions with lower rates of change or to faster inter-
vention to reduce speed. The weighting of the velocities in both axes can also
be set. This is especially important when using this algorithm with a different
starting position or at a different robot geometry. In case of a robot with more
degrees-of-freedom, this can also be used to act against undesired tilting of the
platform which goes hand in hand with a possible loss of the carried load.
For the application of the potential fields an attracting field according to Eq. 19
with origin in the selected target position is defined. The Euclidean distance ρf
is calculated from the known position r p and xf . Influence distance d and the
scaling parameter ξ are selected as shown in Tab. 2. The field generates the vir-
tual force F att (xr ). It is assumed that the space behind the pulleys is a limiting
forces [N] velocity [ ms ]position [m]

1
x
y
0.5
0
5
ẋ
ẏ
0
-5
150 f1
100 f3
50 f4
0
0 0.2 0.4 0.6
time [s]
Fig. 3. Cable failure handling with minimization of kinetic energy
frame. Two repelling fields are defined to avoid collisions with the ground and
the frame opposite to the broken cable. The floor and frame wall are defined
as repulsive objects, causing the forces F rep,B (r p ) and F rep,W (r p ), respectively.
The influence distances of both repulsive fields are chosen such that they just
reach the goal position.Thus the platform is pushed back during overshooting
the goal position on the one hand, and on the other hand the target point itself
is free from virtual forces. In order to further improve the deceleration of the
end effector, a virtual damping based on the current end effector velocity can
be implemented, which overcomes the virtual forces of the potential fields with
increasing velocity and thus limits the total virtual force. The resulting virtual
damping force as a proportional function of the damping constant DP is F DP .
The wrench including all virtual and external forces results in
w(r p , ṙ p )E = F E − F att (r p ) + F rep,B (r p ) + F rep,W (r p ) + F DP (ṙ p ) . (23)

Since the demanded virtual force may not be producible by the cables, an ap-
proximate solution is used as explained in Eq. 6. The desired tension f̃ is chosen
as the mean force in between the force limits. The weighting matrices are chosen
as shown in Tab. 2. The optimization problem is also solved using an interior-
point algorithm and the results are shown in Fig. 4. It can be seen that the
end effector is pulled into the goal after about 0.2s, but since a speed of about
6m/s per axis is built up, it cannot be decelerated enough and thus overshoots.
The position and speed curve show that the end effector is successfully pushed
away from the floor and the wall. It circles around the goal and roughly reaches
there after 0.5s. After about 1s, the cable forces have settled and the robot is
completely stopped. Cable force limits are reached but not exceeded by the al-
gorithm. The desired platform motion can be adjusted by choosing a different
set of parameters, which is also important for usage with different starting po-
forces [N] velocity [ ms ]position [m]

1
x
y
0.5 xf
yf
0
10
ẋ
ẏ
0
-10
150 f1
100 f3
50 f4
0
0 0.5 1
time [s]
Fig. 4. Cable failure handling with potential fields
sitions or at a different robot geometry. In conclusion, both methods are shown

to work. They safely guide the platform to a recovery position. Note that the
results depend very much on which parameters are set. It is not clear whether
the set of parameters has to be readjusted for each other geometrical starting
point. A clarification can be part of future work.
5 Conclusion and Outlook

In this work, the recovery of a cable robot after a cable break was addressed
by two emergency strategies. Both strategies have been implemented in a sim-
plified dynamic simulation model and tested successfully, what allows for a first
conclusion to be drawn about their suitability. The methods themselves bear
great potential for further improvements as part of future work. The predictive
nature of the kinetic energy minimization method allows for predicting the ma-
chine behavior up to a final stable position and optimize the force curve. Also a
combined position- and velocity control could be performed. Continuity in the
produced forces is currently investigated. The use of potential fields can be fur-
ther carried out using fields to avoid collision with obstacles in the workspace.
In terms of computation time it might be also worth a consideration, to try
out other methods such as neural networks or fuzzy logic. The proposed strate-
gies need to be tested with more detailed simulation models. Detailed models of
winches and pulleys including friction and disturbances need to be implemented.
Also a detailed modeling of the cables should be included. Finally, experiments
on hardware need to be carried out. Therefore, the algorithms need to be tested
and eventually adjusted regarding real-time requirements. All results of the work
serve to improve cable robot safety and pave the way for the introduction of cable
robotics into industrial practice.
Acknowledgment
This research received funding from the EFRE.NRW (2014-2020) Joint Research
Funding Programme of the European Union (EFRE) and the Ministry of Econ-
omy, Energy, Industry, and Handicrafts of the German Federal State of North
Rhine-Westphalia (NRW) under grant agreement EFRE-0800365 (ML-1-1-019B,
LEAN).
References
1. Izard, J.B., Dubor, A., Herve, P.E., Cabay, E., Culla, D., Rodriguez, M., Barrado,
M.: Large-scale 3D printing with cable-driven parallel robots. Construction Robotics
1, pp. 69-76 (2017)
2. Bruckmann, T., Lalo, W., Sturm, C.: Application Examples of Wire Robots. In:
Multibody System Dynamics, Robotics and Control. Springer, Vienna (2013)
3. Berti, A., Gouttefarde, M., Carricato M.: Dynamic recovery of cable-suspended
parallel robots after a cable failure. In: Lennarčič J., Merlet JP.(eds) Advances in
Robot Kinematics. Springer, Cham (2016)
4. Passarini, C., Zanotto, D., Boschetti, G.: Dynamic trajectory planning for failure
recovery in cable-suspended camera systems. J. Mechanisms and Robotics 11(2),
(2018)
5. Hiller, M.: Mechanische Systeme: Eine Einführung in die analytische Mechanik und
Systemdynamik. Springer, Heidelberg (1983)
6. Bruckmann, T., Pott, A., Hiller, M.: Calculating force distributions for redundantly
actuated tendon-based Stewart platforms. In: Lennarčič J., Roth B.(eds) Advances
in Robot Kinematics. Springer, Dordrecht (2006)
7. Mikelsons, L., Bruckmann, T., Hiller, M., Schramm, D.: A real-time capable force
calculation algorithm for redundant tendon-based parallel manipulators. In: IEEE
International Conference on Robotics and Automation, pp. 3869-3874. Pasadena,
CA (2008)
8. Côté, A.F., Cardou, P., Gosselin, C.: A tension distribution algorithm for cable-
driven parallel robots operating beyond their wrench-feasible workspace. In: 16th
International Conference on Control, Automation and Systems (ICCAS), pp. 68-73.
Gyeongju (2016)
9. Pott, A.: Cable-driven Parallel Robots: Theory and Application. Springer Tracts in
Advanced Robotics, (2018). - ISBN 978-3-319-76137-4
10. Grüne, L., Pannek, J.: Nonlinear Model Predictive Control: Theory and Algo-
rithms. Springer, London (2011). – ISBN 9780857295019
11. Vivas, A., Poignet, P.: Predictive functional control of a parallel robot. In: Control
Engineering Practice 13(7), pp. 863-874 (2005)
12. Zi, B., Lin, J., Qian, S.: Localization, obstacle avoidance planning and control
of a cooperative cable parallel robot for multiple mobile cranes. In: Robotics and
Computer-Integrated Manufacturing 34, pp. 105-123 (2015)
13. Spong, M.W., Hutchinson, S., Vidyasagar, M.: Robot Modeling and Control. Wiley,
(2005). – ISBN 9780471649908
A Conditional Stop Capable Trajectory Planner
for Cable-Driven Parallel Robots
Patrik Lemmen, Robin Heidel and Tobias Bruckmann
University of Duisburg-Essen, North Rhine Westfalia, Germany,

[email protected], [email protected],
[email protected],
WWW home page: http://www.uni-due.de/mechatronik
Abstract. This paper presents a new method for cable-driven paral-

lel robots (CDPR) to generate point-to-point (PTP) trajectories with
the consideration of CDPR workspace. Similar to existing trajectory
planners that work on position level using a quintic or higher polyno-
mial, here the complete acceleration process is planned referring to a
double-S trajectory, but with quintic polynomial for acceleration. Due
to workspace considerations, the method had to be extended for the
opportunity to trigger well-defined stops. Adapted to practical use of
CDPRs, the method allows to specify distance, maximum velocity and
maximum acceleration while limiting jerk where the latter is indirectly
influenced. A feasibility check for user-given parameters leads to a re-
duction of acceleration and/or velocity to ensure reaching the goal while
limiting the jerk to reduce actuator wear. To decouple the dynamics of
standard trajectories and conditional stops, the braking acceleration is
separately defined.
The method has been tested on a cable robot prototype, called SEGESTA.
The results show the capabilities of the new method especially for the
real-time halt performance and the consideration of workspace bound-
aries.
Keywords: cable-driven parallel robot, trajectory, dynamic limitation,

stop trajectory
1 Introduction and state of the art

For cable-driven parallel robots (CDPR) more and more applications are gen-
erated. The advantages of cable-driven robots compared to the well-established
serial manipulators are potentially large workspaces and lower moved masses rel-
ative to the payload. This leads to higher dynamic capabilities which is beneficial
for many applications.
A CDPR uses a set of tensed cables that are wound on computer-controlled
winches, where all cables are attached to a common payload – typically a plat-
form that serves as the end effector. For control purposes, it is always necessary
to specify a trajectory. A trajectory first defines the geometrical path over time.
Thus, it is possible to specify a set of discretized poses along the path, e.g. per

282 P. Lemmen et al.
time step. This path information is composed of the position as well as the ori-
entation of the end effector, resulting in the pose. Furthermore, time derivatives
of the path describe the dynamic aspects, like velocity or acceleration. Hence,
the robot obtains pose, velocity and acceleration for each point in time.
Considering the workspace of CDPR for valid movements is not very easy
due to non-intuitive shapes of the workspace. In [1] feasible dynamic workspaces
for CDPR are calculated for specific parametric Cartesian trajectories, but not
for real-time applications.
At the University of Duisburg-Essen, a current research topic is to realize a
storage and retrieval system using a cable-driven parallel robot. Therefore, the
limitation of acceleration is needed to ensure the safe transport of the payload on
the end effector. Another application – where constant velocity is required – is
e.g. 3D-printing. Here, directly setting velocity and acceleration in the direction
of motion is convenient. Currently, several studies on 3D-printing with CDPR
(as proposed in [4]) are in progress.
This paper focuses on a trajectory planner that can be used for applications
where the workspace of the cable-driven robot is not known in advance. The
goal is to avoid moving to positions where no feasible cable force distribution
can be found. The dynamic workspace is considered for every point-to-point
motion in space (so only for straight line movements) throughout a real-time
feasibility check. For the case the robot would leave the workspace, a defined stop
trajectory will be initiated. Furthermore, for purposed applications, velocity and
acceleration are needed to be limited. To ensure smooth trajectories and to avoid
actuator wear it is well known that the limitation of jerk is also important, see
[2, 3]. Despite the stop initiated by the workspace boundaries, conditional stops
of the robot might be required for general safety, e.g. upon sensor measurements
or an operator signal. Here, a velocity limit will also be guaranteed as higher
velocity leads to a longer phase of deceleration. Note, that the trajectories should
be be planned within a real-time control system.
The planner allows to know the execution time of each phase of the trajec-
tory to plan the stop trajectory. Existing trajectory algorithms with limited jerk
[5] are taken and enhanced with polynomials of higher grade to obtain more-
over a continuous snap (fourth derivative of path). Additionally, an automatic
adaption of the dynamic parameters will be accomplished if they are too high,
implemented by a double-S curve. Note, in this paper, the motion parameters of
velocity, acceleration and jerk are set in the direction of motion and not in the
joint space.
In the following section, the algorithms will be derived. First, it will be shown
why the use of a double-S curve is reasonable. Second, the possibility to trigger
a stop trajectory for the robot at any time will be implemented. As all parame-
ters of the trajectory are known at every time step, a new stop polynomial can
be calculated for higher maximum values of acceleration or deceleration respec-
tively. This gives the opportunity for stopping the robot in a controlled way, e.g.
depending on the distance of the end effector to an object or to the workspace
A Conditional Stop Capable Trajectory Planner for Cable-Driven… 283
boundaries. This contributes to safety requirements, although it won’t replace a

physical emergency stop. Finally, the algorithms will be tested experimentally.
2 Development of PTP trajectory planners
In this section a new trajectory generator based on time-delayed quintic polyno-

mials on acceleration level will be introduced. Note, the calculation is performed
in direction of the robot’s movement along the straight line between start and
goal, which is necessary for the predictive workspace calculation in the next Sec.
3. Furthermore, it makes the operator’s declaration of dynamic parameters more
intuitively.
2.1 Trajectory planning based on velocity and acceleration limit
In [6], a dividing of the path into parts called segments 1 - 3 is introduced

on velocity level, called double-s trajectory, but can also be transferred to the
acceleration level. In segment 1, a polynomial describes an acceleration phase
from zero to a defined maximum acceleration plateau. Next this acceleration
value will be kept for a given period of time in segment 2. Finally, in segment 3
the first segment is mirrored to a deceleration back to zero. In addition to this,
jerk is limited by derivative-constraints of the acceleration.
Here, to limit velocity and acceleration, a quintic polynomial on acceleration
level is used for segments 1 and 3 due to the quantity of required constraints.
Hence, the equation for acceleration becomes
a(t) = c0 + tc1 + t2 c2 + t3 c3 + t4 c4 + t5 c5 (1)
where the constants ci with i = 1, .., 6 admit six constraints, which are start
acceleration, end acceleration, start jerk, end jerk, start snap and end snap.
The consideration of the snap - to get a continous jerk - is also mentioned
in [6] and requires a ”15-segments double-S trajectory”. Accordingly, the accel-
eration polynomial only needs to be in the second order, but as it’s used here
in fifth order because of the number of constraints the jerk becomes smoother
as a side benefit. The limitation of jerk is not directly implemented by setting a
value but defined by the time for reaching maximum acceleration.
2.2 Time parameters of the segments
In Fig. 1 the time parameters for scheduling the trajectory are shown on velocity
and acceleration level with segments 1 - 3. As mentioned before there is a time
tsa as an operational input that determines the time for reaching Amax . After
that, the maximum acceleration is kept constant for a certain period of time
tda − tsa before it returns to a value of zero within the same time as in segment
1. Thus, tga is composed as
tga = tda + tsa . (2)
Following there is a phase with constant velocity for time twa . The sum of tga
and twa leads to
tdv = tga + twa . (3)
Finally, the time for deceleration phase is described in tga as in segments 1 to 3.
2.3 Calculation of time parameters
The following parameters are set by the operator:
i. maximum velocity vmax

ii. maximum acceleration amax
iii. time tsa for reaching amax
iv. distance Δx between start and end position
To reach the maximum set velocity, the area underneath acceleration curve
(A1 , A2 , A3 in Fig. 1) must be equal to vmax , since
tga
A1 + A2 + A3 = a(t)dt = v(tga ) = vmax . (4)
0
As A1 and A2 are known - due to the fact that they are equal because of the
symmetry of the polynomial - and as also tsa and amax are known, A3 can be
calculated from (4) and tda determined with Eq. 7.
tsa amax
A1 = A2 = . (5)
2
Δx
tdv = . (6)
vmax
vmax A3
tda = = + tsa . (7)
amax amax
The phase of constant velocity twa is determined similarly to Eq. 4, because
the integral of the velocity along the entire time is equal to the accomplished
distance: tdv +tga
V1 + V2 + V3 = v(t)dt = s(tdv + tga ) = Δx. (8)
0
2.4 Review of set parameters
Now there are two critical cases where either vmax or amax cannot be reached.
When calculating the time parameters, first the acceleration phase is build up
to reach maximum velocity. If time parameters coincide after this, Case 1 takes
effect. Second the starting time for deceleration tdv is defined. Case 2 describes
what occurs if tdv doesn’t fit to the other time parameters:
1.5 tf
position [m] final position
1
2 3
0.5 1
0
1.5
tdv tga
velocity [ ms ]
1 vmax
V1 V2
0.5
V2 V3 V1
0
acceleration [ sm2 ]
amax
2 A A2
1
A2 A3 A1
0
tsa
-2 tda
-4 tga twa tga
0 0.5 1 1.5 2
time [s]
Fig. 1. Calculation of the time parameters for a double-S trajectory on acceleration

level
Case 1: Set tsa too long / amax too high. The condition tda < tsa is
checked. If it is true, the acceleration phase would endure too long to reach the
value amax . Thus, the end effector would exceed vmax and overshoot its goal
position (i.e. the areas in (4) would be too large), as tsa can not be undercut.
Therefore, amax has to be reduced in order not to exceed the maximum velocity.
Case 2: Set distance too short / vmax too high. If time tda is suitable
to reach maximum velocity but the distance between start position and chosen
end position is very short, it might be possible that tdv < tga . Accordingly, the
deceleration phase would start before acceleration phase has finished. In this case
the maximum velocity must be reduced until tda exceeds tsa such that tdv ≥ 2tsa .
Reducing of vmax or amax . The sequence of manipulating the maxima of

velocity and acceleration can be traced by Fig. 2. After the values are set by the
operator Case 1 is checked with tda < tsa before Case 2 is considered. For low
motion input values, both requests won’t pertain and the values can be used to
calculated the trajectory.
To understand the following equations properly, say amax is suitable to reach
vmax within time tsa (then tda ≥ tsa and Case 1 doesn’t take effect), but vmax
is too high such that tdv < tga (Case 2). For this purpose, a rule is needed that
calculates vmax according to the given amax but considering tdv ≥ tga , which
can be derived from the edge case:
tdv = tga
Δx
with (2) ⇔ = tda + tsa
vmax
Δx
with (6) ⇔ = tga
vmax
Δx vmax
with (7) ⇔ = + tsa
vmax amax
2
tsa amax tsa amax
⇔ vmax =− ± + Δx amax . (9)
2 2
By altering vmax , all time parameters also change their values and it could
subsequently happen that tdv < tga (Case 1). If this occurs, amax has to be
lowered, but this would again alter vmax as it depends on amax (see Eq. 9)
and this would lead to undesired iterations modifying the maximum values. To
ensure real-time capabilities of the trajectory planner, a new maximum velocity
Δx
vmax = (10)
2tsa
is defined that describes the case of tf = 4tsa which is the lowest possible motion
time for the trajectory to reach the goal position, only depending on tsa . For the
velocity in Eq. 10, only one acceleration is possible:
vmax
amax = . (11)
tsa
This is a solution which is always feasible but it is only used if no higher values
can be calculated analytically.
If Case 1 takes effect before Case 2 is checked, amax is too high, so that
after the minimal time for acceleration (two times tsa ), the maximum velocity
would be exceeded (tda < tsa ). Thus, a lower acceleration is obtained by Eq. 11
regarding to the velocity set by the operator. Either this combination of param-
eters is already feasible, or critical Case 2 occurs with tdv < tga . Again, to avoid
iterative processes, Eq. 10 followed by Eq. 11 describe the maximum dynamic
parameters.
3 Conditional stop trajectory

By using a newly calculated deceleration phase, a stop trajectory can be im-
plemented. In comparison to a hard emergency stop by a hardware switch, the
system will not collapse if the conditional stop is executed, but decelerate to
zero velocity in a defined manner. Based on the previously shown calculations
of the trajectory planner, the new deceleration phase is also defined by a quintic
start
tda < tsa no tdv < tga no
yes yes
Case 1 Case 2
vmax 2
amax = vmax = − tsa a2max ±
t
sa amax
tsa 2 + Δx amax
time calculation time calculation
tdv < tga yes yes tda < tsa
no vmax =
Δx no
2tsa
vmax
amax =
tsa
time calculation
end
Fig. 2. Flow chart for reducing maximum velocity and acceleration
polynomial with the same input parameters as listed in 2.1. The advantage of
the new approach is that the stop trajectory can be initiated at any time while
the robot is moving, indifferent to the conducted trajectory segment.
The difference in the calculation of the conditional stop polynomial com-
pared to the procedure in Sec. 2 is in the settings of section 1 of the trajectory,
because the start acceleration is not necessarily equal to zero (as in Fig. 3). Fur-
thermore, the limits for velocity and deceleration can be set as before to ensure
system components (like maximum acceleration allowed for the payload) are not
violated.
By considering all input parameters, the stop trajectory can be divided into
two segments: In the first segment the acceleration will be shifted to the maxi-
mum deceleration value (or the minimum acceleration value respectively), after
which the acceleration will be shifted to zero in the second segment. Afterwards,
the velocity decreases to zero which is accomplished if the area underneath the
acceleration curve is equal to the velocity value when starting the stop trajec-
tory. It has to be considered that there is a segment with an area above the
abscissa if the end effector has a positive acceleration when the stop trajectory
is initiated. This leads to Eq. 12 illustrated in Fig. 3.
tstop
A2 + A3 + A4 − A1 = a(t)dt = vtstop . (12)
0
acceleration [ ms ]
1
0 2 3 4
-4
-8
0 0.5 1 1.5
time [s]
Fig. 3. Areas of a stop trajectory on acceleration level
An example for a stop trajectory initiated during the constant velocity phase
is illustrated at the first denoted time step in Fig. 4 (trajectory 1). It is discernible
that the robot starts to decelerate immediately with amax,b = −20 sm2 leading to
zero velocity after the stop trajectory is finished.
1
pos. [m]
0.5 Fig. 5
0
1.5
vel. [ ms ]
1
0.5
0
10 1 2 3
feasible [-] acc. [ sm2 ]
0
-10
-20
1
traj. 1
0.5 traj. 2
traj. 3
0
0 0.2 0.4 0.6 0.8 1
time [s]
Fig. 4. Simulation of three characteristic stop trajectories
3.1 Stop trajectory at workspace boundaries

Most considerations so far are not specific to CDPR. However, CDPRs have some
special properties that have to be taken into account when planning motions.
One major criterion is that CDPR are very sensitive to forces due to using cables
that may go slack below a tension fmin or break above a tension fmax . Therefore,
if a path contains points where no cable force configuration can be identified,
the robot needs to stop right before this point is reached. In case of offline path
planning, this is uncritical as the whole trajectory can be checked in advance. If
motions are planned during runtime, realtime conditions have to be met and the
trajectory cannot be checked at all points. Here, a different approach is proposed
where only the halt point (last point) of the stop trajectory is investigated in
every time step as a predictive check. The proposal is based on the following
hypothesis:
As the robot starts a trajectory with zero velocity and acceleration, the stop
trajectory at the starting point has no length. After starting to move, due to
continuous acceleration, the length of the stop trajectory increases continuously.
If now only the last point of each stop trajectory is validated for feasibility with
any braking force needed along the stop trajectory, all previous points of the stop
trajectory were validated in previous time steps as well. This must be done by
a high discretization of the trajectory to ensure no points of the calculated stop
trajectories will be skipped. To validate the last point, the following calculations
are made: Assume a simplified modeling of a fully constrained CDPR with n
degrees-of-freedom and m > n cables. This leads to an actuation redundancy
r = n − m. Using the structure matrix A T which is the transpose of the Jacobian
of the system, and loads including acceleration forces and torques w , the force
equilibrium delivers
AT f + w = 0 (13)
⇔ f = −A +T w + H λ (14)
where 0 < f min ≤ f ≤ f max . (15)
Here, f ∈ Rn denotes the n tendon forces, in the following also called force
distribution. The boundaries of the wire forces specified in Eq. 15 define a m-
dimensional hypercube C ⊂ Rm . All force distributions satisfying Eq. 13 form a r-
dimensional subspace S ⊂ Rm spanned by the kernel H = [h h1h 2 . . . h r ] (satisfies
A T h i = 0 , i = 1, . . . , r) of A T . Hence, if the intersection F of the hypercube C
and the subspace S is not empty, feasible solutions f fulfilling Eq. 15 and Eq. 13
exist, i. e. F = C ∩ S = ∅.
For r = 1 and r = 2, Eq. 14 is a parametric straight line or parametric plane,
respectively, where −A +T w is the support vector. By investigating a single path
point for different accelerations, the only varying value is the required wrench w .
Thus, for varying accelerations, the line or plane, respectively, are only shifted.
As the hypercube C is convex and all accelerations occur in the same direction,
it is sufficient to validate the existence of solutions F only for the minimum and
maximum acceleration values that the stop trajectory defines [8]. It is assumed
that this holds also for higher redundancies r which will be investigated as part
of future work.
According to this hypothesis the required length of the stop trajectory in
every path point is calculated and used to validate the corresponding halt point.
If no feasible cable forces are found, a stop signal will initiate the stop trajec-
tory. Trajectory 2 and 3 from Fig. 4 are examined in more detail in Fig. 5.
Trajectory 3 shows a case where the required stop signal is not triggered. The
feasibility variable ξg exhibits a non-feasible trajectory segment after exceeding
a trajectory length of approximately 0.62m. In comparison to this, trajectory 2
does not have a non-feasible segment, because a stop trajectory was initiated at
ξb = 0.65s due to the predictive check. Here, the robot stops at approximately
0.60m trajectory length with safety margin Δsb which is defined with respect to
the workspace boundary.
0.64
traj. 2
0.62 Δsg traj. 3
feasible [-] pos. [m]
0.6
0.58 Δsb
0.56
1 2
0.54
1
0.5 ξb ξg
0
0.6 0.65 0.7 0.75
time [s]
Fig. 5. Feasibility check along trajectory
Note, all calculations used up to now are performed in the direction of the
robot’s PTP movement. To obtain the joint variables in terms of position, ve-
locity, acceleration, jerk and snap, they have to be transformed by the structure
matrix A T . This allows for a final check to ensure all actuators can follow the
trajectory.
4 Experimental results
At the Chair of Mechatronics at the University of Duisburg-Essen the cable-
driven parallel robot SEGESTA is used to validate the presented trajectory plan-
ner. Therefore, a path is generated to test the trajectory planner. The end effector
starts at a position above the origin with z = 0.5m. To realize a diagonal move-
ment, the first goal position is at r 1 = [−0.05m, −0.20m, 0.60m]T . Afterwards,
the platform moves in y-direction to position r 2 = [−0.05m, 0.20m, 0.60m]T and
a stop signal is set along this way. The forward kinematics of the robot (e. g.
explained in [7]) gives the motion of the robot’s end effector which are seen in
Fig. 6. The parameters are set to:
• vmax,1 = 0.5 ms /vmax,2 = 1.0 ms (for segment 1 and 2)
• amax = 10 sm2 /amax,stop = 20 sm2
• tsa = 0.1s /tsa,stop = 0.01s
In segment 1, the robot moves as intended to reach the goal position r 1 . As
amax = 10 sm2 the acceleration is reduced to not exceed the maximum velocity
(vmax,1 = 0.5 ms ). In segment 2, the new velocity (vmax,2 = 1.0 ms ) is high enough
and the maximum acceleration does not have to be lowered. After moving con-
stantly with vmax,2 for a short period, a stop signal initiates the stop trajectory
(approx. at 2s) with a higher maximum acceleration amax,stop and a lower time
tsa,stop to decelerate to zero velocity immediately. In practice, this signal could
be triggered for example by a distance sensor. After a predefined break, the robot
starts moving again to the previous goal position. Here the remaining distance
is too short to reach maximum velocity vmax,2 and thus the velocity and accel-
eration are lowered by the algorithm.
A testing of the halt performance near the workspace borders on the real system
is still remaining, though it is tested in the simulation (Sec. 3).
0.7
abs. x y z abs. x y z
velocity [ ms ] position [m]
0.3
0
-0.3
1
0.5
0
1
0
-1
1
0
abs. x y z abs. x y z
10
0
acc. [ sm2 ]
-10
-20
20
10
0
100
jerk [ sm3 ]
0
-100
2000
0
0 1 2 3 4 5
time [s]
Fig. 6. Trajectory on the test bed SEGESTA
5 Conclusion
This paper describes the development of a point-to-point trajectory generator
especially for cable-driven parallel robots where the workspace boundaries are
not known in advance.
Parameters like velocity, acceleration or jerk can be limited by the operator
of the system. It is shown how the trajectory is calculated using a double-S
function with quintic polynomials on acceleration level with the required segment
time computations. As an advantage to previous calculations of trajectories for

cable-driven robots, even the jerk has a smooth progression. If velocity and/or
acceleration - given as an input by the operator - cannot be reached due to the
distance between start and goal pose the values will be reduced. Furthermore,
all values are defined in moving direction of the robot which is convenient for
several applications.
In addition to the point-to-point trajectory planning, a stop trajectory was
implemented based on the quintic polynomials calculation. If acceleration and
jerk are set according to limits of the payload, the platform can be stopped
in a defined manner. The approach avoids the computation of the workspace
boundaries in advance and therefore is proposed for real-time conditions.
6 Acknowledgements
This research received funding from the EFRE.NRW (2014-2020) Joint Research
Funding Programme of the European Union (EFRE) and the Ministry of Econ-
omy, Energy, Industry, and Handicrafts of the German Federal State of North
Rhine-Westphalia (NRW) under grant agreement EFRE-0800365 (ML-1-1-019B,
LEAN)].
References
1. Gosselin, C.: Global Planning of Dynamically Feasible Trajectories for Three-DOF
Spatial Cable-Suspended Parallel Robots. Conference Paper of CableCon2012 in:
T. Bruckmann et al. (eds.), Cable-Driven Parallel Robots, Mechanism and Machine
Science 12, Stuttgart 2012.
2. Kyriakopoulos, K. J., Saridis, G. N.: Minimum jerk path generation. In: Proc. IEEE
Int. Conf. Robotics and Automation, 1988, pp. 364-369
3. Piazzi, A., Visioli, A.: Global minimum-jerk trajectory planning of robot manipu-
lators. IEEE Trans. Ind. Electron., vol 47, pp. 140-149, Feb. 2000
4. Barnett, E., Gosselin, C.: Large-scale 3D printing with a cable-suspended robot.
Elsevier B.V., Canada 2015. http://www.elsevier.com/locate/addma
5. Erkorkmaz, K., Altintas, Y.: High speed CNC system design. Part I: jerk limited
trajectory generation and quintic spline interpolation. In: International Journal of
Machine Tools & Manufacture 41, Elsevier, 2001.
6. Biagiotti, L., Melchiorri, C.: Trajectory Planning for Automatic Machines and
Robots. Springer-Verlag Berlin Heidelberg 2008.
7. Reichert, C., Bruckmann, T.: Unified Contact Force Control Approach for Cable-
driven Parallel Robots using an Impedance/Admittance Control Strategy. Confer-
ence Paper: International Federation for the Promotion of Mechanism and Machine
Science World Congress, Taipei 2015.
8. Coxeter, H. S. M.: Regular Polytopes. Book published by Dover Publications, ISBN
978-0486614809, 3rd edition, 1973.
9. Müller, K., Reichert, C., Bruckmann, T.: Analysis of a Real-Time Capable Cable
Force Computation Method. Conference Paper of CableCon2014 in: A. Pott et al.
(eds.), Cable-Driven Parallel Robots, Mechanism and Machine Science 32, Duisburg
2014.
Part VI
Advanced Cable Modeling
JQ/2HBM; Q7 1HbiB+@6H2tB#H2 *#H2b rBi?
hBK2@o`vBM; G2M;i? 7Q` *#H2@.`Bp2M S`HH2H
_Q#Qib
S?BHBTT h2KT2H1 - .QM;rQM G222 - 62HBt h`mir2BM1 - M/ M/`2b SQii1

1
AMbiBimi2 7Q` *QMi`QH 1M;BM22`BM; Q7 J+?BM2 hQQHb M/ JMm7+im`BM; lMBib Aaq-
lMBp2`bBiv Q7 aimii;ì- .@dyRd9 aimiii;ì- :2`KMv-
T?BHBTTXi2KT2H!BbrXmMB@bimii;ìX/2
2
AMbiBimi2 Q7 TTHB2/ J2+?MB+b U*1V- lMBp2`bBiv Q7 aimii;ì- .@dy8eN aimii;ì-
:2`KMv
#bi`+iX *#H2@/`Bp2M T`HH2H `Q#Qib mb2 2HbiB+@~2tB#H2 +#H2b 7Q` QT@

2ìBQM /m2 iQ i?2B` /pMi;2b Qp2` `B;B/@HBMF DQBMibX AM bii2@Q7@i?2@ì
KQ/2HBM; Q7 +#H2@/`Bp2M T`HH2H `Q#Qib- +#H2b `2 KQbiHv FBM2KiB+b@
#b2/ M/ +QMiBM MQ 2tTHB+Bi +QMbB/2ìBQM Q7 i?2B` /vMKB+bX 1tT2`B@
K2MiH Q#b2`piBQMb b?Qr i?2b2 bBKTHB}+iBQMb /Q MQi ?QH/ i`m2 BM p`@
BQmb b+2M`BQb r?2`2 i?2 +#H2@/`Bp2M T`HH2H `Q#Qi #2+QK2b mM+QMi`QH@
H#H2X q2 `2pBbBi i?2 FBM2KiB+b@#b2/ +#H2 KQ/2Hb M/ T`2b2Mi +#H2
KQ/2H 2KTQr2`BM; *Qbb2ì `Q/ i?2Q`v 7Q` r?B+? i?2 /2~2+i2/ +QM};m`@
iBQM Bb 7Q`KmHi2/ i?`Qm;? ?B;?2`@Q`/2` "ûxB2` +m`p2bX LmK2`B+H iBK2@
BMi2;ìBQM Q7 i?2 /vMKB+b Bb T2`7Q`K2/ mbBM; M 2M2`;v@KQK2MimK
+QMb2`pBM; BMi2;ìBQM b+?2K2X h?2 TTHB+#BHBiv Q7 Qm` +#H2 KQ/2H Bb
2t2KTHB}2/ QM THM` +#H2 `Q#Qi rBi? 3 /2;`22b Q7 7`22/QKX
E2vrQ`/b, +#H2@/`Bp2M T`HH2H `Q#Qi- +#H2 KQ/2HBM;- KmHiB@#Q/v bBK@

mHiBQM- bvKTH2+iB+ BMi2;ìBQM
R AMi`Q/m+iBQM
*#H2@/`Bp2M T`HH2H `Q#Qib Ub?Qì +#H2 `Q#QibV- `2 T`HH2H `Q#Qib r?B+? mb2
2HbiB+ M/ ~2tB#H2j +#H2b BMbi2/ Q7 `B;B/ HBMFb b KQiBQM M/ 7Q`+2 i`MbKBbbBQM
2H2K2Mi iQ KQp2 KQ#BH2 THi7Q`KX lbBM; i?Bb b2imT 2M#H2b ?B;? /vMKB+b- ?B;?
TvHQ/b- M/ H`;2 rQ`FbT+2 bBM+2 i?2 KQpBM; BM2ìB MQr QMHv +QKT`Bb2b i?2
KQ#BH2 THi7Q`K M/ i?2 +QKTìBp2Hv HB;?i@r2B;?i +#H2bX >Qr2p2`- i?2b2 7@
pQ`#H2 +?`+i2`BbiB+b #`BM; 7Qì? i?2 /2bB;MǶb BM?2`2Mi ~r, +#H2b +M 2t2ì
QMHv i2MbBH2 7Q`+2b M/ b?Qr+b2 M2;HB;B#H2 `2bBbiM+2 iQ #2M/BM;X *QMb2[m2MiHv-
+#H2b i2MbBQM Kmbi Hrvb #2 F2Ti MQMx2`Q- r?B+? +M M2Bi?2` `mH2 Qmi +#H2b
#2BM; bH+F MQ` pQB/BM; +#H2 pB#ìBQMX qBi? +#H2b #2BM; i?2 KBM +imiBM;
+QKTQM2Mi Q7 +#H2 `Q#Qib- Bi Bb Q7 ;`2i BMi2`2bi 7Q` BKT`QpBM; biBzM2bb- ++m@
`+v- M/ +QMi`QH iQ #2ii2` mM/2`biM/ i?2B` /vMKB+b- 2bT2+BHHv i?2 BMi2`+iBQM
j
AM i?Bb rQ`F- 2HbiB+ +QMbB/2`b i?2 Q#D2+i iQ mM/2`;Q bi`BM /27Q`KiBQM- r?BH2 ~2tB#H2
+QMbB/2`b i?2 Q#D2+i iQ mM/2`;Q #2M/BM; /27Q`KiBQMX

296 P. Tempe et al.
#2ir22M +#H2b M/ KQ#BH2 THi7Q`KX JQ`2 M/ KQ`2 Q7i2M- `2b2`+? 2zQìb `2
/B`2+i2/ i KQ/2HBM; /vMKB+b Q7 +#H2b BM +#H2 `Q#Qib HHQrBM; 7Q` ?QHBbiB+
7Q`KmHiBQM Q7 HH bvbi2K /vMKB+bX
"b2/ QM i?2 biiB+ +i2M`v +#H2 KQ/2H /2`Bp2/ #v A`pBM2 URNd9V- EQxF
2i HX UkyyeV 2ti2M/2/ +#H2 `Q#Qi FBM2KiB+b +QMbB/2`BM; ?27iv +#H2bX GBm 2i HX
UkyRjV //2/ /vMKB+b iQ i?2 bvbi2K #v bmT2`BKTQbBM; i?2 HBM2` rp2 2[miBQM
QM i?2 +i2M`v b?T2 HHQrBM; 7Q` i`Mbp2`bH /2~2+iBQM Q7 i?2 +#H2- ?Qr2p2`-
rBi?Qmi +QMbB/2ìBQM Q7 #2M/BM;@BM/m+2/ 2HbiB+ +?M;2b M/ pHB/ QMHv BM FBM2@
iQbiiB+bX JMFH M/ ;`rH UkyyjV T`2b2Mi2/ KQ/2HBM; Q7 bi2HHBi2 i2i?2`2/
bvbi2Kb mbBM; M TT`Q+? bBKBH` iQ Qm`b mbBM; +QMiBMmmK K2+?MB+b- v2i 2K@
THQvBM; }MBi2 /Bz2`2M+2 TT`QtBKiBQM Q7 i?2 +#H2X aBKBH`Hv- .m M/ ;`rH
UkyR8V 7Q`KmHi2 TmHH2v@BM+Hm/BM; +#H2 KQ/2H #v mbBM; b?T2 7mM+iBQMb v2i
M2;H2+i #2M/BM; 7Q`+2bX
/Bz2`2Mi TT`Q+? Bb T`2b2Mi2/ #v JB+?2HBM 2i HX UkyR8V- r?Q /2`Bp2/
bBKmHiBQM M/ +QMi`QH KQ/2H Q7 M Qp2`+QMbi`BM2/ ?M;BM; +#H2 `Q#Qi mb@
BM; }MBi2 2H2K2Mi MHvbBb rBi? s.1X q?BH2 i?Bb TT`Q+? HHQrb 7Q` /2iBH2/
KQ/2HBM; Q7 i?2 r?QH2 +#H2 `Q#Qi- Bib +QKTmiiBQMH BM2{+B2M+v `2M/2`b Bi MQi
T`+iB+HHv TTHB+#H2X q2 T`2b2Mi2/ bBKBH` TT`Q+? iQ KQ/2HBM; +#H2 `Q#Qib
mbBM; b2;K2MiBx2/ +#H2 KQ/2H- ?Qr2p2`- i?2 KQ/2H /Q2b MQi HHQr 7Q` bBKTH2
BMi2;ìBQM Q7 iBK2@p`vBM; H2M;i?- MQ` Bb Bi MmK2`B+HHv r2HH@+QM/BiBQM2/ /m2 iQ
M2+2bbBiv Q7 KMv b2;K2Mib `2bmHiBM; BM /2Mb2 Kbb Ki`Bt Uh2KT2H 2i HX-
kyR3VX _2+2MiHv- vH *m2pb 2i HX UkyR3V T`2b2Mi2/ KQ/2H 7Q` bBKmHiBM;
THM` +#H2 `Q#Qib iFBM; BMiQ ++QmMi i?2 bTiBH KQiBQM Q7 +#H2b b r2HH b
`22HBM; Q7 i?2 Hii2`X *#H2 /2~2+iBQM Bb /2b+`B#2/ mbBM; TQHvMQKBH #bBb- ?Qr@
2p2`- +#H2 2ti2MbB#BHBiv Bb MQi +QMbB/2`2/ `2M/2`BM; Bi QMHv TìBHHv bmBi#H2 7Q`
TTHB+iBQMX
q?BH2 i?2`2 `2 T`QKBbBM; +#H2 KQ/2Hb rBi? /Bz2`2Mi H2p2Hb Q7 /2iBH- HBKB@
iiBQMb `Bb2 /m2 iQ i?2 bbmKTiBQMb K/2 Q` i?2 KQ/2H +QKTH2tBivX hQ `2K2/v
i?Bb b?Qì+QKBM;- i?Bb TT2` T`2b2Mib KQ/2H 7Q` bBKmHiBQM Q7 +#H2 `Q#Qib
rBi? KBM 7Q+mb QM KQ/2HBM; Q7 i?2 /vMKB+b Q7 i?2 +#H2bX GQM;Bim/BMH M/
i`Mbp2`bH KQiBQM `2 KQ/2H2/ i?`Qm;? +QMbB/2ìBQM Q7 +#H2 2HbiB+Biv M/
~2tB#BHBivX q2 mb2 _vH2B;?@_Bix 7Q`KHBbK 7Q` /vMKB+H bvbi2Kb M/ /Bb+`2iBx2
i?2 KQ/2HǶb bQHmiBQM bT+2 #v b2i Q7 #bBb 7mM+iBQMbX 6BMHHv- i?2 +#H2 KQ/2H Bb
+QK#BM2/ rBi? THM` THi7Q`K rBi? 3 /2;`22b Q7 7`22/QK M/ i?2 MmK2`B+H
MHvbBb M/ `2bmHib `2 T`2b2Mi2/X h?Bb TT2` Bb bi`m+im`2/ b 7QHHQrb, a2+iBQM k
T`2b2Mib Qm` +#H2 KQ/2H BM+Hm/BM; biiBM; Qm` bbmKTiBQMb M/ i?2 7Q`+2b /2@
`Bp2/- /2b+`B#2b i?2 }MH bvbi2K Q7 /Bz2`2MiBH@H;2#`B+ 2[miBQMb U.1V- b
r2HH b i?2 MmK2`B+H BMi2;ìBQM b+?2K2 mb2/ 7Q` Q#iBMBM; bQHmiBQMbX a2+iBQM j
T`2b2Mib `2bmHib 7`QK MmK2`B+H MHvbBb Q7 i?2 bmBi#BHBiv Q7 i?2 +#H2 KQ/2H #mi
HbQ 7`QK MmK2`B+H bBKmHiBQM Q7 i?2 +#H2 `Q#QiX
k .vMKB+b JQ/2HBM;
h?2 +#H2 `Q#Qi mM/2` BMp2biB;iBQM Bb THM` +#H2 `Q#Qi rBi? n = 3 /2;`22b Q7
7`22/QKěi`MbHiBQMH x M/ z- `QiiBQMH β #Qmi Êv @tBbě/`Bp2M #v m = 4
Modeling of Elastic-Flexible Cables with Time-Varying Length… 297
Λ = [0, L]
R χ(·,
X (·
) t )
Ξ = [0, 1] d2
d1
Êx Êx êx ∗
θ
λ∗ λ ∗
êt ξ
0 ξ∗ Êt 0
KP Êt
6B;X R, *#H2 FBM2KiB+b rBi? `272`2M+2 +#H2 +QM};mìBQM QM i?2 H27i bB/2 M/
/27Q`K2/ +QM};mìBQM QM i?2 `B;?i bB/2X
+#H2b BM +`Qbb2/@+#H2 +QM};mìBQMX AM i?2 7QHHQrBM; b2+iBQMb- r2 T`2b2Mi

i?2 +#H2 KQ/2H- i?2 KQ#BH2 THi7Q`K KQ/2H- b r2HH b i?2 +QK#BM2/ /vMKB+b
KQ/2HX
kXR *#H2 JQ/2H
6B;m`2 R b?Qrb i?2 FBM2KiB+ /2b+`BTiBQM Q7 i?2 +#H2 b #b2/ QM 2HbiB+ i?2Q`v
MK2Hv /2}M2/ i?`Qm;? *Qbb2ì `Q/bX Pi?2` i?M 1mH2`@"2`MQmHHB #2K i?2Q`v-
*Qbb2ì `Q/b HHQr 7Q` H`;2 /2~2+iBQMb- r?BH2 mMHBF2 hBKQb?2MFQ #2K i?2Q`v-
Bi Bb HbQ pHB/ 7Q` i?BM #2KbX MiKM Ukyy8V ;Bp2b M 2ti2MbBp2 /2`BpiBQM Q7
i?2 2[miBQMb Q7 KQiBQM Q7 bi`BM;b M/ `Q/b /27Q`KBM; BM bT+2X
G2i mb /2}M2 +m`p2 Q7 `272`2M+2 BM bT+2 #v i?2 irQ@/BK2MbBQMH bKQQi?
+m`p2

χ̃(λ, t) = X(λ, t), Z(λ, t)

URV
7Q` r?B+? r2 bbmK2 λ ∈ Λ := [0, L(t)] +m`pBHBM2` #b+Bbb2 Q7 K2+?MB+H BM@

i2`T`2iiBQM- L(t) i?2 iQiH H2M;i? Q7 Qm` +#H2 BM `272`2M+29 X h?2 irQ +QQ`/BMi2b
Q7 χ̃ /2}M2 TQBMi BM +#H2Ƕb +2Mi2` BX2X- QMƘBib M2mi`H tBb- r2 mb2 ++2Mi ( ˜· ) iQ
/2MQi2 T`K2i`BxiBQM BM λX q2 //BiBQMHHv /2}M2 i?2 iM;2Mi p2+iQ` d1(λ, t)
i χ̃(λ, t)X S2`72+i ~2tB#BHBiv Q7 i?2 bi`BM; `2[mB`2b d1 #2 iM;2Mi iQ +m`p2 χ̃ 7Q`
2+? λ M/ t U+7X MiKM Ukyy8VV- i?mb
χ̃(λ, t)
d1 = , UkV
χ̃(λ, t)
√
r?2`2 r2 /2}M2 (·) b ∂∂λ(·)
X 1[mBpH2MiHv 7QHHQrb d1 = γ̃ χ̃ M/ i?mb γ̃ = γ̃ γ̃
r?2`2 γ̃ = χ̃ X h?2 iM;2Mi iQ d1 - i?mb i?2 MQ`KH iQ χ̃- #2 /2}M2/ bm+? i?i Bi
7Q`Kb M Qì?QMQ`KH #bBb Q7 E2 rBi? d1 BX2X- d1 • d2 ≡ 0 7`QK r?B+? /B`2+iHv
7QHHQrb d2 = d1 X
AM Q`/2` iQ BMi`Q/m+2 +QMbiBimiBp2 Hrb Q7 /27Q`KiBQM- r2 /2}M2 ìBQb Q7
K2bm`2K2Mi Q7 i`MbHiBQMH M/ `QiiBQMH /27Q`KiBQMbX 6Q` i`MbHiBQMH /2@
9
6Q` bF2 Q7 `2/#BHBiv- r2 r`Bi2 L BM TH+2 Q7 L(t) 7Q` i?2 `2KBM/2` Q7 i?2 +QMi`B@
#miBQMX >Qr2p2`- iBK2@/2T2M/2M+2 Q7 i?2 +#H2 H2M;i? `2KBMbX
298 P. Tempe et al.
7Q`KiBQM- r2 BMi`Q/m+2 i?2 ìBQb

γ̃ θ˜ξ 1 det ([χ̃ , χ̃ ])
ε̃ := , M/ κ̃ = = , UjV
Γ̃ Γ̃ Γ̃ γ̃ 2
Q7 /27Q`K2/ H2M;i? γ̃ iQ mM/27Q`K2/ H2M;i? Γ̃ M/ Q7 #2K irBbi θ˜ξ BM Ki2`BH
+QQ`/BMi2bX
>pBM; /2}M2/ i?2 #bB+ FBM2KiB+b Q7 Qm` +#H2 KQ/2H mbBM; 1[bX URV iQ UjV
M/ TT`QT`Bi2 7Q`KmHiBQM Q7 /ǶH2K#2ìǶb T`BM+BTH2- r2 }M/ i?2 /vMKB+ 2[mB@
HB#`BmK Q7 7Q`+2b M/ pBìmH /BbTH+2K2Mib δq Q7 Qm` ;2M2`HBx2/ +QQ`/BMi2b q

0 = f(q, q̇) δq = (f/vM(q̈) − f2ti(q) + fBMi(q)) δq .
bbmKBM; /BbTH+2K2Mi 7mM+iBQM Q7 i?2 bKQQi? +m`p2 χ̃ HBM2` BM b2i
Q7 ;2M2`HBx2/ +QQ`/BMi2b `2/BM; χ̃(λ, t) = χ̃(λ, q(t)) r2 Q#iBM i?2 BMi2`MH
7Q`+2b fBMi - 2ti2`MH 7Q`+2b f2ti - M/ /vMKB+ 7Q`+2b f/vM b
L
∂ r̃(λ, q)
f˜2ti = A ê: dλ , U9V
∂q
0
L
∂ γ̃ ∂ θ˜ξ
f˜BMi = σ̃ + τ̃ dλ , U9#V
∂q ∂q
0
L
∂ r̃(λ, q) ∂ 2 r̃(λ, t)
f˜/vM = A dλ , U9+V
∂q ∂t2
0
BM r?B+? Bb i?2 mMBi /2MbBiv- A Bb i?2 +`Qbb b2+iBQMH `2- E Bb uQmM;Ƕb KQ/mHmb
Q7 2HbiB+Biv- I Bb i?2 b2+QM/ KQK2Mi Q7 `2X h?2 `2biQ`BM; BMM2` bi`2bb σ̃ M/
#2M/BM; KQK2Mib τ̃ 7QHHQr >QQF2Ƕb Hr Q7 2HbiB+Biv bm+? i?i σ̃ = E A (ε̃ − 1)
M/ τ̃ = E I κ̃X AM Q`/2` iQ 7mHHv /2`Bp2 i?2 2[miBQMb Q7 KQiBQM 7`QK 1[X U9V-
bmBi#H2 7Q`KmHiBQM Q7 i?2 /BbTH+2K2Mi 7mM+iBQM r̃(·, ·) Bb M22/2/X
q2 7QHHQr i?2 T`BM+BTH2 Q7 b2TìBQM Q7 p`B#H2b bTHBiiBM; i?2 /BbTH+2K2Mi
7mM+iBQM BMiQ iBK2@BM/2T2M/2Mi b2i Q7 #b2b 7mM+iBQMb Π̃(λ)- M/ BMiQ iBK2@
/2T2M/2Mi p`B#H2b q(t)- r?B+? MQr HbQ +i b Qm` ;2M2`HBx2/ +QQ`/BMi2b
r̃(λ, q(t)) = Π̃(λ) · q(t) . U8V
Ai Kv #2 MQi2/ i?i r2 KF2 QMHv p2`v 72r bbmKTiBQMb QM i?2 2Mi`B2b Q7 Π̃(λ)-
MK2Hv QMHv i?i i?2 b2i Q7 #b2b 7mM+iBQMb Bb irB+2 +QMiBMmQmbHv /Bz2`2MiB#H2
/m2 iQ 1[X UjVX
aBM+2 i?2 mTT2` BMi2;`H #QmM/`B2b Q7 1[X U9V `2 iBK2@/2T2M/2Mi- r2 BM@
i`Q/m+2 M2r iBK2@BM/2T2M/2Mi +m`pBHBM2` #b+Bbb2 ξ = λ/L(t) ∈ Ξ := [0, 1]
+QMp2ìBM; iBK2@p`vBM; bTiBH /QKBM [0, L(t)] BMiQ }t2/ /QKBM [0, 1] M/ iF@
BM; BMiQ ++QmMi i?2 /Bz2`2MiBiBQM Q7 ξ rBi? `2bT2+i iQ Qm` Q`B;BMH +m`pBHBM2`
#b+Bbb2 λ M/ iBK2 t
1 ∂ξ 1 ∂2ξ
ξ= λ, = , = 0, UeV
L ∂λ L ∂λ2
∂ξ L̇ ∂2ξ L̈L − L̇2

= − ξ, 2
=− ξ. Ue#V
∂t L ∂t L2
6`QK 1[X UeV 7QHHQr /B`2+iHv 7Q` i?2 /BbTH+2K2Mi 7mM+iBQM r(ξ, q(t)) M/ Bib
TìBH /2`BpiBp2b rX`XiX λ M/ t
d r̃ L̇ ∂ r ∂ r
r̃(λ, t) = r(ξ, t) , =− ξ + , UdV
dt L ∂ξ ∂t
∂ r̃ 1 ∂r d2 r̃ ∂ 2 r L̈ L − 2 L̇2 ∂ r L̇ ∂ 2 r
r̃ = = , = − ξ − 2 ξ + . . . Ud#V
∂λ L ∂ξ dt2 ∂t2 L2 ∂ξ L ∂ξ∂t
! "2
∂ 2 r̃ 1 ∂2r L̇ ∂2r
r̃ = = + ξ . Ud+V
∂λ2 L2 ∂ξ 2 L ∂ξ 2
am#biBimiBM; 1[bX U8V iQ UdV BMiQ 1[X U9V- mM/2` +QMbB/2ìBQM Q7 i?2 `mH2 7Q`
BMi2;ìBQM #v bm#biBimiBQM BX2X- dλ = L dξ- r2 Q#iBM i?2 7QHHQrBM; i`Mb+`B#2/
7Q`+2b
1
∂ r(ξ, q)
f2ti = A ê: L dξ , U3V
∂q
0
1
∂γ ∂ θξ
fBMi = σ +τ dξ , U3#V
∂q ∂q
0
1
∂ r(ξ, q) ∂ 2 r(ξ, t)
f/vM = A L dξ + . . .
∂q ∂t2
0
1
! "2
∂ r(ξ, q) L̇ ∂ 2 r(ξ, t)
+ A ξ L dξ + . . . U3+V
∂q L ∂ξ 2
0
1
! "
∂ r(ξ, q) L̈ L − 2 L̇2 ∂ r(ξ, t) L̇ ∂ 2 r(ξ, t)
− A ξ + 2 ξ L dξ ,
∂q L2 ∂ξ L ∂ξ∂t
0
LQi2 i?2b2 BMi2;`Hb +MMQi #2 bQHp2/ MHviB+HHv b bm+? :mbbBM [m/ìm`2

Bb 2KTHQv2/ rBi? M TT`QT`Bi2 MmK#2` Q7 [m/ìm`2 TQBMib /2T2M/BM; QM i?2
/2bB`2/ TQHvMQKBH /2;`22X
"v K2Mb Q7 /ǶH2K#2ìǶb T`BM+BTH2 bm#biBimiBM; i?2 7Q`+2b ;Bp2M BM 1[X U3V-
r2 Q#iBM b2+QM/@Q`/2` /Bz2`2MiBH@H;2#`B+ 2[miBQM Q7 i?2 7Q`K
M q̈ = f(q, q̇) − DΦ(q, t) λ ,

UNV
0 = Φ(q, t) .
>QHQMQKB+ +QMbi`BMib Φ(q, t)- rBi? i?2B` C+Q#BM DΦ = ∂∂qΦ - `2bmHi 7`QK
`2[mB`BM; T`QtBKH +#H2 TQBMi (ξ = 0) M/ /BbiH +#H2 TQBMi (ξ = 1) iQ iiBM
300 P. Tempe et al.
;Bp2M TQbBiBQM vB2H/BM;

ΦT(q, t) r(ξ = 0, t) − rT(t)

0= = ,
Φ/(q, t) r(ξ = 1, t) − r/(t)
rBi? rT M/ r/ /2bB`2/ T`QtBKH M/ /BbiH TQbBiBQMX q?2M +QHH2+iBM; HH +#H2b
M/ i?2 KQ#BH2 THi7Q`K BM Qm` +#H2 `Q#Qi 7`K2rQ`F- i?2b2 +QMbi`BMib `2 ;Bp2M
i?`Qm;? i?2 rBM+? TQbBiBQMb M/ i?`Qm;? i?2 KQ#BH2 THi7Q`K M+?Q` TQbBiBQMb-
`2bT2+iBp2HvX
*?QB+2 Q7 "bBb 6mM+iBQMb q2 `2[mB`2 Qm` #b2b 7mM+iBQMb Π(ξ) iQ #2 irB+2

+QMiBMmQmbHv /Bz2`2MiB#H2 U+7X 1[X UjVV BM ξX .Bz2`2Mi #b2b 7mM+iBQMb +M #2
7QmM/ 7Q` mMB7Q`K #2Kb U+7X UhM;- kyyjVV- ?Qr2p2`- i?2B` ;2M2`H TTHB+#BHBiv
Bb HBKBi2/ /m2 iQ i?2 bbmKTiBQMb K/2 2X;X- TBMM2/@7`22 Q` +HKT2/@7`22X KQ`2
bBKTH2 M/ mMBp2`bH b2i Q7 #b2b 7mM+iBQMb +M #2 7QmM/ BM "ûxB2` +m`p2b- r?B+?
`2 p2`v +QKKQM rv iQ /2b+`B#2 bKQQi? +m`p2b BM 2X;X- +QKTmi2` ;`T?B+b Q`
MBKiBQMX "ûxB2` +m`p2b `2 +QKTQb2/ Q7 d + 1 TQHvMQKBHb Q7 /2;`22 d- rBi?
i?2 ν@i? "2`Mbi2BM #bBb TQHvMQKBHb Qp2` ξ ∈ [0, 1] Bb /2}M2/ b

d ν d−ν
bν,d(ξ) = ξ (1 − ξ) ,
ν
r?2`2 νd Bb i?2 #BMQKBH +Q2{+B2MiX irQ@/BK2MbBQMH "ûxB2` +m`p2 Bb ;Bp2M b

i?2 r2B;?i2/ HBM2` +QK#BMiBQM Q7 n+i`H "2`Mbi2BM #bBb TQHvMQKBHb Q7 /2;`22 d

bm+? i?i Bi `2/b
n+i`H

Bd(ξ) = bν,d(ξ) Pν = Π(ξ) P ,
ν=0
r?2`2 Pν ∈ R2×1 - Uν = 1, . . . , n+i`H V- `2 i?2 irQ@/BK2MbBQMH +QQ`/BMi2b Q7 i?2

+m`p2Ƕb +QMi`QH TQBMiběr?B+? r2 BMi2`T`2i b Qm` ;2M2`HBx2/ +QQ`/BMi2bX lHiB@
Ki2Hv- bvbi2K +QKTH2tBiv M/ ++m`+v Bb /2i2`KBM2/ #v i?2 +?Qb2M TQHvMQKBH
/2;`22X
kXk SHi7Q`K JQ/2H

qBi? Qm` 2tKTH2 Q7 THM` +#H2 `Q#Qi Q7 ;2M2`HBx2/ +QQ`/BMi2b qS = [x , z , β] -
i?2 THi7Q`K /vMKB+b KQ/2H 7QHHQrb 7`QK 2X;X- G;`M;BM K2+?MB+b iQ `2/
MS q̈S = wS − DΦ(qS , t) λS ,
URyV
0 = ΦS(qS , t) ,

BM r?B+? wS = [0 , mS g , 0] Bb i?2 2ti2`MH r`2M+?X o2+iQ` Q7 +QMbi`BMib ΦS(qS , t)
Bb +QKTQb2/ Q7 m +QMbi`BMi p2+iQ`b- r?2`2 i?2 i@i? +QMbi`BMi p2+iQ` 7Q` 2+?
THi7Q`K M+?Q` Bb ;Bp2M b
Φi(qS , t) = r + RS bi − ri(t) ,

rBi? THi7Q`K TQbBiBQM r = x, z - THi7Q`K `QiiBQM RS = Rx(β)- M/ ri(t)

+QMbi`BM2/ TQbBiBQM Q7 i?2 i@i? M+?Q`X

kXj *QK#BM2/ .vMKB+b JQ/2H

hQ bBKmHi2 +#H2 `Q#Qi rBi? QM2 THi7Q`K M/ m = 4 +#H2b- r2 +QM+i2@
Mi2 i?2 /vMKB+b Q7 Qm` +#H2 KQ/2H ;Bp2M i?`Qm;? 1[X UNV 7Qm` iBK2b M/ i?2
THi7Q`K /vMKB+b 7`QK 1[X URyVX h?2 +QK#BM2/ bii2 p2+iQ` i?2M `2/b

q+/T` = qT , q+1 q+2 q+3 q+4

7Q` r?B+? r2 7Q`KmHi2 i?2 7mHH bvbi2K /vMKB+b bBKBH` iQ 1[X UNV Q` 1[X URyV-
r?2`2 Kbb Ki`Bt Bb #HQ+F@/B;QMH +QKTQb2 Q7 KQ#BH2 THi7Q`KǶb Kbb K@
i`Bt MKT M/ Q7 i?2 m = 4 +#H2bǶ Kbb Ki`B+2b M+i - (i = 1, . . . , 4)X q2 b2i
mT i?2 p2+iQ` Q7 +QMbi`BMib bm+? i?i TQbBiBQM Q7 i?2 i@i? +#H2Ƕb T`QtBKH TQBMi
BX2X- Φ+i ,T Bb b2i iQ #2 i `Q#Qi 7`K2 M+?Q` ai BX2X-
Φ+i ,T(q+i , t) = r+i (ξ = 0, t) − ai
r?2`2b i?2 +#H2Ƕb /BbiH TQBMi Bb #QmM/ iQ i?2 THi7Q`K i?`Qm;?
Φ+i ,/(q+i , qKT , t) = r+i (ξ = 1, t) − (r + RS bi ) .
j LmK2`B+H _2bmHib
TTHB+#BHBiv Q7 i?2 +#H2 KQ/2H T`2b2Mi2/ BM i?Bb +QMi`B#miBQM Bb ;Bp2M #v +QK@
T`BM; Mim`H 7`2[m2M+B2b Q7 Qm` +#H2 KQ/2H ;BMbi i?Qb2 Q7 /Qm#H2 bmTTQì2/
#2K b r2HH b MmK2`B+H bBKmHiBQM Q7 M TT`QT`Bi2 +#H2 `Q#QiX
jXR LmK2`B+H AMi2;ìBQM a+?2K2

Pm` K2+?MB+H bvbi2K +QKTQb2/ Q7 2HbiB+@~2tB#H2 +#H2b M/ `B;B/ THi7Q`K
Bb bm#Q`/BMi2 iQ /Bz2`2MiBH@H;2#`B+ 2[miBQM U.1V Q7 BM/2t i?`22ěQMHv
+QMbi`BMib QM TQbBiBQM H2p2H `2 2tTHB+BiHv ;Bp2MX q?BH2 p`B2iv Q7 BMi2;`@
iBQM b+?2K2b 2tBbi 7Q` b2+QM/@Q`/2` BM/2t@i?`22 .1b- Bi Bb 7pQ`#H2 iQ mb2
BMi2;ìBQM b+?2K2b /2bB;M2/ 2tTHB+BiHv 7Q` K2+?MB+H bvbi2KbX am+? BMi2;ìQ`b
2Mbm`2 +QMbBbi2Mi BMi2;ìBQM Q7 bm#biMiBH K2+?MB+H T`QT2ìB2b HBF2 2M2`;v Q`
KQK2MimK- /2bTBi2 /Bb+`2iBxiBQM Q7 +QMiBMmQmb .1 rBi? `2bT2+i iQ iBK2X
"2ib+? M/ l?H` UkyydV T`2b2Mi M 2M2`;v@KQK2MimK +QMb2`pBM; BMi2;ìBQM
b+?2K2 i?i HHQrb 7Q` 2tTHB+Bi bQHpBM; Q7 i?2 2[miBQMb Q7 KQiBQM QM i?2 H2p2H Q7
TQbBiBQMb r?BH2 QMHv `2[mB`BM; TQbBiBQMb M/ p2HQ+BiB2bX h?2 BMi2;ìBQM b+?2K2
Bb /2b+`B#2/ b 7QHHQrb, bbmK2 Qm` K2+?MB+H bvbi2K iQ #2 ;Bp2M i?`Qm;? i?2
b2i Q7 /Bz2`2MiBH@H;2#`B+ 2[miBQMb
M q̈ = −∇U(q) + DΦ λ , URRV
0 = Φ(q, t) , URR#V
BM r?B+? ∇U(q) `2 7Q`+2b 7`QK TQi2MiBHbX 6Q` iBK2 BMi2`pH Tn = [tn , tn+1 ] rBi?
[mMiBiB2b (qn , vn ) Q7 iBK2 tn - M/ KB/@TQBMi pHm2 (·)n+ 12 = ((·)n + (·)n+1 )/2-
i?2 QM2@bi2T BMi2;ìBQM b+?2K2 rBi? bi2T bBx2 h = tn+1 − tn - Bb ;Bp2M #v
qn+1 − qn = h vn+ 21 , URkV
302 P. Tempe et al.
M (vn+1 − vn ) = −h ∇U(qn , qn+1 ) − h ∇Φ(qn , qn+1 ) λ̄ , URk#V

0 = ∇Φ(qn , qn+1 ) vn+ 12 , URk+V
BM r?B+? ∇(·) Bb i?2 /Bb+`2i2 /2`BpiBp2 Q7 :QMxH2x URNNeV- /2}M2/ 7Q` 7mM+@
iBQM f : Rn → Rm i Mv irQ TQBMib x, y ∈ Rn rBi? v = y − x = 0
M/ z = (x + y)/2 b
f(y) − f(x) − ∇f(z) v

∇f(x, y) = ∇f(z) + v .
v v
1[miBQM URkV Bb i?2M bQHp2/ 7Q` (qn+1 , vn+1 ) M/ λ̄ #v 2KTHQvBM; i?2 KB/@
TQBMi `mH2 iQ 2HBKBMi2 i?2 mMFMQrM p2HQ+BiB2b vn+1 vB2H/BM; i?2 `2KBMBM; 2[m@
iBQMb
2
0= M (qn+1 − qn ) − 2 M vn + . . .
h URjV
+ h ∇U(qn , qn+1 ) + h ∇Φ(qn , qn+1 ) λ̄ ,
0 = Φ(qn+1 ) . URj#V
1[miBQM URjV TQb2b bvbi2K Q7 MQM@HBM2` 2[miBQMb ;Bp2M BM `2bB/mH p2+@

iQ` 7Q`K 0 = e(qn+1 )- r?B+? Bb bQHp2/ 7Q` qn+1 #v 2KTHQvBM; /KT2/ L2riQMǶb
K2i?Q/X q?BH2 i?Bb Bi2ìBp2 b+?2K2 iQ }M/BM; qn+1 Bb +QKTmiiBQMHHv KQ`2
2tT2MbBp2 i?M 2tTHB+BiHv bQHpBM; 1[X URRV 7Q` q̈- 1[X URjV T`QpB/2b M 2M2`;v
M/ KQK2MimK@+QMb2`pBM;- iBK2@/Bb+`2i2 BMi2;ìBQM b+?2K2 Q7 1[X URRVX *QM@
p2MiBQMH P.1 Q` .1 bQHp2`b /Q MQi BM?2`2MiHv 2Mbm`2 #2BM; K2+?MB+HHv +QM@
b2`pBM; U"2ib+? M/ ai2BMKMM- kyykV- b bm+? QM2 +MMQi 2Mbm`2 +QMbBbi2M+v Q7
i?2 bvbi2KǶb FBM2KiB+ #`M+?X AM //BiBQM- 2tTHB+BiHv ?M/HBM; ?QHQMQKB+ +QM@
bi`BMib /Q2b MQi BMi`Q/m+2 Tb2m/Q@/vMKB+b QM i?2 bvbi2K- r?2`2b +QMbi`BMi
pBQHiBQM ?M/HBM; K2i?Q/b bm+? b "mK;ì2 bi#BHBxiBQM Q` :2`@:mTi@
G2BKFɃ?H2` z2+i Qp2`HH bvbi2K /vMKB+b U`MQH/- kyy9VX
jXk TT`QtBKiBQM Q7 Lim`H 6`2[m2M+B2b
JQ/2H +QKTH2tBiv M/ ++m`+v Bb /2i2`KBM2/ /B`2+iHv #v i?2 MmK#2` Q7 ;2M2`H@

Bx2/ +QQ`/BMi2b Q`- 2[mBpH2MiHv- #v i?2 TQHvMQKBH /2;`22 Q7 i?2 "ûxB2` +m`p2bX
b +#H2 +M #2 +QMbB/2`2/ bH2M/2` `Q/ Q` #2K Q7 HQr tBH M/ ~2tm`H `B;B/@
Biv- +QKT`BbQM ;BMbi Mim`H 7`2[m2M+B2b Q7 #2K bm;;2bib Bib2H7X h?2 KQbi
/B`2+i +QKT`BbQM Q7 Qm` +#H2 KQ/2H Bb irQ7QH/ bBKTH2@bmTTQì2/ #2KƘ7Q`
r?B+? i?2 2[miBQMb Q7 KQiBQM +M #2 bQHp2/ MHviB+HHvX 6`QK i?2`2- i?2 k@i?
2B;2MpHm2 λk vB2H/b i?2 k@i? Mim`H 7`2[m2M+v
2
# kπ EI
ωk = λk = .
L A
*QMp2`;2M+2 Q7 i?2 Mim`H 7`2[m2M+B2b Q7 Qm` +#H2 KQ/2H Bb b?QrM BM 6B;X kX

Ai Bb TT`2Mi i?i ?B;?2` Q`/2` TQHvMQKBHb +?B2p2 #2ii2` TT`QtBKiBQM Q7
LmK2`B+H MHviB+H .Bz2`2M+2
TQHvMQKBH /2;`22 d = 3 TQHvMQKBH /2;`22 d = 4

3
10 103
100 100
Δωk /rad/s
ωk /rad/s
10−3 10−3
10−6 10−6
10−9 10−9
10−12 10−12
10−15 10−15
TQHvMQKBH /2;`22 d = 12 TQHvMQKBH /2;`22 d = 20

103 103
100 100
Δωk /rad/s
ωk /rad/s
10−3 10−3
10−6 10−6
10−9 10−9
10−12 10−12
10−15 10−15
1 4 8 12 16 20 24 1 4 8 12 16 20 24
JQ/2 k/( · ) JQ/2 k/( · )
6B;X k, Lim`H 7`2[m2M+B2b Q7 i?2 T`2b2Mi2/ "ûxB2` +m`p2 +#H2 KQ/2H 7Q` MmK#2`
Q7 +QMi`QH TQBMib n+i`H ∈ { 3, 4, 12, 20 }, i?2 KQ`2 +QMi`QH TQBMib `2 mb2/- i?2
#2ii2` TT`QtBKiBQM Q7 HQr2` Mim`H 7`2[m2M+B2b +M #2 +?B2p2/- ?Qr2p2`-
/`biB+ Qp2`2biBKiBQM Q7 ?B;?2` 7`2[m2M+B2b 2K2`;2b- iQQX
h#H2 R, aBKmHiBQM T`K2i2`b 7Q` 2M2`;v@KQK2MimK +QMb2`pBM; BMi2;ìBQM Q7

i?2 .1 bvbi2KX
S`QT2ìv oHm2 Unit
ai2T bBx2 h 1 × 10−3 s
_2bB/mH iQH2`M+2 εf 5 × 10−9
h`MbBiBQM biì iBK2 a 4 s
h`MbBiBQM 2M/ iBK2 b 5 s
HQr2`@Q`/2` Mim`H 7`2[m2M+B2b Q7 bBKTH2@bmTTQì2/ #2K- r?B+? Bb iQ #2

2tT2+i2/X PM i?2 Qi?2` ?M/- Qp2`2biBKiBQM Q7 ?B;?2`@Q`/2` Mim`H 7`2[m2M@
+B2b +M #2 b22M BM i?2 /i- v2i i?2 MmK#2` Q7 Qp2`2biBKiBQMb U`2HiBp2 2`@
`Q` Δωk ≥ 0.25V `2KBMb rBi?BM `2bQM#H2 #QmM/b- 2p2M rBi? 12@i? /2;`22
TQHvMQKBH U7 Mim`H 7`2[m2M+B2b `2 rBi?BM i?2 #QmM/bVX lHiBKi2Hv- r2 +QM@
+Hm/2 i?i HQr@/BK2MbBQMH bvbi2K rBi? QMHv 8 ;2M2`HBx2/ +QQ`/BMi2b bm{@
+B2MiHv TT`QtBKi2b i?2 }`bi 4 Mim`H 7`2[m2M+B2bX
jXj LmK2`B+H aBKmHiBQM
LmK2`B+H bBKmHiBQM Bb T2`7Q`K2/ bm+? i?i Bi T`QKBM2MiHv T`QpQF2b +#H2 KQ@

iBQM BM bT+2 r?B+? r2 +?B2p2 #v p2ìB+HHv KQpBM; i?2 THi7Q`K i ?B;? /vMK@
B+bX h?2 BMTmi T`Q}H2 BM p2ìB+H /B`2+iBQM 7QHHQrb MQM@MHviB+ bKQQi? HQ;BbiB+b
304 P. Tempe et al.
h#H2 k, :2QK2i`B+H M/ K2+?MB+H T`QT2ìB2b Q7 +#H2b M/ Q7 7`K2 M/

KQ#BH2 THi7Q`K # mb2/ BM bBKmHiBQMX
UV S`QT2ìB2b Q7 +#H2bX U#V S`QT2ìB2b Q7 7`K2 M/ THi7Q`KX
S`QT2ìv oHm2 Unit S`QT2ìv oHm2 Unit

6`K2 SHi7Q`K
.BK2i2` D 6 mm
.2MbBiv 7850 kg/m3 qB/i? 2.00 0.50 m
a2+QM/ KQK2Mi Q7 `2 I 63.62 mm4 >2B;?i 2.00 0.50 m
1HbiB+ JQ/mHmb E 5 MPa Jbb ě 1 kg
UV, iBK2 t = 0 s U#V, iBK2 t = 4.000 s U+V, iBK2 t = 4.500 s

1
o2ìB+H z/m
0.5
−0.5
−1
U/V, iBK2 t = 5.000 s U2V, iBK2 t = 5.500 s U7V, iBK2 t = 7.500 s

1
o2ìB+H z/m
0.5
−0.5
−1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
>Q`BxQMiH x/m >Q`BxQMiH x/m >Q`BxQMiH x/m
6B;X j, aMTb?Qib Q7 bBKmHiBQM `2bmHib i /Bz2`2Mi iBK2b b?QrBM; b;;BM; M/

pB#ìBQM Q7 i?2 2HbiB+@~2tB#H2 +#H2 `Q#Qi KQ/2H b Bi Bb #2BM; KQp2/ mTr`/bX
"H+F +B`+H2b `2T`2b2Mi rBM+?2b- +#H2b `2 /`rM `2/- i?2 KQ#BH2 THi7Q`K /`rM
;`vX
7mM+iBQM +?Qb2M 7Q` `2bQMb Q7 `2/m+BM; MmK2`B+H BMbi#BHBiB2b /m`BM; iBK2 BM@
i2;ìBQM `BbBM; 7`QK D2`Fv KQiBQMX 1z2+iBp2 mMbi`BM2/ +#H2 H2M;i? Li Bb +H@
+mHi2/ mbBM; biM/`/ BMp2`b2 FBM2KiB+b HQQT rBi? rBM+? M+?Q` TQbBiBQMb ai -
THi7Q`K M+?Q`b bi - THi7Q`K TQbBiBQM M/ Q`B2MiiBQM r- RS - `2bT2+iBp2Hv- bm+?
i?i Li = ai − (r + R bi )- M/ mb2/ b biM/`/ 722/ 7Q`r`/ +QMi`QH BMTmiX
aBKmHiBQM T`K2i2`b `2 ;Bp2M BM h#H2 R- K2+?MB+H M/ ;2QK2i`B+H T`QT@
2ìB2b `2 ;Bp2M BM h#H2 kX
"27Q`2 KQiBQM biìb- +#H2 b; Bb TT`2Mi BM TìB+mH` BM i?2 HQr2` +#H2b
Ub22 6B;X kUVV- #mi HbQ 2HbiB+Biv Q7 i?2 mTT2` +#H2b +mbBM; p2ìB+H /2~2+iBQM Q7
i?2 THi7Q`K +2Mi2` Q7 Δz −0.126 mVX qBi? i?2 rBM/BM; biìBM; i t = 4.000 s-
mTT2` +#H2b `2 b?Qì2M2/ +mbBM; i?2K iQ b; H2bb- v2i /m2 iQ i?2B` 2HbiB+Biv-
i?2 THi7Q`K /Q2b MQi KQp2 mMiBH t = 5.000 s Ub22 6B;bX kU#V M/ kU+VVX b i?2
THi7Q`K #2;BMb #2BM; ++2H2ì2/ #v i?2 mTT2` irQ +#H2b- M Qb+BHHiBM; rp2
biìb iQ 7Q`K i i?2 HQr2` +#H2bǶ THi7Q`K ii+?K2Mi TQBMib M/ Bb #2BM;
T`QT;i2/ i?`Qm;? i?2 +#H2X h?Bb mHiBKi2Hv H2/b iQ pB#ìQ`v KQiBQM Q7 i?2
HQr2` +#H2běM 2K2`;BM; rp2 +M #2 b22M BM 6B;bX kU2V M/ kU7V r?B+? HbQ
BMi2`+ib rBi? i?2 THi7Q`KX h?mb- i?2 BM/m+2/ rp2 Q7 KQiBQM Q7 i?2 HQr2`
+#H2b Bb T`QT;i2/ i?`Qm;? i?2 THi7Q`K M/ QMiQ i?2 mTT2` irQ +#H2bX aBM+2
i?2`2 Bb M2Bi?2` K2+?MB+H MQ` MmK2`B+H /KTBM; KQ/2H2/- i?2 rp2 biìb
i`MbKB;ìBM; 7`QK i?2 HQr2` +#H2b i?`Qm;? i?2 THi7Q`K iQ i?2 mTT2`- M/
#+F ;BMX
9 *QM+HmbBQMb
q2 T`2b2Mi2/ +#H2 `Q#Qi 7`K2rQ`F 7Q` bBKmHiBQM Q7 +#H2 `Q#Qib rBi? +@
#H2b #b2/ QM K2+?MB+H +QMiBMmmK 7Q`KmHiBQM #b2/ QM *Qbb2ì `Q/X lHiB@
Ki2Hv- i?2 KQ/2H T`QpB/2b +HQb2 TT`QtBKiBQM Q7 i?2 Mim`H 7`2[m2M+B2b Q7
+#H2 r?BH2 F22TBM; i?2 /2;`22b Q7 7`22/QK HQr UΔω ≤ 5 % rBi? M 8@i? /2;`22
"ûxB2` +m`p2VX //BiBQMHHv- i?2 KQ/2H BM?2`2MiHv +Tim`2b 2HbiB+ /27Q`KiBQM
Ubi`BMV- ~2tB#H2 /27Q`KiBQM U#2M/BM;V- b r2HH b iBK2@p`vBM; H2M;i? r?B+?
mHiBKi2Hv HHQrb 7Q` 7mHH bBKmHiBQM Q7 THM` +#H2 `Q#QiX b bm+?- Bi Bb
MQi2rQì?v 2ti2MbBQM Q7 `2Hi2/ rQ`F BM i?Bb `2 M/ ?QH/b ;`2i T`QKBb2 7Q`
2M#HBM; KQ`2 `2HBbiB+ bBKmHiBQM Q7 +#H2 `Q#QibX *?QQbBM; bvKTH2+iB+ BX2X-
2M2`;v@KQK2MimK +QMb2`pBM; BMi2;ìQ`- r2 +M bbm`2 +QMbBbi2Mi MmK2`B+H BM@
i2;ìBQM rBi?Qmi z2+iBM; i?2 bBKmHiBQM `2bmHi #v r`QM; +?QB+2 Q7 +QMbi`BMi
pBQHiBQM ?M/HBM;X
Pm` BMi2`2bi Bb MmK2`B+HHv 2{+B2Mi BKTH2K2MiiBQM Q7 i?2 +#H2 `Q#Qi bBKm@
HiBQM 7`K2rQ`FX b Q7 iBK2 #2BM;- i?2 bBKmHiBQM T`2b2Mi2/ BM i?Bb +QMi`B#miBQM
+HQ+Fb BM i 15 min- rBi? BM+`2bBM; bBKmHiBQM iBK2 7Q` KQ`2 `2HBbiB+ K2+?M@
B+H T`K2i2`bX h?Bb 2M+mK#`M+2 iQ [mB+F bBKmHiBQM bi2Kb H`;2Hv 7`QK i?2
Bi2ìBp2 L2riQMǶb K2i?Q/ mb2/ iQ bQHp2 i?2 `2bB/mH 2[miBQM Q7 i?2 /vMK@
B+bX 1Bi?2` HQQb2MBM; i?2 iQH2`M+2 #QmM/b Q` BM+`2bBM; bBKmHiBQM bi2T bBx2 rBHH
H2/ iQ [mB+F2` `QQi }M/BM;- ?Qr2p2`- Bi Kv i?2M #2 mM+H2` B7 i?2 +#H2 `Q#Qi
bivb rBi?BM i?2 bK2 FBM2KiB+ #`M+? Bi biì2/ BMX :Bp2M bKHH bi2T bBx2b M/
i?2 2M2`;v +QMbBbi2M+v Q7 i?2 MmK2`B+H BMi2;ìBQM b+?2K2 7`QK a2+iBQM jXR- r2
+M p2`v b72Hv bii2 MQ i`MbBiBQM #2ir22M /D+2Mi FBM2KiB+ #`M+?2b Q++m`bX
GBF2rBb2- p`B#H2 bi2T@bBx2 bQHp2`b b?Qr i?2 bK2 b?Qì+QKBM;b iQ BMi2;ìBM;
bvbi2Kb 2M2`;v@+QMbBbi2Mi b r2HH b 2Mbm`BM; MQ KDQ` +?M;2b Q7 i?2 bvbi2KǶb
FBM2KiB+ bii2X GbiHv- i`MbBiBQMBM; 7`QK Q#D2+i@Q`B2Mi2/ JhG" iQ *f*YY
Qm;?i H2/ iQ KQ`2 2{+B2Mi BMi2;ìBQMX
GbiHv- i?`22@/BK2MbBQMH 7Q`KmHiBQM Q7 i?2 +#H2 KQ/2H Bb i?2 mHiBKi2
;QH- /2bTBi2 Bi #2BM; H2bb i`BpBH i?M i?2 irQ@/BK2MbBQMH +b2X aBM+2 +#H2
#2M/BM; Bb MQ HQM;2` +QM}M2/ iQ QM2 THM2ěBM Qm` +b2 i?2 êt êx @THM2ě- #mi
+QK#BMiBQM Q7 #2M/BM; BM HH i?`22 bTiBH THM2b- Bi `2[mB`2b /Bz2`2Mi K2bm`2b
Q7 bi`BM M/ #2M/BM; /27Q`KiBQMX "v K2Mb Q7 /Bz2`2MiBH ;2QK2i`v- r2 rBHH #2
306 P. Tempe et al.
#H2 iQ /2b+`B#2 i?2 +#H2 b?T2 M/ Bib /27Q`KiBQM K2bm`2b mbBM; i?2 6`2M2i@
a2``2i 7`K2X h?Bb T`QT2ìv Bb 2M+TbmHi2/ BM i?2 *Qbb2ì@`Q/ i?2Q`v- r?B+?
HHQrb 7Q` /2}MBM; ;2QK2i`B+HHv 2t+i #2Kb mM/2` H`;2 /27Q`KiBQMX
+FMQrH2/;K2Mib
h?2 mi?Q`b S?XhXM/ XSXrQmH/ HBF2 iQ i?MF i?2 :2`KM _2b2`+? 6QmM/iBQM
U.6:V 7Q` }MM+BH bmTTQì Q7 i?2 T`QD2+i rBi?BM i?2 *Hmbi2` Q7 1t+2HH2M+2 BM
aBKmHiBQM h2+?MQHQ;v U1s* jRyfkV i i?2 lMBp2`bBiv Q7 aimii;ìX
_272`2M+2b
MiKM- aXaX, LQMHBM2` S`Q#H2Kb Q7 1HbiB+BivX TTHB2/ Ji?2KiB+H a+B2M+2bX aT`BM;2`@
o2`H;- L2r uQ`F Ukyy8V
`MQH/- JX, aBKmHiBQM H;Q`Bi?Kb BM o2?B+H2 avbi2K .vMKB+b Ukyy9V
vH *m2pb- CXAX- G`Q+?2- úX- SB++BM- PX, bbmK2/@JQ/2@"b2/ .vMKB+ JQ/2H 7Q` *#H2
_Q#Qib rBi? LQM@bi`B;?i *#H2bX AM, :Qbb2HBM- *X- *`/Qm- SX- "`m+FKMM- hX- SQii- X
U2/bXV *#H2@.`Bp2M S`HH2H _Q#Qib, S`Q+22/BM;b Q7 i?2 h?B`/ AMi2`MiBQMH *QM72`2M+2
QM *#H2@.`Bp2M S`HH2H _Q#Qib- J2+?MBbKb M/ J+?BM2 a+B2M+2- TTX R8Ĝk8X aT`BM;2`
AMi2`MiBQMH Sm#HBb?BM;- *?K UkyR3V
"2ib+?- SX- ai2BMKMM- SX, .1 TT`Q+? iQ 6H2tB#H2 JmHiB#Q/v .vMKB+bX JmHiB#Q/v
avbi2K .vMKB+b 3UjV- je8Ĝj3N UkyykV
"2ib+?- SX- l?H`- aX, 1M2`;v@KQK2MimK +QMb2`pBM; BMi2;ìBQM Q7 KmHiB#Q/v /vMKB+bX
JmHiB#Q/v avbi2K .vMKB+b RdU9V- k9jĜk3N UkyydV
.m- CX- ;`rH- aXEX, .vMKB+ JQ/2HBM; Q7 *#H2@.`Bp2M S`HH2H JMBTmHiQ`b qBi?
.Bbi`B#mi2/ Jbb 6H2tB#H2 *#H2bX CQm`MH Q7 oB#ìBQM M/ +QmbiB+b RjdUkV- ykR-yky
UkyR8V
:QMxH2x- PX, hBK2 AMi2;ìBQM M/ .Bb+`2i2 >KBHiQMBM avbi2KbX CQm`MH Q7 LQMHBM2`
a+B2M+2 eU8V- 99NĜ9ed URNNeV
A`pBM2- >XJX, aim/B2b BM i?2 biiB+b M/ /vMKB+b Q7 bBKTH2 +#H2 bvbi2KbX *HB7Q`MB
AMbiBimi2 Q7 h2+?MQHQ;v URNd9V
EQxF- EX- ZBM w?Qm- CBMbQM; qM;, aiiB+ MHvbBb Q7 +#H2@/`Bp2M KMBTmHiQ`b rBi?
MQM@M2;HB;B#H2 +#H2 KbbX A111 h`Mb+iBQMb QM _Q#QiB+b kkUjV- 9k8Ĝ9jj UkyyeV
GBm- wX- hM;- sX- a?Q- wX- qM;- GX- hM;- GX, _2b2`+? QM GQM;Bim/BMH oB#ìBQM *?`+@
i2`BbiB+ Q7 i?2 aBt@*#H2@.`Bp2M S`HH2H JMBTmHiQ` BM 6ahX /pM+2b BM J2+?MB+H
1M;BM22`BM; 8UyV- 89d-9Re UkyRjV
JMFH- EXEX- ;`rH- aXEX, .vMKB+ JQ/2HBM; M/ aBKmHiBQM Q7 ai2HHBi2 h2i?2`2/
avbi2KbX AM, oQHmK2 8, RNi? "B2MMBH *QM72`2M+2 QM J2+?MB+H oB#ìBQM M/ LQBb2-
Sìb - "- M/ *- TTX jdNĜj33X aJ1 UkyyjV
JB+?2HBM- JX- "`/i- *X- L;mv2M- .XZX- :Qmii27`/2- JX, aBKmHiBQM M/ *QMi`QH rBi?
s.1 M/ JiH#faBKmHBMF Q7 *#H2@.`Bp2M S`HH2H _Q#Qi U*Q:B_QVX AM, SQii- X-
"`m+FKMM- hX U2/bXV *#H2@.`Bp2M S`HH2H _Q#Qib- J2+?MBbKb M/ J+?BM2 a+B2M+2b-
pQHX jk- TTX dRĜ3jX aT`BM;2` AMi2`MiBQMH Sm#HBb?BM;- *?K UkyR8V
hM;- uX, LmK2`B+H 1pHmiBQM Q7 lMB7Q`K "2K JQ/2bX CQm`MH Q7 1M;BM22`BM; J2+?M@
B+b RkNURkV- R9d8ĜR9dd UkyyjV
h2KT2H- SX- a+?KB/i- X- >b/QMF- "X- SQii- X, TTHB+iBQM Q7 i?2 _B;B/ 6BMBi2 1H2K2Mi
J2i?Q/ iQ i?2 aBKmHiBQM Q7 *#H2@.`Bp2M S`HH2H _Q#QibX AM, w2;?HQmH- aX- _QK/?M2-
GX- G`B#B- JXX U2/bXV *QKTmiiBQMH EBM2KiB+b- J2+?MBbKb M/ J+?BM2 a+B2M+2b-
TTX RN3Ĝky8X aT`BM;2` AMi2`MiBQMH Sm#HBb?BM;- *?K UkyR3V
Static and dynamic analysis of a 6 DoF totally
constrained cable robot with 8 preloaded cables
Damien Gueners1, B. Chedli Bouzgarrou1 and Hélène Chanal1

1
Institut Pascal, UMR6602 UBP/CNRS/SIGMA-Clermont BP10448, F63000 Clermont-Fer-
rand, France
[email protected]
Abstract In this paper, the static and dynamic performances of a cable-driven parallel
robot (CDPR) are analyzed over all its entire workspace. The considered robot has 6
degrees of freedom and it is completely constrained with 8 cables. This paper aims at
highlighting the effects of cable preloads on the robot’s behavior. To this end, the stiff-
ness matrix of the robot is computed from a second order approximation of the defor-
mation energy of preloaded cables. Lagrange formulation is used to obtain dynamic
equations of the robot for vibration mode calculation and trajectory tracking simula-
tions. Static and modal analyses show that, for a given preload, the increase in the ro-
bot’s stiffness and modal frequencies is relatively more significant using soft cables,
made of nylon, rather than stiff cables made of steel. These results are confirmed by
dynamic simulations illustrating the effects of preloaded cables on the accuracy of tra-
jectory tracking performance.
Keywords: Stiffness, Vibration, Cable preload.
1 Introduction
CDPRs [1] are robots for which the fixed base and the mobile platform are connected
by cables. There are no guiding elements except those of the pulleys used to wind and
unwrap the cables. The main advantages of this technology are:
x reduction of the moving masses resulting in a large acceleration capacity of the mov-
ing platform;
x reconfigurability and modularity: the robot can be reconfigured easily by changing
the anchor points of the actuators and adapting its geometric control model;
x reduction of the production costs.
The static and dynamic rigidity of cable robots is generally lower than that of conven-
tional parallel robots due to the flexibility of the cables. The accuracy and vibration
stability of existing cable robots is not high enough to consider them for dynamic load-
ing applications such as the machining process.
CDPRs can be divided into 3 main families:
x Suspended cable driven parallel robots[2] where all the points of attachment are
above the mobile platform. Gravity plays the role of a force cable that pulls down.
x Completely constrained cable driven parallel robots[3] that have cables above and
below the mobile platform.

308 D. Gueners et al.
x Hybrid parallel cable robots[4] that have cables which guide a rigid arm.
Some research has been done on the stiffness and vibration analysis of cable robots.
In [5], it has been shown that the transverse vibrations of cables can be negligible in
relation to the total vibrations of the effector in comparison to their axial vibrations. For
a fully constrained robot, 98.6% of the mobile platform vibrations are caused by axial
vibrations of the cables. These results have also been used to approve the modeling of
the cables by axial springs for totally constrained CDPRs. In [6], the longitudinal and
lateral vibration on a planar and a cable suspended robot is studied by using a finite
element method. In [7, 8], the dynamic stiffness matrix is proposed to analyze the vi-
bration of a cable suspended robot with a sagging cable model. In [9], the control of
reaction wheels mounted on the mobile platform has been used to compensate its vi-
brations. In [10], a controller has been proposed to attenuate the transverse vibrations
of a planar CDPR, which has a lower stiffness in the axis perpendicular to the plane
motion.
The paper cited above did not however study precisely the influence of cable preload
on the performance of cable robots. In this paper we consider a totally constrained cable
robot, with 6 DOFs, driven by 8 preloaded cables. The proposed research is carried out
within the CABFAB project, which addresses the design of a CDPR for additive man-
ufacturing processes. The presented work is part of a preliminary design phase, aimed
at guaranteeing a minimal level of rigidity over the entire robot workspace, to be opti-
mized in terms of accessibility in position and orientation. For additive manufacturing,
a high level of accuracy for tool path tracking is required. A completely constrained
cable robot should have reach a higher level of accuracy than a suspended one [1].
Moreover, since the stiffness depends directly on the cable lengths, a medium-scale
robot of 1m3 size, as that considered in this paper, can have a relatively higher accuracy
compared to the large-scale cable robots.
This paper investigates the influence of cable preloads on the stiffness and vibration
modes of a CDPR. First, the kinematic and dynamic models are presented. The static
stiffness and modal frequencies are then determined for various cable preloads. Next
the static and dynamic simulation of a completely constrained CPDR are analyzed. The
stiffness and natural mode inside the feasible workspace are studied. The static stiffness
and natural frequency are compared for different preloads and cables. The robot’s dy-
namic accuracy is evaluated as a function of the cable preload.
2 Robot modeling
For the considered application, the robot can incorporate thin cables with the aim of
printing medium-sized parts at high speeds. Therefore, the mass of the cable can be
considered as negligible and the effect of sagging is less significant than with large
CDPRs[2]. Thus, the kinematic model does not take into account the sagging effect. In
addition, cable deformations rely on linear elasticity assumptions. The inverse kine-
matic model will be presented in this section.
Static and dynamic analysis of a 6 DoF totally constrained cable… 309
2.1 Inverse Kinematics

The geometric model uses ݉ cables of length ݈௜ with one anchor point on the mounting
frame in ‫ܣ‬௜ , and another anchor point on the end-effector in ‫ܤ‬௜ (fig.1).
࣬ை ൫ܱǡ ‫܍‬௫ೀ ǡ ‫܍‬௬ೀ ǡ ‫܍‬௭ೀ ൯ is the fixed frame of reference and ࣬ா ൫‫ܥ‬ǡ ‫܍‬௫ಶ ǡ ‫܍‬௬ಶ ǡ ‫܍‬௭ಶ ൯ is the
frame attached to the mobile platform at its characteristic point ‫ܥ‬. The kinematic con-
straints of the system can be given by the vector loop-closure equation for each cable:
‫܊ܗ‬଴௜ ൌ ‫ ܋ܗ‬଴ ൅ ‫܊܋‬଴௜ ൌ ‫܉ܗ‬଴௜ ൅ ‫܊܉‬଴௜ ሺͳሻ
Fig. 1. Schematic representation of one cable (left) and robot with 8 cables view (right)
If the sagging effect is ignored, each cable is similar to a rigid rod when in static equi-
librium and its length can be calculated by:
݈௜ ൌ ฮ‫܊܉‬଴௜ ฮଶ ൌ ฮെ‫܉ܗ‬଴௜ ൅ ‫ ܋ܗ‬଴ ൅ ‫ ܀‬ைா ‫܊܋‬ா௜ ฮଶ ሺʹሻ
with ‫ ܀‬ைா as the rotation matrix between the effector frame ࣬ா and the fixed frame of
reference ࣬ை . This equation gives the inverse kinematics of the robot.
Tab. 1. Anchor points position (mm)

࢏ 1 2 3 4 5 6 7 8
ͷͲͲ ͷͲͲ െͷͲͲ െͷͲͲ െͷͲͲ െͷͲͲ ͷͲͲ ͷͲͲ
‫܉ܗ‬૙࢏ ൥ ͷͲͲ ൩ ൥ͷͲͲ൩ ൥ ͷͲͲ ൩ ൥ ͷͲͲ ൩ ൥െͷͲͲ൩ ൥െͷͲͲ൩ ൥െͷͲͲ൩ ൥െͷͲͲ൩
െͷͲͲ ͷͲͲ ͷͲͲ െͷͲͲ ͷͲͲ െͷͲͲ െͷͲͲ ͷͲͲ
Ͷ͹Ǥ͸ Ͷ͹Ǥ͸ െͶ͹Ǥ͸ െͶ͹Ǥ͸ െͶ͹Ǥ͸ െͶ͹Ǥ͸ Ͷ͹Ǥ͸ Ͷ͹Ǥ͸
‫࢏ࡱ܊܋‬ ൥ͶͲǤʹ൩ ൥ ͶͲǤʹ ൩ ൥ ͶͲǤʹ ൩ ൥ ͶͲǤʹ ൩ ൥െͶͲǤʹ൩ ൥െͶͲǤʹ൩ ൥െͶͲǤʹ൩ ൥െͶͲǤʹ൩
ʹͷǤͲ െʹͷǤͲ െʹͷǤͲ ʹͷǤͲ െʹͷǤͲ ʹͷǤͲ ʹͷǤͲ െʹͷǤͲ
Table 1 indicates the anchor point parameters used in this study of a fully constrained
8-cable CDPR. This configuration is similar to that presented in [3].
At the preliminary design stage, it is crucial to have an estimation of the robot stiff-
ness. The following paragraph outlines the computation of the stiffness matrix when
cable preloads are considered.
2.2 Stiffness matrix formulation with preload cables

The 6x8 static Jacobian matrix, mapping the cable tensions to the wrench that cables
apply on the mobile platform, has a rank of 6 and its kernel is 2-dimensional. Therefore,
the static equilibrium of the robot can be achieved with infinite combinations of cable
tensions. In order to formulate the stiffness matrix of the robot, potential energy is used
to compute the stiffness matrix associated with one cable.
Fig. 2. Schematic representation of small variations

For a small variation of the configuration of the end-effector, there is a variation of the
positions of points ‫ܤ‬௜ (figure 2):
෪଴ ‫܊܋‬଴ ൌ ο‫ ܋ܗ‬଴ െ ‫܊܋‬
ο‫܊ܗ‬଴௜ ൌ ο‫ ܋ܗ‬଴ ൅ οી଴ா ൈ ‫ ܀‬ைா ‫܊܋‬ா௜ ൌ ο‫ ܋ܗ‬଴ ൅ οી ෪ప଴ οી଴ா ሺ͵ሻ
ா ௜
With :
ߜ‫ݔ‬௜ ߜ‫ݔ‬஼ ߜ߰ Ͳ െߜ߮ ߜߠ
෪଴ ൌ ൥ ߜ߮
ο‫܊ܗ‬଴௜ ൌ ൥ߜ‫ݕ‬௜ ൩ ; ο‫ ܋ܗ‬଴ ൌ ൥ߜ‫ݕ‬஼ ൩ ; οી଴஼ ൌ ൥ ߜߠ ൩ ; οી Ͳ െߜ߰൩ ;
஼
ߜ‫ݖ‬௜ ߜ‫ݖ‬஼ ߜ߮ െߜߠ ߜ߰ Ͳ
ܺ஻௜ Ͳ ܼ஻௜ െܻ஻௜
෪ప଴ ൌ ൥െܼ஻௜
‫܊܋‬଴௜ ൌ ൥ ܻ஻௜ ൩ ; ‫܊܋‬ Ͳ ܺ஻௜ ൩
ܼ஻௜ ܻ஻௜ െܺ஻௜ Ͳ
We have :
ߜ‫ݔ‬஼
‫ۍ‬ ‫ې‬
ߜ‫ݕ‬஼
ߜ‫ݔ‬௜ ͳ Ͳ Ͳ Ͳ ܼ஻௜ െܻ஻௜ ‫ێ‬ ‫ۑ‬
ߜ‫ݖ‬ ο‫ ܋ܗ‬଴
൥ߜ‫ݕ‬௜ ൩ ൌ ൥Ͳ ͳ Ͳ െܼ஻௜ Ͳ ܺ஻௜ ൩ ‫ ێ‬஼ ‫ ۑ‬՞ ο‫܊ܗ‬଴௜ ൌ ۶௜ ൤ ଴ ൨ ሺͶሻ
ߜ߰ οી஼
ߜ‫ݖ‬௜ Ͳ Ͳ ͳ ܻ஻௜ െܺ஻௜ Ͳ ‫ێ‬
ᇣᇧᇧᇧᇧᇧᇧᇧᇧᇤᇧᇧᇧᇧᇧᇧᇧᇧᇥ ‫ۑ‬
‫ێ‬ ߜߠ ‫ۑ‬
۶೔
‫ے ߮ߜ ۏ‬
ሾሺο‫ ܋ܗ‬଴ ሻ் ሺοી଴஼ ሻ் ሿ் is the infinitesimal displacement screw of the end-effector.
A displacement of ‫ܤ‬௜ from the cable with respect to its static equilibrium configuration
in the reference ࣬௜ ሺ‫ܣ‬௜ ǡ ‫ݑ‬ ሬሬԦ௜ ሻ is given by ο‫܊ܗ‬௜௜ ൌ ሾߜ‫ݑ‬௜ ߜ‫ݒ‬௜ ߜ‫ݓ‬௜ ሿ் .
ሬԦ௜ ǡ ‫ݒ‬Ԧ௜ ǡ ‫ݓ‬
We have the relationship:
ο‫܊ܗ‬଴௜ ൌ ‫ ܀‬ை௜ ο‫܊ܗ‬௜௜ ሺͷሻ
where ‫ ܀‬ை௜ is the rotation matrix between the reference ࣬௜ and the fixed reference ࣬ை :
‫ ܀‬ை௜ ൌ ሾ‫ܝ‬଴௜ ‫ܞ‬௜଴ ‫ܟ‬௜଴ ሿ. We denote by ݈଴ǡ௜ the free length of the non-preloaded cable
݅ሺ݅ ൌ ͳǡ ǥ ǡͺሻ and ݈ ்௜ its length stretched to the static equilibrium configuration. During
small displacements, the current length of the cable is given by:
݈௜ ൌ ටሺ‫் ܮ‬௜ ൅ ߜ‫ݑ‬௜ ሻଶ ൅ ߜ‫ݒ‬௜ ଶ ൅ ߜ‫ݓ‬௜ ଶ ሺ͸ሻ
The deformation energy of a cable is then:

ଵ
ܸ௜ ൌ ݇௜ ሺ݈௜ െ ݈଴ǡ௜ ሻଶ ሺ͹ሻ
ଶ
ாబ ௌ
for a linear behavior of the cable: ݇௜ ൌ , with the elastic modulus ‫ܧ‬଴ and the cross
௟బǡ೔
section ܵ. At the second order, the strain energy variation of the cable can be expressed
as:
ଵ ்
ܸ௜ ൌ ൫ο‫܊ܗ‬௜௜ ൯ ۹ ௜ ο‫܊ܗ‬௜௜ ሺͺሻ
ଶ
where ۹ ௜ is the Hessian matrix at the equilibrium (ο‫܊ܗ‬௜௜ ൌ ૙) of ܸ௜ with respect to the
variables ߜ‫ݑ‬௜ ǡߜ‫ݒ‬௜ ߜ‫ݓ‬௜ Ǥ
݇௜ Ͳ Ͳ
‫ۍ‬ ்೔ ‫ې‬
Ͳ Ͳ‫ۑ‬
۹ ௜ ൌ સસ ் ሺܸ௜ ሻ ൌ ‫ێ‬ ௟೅೔ ሺͻሻ
‫ێ‬ ்೔ ‫ۑ‬
‫Ͳۏ‬ Ͳ
௟೅೔ ‫ے‬
ങ ങ ങ
‫׏‬સ ் ൌ ൣങሺഃೠ೔ሻ ങሺഃೡ ೔ሻ ങሺഃೢ೔ ሻ൧ܶ௜ Ǥ
The deformation energy of a cable can then be expressed as a function of the small
displacements of the mobile platform:
ଵ ் ் ο‫ ܋ܗ‬଴
ܸ௜ ൌ ሾሺο‫ ܋ܗ‬଴ ሻ் ሺοી଴஼ ሻ் ሿ۶௜ ‫ ܀‬ை௜ ۹ ௜ ‫ ܀‬ை௜ ۶௜ ൤
ᇣᇧᇧᇧᇧᇤᇧᇧᇧᇧᇥ ൨ ሺͳͲሻ
ଶ οી଴஼
۹బ೔
Hence the stiffness matrix of the cable robot:

۹ ௥௢௕௢௧ ൌ σ௠
௜ୀଵ ۹ ଴௜ ሺͳͳሻ
The matrix ‫்܀‬ை௜ ۶௜ can be detailed as follows:

் ் ் ෪ప଴ ்
‫ܝ‬଴௜ ‫ܝ‬଴௜ ‫ܝ‬଴௜ ή ‫܊܋‬ ‫ܝ‬଴௜ ሺ‫ ݅Ͳ܊܋‬ൈ ‫ ݅Ͳܝ‬ሻ்
‫்܀‬ை௜ ۶௜ ൌ ൦ ‫ܞ‬௜଴ ் ൪ ൣ۷ଷ ෪ప଴ ൧ ൌ
െ‫܊܋‬ ൦ ‫ܞ‬௜଴ ் ் ෪ప଴ ൪ ൌ ൦ ‫ ܞ‬଴ ்
‫ ܞ‬଴ ή ‫܊܋‬
௜ ௜ ሺ‫ ݅Ͳ܊܋‬ൈ ‫ ݅Ͳܞ‬ሻ் ൪ ሺͳʹሻ
் ் ் ෪ప଴ ்
‫ܟ‬௜଴ ‫ܟ‬௜଴ ‫ܟ‬௜଴ ή ‫܊܋‬ ‫ܟ‬௜଴ ሺ‫݅Ͳ܊܋‬ ൈ ‫ ݅Ͳܟ‬ሻ்
With the Jacobian matrix:

‫ͳͲܝ‬ ‫ڮ‬ ‫݉Ͳܝ‬ ‫ͳͲܞ‬ ‫ڮ‬ ‫݉Ͳܞ‬
۸௨ ൌ ቈ ቉, ۸ ௩ ൌ ቈ ቉, ۸ ௪ ൌ
‫ͳͲ܊܋‬ ൈ ‫ͳͲܝ‬ ‫ڮ‬ ‫݉Ͳ܊܋‬ ൈ ‫݉Ͳܝ‬ ‫ͳͲ܊܋‬ ൈ ‫ͳͲܞ‬ ‫ڮ‬ ‫݉Ͳ܊܋‬ ൈ ‫݉Ͳܞ‬
‫ͳͲܟ‬ ‫ڮ‬ ‫݉Ͳܟ‬
ቈ ቉
‫ͳͲ܊܋‬ ൈ ‫ͳͲܟ‬ ‫ڮ‬ ‫ ݉Ͳ܊܋‬ൈ ‫݉Ͳܟ‬
We can decompose the stiffness matrix with preloaded cables into a sum of three
matrices:
೅భ ೅భ
݇ଵ Ͳ Ͳ ೗೅భ
Ͳ Ͳ ೗೅భ
Ͳ Ͳ
۹ ௥௢௕௢௧ ൌ ۸௨ ൥ Ͳ ‫ڰ‬ Ͳ ൩ ۸௨் ൅ ۸௩ ቎ Ͳ ‫ڰ‬ Ͳ ቏ ۸௩் ൅ ۸௪ ቎ Ͳ ‫ڰ‬ ்
Ͳ ቏ ۸௪ ሺͳ͵ሻ
Ͳ Ͳ ଼݇ Ͳ Ͳ ೅ఴ
೗೅ఴ Ͳ Ͳ ೅ఴ
೗೅ఴ
If the preload is zero, we obtained the same matrix express in [5].
2.3 Mass matrix formulation and vibration modes
Mass matrix formulation can be obtained from the kinetic energy of the mobile plat-
form. The generalized coordinates of the system are denoted by ‫ ܙ‬ൌ ሺ‫ݔ‬ǡ ‫ݕ‬ǡ ‫ݖ‬ǡ ߰ǡ ߠǡ ߮ሻ
with the angles defines by Tait-Bryan XYZ convention. The Kinetic energy is ex-
pressed by:
ଵ ் ଵ ்
ܶ ൌ ݉௘ ‫ܞ‬஼଴ ‫ܞ‬஼଴ ൅ ષா ή ሺ۷஼ா ષா ሻ ሺͳͶሻ
ଶ ଶ
where ‫ܞ‬஼଴
is the translational velocity of the center of mass ‫ ܥ‬of the end-effector, ષா is
the end-effector rotational velocity expressed on the end-effector frame, ݉௘ and ۷஼ா are
respectively the mass of the end-effector and its inertia matrix expressed in the frame
࣬ா ൫‫ܥ‬ǡ ‫܍‬௫ಶ ǡ ‫܍‬௬ಶ ǡ ‫܍‬௭ಶ ൯. For infinitesimal displacements denoted by ߜ‫ ܙ‬around a static
equilibrium configuration, the kinetic energy can be expressed as follows:
ͳ
ܶ ൌ ߜ‫ܙ‬ሶ ் ‫ܙߜۻ‬ሶ
ʹ
where M is given by
݉௘ ۷ଷ ૙૜
‫ۻ‬ൌ൤ ൨
૙૜ ‫ ܀‬ைா ۷࡯ࡱ ‫ࢀ܀‬ைா
Having established the stiffness and the mass matrices ۹ and ‫ۻ‬, undamped vibration
modes can be determined by the resolution of the following eigen problem:
ଶ
ሺ۹ െ ߱଴௜ ‫ۻ‬ሻ‫ܘ‬௜ ൌ ૙ ሺͳͷሻ
where ߱଴௜ ሺ݅ ൌ ͳǡ ǥ ǡ͸ሻ. are the natural pulsations associated with the modal vectors
(mode shapes) ‫ܘ‬௜ .
2.4 Non-linear dynamic modelling

In order to study the behavior of the robot in large displacements (large rotations) and
subject to dynamic loading due to inertia and cable forces, non-linear effects must be
considered such as gyroscopic coupling and cable sagging. The actual performance of
the robot in terms of precision and vibration stability in operating conditions cannot be
predicted precisely by the linear elasto-dynamic model developed in the previous sec-
tion, where infinitesimal displacement assumption have been adopted. A non-linear dy-
namic model is therefore required to simulate the robot’s movements and estimate its
performance during path tacking.
For a totally constrained cable robot with all the cables working under tension, we can
assume that the cable works like a traction spring with a negligible mass. The tension
force generated by the cable is expressed in equation 16 as a function of the following

cable parameters: the elastic modulus ‫ܧ‬଴ , the cross section ܵ, the free length ݈଴ǡ௜ the
current length ݈௜ .
௟೔
ܶ௘௟௔௦ǡ௜ ൌ ‫ܧ‬଴ ܵ ൬ െ ͳ൰ ሺͳ͸ሻ
௟బǡ೔
The relation between the preloaded tension ܶ௜ and its corresponding cable elongation
݈ ்௜ is given by the equation:
௟
ܶ௜ ൌ ‫ܧ‬଴ ܵ ൬ ೅೔ െ ͳ൰ ሺͳ͹ሻ
௟బǡ೔
By replacing the cable free length ݈଴ǡ௜ from the equation ሺͳ͹ሻ into equation ሺͳ͸ሻ, the
tension force becomes :
௟
ܶ௘௟௔௦ǡ௜ ൌ ೔ ሺܶ௜ ൅ ‫ܧ‬଴ ܵሻ െ ‫ܧ‬଴ ܵ ሺͳͺሻ
௟೅೔
However, cables work only under tension. Hence, the force in a cable is null or pos-
itive, which means:
‫ܨ‬௜ ൌ ܶ௘௟௔௦ǡ௜ ܶ௘௟௔௦ǡ௜ ൒ Ͳ
൜ (19)
‫ܨ‬௜ ൌ Ͳܶ௘௟௔௦ǡ௜ ൏ Ͳ
The dynamics of the robot can be formulated with Lagrange equations. We use the
expression of the kinetic energy given by equation ሺͳͶሻ. The generalized force, relative
to each generalized coordinate, is given by:
଴் డ
ܲ௤ೕ ൌ ‫ܨ‬௤ೕ ൅ σ௠௜ୀଵ ‫ܨ‬௜ ‫ܝ‬௜ ή ሺ‫܊ܗ‬଴௜ ሻ ሺʹͲሻ
డ௤ೕ
Where ‫ܨ‬௜ is the force created by the cable tension, ‫ܝ‬଴௜ and the partial derivative of the
డ
vector ‫܊ܗ‬଴௜ dependent on the generalized coordinate ሺ‫܊ܗ‬଴௜ ሻ. ‫ܨ‬௤௝ ሺ݆ ൌ ͳǡ ǥ ǡͺሻ denotes
డ௤
the other external forces projected on the generalized coordinate such as the weight of
the end-effector.
The Dynamic equations are obtained as follows:
ௗ డ் డ்
ࣦ௤ೕ ǣ െ ൌ ܲ௤ೕ ሺʹͳሻ
ௗ௧ డ௤ണሶ డ௤ೕ
This differential equation is solved with ode45 from Matlab software by using ‫ ܇‬ൌ
ሾ‫ܙ ்ܙ‬ሶ ் ሿ் as a state vector and by controlling the cable preload length ݈ ்௜ . ݈௜ is calcu-
lated with the direct kinematic model.
3 Simulation results
The simulations are performed with cable diameter of ͳǤͷ݉݉ and for two cases of
cable material: nylon cables with an elastic modulus of ͵Ǥͻ ൈ ͳͲଽ ܰǤ ݉ିଶ and steel ca-
bles with an elastic modulus of ͳͲʹ ൈ ͳͲଽ ܰǤ ݉ିଶ . The mobile platform is assimilated
to a homogenous parallelepiped of size ሺͻͷǤͳǡ ͺͲǤͷǡ ͷͲǤͲሻ with a mass of
ͳ. In these simulations, the orientation of the mobile platform is maintained constant
at ሺͲιǡ Ͳιǡ Ͳιሻ.
3.1 Static and modal analysis
The static analysis is performed over all the feasible workspace. The stiffness matrix of
the robot is that which relates the infinitesimal displacements twist of the mobile plat-
form to the external force wrench applied to it at its characteristic point ‫ܥ‬. In our case,
the cables are assumed to be permanently preloaded so as to avoid a discontinuity of
the solution. If the cable becomes slack, the analysis is no longer valid.
The mapping of the rigidity terms is shown in Fig. 3 for the layers ‫ ݖ‬ൌ Ͳ݉ and ‫ ݖ‬ൌ
ͲǤʹͷ݉ for the steel cable case. It can be seen that the stiffness depends on effector
position. The translational rigidities in ܺ and ܻ directions, ݇௫௫ and ݇௬௬ , are between
400 kN/m and 550 kN/m. The translational rigidity in ܼ direction, ݇௭௭ , is higher and
varies between 700 kN/m and 900kN/m. The rotational rigidities around ܺ, ܻ and ܼ
directions, ݇టట , ݇ఏఏ and ݇ఝఝ , vary respectively between 2.1kN/rad and 2.8kN/rad,
2.7kN/rad and 5kN/rad, and 0.1kN/rad and 0.9kN/rad. The rotational rigidity ݇ఝఝ is the
lowest because a small rotation around the Z axis produces displacements of the anchor
points almost orthogonal to the cables. In this study, only the diagonal terms of the
rigidity matrix are analyzed.
Fig. 3. Stiffness on the plane xy for z=0m (left) and for z=0.25m (right) for a steel cable
Fig. 4. Natural mode on the plane xy for z=0m (left) and for z=0.25m (right) for a steel cable
The vibration cartography is shown in Fig 4 for the layer ‫ ݖ‬ൌ Ͳ and ‫ ݖ‬ൌ ͲǤʹͷ
for a steel cable. The minimum first mode is closed to 10 Hz when the effector is at the
workspace extremity. In the center of the workspace, the first natural frequency is above
30 Hz.
Tab. 2. Influence of the preload on the stiffness of the position ሺͲǡͲǡͲሻ for a nylon cable
ܶ௜ ݇௫௫ (kN/m) ݇௬௬ (kN/m) ݇௭௭ (kN/m) ݇టట (N/°) ݇ఏఏ (N/°) ݇ఝఝ (N/°)
0N 19.618 20.259 26.415 1.7790 2.2012 0.0225
19.954 20.590 26.702 1.7849 2.2094 0.0546
50N
1.68% 1.61% 1.08% 0.33% 0.37% 58.89%
20.286 20.917 26.985 1.7907 2.2174 0.0863
100N
3.29% 3.15% 2.11% 0.65% 0.73% 73.99%
20.935 21.557 27.539 1.8020 2.2333 0.1484
200N
6.29% 6.02% 4.08% 1.28% 1.44% 84.87%
Tab. 3. Influence of the preload on the stiffness of the position ሺͲǡͲǡͲሻ for a steel cable
ܶ௜ ݇௫௫ (kN/m) ݇௬௬ (kN/m) ݇௭௭ (kN/m) ݇టట (N/°) ݇ఏఏ (N/°) ݇ఝఝ (N/°)
0N 513.09 529,85 690.84 46.5278 57.5688 0.5872
513.43 530,18 691.13 46.5337 57.5771 0.6196
50N
0.07% 0.06% 0.04% 0.01% 0.01% 5.23%
513.77 530,52 691.42 46.5396 57.5853 0.6520
100N
0.13% 0.13% 0.08% 0.03% 0.03% 9.93%
514,45 531,18 692.00 46.5515 57.6018 0.7166
200N
0.26% 0.25% 0.17% 0.05% 0.06% 18.06%
The influence of the preload on the robot stiffness is shown in Table 2 for a nylon
cable and Table 3 for a steel cable at the position ሺͲǡ Ͳǡ Ͳሻ. For the same cross
section, the steel stiffness is higher than the nylon stiffness. On the x axis, the stiffness
is 513.09 kN/m for steel and 19.618 kN/m for nylon without preload. On the x axis, the
stiffness increases by 6.29% for the nylon with a 200N preload. For the steel cable, the
preload has less influence with an increase of 0.26% on the x axis stiffness with a 200N
preload. This may be explained by the fact that a higher preload is required for steel
cable to reach the same relative level of lateral stiffness as in nylon cable.The preload
has a more significant impact on the lesser stiffness direction which is the ߮ orientation
(rotation on z). For the nylon, the stiffness is increased by about 84.87% with a preload
of 200N. For the steel, the stiffness is increased by about 18.06% with a preload of
200N.
Tab. 4. Influence of the preload on the natural frequency of the position ሺͲǡͲǡͲሻ for a nylon cable
ܶ௜ ݂ଵ ሺ‫ݖܪ‬ሻ ݂ଶ ሺ‫ݖܪ‬ሻ ݂ଷ ሺ‫ݖܪ‬ሻ ݂ସ ሺ‫ݖܪ‬ሻ ݂ହ ሺ‫ݖܪ‬ሻ ݂଺ ሺ‫ݖܪ‬ሻ
0N 6.6015 22.2921 22.6531 25.8668 57.6235 58.7635
10.2964 22.4823 22.8377 26.0070 57.7307 58.8605
50N
35.89% 0.85% 0.81% 0.54% 0.19% 0.16%
12.9446 22.6682 23.0183 26.1445 57.8362 58.9560
100N
49% 1.66% 1.59% 1.06% 0.37% 0.33%
16.9720 23.0278 23.3677 26.4117 58.0421 59.1425
200N
61.1% 3.19% 3.06% 2.06% 0.72% 0.64%
The influence of the preload on the natural frequency is shown in the tables 4 and 5.
The first natural frequency is lower for the nylon cable with 6.6015 Hz than the steel
cable with 33.7605 Hz. The preload effect on the frequency is more significant on nylon
cable by an increase of the first mode by 61.1% when preloaded at 200N preload. The
preload has a significant impact on the first mode. For steel cable, the first mode is
increased by 9.48% with a 200N preload, but the other modes are increased by less than
0.13%.
Tab. 5. Influence of the preload on the natural frequency of the position ሺͲǡͲǡͲሻ for a steel cable
ܶ௜ ݂ଵ ሺ‫ݖܪ‬ሻ ݂ଶ ሺ‫ݖܪ‬ሻ ݂ଷ ሺ‫ݖܪ‬ሻ ݂ସ ሺ‫ݖܪ‬ሻ ݂ହ ሺ‫ݖܪ‬ሻ ݂଺ ሺ‫ݖܪ‬ሻ
0N 33.7605 114.0036 115.8499 132.2848 294.6912 300.5215
34.6791 114.0412 115.8864 132.3125 294.7123 300.5406
50N
2.65% 0.033% 0.0315% 0.0209% 0.0072% 0.0064%
35.5736 114.0788 115.9229 132.3402 294.7335 300.5597
100N
5.1% 0.0659% 0.0629% 0.0418% 0.0143% 0.0127%
37.2969 114.1538 115.9957 132.3955 294.7756 300.5979
200N
9.48% 0.13% 0.13% 0.0836% 0.0286% 0.0254%
3.2 Dynamic simulation for a given position
A first dynamic analysis is done on a given position at ሺͲǡ Ͳǡ Ͳሻ. An initial per-
turbation is done by an initial translational velocity of ሺͲǤͷȀǡ ͲǤͷȀǡ ͲǤͷȀሻ and
rotational velocity of ሺͲǤͳȀǡ ͲǤͳȀǡ ͲǤͳȀሻ. Two simulations are carried
out in order to study the induced vibrations. It is not possible to analyze the Fourier
transform of cable working only under traction stress without preload, which may be
explained by the elastic force discontinuity. Cable preload is added with a force distri-
bution algorithm. The close-form algorithm[11] is used to compute the force distribu-
tion with ‫ݐ‬௠௜௡ ൌ ʹͲͲܰ and ‫ݐ‬௠௔௫ ൌ ͵ͲͲܰ. With the preload, the static natural frequen-
cies from table 5 are found. The first mode is presented on the ߮ rotation, which is the
lesser stiffness. The second and the third modes are closed and they are presented re-
spectively on the x and y translations. The fourth mode is on the z translation. The fifth
and sixth modes are respectively on the ߰ and ߠ rotations. These results show that, at
the position ሺͲǡ Ͳǡ Ͳሻ, the two highest modal frequencies correspond to rotations
of the mobile platform around the x and y axes respectively. At this position, move-
ments are clearly decoupled. However, for other positions, the movements are coupled.
The first mode presented mainly a rotation around the z axis coupled with a translation
in the xy plane.
3.3 Dynamic simulation for a circular trajectory

In order to simulate the behavior during the movement of the effector, a circular
trajectory is considered. The angular parameter of the trajectory is calculated with a
polynomial function:
ʹȀܶ ଷ ͳȀܶ ଶ െʹȀܶ ଷ ͳȀܶ ଶ ߙ௜௡௜௧
ߙሺ‫ݐ‬ሻ ൌ ሾ‫ ݐ‬ଷ ‫ ݐ‬ଶ ‫ͳ ݐ‬ሿ ൦െ͵Ȁܶ
ଶ െʹȀܶ ͵Ȁܶ ଶ െͳȀܶ൪ ൦ߙሶ ௜௡௜௧ ൪ ሺʹʹሻ
Ͳ ͳ Ͳ Ͳ ߙ௘௡ௗ
ͳ Ͳ Ͳ Ͳ ߙሶ ௘௡ௗ
where ܶ is the period of the circular trajectory, ߙ௜௡௜௧ is the initial angle position, ߙሶ ௜௡௜௧
is the initial angle velocity, ߙ௘௡ௗ is the final angle position and ߙሶ ௘௡ௗ the final velocity.
The circular trajectory follows the equations:
‫ ݔ‬ൌ ‫ ݎ‬ሺߙሻ ǡ ‫ ݕ‬ൌ ‫ ݎ‬ሺߙሻ ǡ ‫ ݖ‬ൌ Ͳ ሺʹ͵ሻ

The polynomial function ߙሺ‫ݐ‬ሻ is used to be sure that the initial acceleration is null
with an initial velocity ߙሶ ௜௡௜௧ also null. The simulated trajectory is a circle of radius
33mm with a 1-second period that represents a path velocity of 207 mm/s. The move-
ment of the platform is controlled by the variation of cable lengths calculated using the
inverse geometric model. Disturbances are caused solely by the elastodynamic behavior
of the robot.
Fig. 5. Circular error with nylon cable for a circular trajectory
Fig. 6. Circular error with steel cable for a circular trajectory

The position error is shown on figure 5 for a nylon cable and figure 6 for a steel
cable. Simulation without preload and with a preload determined with a force distribu-
tion algorithm are shown. The close-form algorithm[11] was used to compute the force
distribution with ‫ݐ‬௠௜௡ ൌ ͳͲܰ and ‫ݐ‬௠௔௫ ൌ ʹͲܰ. The position error is higher for the ny-
lon cable than steel. The maximum circular error is ͷʹǤͷߤ݉ for steel cable and ͳǤʹ݉݉
for nylon without preload. The error is lower when using the force distribution algo-
rithm. The maximum circular error is ͹Ǥ͸ߤ݉ for steel cable and ͳͻͶߤ݉ for nylon with
preloads. The dynamic performance is better with a preload on the cable.
4 Conclusion
The influence of the preload of the cables on the static and dynamic behavior has been
studied. This has been done by computing the robot stiffness matrix from a second order
approximation of the deformation energy of preloaded cables, expressed as a function
of infinitesimal displacement coordinates.
In terms of static and modal behavior, the preload has a significant impact mainly
on the rotational stiffness around the Z axis and the first natural frequency. For a dy-
namic trajectory, the preload plays a significant role. It keeps the cables under tension
and prevents the discontinuity due to slack cables. Moreover, the preload improves the
trajectory accuracy.
The preload influence is more significant with a low elastic module. For instance,
the preload impact is more significant with the nylon cable than the steel cable. This
result may seem intuitive, but it remains necessary to quantify the gain in terms of ri-
gidity and natural frequencies in order to optimize the choice of cables. In the literature,
rigid cables are generally chosen.
The developed dynamic model can be used to choose the preload on the cable as
well as to determine the minimal preload to avoid slack cable. This study has been
carried out for a constant orientation of the mobile platform, but the developments pre-
sented herein should be exploited to investigate the workspace at different orientations.
However, the dynamic model chosen considers only the cable elasticity. Future works
need to integrate the viscoelastic behavior of the cables. The damping effect can help
to stabilize and reduce the oscillations amplitude, especially for low stiffness cables.
Although this study was undertaken on a totally constrained cable robot, it may be ex-
tended to other CDPR applications.
Acknowledgements
This paper is part of the CABFAB project funded by the Auvergne-Rhône-Alpes
Region as part of the 2017 Ambition Pack Research program.
References
1. Gosselin, C.: Cable-driven parallel mechanisms: state of the art and perspectives. Mech.
Eng. Rev. 1, DSM0004-DSM0004 (2014).
2. Gagliardini, L., Caro, S., Gouttefarde, M., Wenger, P., Girin, A.: A Reconfigurable
Cable-Driven Parallel Robot for Sandblasting and Painting of Large Structures. In: Pott,
A. and Bruckmann, T. (eds.) Cable-Driven Parallel Robots. pp. 275–291. Springer
International Publishing, Cham (2015).
3. Pott, A., Mütherich, H., Kraus, W., Schmidt, V., Miermeister, P., Verl, A.: IPAnema: A
family of Cable-Driven Parallel Robots for Industrial Applications. In: Bruckmann, T.
and Pott, A. (eds.) Cable-Driven Parallel Robots. pp. 119–134. Springer Berlin
Heidelberg, Berlin, Heidelberg (2013).
4. Landsberger, S., Sheridan, T.: Parallel Link Manipulators, (1987).
5. Diao, X., Ma, O.: Vibration analysis of cable-driven parallel manipulators. Multibody
Syst. Dyn. 21, 347–360 (2009).
6. Tourajizadeh, H.: Longitudinal and Lateral Vibration Analysis of Cables in a Cable
Robot using Finite Element Method. 10, (2017).
7. Yuan, H., Courteille, E., Deblaise, D.: Static and dynamic stiffness analyses of cable-
driven parallel robots with non-negligible cable mass and elasticity. Mech. Mach.
Theory. 85, 64–81 (2015).
8. Yuan, H., Courteille, E., Gouttefarde, M., Hervé, P.E.: Vibration analysis of cable-
driven parallel robots based on the dynamic stiffness matrix method. J. Sound Vib. 394,
527–544 (2017).
9. Weber, X., Cuvillon, L., Gangloff, J.: Active vibration canceling of a cable-driven
parallel robot using reaction wheels. 2014 IEEE/RSJ Int. Conf. Intell. Robot. Syst.
1724–1729 (2014).
10. Rushton, M., Khajepour, A.: Transverse vibration control in planar cable-driven robotic
manipulators. Mech. Mach. Sci. 53, 243–253 (2018).
11. Pott, A., Bruckmann, T., Mikelsons, L.: Closed-form Force Distribution for Parallel
Wire Robots. In: Computational Kinematics. pp. 25–34 (2009).
Slackening Effects in 2D Exact Positioning in
Cable-Driven Parallel Manipulators
Erika Ottaviano1 , Andrea Arena2 , Vincenzo Gattulli2 , and Francesco Potenza3

1
University of Cassino and Southern Lazio, Department of Civil and Mechanical
Engineering, Cassino, FR, 03043, Italy, [email protected],
2
Sapienza University of Rome, Department of Structural and Geotechnical
Engineering, Rome, 00184, Italy, [email protected],
[email protected],
3
University of L’Aquila, Department of Civil Architectural and Environmental
Engineering, L’Aquila, 67100, Italy, [email protected]
Abstract. An analytical model of elastic cables is presented to eval-

uate the influence of the cables sag in the exact positioning of a point
mass end-effector in a two-dimensional space within the context of cable-
driven parallel manipulators. Direct and inverse kinematic problems are
formulated by means of 3n nonlinear equations expressing compatibil-
ity, equilibrium, and targeting conditions. Examples with an increasing
number of cables are proposed to show that, in specific regions of the
workspace, solving the inverse kinematics problem implies very slack
configurations of redundant cables.
Keywords: cable-driven parallel robot, sag cables, inverse kinematics,

direct kinematics
1 Introduction
Cable Driven Parallel Manipulators (CDPMs) have been developed since few
decades and are currently studied because of their inherent advantages with re-
spect their classical counterparts composed by rigid links only. The main element
to consider in the study of the CDPMs is the cable and its correct modeling.
The complexity is related to the use of cables as tools for actuation and ma-
nipulation of payloads. Indeed, cables have the characteristic of being only able
to work in tension, and not in compression leading to many difficulties with
kinematic, static, dynamic analyses and control. Some works are related to the
low inertia of such mechanisms [7]. Applications taking advantage of the large
working volume are the NIST Robocrane [4] and the systems proposed in [13,
2]. When the CDPMs are in a crane configuration, where gravity acts like an
additional cable, not all the end-effector DOFs can be controlled.
Underconstrained cable robots offer several advantages with respect to fully
constrained ones. A smaller number of cables means reduced number of actua-
tors, overall costs and setup time, improved ease of assembly and a lower likeli-
hood of cable interference. Within the context of the underconstrained CDPMs

320 E. Ottaviano et al.
kinematics, the challenging problem is that the exact positioning of the end-
effector can be achieved with difficulty. Therefore, loop-closure or kinematic
compatibility and equilibrium equations must be simultaneously solved as pro-
posed in [14] and traditional displacement-analysis problem has been defined as
geometrico-static also expressing some connections between stability and energy
[3]. In [10] is solved the Inverse Kinematics (IK) problem considering the end-
effector pose and evaluating the cables forces and lengths. The IK problem is
solved in [8] with the hypothesis of inextensible cable first, and then using a
parabolic cable profile equation. In [12] both Direct Kinematics (DC) and IK
problems are solved within the context of sagged elastic cables. In [15] an ap-
proach based on energy and optimization was proposed to solve the DK problem.
The energy method was also adopted in [5] to formulate the equations of motion
of CDPMs to take into account cables vibrations. The importance of vibrations
in cable-suspended robots and their mitigation via active control were discussed
in [6]. When dealing with non-trivial cable models and underconstrained cable
robots the main challenge is related to the real-time analysis capability of the
controller. Indeed, simplified models allow fast computation but leading to in-
accurate solutions, while non-trivial cable modeling may lead to ill-conditioned
behavior at certain end-effector positions in the workspace.
In this paper an analytical model for solving numerically the DK and the
IK problems in CDPMs working in a 2D space is presented. In particular, for a
given, arbitrary number n of elastic cables the solution of the IK and equilibrium
problems is calculated starting from the solution of the DK approach and it is
found in terms of the cables incremental lengths and tensions so as to provide
the exact positioning of a point-mass end-effector across the workspace. To this
aim, a kinematic descriptor, namely the vectorial distance between the end-
effector position evaluated through a DK analysis and the target position, is
minimized. It is also shown that there are specific regions in the 2D workspace
where compatible and equilibrium solutions, ensuring the exact positioning of
the end-effector, exist but entail very slack configurations of at least one cable.
2 Problem Formulation
In the two-dimensional (2D) Euclidean space, a system of n elastic cables sub-
jected to fixed boundary condition at one end and connected together at their
free boundary is here considered to study the equilibrium configurations reached
by the cables for assigned target positions of an end-effector point mass M col-
located at the cables free boundaries.
2.1 Continuum Model of Elastic Cables

The analytical model adopted to describe the cables overall 2D kinematics,
illustrated in the schematic representation in Fig. 1, is parametrized by the
space coordinate along the unstretched arclength Si (i = 1, . . . , n) of each ca-
(i) (i)
ble. By introducing n fixed Cartesian frames (ex , ey ) having their origin Oi
Slackening Effects in 2D Exact Positioning in Cable-Driven… 321
(i)
in the fixed boundary of the ith cable, where ey represents the upward ver-
(i)
tical direction and ex is collinear with the horizontal direction, the reference,
unstretched configurations of the cables are described by the position vectors
(i) (i)
pi (Si ) = x(Si )ex + y(Si )ey , while the direction of the cables arclength is given
(i) (i)
by the unit vector ai = ai,x ex + ai,y ey . On the other hand, the unstretched
cables configuration is assigned a priori by assuming a trial collocation of the
(1) (1)
mass M whose position, in the frame (ex , ey ) of the first cable, is given by
(1) (1)
the position vector p0M,1 = l1 a01 , where l1 and a01 = a01,x ex + a01,y ey are the
distance and the direction, respectively, between the cable fixed end and the
mass M . Accordingly, the position of the fixed end of the ith cable is known and
(1) (1)
it is assigned through the vector rOi = d1i ex + h1i ey (i = 1, . . . , n), where
d1i and h1i are the horizontal and vertical distances, respectively, between the
fixed ends of the first and the ith cable. The remaining n − 1 expressions of the
mass position vectors p0M,i with respect to the ith frame, can be determined by
the n − 1 vectorial relations rOi + p0M,i = p0M,1 ; the latter, can be written in
component form as d1i + li a0i,x = l1 a01,x , h1i + li a0i,y = l1 a01,y , which, together
with the normalization condition (a0i,x )2 + (a0i,y )2 = 1, provides the expressions
of li , a0i,x , and a0i,y (i = 2, . . . , n), respectively.
H\ GL
9 H\ L

2 H[ [ 6 9
+ U2L 2 +
9L KL
V [ L 6L +L
6 \ 6
S0L OLDL 2L H[ L

S0 OD \L 6L VL
/
0
/ /L
PJ
SL /L
S / /L
'SL
0J
0
Fig. 1. Schematic representation of the CDPM.
In the continuum formulation here adopted, the flexural deformation mode

of the cables [1] is neglected, thus the only generalized strain of the ith cable is
provided by the stretch νi . The latter, can be defined as νi = dsi /dSi , where si
represents the arclength of the ith cable in the equilibrium configuration and d
stands for total differentiation. The stress state in the ith cable is given by the
vector ni = Ni ai , where, by considering a linear elastic constitutive behavior,
the tension Ni can be expressed in terms of the cable stretch as Ni = EAi (νi −1),
being EAi the cable axial stiffness.
By introducing the horizontal Hi and the vertical Vi components of the cable
tension in the ith reference frame, the following equilibrium equations for the
ith cable holds
Ni ai,x = Hi ,
$S (1)
Ni ai,y = −Vi + 0 i m g dSi .
where the integral term in the right-hand side of the second equation in (1) rep-
resents the weight of the ith cable having unstretched length Si , i.e., m g Si =
$ Si
0
m g dSi , being m the cable mass per unit unstretched length and g the gravity
acceleration. By solving Eq.(1) in terms of the directors ai,x and ai,y , and provid-
ing the normalization condition a2i,x + a2i,y = 1, it is possible to obtain th expres-
sion of the tension in the ith cable in the form Ni = Si2 + Hi2 + Vi2 − 2Si Vi .
#
By now enforcing the definition of the stretch νi , which implies that dsi =
νi dSi , the directors of the ith unit vector ai can be cast in the following form:
ai,x = dxi /dsi = dxi /νi dSi , ai,y = dyi /dsi = dyi /νi dSi . By combining the
obtained expressions of the directors ai,x and ai,y and the constitutive equation
is then possible to obtain the differential relations between the components of
the position vector pi (Si ) of the ith cable and the unstretched arclength Si in
the form

Ni Ni
dxi = ai,x + 1 dSi , dyi = ai,y + 1 dSi , (2)
EAi EAi
with: ai,x = Hi /Ni and ai,y = (m g Si − Vi ) /Ni . Equation (2), integrated in

[0, Si ], provides the equilibrium configuration of the ith cable in terms of the
horizontal and the vertical components of the position vector pi (Si ).
Nondimensional form. The nondimensional form of Eq.(2) and of the equilib-

rium equations (1) are obtained here to identify the main mechanical parameters
governing the static configuration of the ith elastic cable. To this end, the span
l1 of the first cable is adopted as characteristic geometric parameter for the
nondimensionalization. Thus, ξi = Si /l1 is the nondimensional arclength in the
unstretched configuration, while x̄i = xi /l1 and ȳi = yi /l1 represent the horizon-
tal and vertical components of the ith cable position vector in the equilibrium
configuration, respectively. Moreover, by assuming f c = m g l1 as characteris-
tic force, the following nondimensional expressions of the components of the
cables tension can be provided: αi = Hi /m g l1 , βi = Vi /m g l1 . Finally, the
nondimensional tension in the ith cable can be written as N̄i = Ni /m g l1 =
κi (νi −
#1), where κi = EAi /m g l1 is the nondimensional cable stiffness, thus
N̄i = ξi2 + αi2 + βi2 − 2ξi βi . By now integrating Eq.(2), the equilibrium config-
uration of the ith cable, in its nondimensional form, can be calculated as [9]
% &
x̄i (ξi ) = x̄i (0) + αi κξii + sinh−1 ξiα−β i
i
+ sinh −1 βi
αi ,
ξi (ξi −2βi )
(3)
+ ξi + αi + βi − 2ξi βi − αi2 + βi2 ,
# #
ȳi (ξi ) = ȳi (0) + 2 2 2
2κi
where sinh−1 is the inverse hyperbolic sine function. Finally, to ensure the fixed
boundary conditions at the cables ends, x̄i (0) = ȳi (0) = 0 (i = 1, . . . , n).
2.2 Equilibrium and Kinematically Compatible Configurations

The system (3) provides the nondimensional expression of the configuration of
the ith cable as function of the elastic stiffness κi and of the internal forces αi
and βi . Although, to ensure that the n cables are in equilibrium under the effect
of the external forces acting on them, the following two relations hold
n
n
n

αi = 0, βi − ηi − W̄ = 0, (4)
i=1 i=1 i=1
where W̄ = M g/m g l1 is the nondimensional weight of the point mass M ,

while ηi = m g L0,i /m g l1 represents the nondimensional weight of the ith cable,
which, in turn, corresponds to the ratio between the unstretched length L0,i of
the ith cable and the span l1 of the first cable, i.e., ηi = L0,i /l1 .
In addition, kinematic compatibility conditions must be provided in order to
ensure that the free boundaries of the n cables are connected to each other. To
this aim, the following n − 1 vectorial relations hold
r̄Oi + p̄i (ηi ) − p̄1 (η1 ) = o, (i = 2, . . . , n) (5)
where r̄Oi = rOi /l1 , p̄i (ηi ) and p̄1 (η1 ) are the nondimensional position vectors
of the mass M with respect to the fixed end of the ith cable (i.e., when ξi = ηi )
and the first cable (i.e., when ξ1 = η1 ), respectively, and o is the trivial vector
with null components.
3 Direct vs Inverse Approach

Equation (5), in component form, provides 2(n − 1) compatibility conditions
which, together with the 2 equilibrium equations (4), give the 2n equations that
allow to calculate the 2n unknowns αi and βi (i = 1, . . . , n) ensuring that the
mechanical system is kinematically compatible and in equilibrium. The above
mentioned system of 2n nonlinear equations in 2n unknowns can be solved nu-
merically for selected values of the cables unstretched lengths L0,i , assigned
through the nondimensional parameters ηi , and for a set of trial positions of the
mass M , which, in turn, are provided by the nondimensional position vectors
p̄0M,i = p0M,i /l1 . Although, due to the elastic feature of the cables, the final po-
sition of the mass (given in the ith reference frame by the vector p̄i (ηi )) will not
be coincident to the assigned trial position.
Finding a closed-form solution of the system of the 2n nonlinear equations is

challenging and out of the scope of the present work; nevertheless, approximate
solutions can be found numerically by using minimization methods. To this end,
a numerical algorithm for constrained global optimization implemented in the
software Mathematica was used. In particular, the Random Search algorithm ac-
counting for a large number of search points is adopted to minimize the objective
function given by the squared norm of the (2n×1) vector collecting the left-hand
side (LHS) of the system of 2n nonlinear equations. To overcome the limits of
the direct kinematic approach in finding the cables configurations leading to the
mass exact positioning, an inverse kinematic approach is adopted to drive the
position of the end-effector to target collocations. To this end, the target position
(1) (1)
vector p̄T , whose components are given in the fixed frame (O1 , ex , ey ) of the
first cable, is introduced to define the difference vector between the position of
the mass evaluated through a direct kinematic analysis and the target position
vector. Since the mass position, in the stretched cables configurations, can be
arbitrary provided with respect to each cable by the sum vector r̄Oi + p̄i (ηi ),
therefore, it is possible to evaluate n difference vectors Δp̄i (i = 1, . . . , n) as
Δp̄i = r̄Oi + p̄i (ηi ) − p̄T ; the latter turn out to be Δp̄i = o in the configu-
rations achieved through a direct kinematic approach. On the other hand, the
mass target position is reached when Δp̄i = o.
Starting from the known configurations attained through the DK, equilibrium
equations (4) and compatibility equations (5) can be written in their incremental
form by introducing the following expressions: αi = αi0 + Δαi , βi = βi0 + Δβi
and ηi = ηi0 + Δηi , where α0 and βi0 are solution of the direct problem, while ηi0
are the assigned parameters providing the unstretched cables lengths. Therefore,
3n incremental unknowns, namely Δαi , Δβi , and Δηi (i = 1, . . . , n), are intro-
duced and the solution of the inverse kinematic problem can be then numerically
searched by simultaneously solving the set of 3n equations given by
'n 0
i=1 αi + Δαi = 0,
'n 'n
βi + Δβi − i=1 ηi0 + Δηi − W̄ = 0,
0
i=1
(6)
r̄Oi + p̄i (ηi0 + Δηi ) − p̄1 (η10 + Δη1 ) = o, (i = 2, . . . , n),
||Δp̄i || = 0, (i = 1, . . . , n),
where the last of (6) represents n target equations. In accordance with the numer-
ical procedure performed to find the approximate solution of the direct problem,
which depends largely on a good initial guess [12], the numerical global opti-
mization based on the Random Search algorithm is adopted to minimize the
objective function given, for the case of the inverse kinematics, by the squared
norm of the (3n × 1) vector collecting the left-hand side (LHS) of the nonlinear
equations system (6).
3.1 Mechanical Parameters of the Selected CDPMs
Two cables configurations are investigated, namely, case-study A (CSA) and

case-study B (CSB), respectively. In particular, CSA accounts for three sus-
pended cables having their fixed end at the same height and being symmetri-
cally positioned along the horizontal direction, while in CSB are considered four
cables nonsymmetrically positioned in the workspace. By referring to the me-
chanical parameters reported in [14], in both cases the cables are characterized
by the same axial stiffness EAi = 628.319 N (i = 1, . . . , n) and the end-effector
mass is M = 5.326 × 10−2 kg. Moreover, the positions of the cables fixed end
are assumed as follows: in CSA, d12 = 1 m, d13 = 2 m and h12 = 0 = h13 , while
in CSB, d12 = 0.5 m, d13 = 1.25 m, d14 = 2 m, and h12 = −0.1 m , h13 = 0.25
m, h14 = 0.15 m.

Fig. 2. Cables equilibrium configurations (Λ0i = 1.01) for CSA: direct approach (blue
lines) vs. inverse approach (red lines). The gray dashed lines indicate the distances li
(i = 1, . . . , n); dimensions in [m].
The reference configuration of the ith cable is provided by assigning the cable
aspect ratio Λ0i [9, 11], that is, the ratio between the unstretched cable length L0,i
and the distance li between its fixed end and the mass target position. Conse-
quently, the nondimensional unstretched length of the ith cable can be calculated
as ηi0 = Λ0i li /l1 . As demonstrated in [14], the sensitivity of the end-effector mass
positioning with respect to the prescribed parameter Λ0i is a study of valuable
interest in the case of direct kinematic approach, where the accurate tuning of
Λ0i may provide an end-effector mass positioning close, although not coincident,
to the target position. On the other hand, in the inverse kinematic approach,
the final cables aspect ratios are provided
by the minimization procedure and
can be calculated as Λi = ηi0 + Δηi l1 /li . Therefore, in both case-studies here
investigated, Λ0i = 1.01 corresponds to the initial, tentative, aspect ratio for each
cable (i.e., i = 1, . . . , n).
Table 1. CSA: coordinates (in [m]) of the target points compared with the end-effector
mass position reached through the direct and the inverse approach, respectively. Nu-
merical approximation of the LHS of Eq. (6).
Case xT yT x D yD xI yI ||LHS||2 ||LHS||

(a) 0 -0.2 0.0172 -0.2014 0.00169 -0.1999 1.43×10−4 1.19×10−2
(b) 0.3 -1 0.2966 -1.0126 0.3 -1 8.09×10−29 8.99×10−15
(c) 0.5 -1.6 0.5006 -1.6183 0.5 -1.6 2.57×10−29 5.07×10−15
(d) 1 -2 1 -2.0214 1 -2 1.68×10−21 4.09×10−11
In Fig. 2 are shown, for the CSA, the equilibrium configurations undergone
by the cables for 4 selected end-effector positions across the grid of target points
investigated. Accordingly, Tab. 1 shows for the cases displayed in Fig. 2 the final
positions of the end-effector mass attained through the direct approach (i.e.,
xD and yD ) and through the inverse approach (i.e., xI and yI ) compared with
the selected target positions (i.e., xT and yT ). The last two columns of Tab.
1 report the approximation of the numerical solution of the nonlinear system
(6) in terms of the squared norm ||LHS||2 and of the norm ||LHS|| of its left-
end side expressions. Due to the superimposed value of Λ0i = 1.01, the cables
configurations attained by the direct kinematics approach (blue lines in Fig. 2)
are non-slack in the whole workspace, although, the final end-effector position
does not match the target points due to the effect of the cables elasticity, as shown
in Tab. 1. On the other hand, the formulation based on the inverse kinematics
(red lines in Fig. 2) allows to obtain the cables elastic configurations such as
to provide the exact mass positioning (see xI and yI in Tab. 1). Nevertheless,
as depicted in Fig. 2 (Case (a)), several target points cannot be reached by
ensuring non-slack Λi ≤ 1 or moderately slack configurations for all the cables.
This results in a higher residual of the LHS of Eq. (6) and in larger cables
aspect ratios Λi >> 1, as shown in Tab. 2 where, together with the values of
Λi , are reported the cables axial forces Ni in the proximity of the mass and the
cables stretched lengths Li . Finally, in Fig. 3, are reported the axial forces in
the proximity of the end-effector and the cables total length for all target points
reached through non-slack or moderately slack configurations.
Table 2. CSA: results in terms of axial forces Ni (in [N]), cables total stretched lengths
Li (in [m]), and cables aspect ratio Λi .
Case N1 N2 N3 L1 L2 L3 Λ1 Λ2 Λ3
(a) 0.588 0.024 0.042 0.199 4.565 7.854 0.999 4.48 3.91
(b) 0.425 0.119 0.057 1.044 1.221 1.980 0.999 0.999 1.004
(c) 0.314 0.213 0.037 1.676 1.676 2.206 0.999 0.999 1.006
(d) 0.043 0.451 0.043 2.240 2.000 2.240 1.002 0.999 1.002
Fig. 3. Axial forces Ni (in [N]) and total lengths Li (in [m]) for each target position
of CSA (coordinates are in [m]).
To show the capability of the proposed model to deal with generic cables ar-
rangements, the case of four, nonsymmetrically positioned cables (i.e., the CSB)
is investigated next. Figure 4 shows the cables configurations for selected the
target points; in particular, case (a) shows very slack configurations attained in
a selected target point positioned along the boundary of the workspace (i.e., gray
points at x = 0 m) by the three cables positioned far away from the boundary,
while cases (b), (c) and (d) show the non-slack and the moderately slack config-
urations undergone by all cables to reach the target points inside the workspace.
Blue and red lines represent the configurations obtained via the DK and the IK
approach, respectively. In Tab. 3 and Tab. 4 are reported the numerical errors
and the cables mechanical parameters for the selected four cases of CSB. In Fig.
5 are shown, for CSB, the axial forces in correspondence with the end-effector
and the cables total length for all target points reached through non-slack or
moderately slack configurations.
Table 3. CSB: coordinates (in [m]) of the target points compared with the end-effector
mass position reached through the direct and the inverse approach, respectively. Nu-
merical approximation of the LHS of Eq. (6).
Case xT yT xD yD xI yI ||LHS||2 ||LHS||

(a) 0 -0.5 0.0396 -0.5039 0.0034 -0.4999 1.1×10−4 1.1×10−2
(b) 0.25 -0.975 0.2525 -0.9854 0.25 -0.975 5.9×10−28 2.4×10−14
(c) 0.5 -1.2125 0.5056 -1.2244 0.5 -1.2125 2.7×10−15 5.2×10−8
(d) 1.75 -1.925 1.7387 -1.9463 1.75 -1.925 1.6×10−27 4.1×10−14

Fig. 4. Cables equilibrium configurations (Λ0i = 1.01) for CSB: direct approach (blue
lines) vs. inverse approach (red lines). The gray dashed lines provide the distances li
(i = 1, . . . , n); dimensions are in [m].
A deeper investigation of the cables configurations in the neighborhood of

the target points positioned along the workspace boundaries (i.e., xT = 0 m and
xT = 2 m) was performed next to define the workspace of the CDPM where the
cables are non-slack or moderately slack (i.e., Λi ≈ 1). For both case-studies (i.e.,
CSA and CSB) a thicker grid of target points was used to determine, via the
inverse kinematic approach, the end-effector positions where all cables exhibit
a moderately slack configuration, i.e., Λi < 1.1. In Fig. 6 are shown in red
Table 4. CSB: results in terms of axial forces Ni (in [N]), cables total stretched lengths
Li (in [m]), and cables aspect ratio Λi .
Case N1 N2 N3 N4 L1 L2 L3 L4 Λ1 Λ2 Λ3 Λ4
(a) 0.703 0.034 0.066 0.0799 0.499 6.634 12.78 15.2 0.999 10.36 8.767 7.226
(b) 0.376 0.139 0.019 0.042 1.007 0.910 1.594 2.095 0.9994 0.9998 1.0079 1.0071
(c) 0.083 0.424 0.015 0.027 1.312 1.112 1.654 2.048 1.0 0.9993 1.0063 1.0106
(d) 0.041 0.020 0.025 0.467 2.618 2.235 2.234 2.090 1.0062 1.0104 1.0009 0.9992
Fig. 5. Axial forces Ni (in [N]) and total lengths Li (in [m]) for each target position
of CSB (coordinates are in [m]).
the target points for which at least one cable results very slack (i.e. Λ > 1.1),
therefore redundant, while black points indicate the target positions reached by
non-slack or moderately slack cables configurations.

y
y

x x
Fig. 6. Target points for CSA (left) and CSB (right) where at least one cable possesses
a very slack configuration (red) and where all cables are non-slack or moderately slack
(black); coordinates are in [m].
4 Conclusions
The studies performed in this paper show that the exact positioning of the
end-effector in specific area of the workspace of CDPMs may imply a large sag
in “redundant” cables. This occurrence is due to the small contribute of these
cables to the equilibrium of the end-effector and cannot be studied with simplified
models of CDPMs. The presented model permits to determine reachable portion
of the workspace when the maximum cable sag is given.
Acknowledgements
This work was partially supported by EU Call RFCS-2017 through the research
project DESDEMONA (grant agreement number 800687).
References
1. Arena, A., Pacitti, A., Lacarbonara W.: Nonlinear response of elastic cables with
flexural-torsional stiffness. Int J Solids Struct, 87, 267–277 (2016).
2. Bosscher, P.: Cable-suspended robotic contour crafting system. Automat Constr 17:
45–55 (2006).
3. Carricato, M., Merlet, J-P.: Stability analysis of underconstrained cable-driven par-
allel robots. IEEE Trans. on Robotics 29(1):288–296 (2013).
4. Castelli, G., Ottaviano, E., Rea, P.: A Cartesian Cable-Suspended Robot for im-
proving end-users’ mobility in an urban environment. Robot Cim-Int Manuf 30(3):
335–343 (2014).
5. Du, J., Agrawal, SK.: Dynamic Modeling of Cable-Driven Parallel Manipulators
with Distributed Mass Flexible Cables. J Vib Acoust 137(2): 1–8 (2015).
6. Gattulli, V., Alaggio, R., Potenza, F.: Analytical Prediction and Experimental Val-
idation for Longitudinal Control of Cable Oscillations. Int J Nonlin Mech 43: 36–52
(2008).
7. Gonzalez-Rodriguez, A., Castillo-Garcia, F.J., Ottaviano, E., Rea, P., Gonzalez-
Rodriguez, A.G.:On the effects of the design of cable-Driven robots on kinematics
and dynamics models accuracy. Mechatronics 43: 18–27 (2017).
8. Gouttefarde, M., Collard, J., Riehl, N., Baradat, C.: Simplified static analysis of
large-dimension parallel cable-driven robots. IEEE International Conference on
Robotics and Automation, Saint Paul, MN: 2299–2305 (2012).
9. Irvine, H.: Cable Structures. Cambridge, MA: MIT Press (1981).
10. Kozak et al.: Static Analysis of Cable-Driven Manipulators with Non-Negligible
Cable Mass. IEEE Trans. on Robotics 22(3):425–433 (2006).
11. Lepidi, M., Gattulli, V.: Static and Dynamic Response of Elastic Suspended Cables
with Thermal Effects. Int J Solids Struct 49(9): 1103–1116 (2012).
12. Merlet, J.-P., dit Sandretto, A.: The Forward Kinematics of Cable-Driven Parallel
Robots with Sagging Cables. In Cable-Driven Parallel Robots: 3–15 (2014) Springer.
13. Nan, R., Peng, B.: A chinese concept for 1km radio telescope. Acta Astronaut
46(10-12): 667–675 (2000).
14. Ottaviano, E., Gattulli, V., Potenza, F.: Elasto-Static Model for Point Mass Sagged
Cable-Suspended Robots. Advances in Robot Kinematics 4:351–359 (2018).
15. Pott, A, Tempel, P.: A Unified Approach to Forward Kinematics for Cable-Driven
Parallel Robots Based on Energy. Advances in Robot Kinematics 8:401–409 (2019).
Part VII
Calibration and Identification
Automatic Self-Calibration of Suspended
Under-Actuated Cable-Driven Parallel Robot using
Incremental Measurements
Edoardo Idá1 , Jean-Pierre Merlet2 , and Marco Carricato1

1 Department of Industrial Engineering (DIN), University of Bologna, Bologna, Italy
2 French National Institute for Research in Computer Science and Control (INRIA),
Sophia-Antipolis, France
[email protected]
Abstract. This paper focuses on the problem of the initial-pose estimation by

means of proprioceptive sensors (self-calibration) of suspended under-actuated
Cable-Driven Parallel Robots (CDPRs). For this class of manipulators, the initial-
pose estimation cannot be carried out by means of forward kinematics only,
but mechanical equilibrium conditions must be considered as well. In addi-
tion, forward kinematics solution is based on cable-length measurements, but
if the robot is equipped with incremental sensors cables’ initial values are un-
known. In this paper, the self-calibration problem is formulated as a non-linear
least square optimization problem (NLLS), based on the direct geometrico-
static problem, where only incremental measurements on cable lengths and
on swivel pulley angles are required. In addition, a data acquisition algorithm
and an initial value selection procedure for the NLLS are proposed, aiming at
automatizing the self-calibration procedure. Simulations and experimental re-
sults on a 3-cable 6-degree-of-freedom robot are provided so as to prove the
effectiveness of the proposed methodology.
Keywords: Cable-driven parallel robots, Underconstrained robots, Underac-

tuated robots, Homing, Self-calibration
1 Introduction
Cable-driven parallel robots (CDPRs) are a class of parallel manipulators which em-
ploys flexible cables instead of rigid links in order to control their end-effector (e-e)
pose. A CDPR is completely-constrained when the pose of its e-e can be fully deter-
mined if actuators are locked and, thus, all cable lengths are known. On the contrary,
a CDPR is under-constrained if the e-e preserves some degrees of freedom (DoFs)
when the cable lengths are assigned. This may originate from the e-e being con-
strained by a number of cables smaller than its DoFs or from some cables being
slack [4]. This paper focuses on suspended CDPRs, in which all cables are located
above the e-e, so that their tension state only depends on the action of gravity and
inertia forces. These robots are inherently under-constrained when they are under-
actuated, that is, the number of actuators is less than the number of variables needed

334 E. Idá et al.
to fully describe the manipulator. A growing number of studies are being conducted
on under-actuated CDPRs [1, 2, 4, 6, 7, 9, 17].
A major issue in the practical use of under-actuated CDPRs is the estimation of
the e-e initial-pose. When the machine is switched on in a generic start-up condi-
tion, the e-e pose is generally unknown, but its knowledge is fundamental for any
subsequent operation. In order to directly measure the e-e pose, external measure-
ment devices such as laser trackers [13] or high-resolution cameras [5] can be em-
ployed. On the other hand, an indirect estimation of the pose can be performed by
measuring some of the robot’s internal joint variables, followed by the solution of
the direct kinematic problem [10, 14]. When this approach is used for pose estima-
tion in start-up conditions, the solution to this problem is sometimes referred to as
self-calibration [3] or internal-calibration [8] of the e-e initial-pose.
In [3], a self-calibration procedure for initial-pose estimation of a 2-DoF 4-cable
over-constrained robot is proposed, which is based on cable tension and length in-
crement measurements. The proposed method relies on the over-constrained nature
of the robot. In [12] a two-stage calibration procedure for generic over-constrained
CDPRs is introduced, aiming at both optimizing robot static parameters and deter-
mining the initial-pose of the e-e. Ref. [8] shows how to perform initial-pose estima-
tion by means of a manual self-calibration procedure for over-constrained robots,
only relying on cable length increment measurement.
This paper extends the method introduced in [8] by proposing an automatic pro-
cedure to estimate the initial-pose of a generic suspended under-actuated CDPR,
e.g. a 6-DoF CDPR actuated by n < 6 cables, that only relies on incremental mea-
surements of cable length and orientation. An automatic data acquisition procedure
is exploited in order to reliably and autonomously collect the information required
for calibration purposes. The initial-pose estimation is formulated as a non-linear
least square optimization problem (NLLS), whose initial guess is generated automat-
ically according to the data acquisition algorithm. In Section 2, the geometrico-static
model of an under-actuated CDPR is developed. In Sections 3 and 4, the NLLS opti-
mization problem is formulated and the data acquisition algorithm employed for its
solution is discussed. Finally, simulation and experiments are presented.
2 Geometrico-Static Modelling
2.1 Kinematics
A generic 6-DoF under-actuated CDPR consists in a mobile platform connected to

the base by n < 6 cables, which are actuated by motorized winches (Fig. 1). Ox y z is
an inertial frame, whereas P x y z is a mobile frame attached to the e-e. The platform
pose is described by the position vector pP of P , and the rotation matrix R(φ, θ, χ) is
parametrized by Tait-Bryant angles = [φ, θ, χ]T according to the x y z convention.
The platform generalized coordinates are thus p = [pP , ]T ∈ R6 .
Cables are modelled as massless and inextensible, and are guided by swivel pul-
leys [15] to their attachment points A i , i = 1, . . . , n on the platform (Fig. 2a). G is the
Automatic Self-Calibration of Suspended Under-Actuated Cable-Driven… 335
Fig. 1: Geometry of the cable-driven robot
(a) Global view (b) Local view
Fig. 2: Geometry of the swivel pulley
platform center of mass. If the coordinates of G and A i in the mobile frame are de-

scribed by vectors P pG and P ai , their coordinates in the inertial frame are:

pG = pP + pG = pP + R P pG , ai = pP + ai = pP + R P ai (1)
The i -th swivel pulley has center C i , radius r i , and is mounted on a hinged sup-
port, whose swivel axis z i is tangent to the pulley (Fig. 2a). Vector di defines the fixed
position of point D i where the i -th cable enters the pulley’s groove and which is on
the z i -axis. Unit vectors ii , ji , ki , associated with an additional fixed reference frame
D i x i y i z i attached in D i , describe the orientation of each swivel pulley in Ox y z.
The i -th cable exit point from the pulley groove is denoted by B i and the vector
ρ i = A i − B i is tangent to the pulley.
In static conditions, if friction in the pulley hinges is negligible, vectors zi and ρ i
define the pulley plane, forming the swivel angle σi with the coordinate plane x i z i .
336 E. Idá et al.
If ui and wi are unit vectors orthogonal to ki and to each other:
wi = − sin σi ii + cos σi ji , ui = cos σi ii + sin σi ji (2)
the aforementioned static constraint can be analytically expressed as:
wi · (ai − di ) = 0 (3)
which can be solved for σi (p), if the pose is assigned.

We define ψi ∈ (−π, π) as the angle between ui and B i − C i (Fig. 2b). It is conve-
nient to define two more unit vectors, orthogonal to wi and to each other:
ni = cos ψi ui + sin ψi ki , ti = sin ψi ui − cos ψi ki (4)
so that the i -th cable vector can be expressed as:
ρ i = ai − di − r i (ui + ni ) = ρ i ti (5)
where
ρ i = l i − r i (π − ψi ) (6)
is the rectilinear cable length, and l i is the i -th total cable length, comprising the
rectilinear part and the wrapped part B i D i . The tangency constraint between ρ i and
the pulley can be expressed as:
ni · ρ i = 0 (7)
which can be solved for ψi (p), if the pose (and thus the swivel angle) are known.
Once ψi (p) and σi (p) are determined, l i (p) can be found from the geometrical
constraint that the i -th cable imposes on the platform:
ρ Ti ρ i − ρ i 2 = 0 (8)
where ρ i is evaluated as in Eq. (5) and ρ i as in Eq. (6).
2.2 Statics
For the purpose of this work, we will consider that the e-e is acted upon by a constant
force only, e.g. gravity, which is parallel to the z axis and applied at point G. Thus, if

the e-e mass is m, the gravitational acceleration is g , and p̃G is the skew-symmetric

representation of the vector product, namely pG × = p̃G , the external wrench applied

to the reference point P is W = −mg [k; p̃G k].
If all cables are taut, the static equilibrium equations of the e-e is:
JT T − W = 0 (9)
where T = [T1 , · · · , Tn ]T , Ti is the cable tension, and J ∈ Rn×6 is the Jacobian matrix
of the constraints in Eq. (8), whose i -th row is Ji = [tTi − tTi ãi ]. It is convenient to
partition J in a n × n matrix Ja and a n × (6 − n) matrix Ju , namely J = [Ja Ju ], and

vector W as W = [Wa ; Wu ], so that Eq. (9) can be rewritten as:
JTa T − Wa =0 (10)
JTu T − Wu =0 (11)
The first part can be solved for T, thus yielding:
T = J−T
a Wa (12)
and the result can be substituted in the second part, thus obtaining the static equi-
librium constraint of the platform:
JTu J−T
a Wa − Wu = 0 (13)
2.3 Forward Geometrico-Static Problem

The aim of the forward geometrico-static problem is to find the e-e pose, once the
value of a suitable set of the robot’s internal joint variables is known. In the case that
only cable length measurements are available, it is possible to formulate such a prob-
lem by considering the n constraints in Eq. (8) and the 6 − n equations in Eq. (13) [1].
The resulting system of nonlinear equations is completely determined and, thus, can
be numerically solved by using nonlinear solvers. Alternatively, if redundant mea-
surements are available, one can (possibly) neglect the static constraint equations
and replace them with additional kinematic constraint equations that are explicitly
dependent on the measured variables [10]. This approach leads to a system of equa-
tions that, depending on the number and the type of redundant measurements, can
be completely determined or overdetermined (the system is never underdetermined,
because static equilibrium constraints can always be accounted for).
In this paper, we assume that both cable lengths and swivel pulley angles can
be measured. For the sake of generality, we formulate the forward geometrico-static
problem by considering both 2n kinematic constraints, derived from Eqs. (3) and
(8), and the 6−n static equilibrium constraints of Eq. (13), thus leading to an overde-
termined system of 6 + n equations in 6 unknowns (namely, the generalized pose
coordinates in p). By letting:
σ1 (p) − σ∗1 l 1 (p) − l 1∗

⎡ ⎤ ⎡ ⎤
.. ..
F1 (p) = ⎣ ⎦, F2 (p) = ⎣ ⎦, F3 (p) = JTu J−T
a Wa − Wu (14)
⎢ ⎥ ⎢ ⎥
. .
σn (p) − σ∗n l n (p) − l n∗
where σi (p) and l i (p) can be easily calculated from Eqs. (3) and (8), and σ∗i and l i∗
are the corresponding measured values, the forward geometrico-static problem can
be formulated as: ⎡ ⎤
F1 (p)
F(p) = ⎣F2 (p)⎦ = 0 (15)
F3 (p)
This formulation leads to a fairly simple analytical formulation of the first-order dif-
ferentiation of Eq. (15), as described in Section 3.
338 E. Idá et al.
3 Initial-Pose Estimation Problem
In this section, the initial-pose estimation problem will be formulated by extending

the work presented in [8]. If the under-actuated CDPR is equipped with incremental
measurement devices on motors and swivel axes, i.e. incremental encoders, cable
lengths and swivel angles at a generic pose pi can be measured relatively to the initial
values σ0i and l i0 at pose p0 , namely:
σ∗i = σ0i + Δσ∗i (16)

l i∗ = l i0 + Δl i∗ (17)
While Δσ∗i and Δl i∗ are measures provided by the encoders, σ0i and l i0 are generally
unknown and are the objective of the self-calibration procedure. The direct geometrico-
static problem in Eq. (15) can thus be expressed as:
F(σ0 , l0 , p) = 0 (18)
where σ0 = [σ01 , · · · , σ0n ]T and l0 = [l 10 , · · · , l n0 ]T . This problem has 6 + n equations and

6 + 2n unknowns (σ0 , l0 , p), and it has generally an infinite number of solutions.
However, by assuming that λ different measurement sets are available, that is:
σ∗i ,k = σ0i + Δσ∗i ,k

k = 1, · · · , λ (19)
l i∗,k = l i0 + Δl i∗,k
the following system of equations is obtained:
G(X) = G(σ0 , l0 , p1 , · · · , pλ ) = [F(σ0 , l0 , p1 ); · · · ; F(σ0 , l0 , pλ )] = 0 (20)
where X = [σ0 ; l0 ; p1 ; · · · ; pλ ] ∈ R6λ+2n .

The system (20) has a total of λ(6 + n) equations and 6λ + 2n unknowns. Thus, if
λ > 2, the initial-pose estimation problem is overdetermined and can be formulated
as a non-linear least-square optimization:
Xopt = arg min G(X)2 (21)

X
This problem can be solved by employing numerical techniques, such as the Levenberg-
Marquardt algorithm. The efficient solution of Eq. (21) relies on a reasonable initial
solution guess Xg uess (see Section 4), and an analytical formulation of the Jacobian
matrix in Eq.(20). While the former is fundamental for both the solution accuracy
and the algorithm rapidity, the latter is critical only in terms of computational time.
The Jacobian ∂G/∂X can be expressed analitically as:
−In×n 0n×n ∂σ(p1 )/∂p ···

⎡ ⎤
0n×6 0n×6
⎢ 0
⎢ n×n −In×n ∂l(p1 )/∂p 0n×6 ··· 0n×6 ⎥
⎥
⎢0(6−n)×n 0(6−n)×n ∂F3 (p1 )/∂p 0(6−n)×6 · · · 0(6−n)×6 ⎥
⎢ ⎥
⎢ −In×n 0n×n 0n×6 ∂σ(p2 )/∂p · · · 0n×6
⎢ ⎥
⎥
−In×n ∂l(p2 )/∂p · · ·
⎢ ⎥
∂G ⎢⎢ 0n×n 0n×6 0n×6 ⎥
= ⎢0(6−n)×n 0(6−n)×n 0(6−n)×6 ∂F3 (p2 )/∂p · · · 0(6−n)×6 ⎥ (22)
⎥
∂X ⎢ ⎥
⎢ .. .. .. .. .. .. ⎥
⎢
⎢ . . . . . . ⎥
⎥
⎢ −In×n 0n×n 0n×6 0n×6 · · · ∂σ(pλ )/∂p ⎥
⎢ ⎥
⎣ 0n×n −In×n 0n×6 0n×6 · · · ∂l(pλ )/∂p ⎦
⎢ ⎥
0(6−n)×n 0(6−n)×n 0(6−n)×6 0(6−n)×6 · · · ∂F3 (pλ )/∂p
where In×n and 0n×n are the n×n identity and zero matrices, σ(pk ) = [σ1 (pk ); · · · ; σn (pk )]
and l(pk ) = [l 1 (pk ); · · · ; l n (pk )]. Manipulating Eqs. (3),(7) and (8) yields:
∂σi wT wTi ãi H ∂l i

= T i − , = tTi −tTi ãi H (23)
∂p ui (ai − di ) uTi (ai − di ) ∂p
where, respectively denoting cos x and sin x by c x and s x :

⎡ ⎤
1 0 sθ
H = ⎣0 c φ −s φ c θ ⎦ (24)
0 sφ cφ cθ
Finally, by considering that:
∂JTi ∂ti /∂p ∂W 03×1

= , = −mg (25)
∂p ãi (∂ti /∂p) − t̃i (∂ai /∂p) ∂p k̃r̃ H
and:
∂ti ∂σi ∂ψi ∂ai
= wi sin ψi + ni , = 03×3 −ãi H (26)
∂p ∂p ∂p ∂p
∂ψi nTi nTi ãi H
= − (27)
∂p ρ i ρ i
∂F3 /∂p can be evaluated according to the partition defined in Section 2.2 as:
∂F3 ∂JTu ∂JTa −T ∂Wa ∂Wu

= − JTu J−T
a J Wa + JTu J−T − (28)
∂p ∂p ∂p a a
∂p ∂p
4 Data Acquisition Algorithm
It is beyond the scope of this work to determine an optimal data acquisition algo-
rithm. However, a practical one, which enables autonomous and safe operation of
the CDPR during calibration, is provided hereafter. For this aim, cable tensions or
alternatively motor torques are assumed to be measurable or at least estimated, and
actively controlled by a suitable feedback system.
340 E. Idá et al.
In the instant the robot is switched on, its pose is generally unknown. It is possibly
unsafe to start the self-calibration process in this configuration. Then, it is useful to
pre-determine a safe start configuration, in which every cable is taut and sufficiently
long, so that it can be coiled and uncoiled, and the e-e may attain different poses. By
assigning a start cable tension vector T0 , the static problem (9) can be solved for p as
a non-linear system of six equations in six unknowns. Because of the non-linearity
of the problem, a finite set of real solutions can be determined: this calculation may
be done off-line and just once, i.e. during robot parameter calibration. Only stable
solutions [4] among the possibly many available should be considered. Additionally,
T0 may be selected so that only one stable solution exist.
In the following, we will consider a start cable tension vector leading to a unique
stable solution of problem (9). Accordingly, a (computed) start pose p0,comp is unam-
biguously determined, as well as start cable lengths l0comp and swivel angles σ0comp .
The real start pose p0 attained by the CDPR can be fairly different from the ideal one
p0,comp , and its determination is the aim of the self-calibration procedure. A maxi-
mum cable tension vector Tm should be set as well. The data acquisition algorithm
objective is to ensure that every DoF of the e-e is varied during measurements, so
that problem (21) is always well conditioned. The procedure workflow can be sum-
marized as follows.
1. Start phase: command the CDPR so that cable tensions (or motor torques) quasi-
statically reach the assigned value T0 . When T0 is reached and static conditions
are attained3 , the j -th actuator is assigned an incremental cable tension (or mo-
tor torque) set-point, starting from j = 1. The change in a single actuator set-
point ensures that the pose of the end-effector is different at any iteration, thus
being effective, as well as practical and easy to implement;
2. Tensioning phase: quasi-statically move the CDPR by assigning λ/(2n) positive
increments of magnitude ΔT = 2n(T j ,m − T j ,0 )/λ to the tension set-point of the
j -th actuator, namely T j ,k = T j ,k−1 + ΔT , where T j ,0 is the j -th component of
T0 . After each assignment k, the CDPR e-e could possibly oscillate during the
transition. When static conditions are attained, record measurements Δσ∗i ,k and
Δl i∗,k , for i = 1, · · · , n;
3. Detensioning phase: assign λ/(2n) negative increments of magnitude ΔT to the
tension set-point of the j -th actuator, namely T j ,k = T j ,k−1 − ΔT . When static
conditions are attained, record measurements Δσ∗i ,k and Δl i∗,k , for i = 1, · · · , n.
During the detensioning phase, the robot follows exactly the same cable tension
(motor torques) set-points as in the tensioning phase: on a real machine, due
to repeatability errors, this could lead to different cable lengths and swivel an-
gles, which are possibly useful in the calibration procedure in order to minimize
the repeatability error of the robot. λ/n measurement sets are thus obtained by
varying a single actuator set-point;
3 Notice that cable tensions are only used to lead the platform to poses where the robot is
stable and kinematic measures can be accurately performed. They have no role in the op-
timization problem solution, since they do not appear as variables in Eq. (21). They may
be affected by appreciable errors without compromising the procedure, whereas platform
stability plays a key role in the data acquisition process.
Table 1: Actuation unit properties

i di [m] r i [m] P a [m] xi yi zi
T i T
1 −2.030 −0.170 0.806 0.025 0.020 −0.287 0.250 e y −ez −ex
T T
2 0.066 0.920 0.757 0.025 0.251 0.153 0.250 −ex −ez −e y
T T
3 −2.043 2.241 0.738] 0.025 0.211 0.153 0.250 −e y ez −ex
4. Increment phase: j = j + 1; if j ≤ n then go to point 2, otherwise the algorithm is

finished because λ measurement sets have been recorded.
The initial guess for the solution of problem (21) is computed as:
Xg uess = σ0comp ; l0comp ; p1,comp ; · · · ; pλ,comp (29)
where pk,comp , k = 1, . . . , λ, can be evaluated by solving the static problem (9)

with assigned tension Tk . Finally, by employing Xg uess , Δσ∗k = [Δσ∗1,k · · · Δσ∗n,k ]T and
Δl∗k = [Δl 1,k
∗ ∗ T
· · · Δl n,k ] for k = 1, · · · , λ, it is possible to determine Xopt as a solution of
(21). Ideally, σ0opt and l0opt should converge to σ0 and l0 , respectively.
5 Simulations and Experimental Results
A simulation example, developed in MATLAB®, aims at testing the efficiency of the

optimization routine as a function of λ and the different formulations of the Jaco-
bian. Experimentation on a prototype shows the application of the method in a real-
word scenario. The geometrical properties of a 3-cable 6-DoF under-actuated CDPR
used for both simulation and experimentation are summarized in Table 1, where
ex = [1 0 0]]T , e y = [0 1 0]T and ez = [0 0 1]T are elements of the canonical basis of
T
SO(3). The platform mass is m = 8 Kg, and P pG = 0 0 0.182 m.
5.1 Simulation results

T
If the start tension vector T0 = 40 40 40 N is assigned, the only stable solution of
T
the static problem (9) is p0 = −1.363 0.963 −0.460 −0.056 −0.069 −0.557 m, l0 =
T T
1.470 1.509 1.543 m, and σ0 = 0.738 0.607 −0.799 rad. Since the simulation is
based on a perfect data acquisition process, the above (theoretical) values of p0 , l0
and σ0 are the expected output of the optimization algorithm.
According to the procedure proposed in Section 4, several simulations were
performed with different values of λ, i.e. λ = 12 ÷ 372. The initial guess Xg uess
for the optimization solver was generated by considering a perturbed start ten-
T
sion T0,per t = 46.9 23.9 32.6 N. In this case, the only stable solution of the
T
static problem (9) yields p0,comp = −1.648 0.740 −0.732 −0.246 0.179 −0.523 m,
342 E. Idá et al.
(a) Computational time t (b) Solution error norm e
Fig. 3: Simulation results comparing FD and J formulations
T T
l0comp = 1.468 1.914 1.802 m, and σ0comp = 0.926 0.684 −0.849 rad. Further-
T
more, Tm = 80.0 80.0 80.0 N was set. Complete numerical data regarding Xg uess ,
for k = 1, · · · , λ, are omitted due to space limitations.
In the case of ideal measurements, a solution Xopt to the problem (21) is always
found that is not only a minimum, but a real numerical zero. Solutions were com-
puted in Matlab by using the lsqnonlin solver. Figure 3 shows the computational time
and the norm of the error e = p0 −p0,opt for a variable number λ of measurement sets,
in the case a finite-difference estimation (FD) or an analytical formulation (J) of the
Jacobian. It is clear that the analytical formulation is crucial for the algorithm time-
performance, whereas there is no noticeable difference in the numerical precision
achieved by the two numerical formulations. Simulations also show that there are
no advantages in acquiring an increasing number of data points for the initial-pose
estimation. The norm of error e is consistent with the numerical tolerance.
5.2 Experimental results
The proposed data acquisition strategy and calibration method was tested on a pro-
totype. Swivel angles were measured by 16-bit incremental encoders, mounted di-
rectly on the swivel axes of pulleys, whereas cable lengths were estimated by using
20-bit incremental encoders on each motor axis and a kinematic model of the winch.
Swivel pulleys were manufactured by FDM technology, thus limited, but not negli-
gible, errors in their geometry and elasticity exist. Cables were coiled on IPAnema
winches [16]. Clearance and elasticity in the winch components, as well as cable elas-
ticity itself, are possible sources of error in the estimation of cable lengths.
In order to measure the real pose of the platform during experiments, 8 VICON
Motion Capture Systems cameras were employed to track the position of 4 markers
mounted on the robot platform.The accuracy of the measure is ±0.2mm for each di-
mension of the marker (x, y, z), according to manufacturer specifications. In the end,
the position of the reference point and the platform orientation were reconstructed
from the recorded position of each marker.
Table 2: Experiments results

i p0 [mm, ◦ ] p0,opt [mm, ◦ ] ep [mm] e [◦ ]
T T
1 −1314 900 −355 −5.0 −4.8 −32.0 −1324 906 −361 −4.5 −4.9 − 32.1 13.4 0.5
T T
2 −1345 922 −322 −4.0 −3.6 −31.9 −1347 913 −325 −4.2 −4.0 − 32.0 9.9 0.4
T T
3 −1348 917 310 −4.1 −3.4 −32.1 −1357 895 −310 −5.0 −3.6 −32.0 23.4 0.9
T T
4 −1349 917 −310 −4.1 −3.4 −32.0 −1359 914 −307 −4.2 −3.5 −31.9 11.5 0.2
T T
5 −1344 913 −314 −4.2 −3.6 −32.0 −1348 918 −315 −4.0 −3.9 −31.9 5.7 0.4
Because of the lack of force sensors in the robot set-up, motor torques were em-
ployed instead of cable tensions for the implementation of the algorithm presented
in Section 4. The start tension vector and maximum cable tensions were set to T0 =
T T
40.0 40.0 40.0 N and Tm = 80.0 80.0 80.0 N, respectively, and converted in mo-
tor torques according to static equilibrium of the cable transmissions. In the end,
λ = 60 was chosen as a trade-off between accuracy of the initial-pose estimation and
data-acquisition speed.
The results of five experiments are reported in Table 2, where p0 is the real start-
ing pose, as measured by the motion tracking system, p0,opt is the estimated start-
ing pose resulting from the solution of the problem (21), ep = p0 − p0,opt is the
reference position error norm and e = 0 − 0,opt is the norm of the error of the
orientation parameters. Positions are expressed in millimeters and angles in degrees.
The execution of the calibration procedure required, on average, 4 min for the data
acquisition procedure, and 2.5 s for the initial-pose estimation by using the analytical
formulation of the Jacobian (16 min by using a finite-difference estimation).
During experiments, it was observed that the orientation of the swivel pulley axes
plays a crucial role for the conditioning of problem (21). Pulley orientations were set
in order to achieve the best possible results with the robot architecture at hand, but
are not optimal. Nonetheless, the results are satisfactory by considering the mod-
elling simplification and the hardware at hand.
6 Conclusions
This paper presented an automatic self-calibration method for suspended under-

actuated CDPRs, extending the method proposed in [8]. An analytical formulation of
the problem was developed and an autonomous procedure to calibrate the initial-
pose of the CDPR by means of incremental measurements of cable lengths and
swivel angles was shown. Experimental results on a prototype show that the applica-
tion of the proposed method is promising, but it needs refining in order to achieve a
higher accuracy. In particular, the optimal orientation of the swivel pulleys and the
effect of uncertainties in the measurements will be investigated in the future.
344 E. Idá et al.
References
1. Abbasnejad, G., Carricato, M.: Direct geometrico-static problem of underconstrained
cable-driven parallel robots with n cables, IEEE Transactions on Robotics, vol. 31, no.
2, pp. 468-478 (2015).
2. Berti, A., Merlet, J.P., Carricato, M.: Solving the direct geometrico-static problem of under-
constrained cable-driven parallel robots by interval analysis, The International Journal of
Robotics Research, vol. 35, no. 6, pp. 723-739 (2016).
3. Borgstrom, P.H., Jordan, B.L., Borgstrom, B.J., Stealey, M.J., Sukhatme, G.S., Batalin, M.A.,
Kaiser, W.J.: Nims-pl: A cable-driven robot with self-calibration capabilities, IEEE Trans-
actions on Robotics, vol. 25, no. 5, pp. 1005-1015 (2009).
4. Carricato, M., Merlet, J.P.: Stability Analysis of Underconstrained Cable-Driven Parallel
Robots, IEEE Transactions on Robotics, vol. 29, no. 1, pp. 288-296 (2013).
5. Daney, D., Andreff, N., Chabert, G., Papegay, Y.: Interval method for calibration of parallel
robots: Vision-based experiments, Mechanism and Machine Theory, vol. 41, no. 8, pp.
929-944, (2006).
6. Hwang, S.W., Bak, J.H., Yoon, J., Park, J.H., Park, J.O.: Trajectory generation to suppress
oscillations in under-constrained cable-driven parallel robots, Journal of Mechanical Sci-
ence and Technology, vol. 30, no. 12, pp. 5689-5697 (2016).
7. Idá, E., Berti, A., Bruckmann, T., Carricato, M.: Rest-to-rest trajectory planning for planar
underactuated cable-driven parallel robots, in Cable-Driven Parallel Robots, C. Gosselin,
P. Cardou, T. Bruckmann, and A. Pott, Springer, pp. 207-218 (2018).
8. Lau, D.: Initial length and pose calibration for cable-driven parallel robots with relative
length feedback, in Cable-Driven Parallel Robots, C. Gosselin, P. Cardou, T. Bruckmann,
and A. Pott, Springer, pp. 140-151 (2018).
9. Lin, J., Liao, G.: Design and oscillation suppression control for cable-suspended robot, in
2016 American Control Conference, pp. 3014-3019 (2016).
10. Merlet, J.P.: Direct kinematics of cdpr with extra cable orientation sensors: The 2 and 3
cables case with perfect measurement and ideal or elastic cables, in Cable-Driven Parallel
Robots, C. Gosselin, P. Cardou, T. Bruckmann, and A. Pott, Springer , pp. 180-191 (2018).
11. Miermeister, P., Pott, A.: Auto calibration method for cable-driven parallel robots using
force sensors, in Latest Advances in Robot Kinematics, J. Lenarcic and M. Husty, Springer,
pp. 269-276 (2012).
12. Miermeister, P., Pott, A., Verl, A.: Auto-calibration method for overconstrained cable-
driven parallel robots, in ROBOTIK 2012; 7th German Conference on Robotics, pp. 1-6
(2012).
13. Nubiola, A., Slamani, M., Joubair, A., Bonev, I.A.: Comparison of two calibration methods
for a small industrial robot based on an optical CMM and a laser tracker, Robotica, vol.
32, no. 3, pp. 447-466 (2014).
14. Pott, A.: An algorithm for real-time forward kinematics of cable-driven parallel robots,
in Advances in Robot Kinematics: Motion in Man and Machine, J. Lenarcic and M. M.
Stanisic, Springer, pp. 529-538 (2010).
15. Pott, A.: Influence of pulley kinematics on cable-driven parallel robots, in Latest Advances
in Robot Kinematics, J. Lenarcic and M. Husty, Springer, pp. 197-204 (2012).
16. Pott, A., Mütherich, H., Kraus, W., Schmidt, V., Miermeister, P., Verl, A.: IPAnema: A fam-
ily of Cable-Driven Parallel Robots for Industrial Applications,in Cable-Driven Parallel
Robots, T. Bruckmann and A. Pott, Springer, pp. 119-134 (2013).
17. Zarei, M., Aflakian, A., Kalhor, A., Masouleh, M.T.: Oscillation damping of nonlinear con-
trol systems based on the phase trajectory length concept: An experimental case study
on a cable-driven parallel robot, Mechanism and Machine Theory, vol. 126, pp. 377-396,
(2018).
Eye-on-Hand Calibration Method for
Nicolas Tremblay1 , Kaveh Kamali2 , Philippe Cardou1 , Christian Desrosiers2 ,

Marc Gouttefarde3 , and Martin J.-D. Otis4
1
Laboratoire de Robotique, Département de génie mécanique, Université Laval,
Québec City, QC, Canada
2
École de technologie supérieure, Montréal, QC, Canada
3
[email protected]
4
Laboratoire d’automatique et d’interaction 3D multimodale intelligente (LAIMI),
Département des Sciences Appliquées, Université du Québec à Chicoutimi (UQAC),
Chicoutimi, QC, Canada
martin [email protected]
Abstract. Estimating the geometric parameters of a cable-driven par-

allel robot (CDPR) can be a labour intensive process or one that requires
expensive sensors. This paper presents a low-cost method for estimating
initial cable lengths and fixed cable attachment points of CDPRs. The
proposed approach relies on the detection, mapping and localisation of
fiduciary markers in the robot environment using a camera attached to
the end effector. This paper additionally tackles the generation of a list
of reachable calibration poses and presents a control scheme allowing
the CDPR to reach those. Experiments are also carried out to assess the
performance of the proposed calibration method. It appears that the pro-
posed eye-on-hand method is more accurate than a previously reported
method relying solely on cable-length measurements.
Keywords: Cable-driven parallel robot, parameter identification, ex-

perimental testing, vision-based calibration
1 Introduction
Cable-driven parallel robots (CDPRs) are a type of parallel robots where rigid
links are replaced by cables, known for their advantageous payload to weight ra-
tio which allows for higher dynamic capabilities, potentially larger translational
workspace, portability and reconfigurability. In order to fully benefit from these
capabilities, the CDPR geometric parameters have to be determined accurately,
which can prove to be difficult on a medium to large scale. Indeed, a poor knowl-
edge of these parameters inevitably degrades the positioning accuracy of the end
effector.

346 N. Tremblay et al.
Excluding conceptually simple but impractical direct measurement of the

geometry, some techniques have been developed for the Gough-Stewart platform
calibration, but also apply to CDPRs. A popular and straightforward method
uses measurements of the end effector pose in order to minimise inverse kinematic
residuals [16]. The required measurements are often obtained using external
sensors, such as a redundant leg [12], a theodolite [15] or a camera [13]. The
calibration can also be performed by imposing additional kinematics constraints
to render the internal sensors redundant [5].
Methods were proposed specifically for calibrating fully-constrained CDPRs,
without the need for external sensors [1, 2, 6]. The use of force sensors, allowing
for simultaneous stiffness identification, is also addressed in [10]. A few different
calibration algorithms are also tested experimentally in [1]. However, because
they rely on measurement redundancy, these methods algorithms cannot be ap-
plied to cable-suspended CDPRs. Without proper initial estimation of the so-
lution to the geometric identification problem, the calculation time for a large
number of calibration poses can also be considerable and the accuracy of the
solution cannot be guaranteed.
An alternative to these self-calibration methods consists in resorting to exter-
nal sensors to measure directly the end effector pose. Over the large workspace
of CDPRs, one could use traditional surveying sensors such as theodolites or
total stations but their use is time-consuming. This problem can be resolved by
resorting to automated measurement devices such as laser trackers [4] or motion
capture systems, which can be used to perform different measurements either to
directly measure the geometry or to establish the robot posture. However, while
these solutions currently seem to prevail within academia, the cost of such sys-
tems may be of the same order of magnitude as that of the CDPR itself. Hence,
there is a true incentive to find an inexpensive methodology for the calibration
of CDPRs.
It is well known that a calibrated camera can be used to estimate the position
and orientation of a size-known object expressed in its own reference frame
and vice versa. Eye-on-hand, or eye-in-hand calibration is performed using a
moving camera, often attached to the robot end effector and a fixed target. This
method is used in particular to calibrate serial robots [7]. In the case of a CDPR,
one advantage of this method is that only one camera is needed to perform a
calibration over a large workspace. Moreover, the placement of the camera is
simple and space efficient, considering that the camera on the end effector can
move across the whole workspace. This avoids having to position one or more
fixed cameras in such a way that it is possible to accurately track targets fixed on
the end effector. Considering those potential benefits, the present paper outlines
a new approach to CDPR calibration using this technique.
In section 2, the calibration process and the generation of a list of poses used
to gather data is addressed. Then, in section 3, a control scheme for reaching the
calibration poses is described. Finally, experiments are presented and the results
are discussed in section 4.
Eye-on-Hand Calibration Method for Cable-Driven Parallel Robots 347
2 Calibration
The goal of the calibration process is to identify the geometry of a given CDPR.
In the current case, we work under the hypothesis of massless and inextensible
cables. Fig. 1 shows the general geometric model of a CDPR driven by m cables.
We assume that the cable lengths are measured using relative encoders, so that
the total length of the ith cable is expressed as ρi = Δρi + ρi,0 , where ρi,0 is the
cable length when the robot controller is first powered on after a reconfiguration.
The geometric loop-closure equation, for a given pose X = {p, Q} takes the form:
(Δρi + ρi,0 )2 = (p + Qbi − ai )T (p + Qbi − ai ), (1)
where p and Q are respectively the position vector and orientation matrix of
the end effector. In the case of a CDPR, the unknown parameters are usually
the position of the cable output points and the initial cable lengths, namely ai
and ρi0 . A full calibration would include an identification of the end effector
attachment points bi but since the latter is generally small, we consider that
these parameters can be established beforehand using conventional measuring
techniques. Regardless of the value of m, if a spatial mechanism is considered,
the constraint n ≥ 4 must be satisfied in order to avoid underdetermination of
the system, where n is the number of calibration poses.
pce , Qce
Oe
Oc
Q bi
p Bi pcm , Qcm
ρi
ai Ai
Ow Om
pmw , Qmw
Fig. 1. Cable robot geometric model for eye-on-hand calibration
The residual of the loop-closure equation, for each measurement j, is written

ui,j = ||pj + Qj bi − ai ||22 − (Δρi,j + ρi,0 )2 . (2)
where Δρi,j is directly obtained from the relative encoders and pj and Qj are
provided by an external sensor. The calibration problem is then written as a
nonlinear least squares optimisation, namely
m
n
minimise u2i,j (3)
ai ,ρi,0
i=1 j=1
when considering all cables i = 1, ..., m. The next section provides details con-
cerning the external measurements for every calibration pose.
2.1 Eye-on-Hand Calibration

In the work reported in this paper, this external sensor is an embedded camera,
which must provide enough data to estimate the end effector pose (pj , Qj ). The
detection of fiduciary markers is the chosen approach because it is simple and it
allows for fast processing and reliable positioning.
As the end effector is intended to cover a large workspace, the use of a single
marker would result in poor positioning accuracy and would be limited to only a
small part of the workspace. This problem could be solved by using an array of
different markers of known relative poses but the manufacturing of such objects
would be impractical or prone to errors. Instead, we choose to spread uniquely
identified markers throughout the robot workspace, without prior knowledge
of their relative pose. The existing ArUco C++ library [3, 14] can be used for
marker generation and detection and the corresponding marker mapper [11] tool
can easily be adapted, which makes it well suited for the current application.
This library allows for localisation and mapping of the markers. The resulting
map can then be used to estimate the camera pose. In such a case, the markers
can be removed after the calibration is completed.
The end effector pose can be obtained from the measured camera pose
through the rigid transformations pce and Qce , namely,
p = pmw + Qmw pcm + Qmw Qcm pce , (4a)

Q= Qmw Qcm QTce , (4b)
where pcm and Qcm are obtained using the camera pose reconstructed by the
marker mapper. If pmw and Qmw are not known beforehand, it is possible to
choose Ow such as it is coincident with Om , which leads to pmw = 0 and
Qmw = 13×3 , thus eliminating any possibility of measurement errors between
the two reference frames.
The constant values pce and Qce must be determined prior to calibration.
These can be obtained by observing a marker with the end effector camera and
comparing the resulting pcm and Qcm measurements to the measured position
and orientation of the end effector with respect to the marker reference frame. For
fully-constrained CDPRs, this transformation can also be estimated alongside
the calibration by simply adding the parameters pce and Qce to the decision
variables of the optimisation problem of Eq. 3.
2.2 Generation of a List of Reachable Calibration Poses

In order to perform an autonomous calibration, it is essential to generate a pool of
poses Xj = {pj , Qj }, which can be reached from an initial pose X0 = {p0 , Q0 },
located inside the workspace. To be feasible, the trajectory must be included
within the wrench-feasible workspace and free of collision between cables. The
proposed method works by projecting Xr , a pose randomly generated inside a

6-dimensional box, towards X0 , on the workspace of the robot.
However, in order to find reachable poses, an initial estimation of the param-
eters to calibrate is needed. In the context of this paper, the latter is obtained
using a simple measuring tape, but other viable and affordable solutions are
available, such as ultra-wideband positioning systems and inclinometers.
Let us define the trajectory between X0 and Xr as a straight-line motion of
point p0 combined with a constant angular velocity of the end effector. Such a
motion is parameterised as
p(λ) = p0 + λ(pr − p0 ), (5a)

Q(λ) = Qt (λ)Q0 , (5b)
where 0 ≤ λ ≤ 1 and
Qt (λ) = eeT + cos(λφ)(13×3 − eeT ) + sin λφE (6)
is the rotation undergone by the end effector along its trajectory from its attitude
at λ = 0 to that at λ. We use
∂(x × y)
E = cpm(e), where cpm(x) ≡ (7)
∂y
so that e and φ respectively represent the rotation axis and the rotation angle.
Upon solving Eq. (5b) for Qt (λ) and setting λ = 1, we obtain
Qt (1) = Qr QT0 . (8)
Those axis and angle values can be then found first by defining
q − qt23
⎡ ⎤
1 ⎣ t32
vect(Qt (1)) ≡ qt13 − qt31 ⎦ and tr(Qt (1)) ≡ qt11 + qt22 + qt33 (9)
2
qt21 − qt12
and then using the properties
tr(Qt (1)) = 1 + 2 cos φ, (10a)

vect(Qt (1)) = sin φe. (10b)
For the purpose of satisfying the wrench-feasible criterion, we aim to find the
maximum λ = λs value that can be reached without any constraint violation.
This can be ensured first by defining the cable direction for each cable
ai − p(λs ) − Q(λs )bi
ci (λs ) = (11)
||ai − p(λs ) − Q(λs )bi ||
and constructing the end effector pose-dependant wrench matrix
···

c1 (λs ) cm (λs )
W(λs ) = . (12)
Q(λs )b1 × c1 (λs ) · · · Q(λs )bm × cm (λs )
By imposing bounds over the cable tensions in the cables tmin ≤ ti ≤ tmax and
the static equilibrium constraints, we then formulate the problem as follows
maximise λs
subject to tmin ≤ t ≤ tmax ,
(13)
W(λs )t + we = 0,
0 ≤ λs ≤ 1
where t is the array of cable tensions and we is the wrench applied on the end
effector by its environment.
A similar approach is used for considering the cable-cable interferences. Let
ζ(pk1 , pk2 , pl1 , pl2 ) be a function computing the distance between two line seg-
ments defined by their four end-points pk1 , pk2 , pl1 and pl2 , as presented in [9].
The cables k and l are therefore represented by the line segments pk2 − pk1 and
pl2 − pl1 . These points can then be rewritten as functions of the geometry and
of the trajectories described in Eqs. (5a) and (5b):
pk1 = ak , (14a)
pk2 = p(λk,l ) + Q(λk,l )bk , (14b)
pl1 = al , (14c)
pl2 = p(λk,l ) + Q(λk,l )bl . (14d)
A collision between two real cables occurs when ζ(pk1 , pk2 , pl1 , pl2 ) < , where
is the diameter of the cables. For the sake of simplicity, we consider cables of
a negligible diameter, i.e., = 0. The roots of the function are then found using
minimise ζ(pk1 , pk2 , pl1 , pl2 )
λk,l
(15)
subject to 0 ≤ λk,l ≤ 1
2 (m − 1) possible pairs of cables, i.e., for k, l ∈ {1, ..., m}, k < l. If

for the m
no root is found, then λk,l = 1. Ultimately, when considering simultaneously
wrench feasibility and the possibility of interference between any cable pair, the
maximum point that can be reached on the trajectory is obtained as:
λ = μ min{λs , λk,l } (16)
where 0 ≤ μ ≤ 1 is a factor introduced to account for discrepancies between the

initial estimate of the robot geometry and its true geometry. It can be adjusted
according to the level of trust in the initial approximation of the parameters,
a high value representing confidence, and a low one, uncertainty. The resulting
pose X can finally be obtained using Eqs. (5a) and (5b). Since the optimisations
(13) and (15) both starts from X0 , known to be located within the workspace of
the robot, a feasible solution is always found, the worst case scenario being that
Xj = X0 , i.e., λ = 0.
Once the list of poses and their associated trajectories is generated, the focus
can shift to the problem of tracking them with the robot.
3 Control
In order to move the end effector to the computed calibration poses before the
robot is fully calibrated, a control scheme capable of dealing with errors in the ini-
tial estimate of the geometric parameters is necessary. Thus, control is achieved
by a simple low-level velocity controller, inspired by [2], driven by a Cartesian
position controller. The block diagram of the control algorithm is presented in
Fig. 2. To ensure valid cable measurements, this controller purpose is to keep
the cables thaught, while trying to reach the desired pose.
p̃, Q̃ Forward ρ̃
Kinematics
Cogging θ̃
we Compensation
tc
ρ̃, ρ̃˙
Tension
Distribution
te
Kτ tτ
+
+
+
t
Robot
ṗ, ω
p, Q
Cp J
ρ̇ +
− Kv tv
ρ̃˙
Fig. 2. Control scheme used to perform the calibration
The controller Cp generates trajectories with trapezoidal speed profiles ṗ

and ω in order to reach the desired Cartesian position p and orientation Q. The
matrix J = −WT is then used to transform the desired Cartesian and angular
speeds into articular speeds. A set of feasible cable tensions τe is generated
from we using quadratic programming. The algorithm is concretely implemented
using a C program generated with CVXGEN [8]. The gain Kτ is derived from
the geometry of the winches and used to link the desired cable tension to the
motor torque, while Kv is the gain of the low-level proportional speed controller.
It should be noted than Kτ could be replaced by a closed loop tension controller.
To address undesirable speed variation issues caused by the cogging ripple of the
permanent magnets motors used in section 4, a cogging compensation block was
introduced in the model. Even though this addition produced satisfying results,
it will not be discussed in detail in this paper. The controlled motor torque is
then given by the sum of the cogging compensation torque tc , the torque tτ
needed to achieve the desired set of tensions and the torque tv generated by the
low-level proportional speed controller. A saturation is placed on tv to prevent
any excessive unwinding of the cables that could be caused by modeling errors,
since the mechanism that will next be presented for validation is backdrivable.
4 Experiment
To test the performance of the proposed method, a validation is carried out
on an existing CDPR. The setup consists of eight winches, each mounted on a
motorised linear axis, allowing it to be easily reconfigured. The winches were

designed to operate in industrial-like conditions and to withstand a large pay-
load. They are driven by Parker MPP0923D motors coupled to 4:1 gearboxes.
The size of the gantry is approximately 5 m × 4 m × 3 m and the winches are
brought as close as physically possible to the corners of the structure.
The end effector is a 0.2 m, × 0.2 m × 0.6 m prism. The cables are made
of 5 mm Dyneema SK75 synthetic rope. As can be seen in Fig. 3, they are
attached to the end effector in a way that allows the control of its orientation. A
self-contained vision module consisting of a 1.3 megapixel Basler ace acA1300-
30gm camera, a battery and a Raspberry Pi computer is located inside the latter.
The camera is oriented towards the markers, placed on the ground for practical
reasons. For generality’s sake, the markers are not considered to be located on
a common plane. One marker is chosen as the reference for the whole map.
For the measurements, the ground truth is provided by a Faro X130 laser
scanner with a ranging error of ± 2 mm (at 1σ).
4.1 Trials and results

At first, the camera is calibrated using the Aruco calibration board, laminated on
a flat plastic surface, resulting in a mean reprojection error is 0.51 pixels. Then,
three different lists of random calibration poses (A, B and C) are generated using
μ = 0.8 and a length of n = 100 samples. The practical choice of Ow coincident
with Om is made. Data is then gathered in the form of relative cable length and
pictures of the marker map are taken using the camera. Once the robot covered
all calibration poses, the marker map geometry is established. Fig. 4 shows an
example of the reconstructed camera poses and markers locations. Using the
reconstructed camera poses, the calibration is performed by solving Eq.(3) using
the trust-region-reflective algorithm provided by the lsqnonlin MATLAB func-
tion. The resulting geometric parameters are then updated, and the accuracy is
tested for ten witness poses, shown in Fig. 5.
A second calibration, for comparison purposes, is performed again but exclud-
ing the camera data, i.e., using only cable measurements. The second method,
described in [2, 6], relies exclusively on the motor encoders of fully constrained
Fig. 3. Experimental setup: on the left, an overall view of the gantry and markers
located on the ground. On the right, the end effector and the embedded vision system
2000
1500
1000
500
1000
1000
0
0
-1000
-1000
Fig. 4. Reconstructed marker map and camera poses
2500
2000
1500
1000 1000
-500 0
0
500
1000 -1000
Fig. 5. Prescribed poses for calibration validation
CDPR architecture and as stated above, it can only be applied to robots with
more cables than degrees of freedom. Without an external reference, the mech-
anism geometry is unconstrained. To resolve this indeterminacy and to ensure
consistancy, six components of the ai vectors are prescribed from the values ob-
tained in the previous step. The end effector pose previously computed using the
camera is also provided to the optimisation algorithm as an initial estimation of
the solution to avoid convergence to a different local minimum. Upon completion
of the calibration, the accuracy is tested for the same ten witness poses.
The error is determined by the difference between the forward kinematics
problem (FK) result and an external measurement between the end effector and
the reference marker, provided by the laser scanner. The optimisation conver-
gence is expressed by r, which is defined as the fourth root of the residual of
Eq. (3), divided by the number of calibration poses n.
Table 1 presents the compiled performance results. The pose is measured
relative to the marker map (R.t. marker) used for calibration. To ensure proper
comparison between the two calibration methods, and to prevent any bias caused
by the camera from degrading the accuracy over the entire workspace, the pose
is also measured relative to the first pose (R.t. 1st pose). It should be noted
that according to the ground truth measurements, errors in both position and
orientation do not seem to be predominant in one particular direction.
Table 1. Calibration results
Data set r (mm) Position error (mm) Orientation error (deg)

R.t. marker R.t. 1st pose R.t. marker R.t. 1st pose
Mean Std Mean Std Mean Std Mean Std
Proposed A 12.9 60.9 8.1 25.8 7.8 2.1 0.8 1.3 0.9
method B 14.1 64.5 14.0 27.4 8.2 2.3 0.5 1.3 0.7
(with camera) C 15.6 59.5 17.2 30.5 11.1 2.9 0.6 1.6 1.2
A 3.8 278.3 22.0 28.6 11.2 4.6 1.8 2.9 2.0
Cable lengths
B 4.2 268.2 24.0 28.7 16.0 3.2 1.2 2.7 1.0
only [2, 6]
C 4.0 984.5 25.6 110.1 32.1 7.6 4.4 5.7 1.1
Additionally, for each of the 10 validation poses of the eye-on-hand calibra-

tions, a snapshot of the marker map is taken using the camera. The resulting
images can be compared to the previously established marker map geometry for
the purpose of validating the camera pose reconstruction. Because the pose of
the camera cannot be directly picked up, we measured instead the end effector
pose. Table 2 summarises the errors obtained using the laser scanner data as a
reference while Table 3 presents the discrepancy between the poses obtained with
the camera measurements alone, and those obtained through the cable lengths,
after calibration. In the following section, relationships between these data and
the calibration results will be discussed.
Table 2. Camera pose errors, relative to ground-truth
Data set Position error (mm) Orientation error (deg)

Mean Std Mean Std
A 59.2 9.1 1.8 0.4
B 61.5 12.4 1.9 0.4
C 30.8 9.6 1.5 0.4
4.2 Discussion
From the results of Table 1, it appears that the proposed eye-on-hand calibration
method did improve the accuracy of the robot relative to the first pose, but this
improvement is only statistically significant for some data sets. It can also be
Table 3. Camera pose inconsistencies, relative to FK solution
Data set Position error (mm) Orientation error (deg)

Mean Std Mean Std
A 8.4 4.2 1.3 0.4
B 8.2 3.5 1.2 0.5
C 32.2 10.3 2.3 0.5
seen by looking at the data set C that, when compared to the calibration method
using only measurement redundancy, the use of the camera can prevent excessive
divergence of the geometric parameters. Upon one-way analysis of variance using
a significance level of 0.05, the positioning accuracy was improved only for the
dataset C, while the orientation accuracy was improved for all datasets.
It can be observed from Table 3 that the FK solution is generally consistent
with the reconstructed camera pose, especially in the case of the mean position
error of data sets A and B. However, Table 2 shows that the same camera pose
estimation is subject to errors in an order of magnitude similar to those mea-
sured after calibration, both in orientation and in translation. This suggests that
the precision of the camera pose reconstruction is limiting the accuracy of the
calibration. Improving the resolution and calibration of the vision system could
therefore lead to improvements in calibration and thus in positioning accuracy.
5 Conclusion
This paper presents an investigation into the use of an eye-on-hand calibration
method, for determining the geometric parameters of CDPRs. To this end, a list
of reachable poses was generated based on an estimate of the CDPR geometry.
A control scheme was adapted to allow the end effector to reach each of these
poses. Finally, the method was experimentally validated on a medium sized fully
constrained CDPR. While the worst relative mean position and orientation errors
measured were respectively of 30.5 mm and 1.6◦ , the results suggest that the
positioning accuracy of the calibrated CDPR could be increased by improving
the accuracy of the reconstruction of the camera pose.
Acknowledgments
This work is financially supported by the Fonds de recherche du Québec - Nature
et technologies (FRQNT), under the grant number 2016-PR-188869.
References
1. J. A. Dit Sandretto, D. Daney, and M. Gouttefarde. Calibration of a fully-
constrained parallel cable-driven robot. In 19th CISM-IFToMM Symposium on
Robot Design, Dynamics, and Control (ROMANSY 2012), Paris, France, 2012.
2. A. Fortin-Côté, P. Cardou, and C. Gosselin. An admittance control scheme for

haptic interfaces based on cable-driven parallel mechanisms. In Proceedings - IEEE
International Conference on Robotics and Automation, pages 819–825, 2014.
3. S. Garrido-Jurado, R. Muñoz-Salinas, F. J. Madrid-Cuevas, and R. Medina-
Carnicer. Generation of fiducial marker dictionaries using Mixed Integer Linear
Programming. Pattern Recognition, 51:481–491, 2016.
4. J.-B. Izard, M. Gouttefarde, M. Michelin, O. Tempier, and C. Baradat. A Reconfig-
urable Robot for Cable-Driven Parallel Robotic Research and Industrial Scenario
Proofing. In T. Bruckmann and A. Pott, editors, Cable-Driven Parallel Robots,
pages 135–148. Springer Berlin Heidelberg, Berlin, Heidelberg, 2013.
5. W. Khalil and S. Besnard. Self calibration of Stewart-Gough parallel robots with-
out extra sensors. IEEE Transactions on Robotics and Automation, 15(6):1116–
1121, 1999.
6. D. Lau. Initial Length and Pose Calibration for Cable-Driven Parallel Robots with
Relative Length Feedback. In C. Gosselin, P. Cardou, T. Bruckmann, and A. Pott,
editors, Cable-Driven Parallel Robots, pages 140–151, Cham, 2018. Springer Inter-
national Publishing.
7. R. K. Lenz and R. Y. Tsai. Calibrating a Cartesian robot with eye-on-hand con-
figuration independent of eye-to-hand relationship. IEEE Transactions on Pattern
Analysis and Machine Intelligence, 11(9):916–928, sep 1989.
8. J. Mattingley and S. Boyd. CVXGEN: A code generator for embedded convex
optimization. Optimization and Engineering, 13(1):1–27, 2012.
9. R. Meziane, P. Cardou, and M. J. Otis. Cable interference control in physical
interaction for cable-driven parallel mechanisms. Mechanism and Machine Theory,
132:30–47, 2019.
10. P. Miermeister and A. Pott. Auto Calibration Method for Cable-Driven Parallel
Robots Using Force Sensors. In J. Lenarcic and M. Husty, editors, Latest Advances
in Robot Kinematics, pages 269–276. Springer Netherlands, Dordrecht, 2012.
11. R. Muñoz-Salinas, M. J. Marı́n-Jimenez, E. Yeguas-Bolivar, and R. Medina-
Carnicer. Mapping and localization from planar markers. Pattern Recognition,
73:158–171, 2018.
12. A. J. Patel and K. F. Ehmann. Calibration of a hexapod machine tool using a re-
dundant leg. International Journal of Machine Tools and Manufacture, 40(4):489–
512, mar 2000.
13. P. Renaud, N. Andreff, F. Marquet, and P. Martinet. Vision-based kinematic
calibration of a H4 parallel mechanism. 2003 IEEE International Conference on
Robotics and Automation (Cat. No.03CH37422), 1:1191–1196, 2003.
14. F. J. Romero-Ramirez, R. Muñoz-Salinas, and R. Medina-Carnicer. Speeded up
detection of squared fiducial markers. Image and Vision Computing, 76:38–47,
2018.
15. H. Zhuang, O. Masory, and J. Yan. Kinematic calibration of a Stewart platform
using pose measurements obtained by a single theodolite. Intelligent Robots and
Systems 95. ’Human Robot Interaction and Cooperative Robots’, Proceedings. 1995
IEEE/RSJ International Conference on, 2:329–334 vol.2, 1995.
16. H. Zhuang, J. Yan, and O. Masory. Calibration of Stewart platforms and other
parallel manipulators by minimizing inverse kinematic residuals. Journal of Robotic
Systems, 15:395–405, 1998.
On the automatic calibration of redundantly
actuated cable-driven parallel robots
Han YUAN1 , Yongqing ZHANG1 , and Wenfu XU1,2

1
School of Mechanical Engineering and Automation, Harbin Institute of Technology
(Shenzhen), 518055 Shenzhen, P.R. China, [email protected]
2
The State Key Laboratory of Robotics and System, Harbin Institute of Technology,
150001 Harbin, P.R. China, [email protected]
Abstract. Calibration is vital to improve robot accuracy. Automatic

calibrations that require no extra devices have lots of conveniences, which
is especially significant for cable-driven parallel robots that usually have
the reconfigurable ability. This paper proposes a new automatic calibra-
tion method that is applicable for a general kind of redundantly actuated
cable-driven parallel robots. The key point of this method is to estab-
lish the mapping between the unknown parameters to be calibrated and
the parameters that could be measured by the inner sensors, and then
use least squares algorithm to find the solutions. Specifically, the un-
known parameters herein are the coordinates of the attachment points
and the measured parameters are the lengths of the redundant cables.
Simulations are performed on a 3-DOF parallel robot driven by 4 cables
for validation. Results show the proposed calibration method could pre-
cisely find the real coordinate values of the attachment points, errors less
than 10−12 mm. Trajectory simulations also indicate that the positioning
accuracy of the CDPR could be greatly improved after calibration using
the proposed method.
Keywords: self-calibration; redundantly actuation; parallel robots; ca-

ble robots
1 Introduction
Cable-driven parallel robot (CDPR) is a new kind of parallel kinematic machine

system who is driven by flexible cables [1]. The main structure of CDPRs con-
sists of a base frame, a moving platform, a certain number of driving cables and
winch modules. Such simple structure makes CDPRs convenient for quick assem-
blage and reconfiguration. Therefore, CDPRs become more flexible than other
kinds of robots on changing themselves to adapt new tasks and environments.
Moreover, CDPRs also have some common characteristics of traditional parallel
robots made by rigid links, such as fast moving speed, large payload capacity,
high precision, and good structural stiffness[2, 3]. In addition, CDPRs also have
their own advantages such as low moving mass, large workspace and low manu-
facturing cost. Because of these superior characteristics, CDPRs are widely used

358 H. Yuan et al.
in disaster rescues, large space motion, microgravity simulation, flight simulator,

virtual reality, rehabilitation, etc[4–12].
Geometric calibration is obligated for any robot to improve the kinematic
accuracy. Generally, geometric calibration methods can be classified into two
types, i.e. with and without external measurements. The previous ones should
use extra sensors or measuring systems, such as the inclinometer and the laser
tracker to obtain the required measuring data[13–15]. The later ones do not
need any external sensor or measuring equipment, while only use the internal
information to complete the calibration, thus these methods are also called auto-
calibration or self-calibration methods[16–18, 21].
For traditional rigid-link robots including both serial and parallel ones, cali-
brations are usually done by manufacturers before the robots leave their facto-
ries. Once calibrated, these robots do not need calibration again for normal use.
However, it is totally different for CDPRs. Since most CDPRs are assembled at
the users places and need reconfiguration for different tasks, calibration of CD-
PRs are usually needed in daily use. Therefore, auto-calibration/self-calibration
methods are especially significant for CDPRs.
A variety of methods have been proposed on the calibration of CDPRs. For
example, Joshi et al. proposed a kinematic calibration methodology by using two
inclinometers for 6-DOF CDPRs[13]. Xuechao Duan et al. using a triple-level
spatial positioner to obtain measuring data when calibration[14]. An interval
based approach has been reported in [19], and measurements achieved accord-
ing to external equipment similarly. There are also some auto/self-calibration
method has been proposed. Philipp Miermeister and Andreas Pott presented
an auto calibration method for overconstrained CDPRs using internal position
and force sensors to achieved measurements[16]. A differential kinematic and
force based forward kinematics for CDPRs is introduced in [20]. Per Henrik
Borgstrom et al. presented 2 novel self-calibration method, based on incremental
displacement and cable tension respectively[17]. Some researchers have proposed
a number of calibration methods for CDPRs. However, most of them need to be
done with external measuring equipment, like laser tracker, inclinometers, spa-
tial positioner, and etc [13, 14, 19]. For some self-calibration methods[16, 18], they
are only proposed for some specific mechanism CDPRs, as for example 6-DOF
CDPRs driven by 8 or 7 cables. There is no general automatic self-calibration
method has been proposed for all kinds of redundantly CDPRs so far.
This paper proposes a general self-calibration method which is applicable
for redundantly actuated CDPRs. The encoders of the servo motors are used
to obtain the driving cable lengths that are required by the calibration model.
Compared to traditional calibration methods with external measurements, the
proposed method uses the internal sensors of CDPRs to generate measuring
data. In addition, a hybrid force-position control algorithm is integrated in the
robot motion control scheme, in order to automatically complete the calibration
process. Therefore, the proposed auto-calibration method is convenient for the
field calibration of CDPRs, and also quite useful for reconfigurable CDPRs.
On the automatic calibration of redundantly actuated cable-driven… 359
This paper is organized as follows. The kinematic derivation of CDPRs is

detailed in section 2. The automatic calibration method for general redundantly
actuated CDPRs is proposed in section 3. The workflows of the proposed self-
calibration method including the hybrid force-position control are described in
section 4. In section 5, a numerical example is presented through a 3-DOF CDPR
driven by 4 cables to validate the proposed method. Finally, conclusions are made
in section 6.
2 Kinematic Modeling
2.1 General structure of CDPRs
Most CDPRs have a similar composition, mainly consisting of a base body, a

moving platform, and a number of driving cables. The schematics of general
CDPRs is shown in Figure 1. There are m attachment points Bi (i = 1, ...m) on
the fixed base, and m end points Ai (i = 1, ...m)located at the moving platform,
and m driving cables connecting the moving platform and the fixed base. The
moving platform could change its positions and orientations by changing the
cable lengths According to the relationship between the driving cable number m
and the controllable DOF of the moving platform n, CDPRs could be classified
into two groups, i.e. redundantly (m > n) and non-redundantly (m <= n)
actuated CDPRs. In this paper, the proposed automatic calibration method
aims redundantly actuated CDPRs, i.e. m > n.
%M
% %L
%
%
O
O OL
O
$M
]
$P
$ $ 2H [ $N
M
$M \
$L
OP $
$
OM OM ON
]*
%P
%M %N 2*
[*
%M \*
Fig. 1. Structural schematics of general CDPRs.

360 H. Yuan et al.
2.2 Inverse geometrics
In this section, the inverse geometric model of CDPRs is presented. Firstly, two
coordinate systems are defined, i.e. the local frame Re {Oe xe ye ze } fixed on the
moving platform and the global frame RG {OG xG yG zG }. Then, the homogeneous
transformation matrix T from Re to RG could be defined as a function of the
pose of the moving platform pj (j = 1, ...n). For example, when n = 6, P =
[x, y, z, α, β, γ]T ,
cαsβcγ − sαsγ −cαcβsγ − sαcγ cαsβ x

⎡ ⎤
⎢ sαcβcγ + cαsγ −sαcβsγ + cαcγ sαsβ y ⎥
T=⎢ (1)
−sβcγ
⎥
⎣ sβsγ cβ z ⎦
0 0 0 1
Then, any attachment point on the moving platform Ai could be expressed
in the global frame:
Ai = T · Ai
G e
(2)
Finally, the inverse geometrics of CDPRs could be formed as:
Ai − Bi , (i = 1, ...m)
G G
li = (3)
It should be noted that the driving cable lengths are usually controlled by
winches and motors. So it is necessary to introduce a simple equation to substi-
tute the cable length li with the winch rotatory angle θi .
li = θi · di /2 (4)
Where di is the winch diameter. Then, the inverse geometric model can be
reformed as:
fi (p1 , p2 , · · · pn , e Ai , G Bi , θi , di ) = 0 (i = 1, · · · m) (5)
3 Calibration Method
In this section, the automatic calibration method is detailed for general redun-
dantly actuated CDPRs with n DOFs and driven by m cables.
3.1 Differential kinematics
Firstly, the differential kinematics of CDPRs is presented. Through making total

differentiation on the inverse kinematic equation by Eq.5, we can obtain:
∂fi ∂fi ∂fi ∂fi ∂fi ∂fi ∂fi

dp1 + dp2 +· · ·+ dpn + e de Ai + G dG Bi + dθi + ddi = 0
∂p1 ∂p2 ∂pn ∂ Ai ∂ Bi ∂θi ∂di
(6)
Noting dX = [dp1 , dp2 , · · · , dpn ]T , the above differential equation can be

reorganized as:
J · dX=B (7)
Where J is the classical Jacobian matrix and B is a column vector, which
can be expressed as:
∂fi
Jij = , (i = 1, ...m; j = 1, ...n) (8)
∂pj
∂fi e ∂fi ∂fi ∂fi

Bi = − e
d Ai + G d G B i + dθi + ddi (9)
∂ Ai ∂ Bi ∂θi ∂di
3.2 The calibration model
Since the proposed calibration method aims at redundantly actuated CDPRs,

i.e. the driving cable number m is larger than the robot DOF n, the differential
equations Eq.7 can be separated into two parts:
JC dX = BC , (10)
JM dX = BM , (11)
where:
∂fi
JCij = , (i = 1, ...n; j = 1, ...n), (12)
∂pj
∂fi
JM ij = , (i = n + 1, ...m; j = 1, ...n), (13)
∂pj
BCi = − d Ai + G dG Bi + dθi + ddi , (i = 1, ...n), (14)
∂ e Ai ∂ Bi ∂θi ∂di
BM i = − d Ai + G dG Bi + dθi + ddi , (i = n + 1, ...m).
∂ e Ai ∂ Bi ∂θi ∂di
(15)
It is worth noting that the winch diameters di (i = 1, ...m) are constant. Thus,
the ddi in Eq.11 is zero. However, two cases should be considered for the motor
rotary angles θi . For angles θ1 ∼ θn , they are controlled by the robot controller.
Assuming the control error is negligible, dθ1 ∼ dθn in Eq.11 should be zeros.
dθn+1 ∼ dθm are approximate instead of difference between the theory length
of cables and real length of measuring cables which measured by internal motor
encoders. But for angles θn+1 ∼ θm , they are the measuring data obtained by
the encoders of these corresponding motors. Thus the θn+1 ∼ θm in Eq.11 must
remain. Therefore BC and BM can be simplified to:
∂fi e ∂fi
BCi = − e
d Ai + G dG Bi , (i = 1, ...n) (16)
∂ Ai ∂ Bi
362 H. Yuan et al.
∂fi e ∂fi ∂fi

BM i = − d Ai + G dG Bi + dθi , (i = n + 1, ...m) (17)
∂ e Ai ∂ Bi ∂θi
Then, substitute Eq.(10) to (11):
JM · JC −1 · BC = BM , (18)
where JM , JC −1 , BC and BM are detailed in Eqs. (12), (13), (15) and (16). By
noting dq as a vector containing all the unknown parameters to be calibrated, i.e.
the coordinates of the attachment points on the fixed base and the moving plat-
T
form dq = de A1 , de A2 , de A3 , · · · , de Am , dG B1 , dG B2 , dG B3 , · · · , dG Bm , Eq.18

can be rearranged to the following form:
Jerr · dq = dθ (19)
Where Jerr is called the error Jacobian matrix which has (m − n) lines and
r columns, and r represents the number of the unknown parameters to be cali-
brated. The θ represents the vector of motor angles that used for measurement,
which consists of m − n components.
Finally, the calibration model can be obtained from Eq.19:
Jerr · Δq ≈ Δθ (20)
Where, Δq = [Δe A1 , Δe A2 , Δe A3 , · · · , Δe Am , ΔG B1 , ΔG B2 , ΔG B3 , · · · , ΔG Bm ]T ,
Δθ is the error vector consisting of m − n components which are the differences
of the motor rotary angles between the theoretical values and the measuring
counterparts.
In fact according to Eq.20, m−n equations can be obtained from each collec-
tion pose of the CDPR in its workspace. Since there are r unknown parameters
to be solved, at least r/(m − n) different poses should be collected. If k different
poses are used for calibration, there are totally k(m − n) equations. Which can
be written as:
⎡ ⎤ ⎡ ⎤
Jerr1 dθ1
⎢ Jerr2 ⎥ ⎢ dθ2 ⎥
⎢ .. ⎥ · Δq = ⎢ .. ⎥ (21)
⎢ ⎥ ⎢ ⎥
⎣ . ⎦ ⎣ . ⎦
Jerrk dθk
⎡ ⎤ ⎡ ⎤
Jerr1 dθ1
⎢ Jerr2 ⎥ ⎢ dθ2 ⎥
Noting Jclbr = ⎢ . ⎥, dθclbr = ⎢ . ⎥, the k(m − n) equations can be
⎢ ⎥ ⎢ ⎥
⎣ .. ⎦ ⎣ .. ⎦
Jerrk dθk
written as:
Jclbr · Δq = dθclbr (22)

Then, error of difference between real structure parameter and theory struc-
ture parameter can be obtained from calibration model.
Δq = Jclbr −1 · dθclbr (23)

−1
It is obviously that the Jclbr is the core of calibration model. And the Jclbr
maybe not the positive definite matrix. Therefore, solving this problem by using
least squares optimization method or pseudo inverse. Then, Δq can be solved.
4 Numerical Example
In order to verify the proposed automatic calibration method, a numerical ex-

ample is detailed on a 3-DOF CDPR driven by 4 cables. The structural diagram
of the CDPR is shown in Fig. 2.
\*
2*
[*
% %
O O
\
$ $
[ \ D
2H [
$ $
O O
% %
Fig. 2. Diagram of the 3-DOF planar CDPR.
As shown in Fig. 2, the moving platform is a rectangle whose length and width
are respectively 120 and 80 mm. The local coordinate frame Re {Oe xe ye } fixed
on the center of the rectangular platform. The origin of the global coordinate
frame RG {OG xG yG } coincides with the attachment point B4 , which means that
the global coordinates of B4 are 0. The nominal values of the other 3 attachment
points B1 ∼ B3 are listed in Table 1.
Generally, the attachment points on the moving platform are usually con-
stant for different CDPR configurations. Thus the coordinates of the attachment
points A1 ∼ A4 are assumed to be accurate, which do not need calibration. More-
over, the global frame is set on point B4 . Therefore, the unknown parameters
remained to be calibrated are the coordinates of B1 ∼ B3 . According to the pro-
posed method, in the calibration procedures, the servo motors for the driving
cables l1 ∼ l3 work at position mode, while the motor of cable l4 works at torque
364 H. Yuan et al.
mode. Thus the cable lengths l1 ∼ l3 are controlled by the servo motors and the
length l4 could be measured by the encoder of the corresponding motor.
At the beginning of the simulation, a group of random deviations (range:
±20 mm) are added to the nominal values of the coordinates of the attachment
points B1 ∼ B3 , and the results are took as the real positions of these attach-
ment points, which are listed in Table 1. Then, 200 poses of the moving platform
are randomly generated in the workspace of the CDPR. For each pose, a inde-
pendent equation can be calculated according to Eqn. (20). Thus Δq could be
obtained through the 200 linear equations according to Eqn. (23). After that, the
coordinates of the attachment points are revised according to Δq obtained in the
previous step. Then, iterations are made to recalculate Δq based on the revised
coordinate values, until Δq is small enough which is 10−6 mm in this simulation.
Finally, the calibrated coordinates of the attachment points are listed in Table.
1, where it is shown that the difference between the real coordinate values and
the calibrated ones are quite small, i.e. less than 10−12 mm.
Table 1. Nominal, real, and calibrated coordinate values of the attachment points
Attachment point Nominal value (mm) Real value (mm) Calibrated value (mm)
x -850 -861.69 −861.69 − 0.341 × 10−12
B1
y 0 -18.38 −18.38 + 0.277 × 10−12
x -850 -856.23 −856.23 − 0.114 × 10−12
B2
y -770 -751.43 −751.43 − 0.227 × 10−12
x 0 -13.97 −13.97 − 0.011 × 10−12
B3
y -770 -787.36 -787.36
In order to further examine the proposed calibration method, two typical

trajectories of the moving platform are performed by simulation. The first tra-
jectory is a straight line and the second is a circle. Figures 3 and 5 show the
simulation results, where the red solid lines represent the ideal trajectories, and
the blue dash lines represent the trajectories obtained by the initial coordinate
values before calibration, and the green dot lines represent the trajectories ob-
tained by the corrected coordinate values after calibration. As is shown in Figs.
3 and 5, the trajectories obtained by the calibrated coordinates are much closer
to the ideal trajectories. These trajectory errors are further shown in Fig. 4 and
6. As we can see, the positioning errors of the moving platform have been greatly
reduced after calibration by the proposed calibration method.
5 Conclusion
This paper presents a new automatic calibration method for a general kind of re-
dundantly actuated CDPRs. This proposed method uses the length information
of the redundant driving cables which can be easily obtained by the encoder of
Fig. 3. Trajectories of the mov-Fig. 4. Trajectory errors relative to ideal situation:

ing platform: circular case. circular case.
Fig. 5. Trajectories of the mov-Fig. 6. Trajectory errors relative to ideal situation:

ing platform: linear case. linear case.
the servo motors, and does not need any extra measuring equipment. Therefore,
this method brings great convenience to the calibration of CDPRs, especially for
reconfigurable CDPRs. Simulation results show that this method could find the
real coordinate values of the attachment points with extremely small errors, i.e.
less than 10−12 mm. Results also prove that the positioning error of the CDPR
along given trajectories could be greatly reduced after calibration using the pro-
posed method. In future work, a prototype of the CDPR will be fabricated and
the proposed calibration method will be further verified by experiments.
6 Acknowledge
The authors would like to thank the National Natural Science Foundation of
China (Grant No. 61803125), the Natural Science Foundation of Guangdong
Province China (Grant No.2018A030313247), and the Science and Technology
Innovation Committee of Shenzhen (grant number JCYJ20170811155308246 and
JSGG20170413164102635) for their financial supports on this work.
References
1. Merlet JP. Parallel robots, vol 128. Springer Science & Business Media
2. Mario Luces, James K Mills, and Beno Benhabib. A review of redundant parallel
kinematic mechanisms. Journal of Intelligent & Robotic Systems, 86(2):175–198,
2017.
366 H. Yuan et al.
3. Han Yuan, Eric Courteille, and Dominique Deblaise. Static and dynamic stiffness
analyses of cable-driven parallel robots with non-negligible cable mass and elasticity.
Mechanism and Machine Theory, 85:64–81, 2015.
4. B. Zi, B. Y. Duan, J. L. Du, and H. Bao. Dynamic modeling and active control of
a cable-suspended parallel robot. Mechatronics, 18(1):1–12, 2008.
5. Paul Bosscher, Robert L Williams, and Melissa Tummino. A concept for rapidly-
deployable cable robot search and rescue systems. In ASME 2005 International
Design Engineering Technical Conferences and Computers and Information in En-
gineering Conference, pages 589–598. 2005.
6. Andreas Pott, Hendrick Mütherich, Werner Kraus, Valentine Schmidt, Philipp Mier-
meister, and Alexander Verl. Ipanema: a family of cable-driven parallel robots for
industrial applications. In Cable-Driven Parallel Robots, pages 119–134, 2013.
7. Sadao Kawamura, Hitoshi Kino, and Choe Won. High-speed manipulation by using
parallel wire-driven robots. Robotica, 18(1):13–21, 2000.
8. Vincent Nabat. Robots parallèles à nacelle articulée: Du concept à la solution in-
dustrielle pour le pick-and-place. PhD thesis, Montpellier 2, 2007.
9. NAN and Rendong. Five hundred meter aperture spherical radio telescope (fast).
Science in China, 49(02):3–22, 2006.
10. J. P. Merlet. Marionet, a family of modular wire-driven parallel robots. Advances
in Robot Kinematics Motion in Man & Machine, 169(2):53–61, 2010.
11. Ying Mao and Sunil Kumar Agrawal. Design of a cable-driven arm exoskeleton
(carex) for neural rehabilitation. IEEE Transactions on Robotics, 28(4):922–931,
2012.
12. D. Surdilovic, Jinyu Zhang, and R. Bernhardt. String-man: Wire-robot technology
for safe, flexible and human-friendly gait rehabilitation. In IEEE International
Conference on Rehabilitation Robotics, pages 446–453, 2007.
13. SA Joshi and A Surianarayan. Calibration of a 6-dof cable robot using two incli-
nometers. Performance metrics for intelligent systems, pages 3660–3665, 2003.
14. Xuechao Duan, Yuanying Qiu, Qingjuan Duan, and Jingli Du. Calibration and
motion control of a cable-driven parallel manipulator based triple-level spatial po-
sitioner. Advances in Mechanical Engineering, 6:368018, 2014.
15. M Saeed Varziri and Leila Notash. Kinematic calibration of a wire-actuated parallel
robot. Mechanism and Machine Theory, 42(8):960–976, 2007.
16. Philipp Miermeister and Andreas Pott. Auto calibration method for cable-driven
parallel robots using force sensors. In Latest Advances in Robot Kinematics, pages
269–276. Springer, 2012.
17. Per Henrik Borgstrom, Brett L Jordan, Bengt J Borgstrom, Michael J Stealey,
Gaurav S Sukhatme, Maxim A Batalin, and William J Kaiser. Nims-pl: A cable-
driven robot with self-calibration capabilities. IEEE Transactions on Robotics, 25
(5):1005–1015, 2009.
18. Shabbir Kurbanhusen Mustafa, Guilin Yang, Song Huat Yeo, Wei Lin, and I-Ming
Chen. Self-calibration of a biologically inspired 7 dof cable-driven robotic arm.
IEEE/ASME transactions on mechatronics, 13(1):66–75, 2008.
19. Julien Alexandre dit Sandretto, Gilles Trombettoni, David Daney, and Gilles
Chabert. Certified calibration of a cable-driven robot using interval contractor
programming. In Computational Kinematics, pages 209–217. Springer, 2014.
20. Philipp Miermeister, Werner Kraus, and Andreas Pott. Differential kinematics for
calibration, system investigation, and force based forward kinematics of cable-driven
parallel robots. In Cable-Driven Parallel Robots, pages 319–333. Springer, 2013.
21. Andreas Pott. Cable-Driven Parallel Robots: Theory and Application. Springer,
2018.
Towards a Precise Cable-Driven Parallel Robot -
A Model-Driven Parameter Identification
Enhanced by Data-Driven Position Correction
Marcus Hamann1 , Pauline Marie Nüsse, David Winter, and Christoph Ament
University of Augsburg, 86159, Augsburg, GER

[email protected],
WWW home page: https://www.informatik.uni-augsburg.de/lehrstuehle/rt/
Abstract. The objective of this work is to improve an existing model-

driven method for the identification of geometric parameters of a recon-
figurable parallel 3-DoF cable robot. Therefore, a procedure based on
radial basis functions (RBFs) is introduced and presented. The required
measurement is performed with the help of a laser tracker.
Keywords: cable-driven parallel robot, identification, radial basis func-

tions
1 Introduction
A flexible manipulation system is to be designed in order to assist and automate
manufacturing tasks performed in an environment with smaller batch sizes and
changing requirements. This system should be able to adapt to changing work
conditions, it should be dynamically fast and harmless to humans. As represen-
tatives of parallel kinematics, cable robots offer numerous advantages over serial
kinematics traditionally used in industry in order to achieve the objectives set
out above. Due to the often negligible weight of the cables and the low weight
of the endeffector, high accelerations in the working area can be reached. In ad-
dition, the workspace covered by cable robots is usually much larger than that
of serial kinematics.
However, there are also some disadvantages. For example, human robot inter-
action is difficult to achieve if the endeffector is connected to both the bottom
and the top of the workspace, as the safe range for humans is severely restricted.
Furthermore, the workspace of these robots – as well as for serial kinematics –
remains unchangeable after setup. After all, many cable robots available today
do not work precisely.
The MoCaRo (”Modular Cable Robot”) of the University of Augsburg counter-
acts these disadvantages by the fact that the endeffector is tensioned exclusively
upwards. Furthermore, its modular structure offers a flexibility with regard to
adaptation to new work environments due to its reconfigurability. All modules
consisting of motors, cable drums and cables can be moved flexibly along an
aluminium profile. Position errors are detected and corrected by a laser tracker

368 M. Hamann et al.
mounted on a 3D positioning system in order to increase precision. The posi-

tioning system is placed on an optical table to compensate for vibrations. Both
systems the cable robot and the laser tracker are depicted in Fig. 1.
Fig. 1. Cable-robot and laser tracker
The flexibility of this robot requires a systematic concept for identification of

geometric and dynamic parameters, since both the dynamics and the precision
of the handling should not be restricted by a modified workspace. In the case
of the 3-DoF robot MoCaRo, these parameters are three pulley positions and
three cable lengths. The attachment points of the cables at the endeffector are
assumed to be punctiform. To achieve this goal, the feasibility of an automated
model-based calibration of geometric parameters using a laser tracker was pre-
sented in [1].
Other researchers have already dealt with the calibration of geometric parame-
ters such as [2], [3]. In some of these publications also a laser tracker was used
[4], [5].
The model-driven approach presented in [1] shows that a positioning accuracy of
15 mm and less can be achieved for a working range of several meters. However,
if a higher precision is required, this can be provided by a data-driven proce-
dure. This method improves the calibration without exact knowledge about the
influence of all error sources.
The following work presents a data-driven approach, which further improves the
position accuracy of the cable robot. This approach aims to reduce the residual
position error resulting from the model-driven identification by correction of a
preset position. In contrast to the first part, this second part of the calibration
Towards a Precise Cable-Driven Parallel Robot - A Model-Driven Parameter… 369
does not refer to an underlying model, but is solely data-driven by measure-

ments.
For the procedure, a retroreflector is attached to the endeffector. This reflector
is then tracked by a laser tracker. During identification, the beam contact be-
tween laser source and endeffector or retroreflector respectively must not break
off. In this way, a relative laser length measurement can be carried out using an
interferometer. This measures laser length variations very precisely from a reset
position.
For the presentation of the procedure the model of the cable robot as well as
the laser tracker will be presented first (section 2). This is used to give a short
summary of the existing model-based procedure (section 3) in order to present
the data-driven optimization afterwards (section 4). Finally, simulative and ex-
perimental results are shown (section 5) and a short conclusion is drawn.
2 Modeling
2.1 Robot Modeling
For the first, model-driven step of identification, the inverse kinematics of the
cable robot and the measuring system – the laser tracker – are required. A
derivation of this inverse kinematics which considers the pulleys was given in [1].
The inverse kinematics results in two expressions. The first expression describes
the modeled cable lengths L̂(XS , XE ) as a function of suspension points XS and
the endeffector position XE . The second expression describes the modeled laser
lengths Ŝ(XE , XL ) as a function of the endeffector position and the laser source
position XL .
If we now take the measuring system into account, the cable and laser lengths
are composed of an offset and a length variation. In the case of cable lengths L,
the measured cable length results in
L = Loff + ΔL . (1)
The offset Loff is unknown. The cable length variation ΔL is determined using
rotary encoders. The measured laser length S results in
S = Soff + ΔS . (2)
The offset Soff is also unknown here. Since a interferometer is used the variation
in laser lengths ΔS can also be measured very accurately.
2.2 Radial Basis Functions
For the data-driven description of a spatial correction function later on, we will
introduce radial basis functions (RBFs). This chapter is a general description of
RBFs and is applied in chapter 4. The fact that this approach results in a linear
function of model paramters will simplify parameter identification from measured
data. RBFs are characterized by the fact that they are radially symmetrical. This
means their function value depends solely on a distance r from x to the origin
or the center c of the function respectively. The distance is usually described by
a norm.
θ (x, c) = θ (r) (3)
with
r = x − c (4)
The function value should be monotonously increasing or decreasing depending
on the chosen basis function and should not be negative. In the following the
function is defined as a bell-shaped function
2
θ (r) = e−(r) . (5)
In order to approximate a nonlinear function by a set of radial basis functions,

n of these functions are superposed with different weights:
n

y(x) = ai · θ (x − ci ) (6)
i=1
Parameters ai now can be identified by a linear optimization. The output of a

model results in
n

ŷ (k, p̂) = ai · θ (u(k)) (7)
i=1
⎡
⎤
a1
⎢⎢ a2 ⎥
⎥
= θ1 (u(k)) θ2 (u(k)) . . . θn (u(k)) · ⎢ . ⎥

(8)
⎣ .. ⎦
m(k)
an
T
with parameter vector p̂ = [a1 a2 . . . an ] and vector of function values m(k).
For a set of N training samples the vector of function values becomes a matrix
and can be expressed as
ŷ (p̂) = M · p̂ (9)
and the parameter vector can be calculated by
−1
p̂ = MT · M · MT · y .

(10)
3 Model-driven Calibration
For the first step of the identification two optimization problems have to be
solved. Each of them compares the modeled inverse kinematics with measure-
ments received from an interferometer or rotary encoders. First, endeffector po-
sitions are comuputed and, second, all three pulley positions are evolved. Both
steps are computed within an optimization given by the following equations:

P
2
min JS,k = min Ŝj,k − ΔSj,k (11)
j=1
N
2
min JL,i = min L̂i,k − ΔLi,k (12)
k=1
Equation (11) considers the k-th endffector position and j-th position in reference
frame which is defined by the laser tracker. For more information on this reference
frame, see [1]. Similar to the first one, the second optimization considers the k-th
endeffector postion and the i-th pulley position. Ŝj,k and L̂i,k are the inverse
kinematics. ΔSj,k and ΔLi,k are measured laser or cable lengths respectively.
For optimization we use a Levenberg-Marquardt Algorithm.
4 Data-Driven Correction
With regard to a control system to be implemented later, a high spatial accuracy

is aspired. Accordingly, a controller with input r∗ and output y will be used. In
our case, the control variable y corresponds to the endeffector position XE . The
target position is determined by the reference signal r∗ . However, the positioning
accuracy mainly depends on a correct identification of the geometric parameters.
At this point, the control loop is extended by a Two-Degree-of-Freedom Control
[6]. For this type of control a closed-loop controller is extended by an open-loop
controller. For this purpose, the output variable y of the system – the position
of the endeffector – is measured by a laser tracker and is adjusted with the help
of an adaptation. This adaptation is realized by the RBFs. Fig. 2 shows the
procedure at a glance.
Adaptation Laser Tracker
r Feedforward r∗ Feedback y
Cable Robot
Control Control
Fig. 2. Two-Degree-of-Freedom Control structure
In this way, the former reference variable r∗ will be corrected by a feedforward

controller. The new reference variable is now r. The adjustment using RBFs is
based on the Cartesian coordinates of the endeffector. For reasons of simplifica-

tion, a planar cable robot is assumed at first. For a correction of two dimensions
x and y two nonlinear functions f1 (x, z) and f2 (x, z) are required. These func-
tions are described – as explained in subsection 2.2 – by a set of RBFs and are
used to approximate a deviation for both dimensions x and y. The goal in the
following is to adapt weights on the basis of a training example. These weights
are subsequently used to correct the following positions.
A deviation between the target position xs and the measured position xm can
be determined as
Δx = xs − xm . (13)
The idea of a first training step is to measure xm and obtain Δx using an
approximation. In this way we obtain a corrected position xc . For this purpose,
an endeffector position is measured with the laser tracker and then compared
with the reference position. Fig. 3 shows the planar configuration. As can be
seen in the figure, 9 RBFs are used. These are distributed symmetrically in the
two-dimensional workspace. The positions of the training example used for the
simulation are also shown.
z
Winch
RBF center
Training point
Fig. 3. Two-Degree-of-Freedom Control structure
5 Results
5.1 2D Simulative Results
For the simulation, two winch positions are assumed as the initial configuration,
see Fig. 3. To simulate a distortion the right winch is then displaced to the bot-
tom left. This scenario is intended to show that in practice the winch positions
can never be known exactly. Nevertheless, the endeffector should move to the
correct position with a given target position r. Therefore, a correction of the
target position is required. Two values are now required for the simulation, xs
and xm . In the first winch configuration, the target cable lengths are calculated
using the inverse kinematics. In the second winch configuration, the ”measured”
cable lengths can now be determined using direct kinematics. For the calculation
of the direct kinematics, a simplified model of the cable robot with punctiform
pulleys can be assumed, since the procedure is data-driven. Based on the simu-
lated values for xs and xm , the deviations can now be determined and used for
a correction.
Fig. 4 shows the vector field of displacements as quiver plot that are required to
move the endeffector to a correct position. The beginning of an arrow indicates
the position to be set. The end of the arrow points to the position that must
be set after the shift in order to actually reach the position. The grey arrows
indicate the displacement that is required and the black arrows indicate the dis-
placement optimized by the RBFs. It turns out that these arrows are mostly
identically. In the upper part of the figure there are noticeable differences.
Fig. 4. Simulative results of adaptation in 2D; grey arrows indicate required displace-
ments, black arrows indicate optimized displacements
5.2 2D Experimental Results
For the experiment a similar scenario as in the simulation was chosen. The planar
cable robot was suspended with two winches. The winches have an initial distance
of 1.7 m. In this configuration the endeffector was moved within a grid of 3 x 3
points to defined positions. In Fig. 5 these positions are shown as circles. The
cable lengths required for each endeffector position were saved. Subsequently, the
winch distance was increased to 1.9 m. In the second configuration, the saved
cable lengths from the initial configuration were used to move the endeffector to
9 different positions. Obviously, endeffector positions are no longer in the grid
shape of the the initial step. The new positions are shown as squares in Fig. 5.
The objective of the data-driven approach is now to provide a setpoint correction
for the coordinates x and z. In this way, the endeffector can be moved to the
target positions despite the shift of the winch..
The result of the optimization by RBF is shown in Fig. 5 by crosses. As it can be
seen, these points are very close to the initial points. The root mean squared error
for a displacement of 200 mm without correction was about RMSE = 139.2 mm.
After optimization a root mean squared error of RMSEcorr = 10.3 mm was
obtained.
Fig. 5. Experimental results of adaptation in 2D
5.3 3D Simulative Results
The procedure tested for the two-dimensional case can also be applied to the
three-dimensional case. This was done simulatively. For this purpose, a configu-
ration of three winches was assumed. After that, all three winches were moved by
several centimeters. The quiver plot (see Fig. 6) shows the displacements which
are necessary to make an appropriate setpoint correction.
The grey arrows now show the calculated displacement and the black arrows
show the optimized displacements. There is also a high correspondence between
the true setpoint corrections and the corrections resulting from the optimization.
Fig. 6. Simulative results of adaptation in 3D; grey arrows indicate required displace-
ments, black arrows indicate optimized displacements
This work presented a two-stage approach for the identification of geometric

model parameters and a position correction of a 3-DoF cable robot. In a first
step, a model-driven approach based on the inverse kinematics of the robot was
used. The resulting outcomes are still subject to inaccuracies. These can be
improved by a second data-driven approach with the help of optimization by
RBF.
The simulative results of the 3-DoF cable robot are currently being tested in
an experiment. We also want to show for the three-dimensional case that the
data-driven optimization of geometric parameters – as in the two-dimensional
case – provides a siginificant improvement. Another idea is the possibility of
online identification using RBFs. After each new training point an optimization
is performed. In this way it would be possible to adjust the model parameters
during operation. This is particularly useful in view of external influences such
as temperature impacts on the geometric parameters. However, the sequence
of points to be moved to is crucial. It would be useful to perform a sensitivity
analysis within the working area to determine which next point is optimal for
identification.
The data-driven approach is independent of the underlying kinematic chain.
Both the data-driven and model-driven approach can be transferred to other
parallel and serial robots in order to identify geometric parameters or improve
their estimation.
References
1. Marcus Hamann and Christoph Ament. Calibration procedure for a geometrically
reconfigurable 3-dof cable-driven parallel robot. In Modelling, Simulation and Iden-
tification, 7/16/2018-7/17/2018.
2. Philipp Miermeister, Andreas Pott, and Alexander Verl. Auto-calibration method
for overconstrained cable-driven parallel robots. In ROBOTIK 2012. IEEE, Piscat-
away, NJ.
3. P. H. Borgstrom, B. L. Jordan, B. J. Borgstrom, M. J. Stealey, G. S. Sukhatme,
M. A. Batalin, and W. J. Kaiser. Nims-pl: A cable-driven robot with self-calibration
capabilities. IEEE Transactions on Robotics, 25(5):1005–1015, 2009.
4. XueJun Jin, Jinwoo Jung, Sukho Park, Jong-Oh Park, and Seong Young Ko. Ge-
ometric parameter calibration using a low cost laser distance sensor for a planar
cable robot: Matlab simulation. In 2016 13th International Conference on Ubiqui-
tous Robots and Ambient Intelligence (URAI), pages 534–537.
5. Julien Alexandre dit Sandretto, David Daney, and Marc Gouttefarde. Calibration
of a fully-constrained parallel cable-driven robot. In Friedrich Pfeiffer, Franz G.
Rammerstorfer, Jean Salençon, Bernhard Schrefler, Paolo Serafini, Vincent Padois,
Philippe Bidaud, and Oussama Khatib, editors, Romansy 19 – Robot Design, Dy-
namics and Control, volume 544 of CISM International Centre for Mechanical Sci-
ences, pages 77–84. Springer Vienna, Vienna, 2013.
6. William S. Levine. Control System Fundamentals. Chapman and Hall/CRC, Boca
Raton, 2018.
Part VIII
Applications
Design, Implementation and Long-Term
Running Experiences of the Cable-Driven
Parallel Robot CaRo Printer
Andreas Pott, Philipp Tempel, Alexander Verl and Frederik Wulle
Institute for Control Engineering of Machine Tools and Manufacturing Units (ISW),
University of Stuttgart
[email protected].
Abstract. Additive manufacturing has attracted a lot of attention in

the recent years as it allows to effectively manufacture objects with com-
plex shape in batch size one. Extrusion-based additive processes employ
manipulators, such as robots, to move the printing head along a pre-
defined path. This paper deals with the design, implementation, and
experimental evaluation of a new cable-driven parallel robot for addi-
tive manufacturing called CaRo printer. We compare the proposed robot
structure with other cable robots and present technical details of the
evaluation. Technical details on the mechanical and controller design are
given. A special focus is laid on practical aspects and observations made
from long-term operation of the demonstrator. We present measured data
from the long-term operation at the exhibition.
Keywords: cable-driven parallel robot, additive manufacturing
1 Introduction
Large-scale printing as a process in building construction is an interesting yet

challenging application. Improvements in terms of efficiency have a high impact
on economy and the environment, as construction is the largest section amongst
all industrial activities worldwide. The use of robots and automation in general
is, therefore, a promising approach. Cable-driven parallel robots (abbreviated as
cable robots hereafter) are proposed for construction as early as 1993 by Albus
et al. by who introduced the RoboCrane system [1]. Later, the application of
cable robots is discussed in relation to additive manufacturing technologies [2].
The main relevant advantages of cable robots are excellent scalability in terms
of size, payload, and dynamics. The scalability of cable robots is well demon-
strated by the gigantic FAST systems [3] spanning some five hundred meters
to support the collector of the world’s largest radio telescope. Also, the CoGiRo
system [4], the IPAnema robot family [5] and the cable robots shown during
the world exhibition Expo 2015 in Milan [6] underline the technical feasibility of
large-scale cable robots for different applications. Recently, the CableRobot Sim-
ulator [7] proofed that cable robots can be used for high dynamics taking into

380 A. Pott et al.
account strict safety standards for person transportation. The research on cable
robots in the field of construction robotics raised, where one can take advantage
from the large workspace and the high payload. Bosscher proposed to use cable
robots for counter crafting [8]. For example, where a mobile frame was designed
to move between construction sites. In the filed patent [9], Bosscher describes a
cable robot with up to twelve cables and a group of eight lower cables mounted
on vertical guideways.
Large-scale printing with cable robots is addressed by Izard [10] who pre-
sented the eight cable suspended robot CoGiRo for concrete extrusion printing.
The experimental results are encouraging, yet the robot employs eight actuators
for manipulating the three required degrees of freedom essential for printing. A
large-scale printing application is described in Barnett and Gosselin [11], where
foam is printed using a six-degrees-of-freedom cable-suspended robot. A more
in-depth analysis of this robot is most recently proposed by Vu et al. [12]. This
robot employs a planar platform where the pairs of cables act on the diametral
sides of a hexagon. In the paper, a significant improvement in the size of the
workspace is shown. Other processes typical for construction are handling and
assembly. A cable robot for large-scale products like collectors for concentrated
solar power (CSP) plants is studied by Pott [13] and was presented during the
trade fair Automatica 2010 in Munich, Germany. More recently, Izard studied
the installation of a cable robot on the facade of a building [14].
In this paper, a robot design with four pairs of cables on a spatial platform
is applied to achieve a similar effect for application in additive manufacturing.
The rest of this paper is structured as follows: First, we present the conceptual
design of a cable robot for additive manufacturing. Secondly, details on the design
and implementation are given including an analysis of the expected effect of
the aforementioned simplification. Thereafter, design and implementation details
on the CaRo printer are provided along with measurement and experimental
results from long-term running experiments. We close the paper with concluding
remarks.
2 Conception of Printing Cable Robot
The kinematic structure used for the CaRo printer is shown in Fig. 1. Here
the workspace does not suffer from potential interference of the cables with the
printed objects in comparison to fully-constrained robots. Furthermore, these
robots require actuation of each cable, which causes increased costs due to the
number of motors. In contrast, the suspended cable robot CaRo printer facilitates
pairs of cables and thus reduces the efforts, which are necessary for actuation of
the mobile platform.
The kinematic analysis of such robots is involved as the parallel cables render
the robot’s kinematics architecturally singular. However, the performance of the
prototype is quite satisfactory. We call such robot designs generic, as small
changes in the geometry (i.e., breaking the parallelism in the cables) significantly
change the kinematic properties. Generic constraints in the robot design are
Design, Implementation and Long-Term Running Experiences… 381
Fig. 1: Archetypical structure of the CaRo cable robot: suspended cable robot
with eight cables and four actuators.
often used to simplify the kinematic analysis. However, such reduced models
may conceal important properties in the kinematics and dynamics of the robot.
There are different ways to overcome the numerical problems arising from
the kinematic model of such generic designs. First, one can add a little distur-
bance to the geometric parameters ai and bi to deal with the ill-conditioned
kinematic model. This disturbance can be understood as manufacturing and as-
sembly tolerances of the parts. Second, one can analyze the (likewise) simplified
generic model of a 3T robot, where one computes the workspace and mobility
of a point-mass platform instead. In this case, the lock of rotation is concluded
from geometry considerations and thus conceals the physical model of stiffness
and the platform’s actual tendency for parasitic tilting. As 3T robots have been
widely studied in the literature (see e.g., [15, 16]), the presented results in this
paper are computed using a slightly disturbed geometry of the robot. Using cable
pairs that form parallelograms, one can build a cable robot that has a passive
stiffness in the rotational degrees of freedom using only three or four motors.
Such a design requires a kinematic analysis, which is described below.
For a better understanding in this paper, the kinematic foundation of cable
robots is briefly summarized. The geometry of a spatial cable robot has m cables,
where the vectors ai denote the proximal anchor points on the robot base, the
vectors bi are the relative positions of the distal anchor points on the mobile
platform and li denotes the vectors of the cables. The length of the cables is
abbreviated by li = ||li ||2 . The closure constraint follows from a vector loop and
reads
ai − r − Rbi − li = 0 for i = 1, . . . , m , (1)

382 A. Pott et al.
0,06
0,05
0,04
0,03
0,02
0,01
0
-0,1 -0,08 -0,06 -0,04 -0,02 0 0,02 0,04 0,06 0,08 0,1
x [m]
|Δβ| |Δx|
Fig. 2: Difference between the planned trajectory and the executed trajectory in
terms of horizontal position error |x| [m] and orientation error |Δβ| [rad].
where the vector r is the Cartesian position of the platform and the rotation
matrix R represents the orientation of the platform. We denote the full pose
composed of position and orientation of the mobile platform by y. From Eq. (1)
we obtain m nonlinear equations to be used for forward kinematics and the
closure conditions νi for the i-th cables reads
νi (l, r, R) = ||ai − r − Rbi ||2 − li2 = 0 for i = 1, . . . , m . (2)
This over-constrained system can be solved by using minimization techniques

as shown by [13]. Employing such a numerical procedure to solve the forward
kinematics, one computes the forward kinematic function
m

φ(l) = min Ψi (l, r, R) , (3)
r,R
i
where the given vector l = [l1 , . . . , lm ] is the vector containing the cable lengths.
Then, the function φ(l) yields the values r ∗ , R∗ that minimize the right hand
side of Eq. (3).
Now, the suspended cable robot CaRo printer is designed to execute a pure
translational motion thanks to the parallelograms. However, in practice, such
errors in geometry are due to manufacturing and assembly errors as well as
accepted displacements for mounting two panning pulleys on the same rotation
axis where the cables have little spatial displacement to avoid collision. Thus, the
lengths of the cable pairs are subject to errors that gives the robot the behavior
of a system with eight independent cables where the cable length are l1 l2 , l3
l4 , l5 l6 , and l7 l8 .
The error in the orientation Δβ = β(R̂) − β(R) is determined from the

rotation between the computed orientation R̂ by forward kinematics and the
pre-programmed rotations R (in the example simply the identity matrix R = I)
as
trR − 1 R11 + R22 + R33 − 1
β(R) = arccos = arccos , (4)
2 2
where tr(·) is matrix trace. Note that the given formula only applies to the
non-singular case. The determined angle β is equivalent to the rotation angle in
the Rodrigues’ rotation formula and it is a proper norm to measure distance in
the special orthogonal group SO3 . Numerical results for a test trajectory from
point p1 = [−0.1, −0.1, 0.25] m to p2 = [0.1, −0.1, 0.25] m are shown in Fig. 2.
A systematic position error of 5 mm and an error in orientation of up to 0.048 rad
is caused by a deflection of 25 mm of one proximal cable anchor along the axis
under investigation.
2.1 Workspoace and Interference

The workspace is evaluated based on the standard kinematic model derived
above. The direction vector of the i-th cable reads
li
ui = , (5)
||li ||
and thus the static equilibrium is expressed by the structure matrix A
f1
⎡ ⎤
u1 . . . um fP

⎢ .. ⎥
⎣ . ⎦+ = 0, (6)
b1 × u1 . . . bm × um τP
fm
A (r,R) wP
f

where f = [f1 , . . . , fm ] is the vector of the cable forces and the wrench wP
composed of the applied force fP and the applied torque τP . The rank of A is
6. The wrench-feasible workspace is computed as shown in Fig. 3. The workspace
is computed based on the boundary method [17] where the advanced closed-form
solution [18] for the force distribution problem is used to decide if a given pose
belongs to the workspace. To execute the computation the applied wrench is set
to wP = [0, 0, −60, 0, 0, 0]. An assessment of the cable-cable interference is done
with the method proposed by Perreault [19] for the CaRo printer where it reveals
no interference within the robot frame due to the parallel cable arrangement and
the suspended configuration. The problem for fully-constrained robots is much
more involved and not discussed in this contribution.
3 Design and Implementation

3.1 Controller Structure
The controller architecture used for the CaRo printer is an adaption of the con-
troller architecture used for the IPAnema robots [5]. At its core, the controller is
384 A. Pott et al.
S2
z
y
x γ = 30◦
S1
γ = 0◦
Fig. 3: The hull of the wrench-feasible workspace of the CaRo printer for a plat-
form mass (with load) of mp = 6 kg
based on a CNC motion kernel that allows parsing and executing programs writ-
ten in G-code. Therefore, the robot has a real-time standard kinematic transfor-
mation in a C++-task regarding the upper cables where the motors of the winches
are position controlled. However, the process of printing requires a path-velocity
synchronized operation of the printing subsystems. The process-related informa-
tion for printing is encoded in the G-code program and synchronously executed
in the PLC tasks. An important advantage of the modular control approach
allows the execution of printing programs. For long-term running analysis, the
CaRo printer control system is connected to a database that saves parameters
e.g., winch set-point values (Θs,i ) and actual winch values (Θi , i = 1, . . . , 4).
3.2 Hardware Design

The CaRo printer has an aluminum profile frame with a size of 1.20 m × 1.20 m × 2.13 m
(exact dimensions in Tab. 1) An integrated electrical cabinet and the internal
cable mechanic are installed in the actor space. On a metal plate, four winches
with their servo drives are placed symmetrically. The geometrical design requires
vertical parallelograms formed by the cable pairs, the mobile platform and the
proximal anchor points. Here, each synchronized cable pair runs on one actua-
tion unit with two drums, one left-hand thread and one right-hand thread (see
Fig. 4). Thereby, the cable lengths between the point of attack on the drums to
the parallel arranged lower pulleys are equidistant.
The cables of each pair reside at the lower pulley about 40 mm apart from
each other. In order to trim the cable pair’s single cables identically, the Z-
position of the lower pulleys can be slightly changed. From the lower pulleys the
cables are guided to the two upper pulleys (proximal anchor points), which syn-
chronously rotate about the Z-axis. Within each cable pair at the upper pulleys
Table 1: Geometry data of the CaRo printer in terms of the proximal anchor
points ai and the distal anchor points bi .
ai /m bi /m
i x y z x y z
1 0.5 −0.5 0.96 0.0495 −0.0495 0.115
2 0.4975 −0.4975 0.895 0.0470 −0.0470 0.050
3 −0.5 −0.5 0.96 −0.0566 −0.0566 0.115
4 −0.4975 −0.4975 0.895 −0.0541 −0.0541 0.050
5 −0.5 0.5 0.96 −0.0495 0.0495 0.115
6 −0.4975 0.4975 0.895 −0.0470 0.0470 0.050
7 0.5 0.5 0.96 0.0566 0.0566 0.115
8 0.4975 0.4975 0.895 0.0541 0.0541 0.050
Table 2: Overview of the specifications of the CaRo printer

parameter and symbol value/component unit
Number of cables m 8 −
Number of actuators 4 −
Degrees of freedom n 3, 3T, suspended −
Size of the robot frame 1.155 × 1.155 × 1.145 m
Size of the mobile platform 0.05 × 0.07 × 0.14 m
Rated cable force fmin , fmax 2 to 130 N
Max cable velocity l˙max 0.59 m s−1
Drive servo drives AM 8121 Beckhoff −
Drive power 0.157 kW
Gear ratio νP G 1:16 −
Drum diameter dD 0.06 m
Cable type LIROS-D-Pro Static −
Cable diameter dC 3 mm
Control system Beckhoff TwinCAT 3.1 −
the cables have a distance of 85 mm in Z-direction and 3 mm in the direction

of li (last one for cable collision avoidance). This distance is parallelly mapped
to the platform joints. The cables introduce forces into the platform through
universal joints at the distal anchor points. The parameters resulting of the de-
sign considerations are given in Tab. 1 and Tab. 2, respectively. For calibration,
the platform is rigidly attached to the origin of the coordinate system, and all
cables are tensed with a winch torque of 3 N m. After calibration, the platform
is released and becomes a suspended configuration. Due to the platform being
a fully restrained cable robot during calibration and subsequent release of the
fixation, the robot changes its kinematic configuration, however, no major draw-
backs or large deflections are emerging from this configuration change which is
likely to be imputable to the low cable inertia and low induced cable tension.
386 A. Pott et al.
Fig. 4: Prototype of the CaRo Printer with printing head
For printing operations a modular printing head is screwed. The CaRo printer
was designed with a strong focus on long-term running in public environment.
The robustness and safety precautions against unauthorized interventions are
c
realized by wood and Plexiglas coverings, which encapsulate the robot. For
printing applications skilled staff is controlling the machine by a GUI, whereas
in non-printing running cycles museum visitors are actuating the robot. Five
mode buttons are loading pre-defined G-codes with a cycle time of 22 s to 30 s
for each code.
4 Experimental Results
In this section, we present experimental accuracy and printing results and ex-
periences with the long-term operation of the cable robot CaRo printer which
has been operated for six months in an exhibition room. After installation and
first calibration, the CaRo printer has been subjected to endurance tests of con-
tinuous movement about one week. The subsequent calibration has shown that
the cables creep some 0.2 mm. Further creep could not be detected in regular
(approx. monthly) calibration procedures. The cause of no change in length is
probably the low platform mass.
Position measurements according to DIN ISO 230 standard (5 times per
position point) were carried out in order to examine the position accuracy. The
measurement was done from two central positions (S1 = [0, 0, 0] m , S2 =
[0, 0, 0.25] m) in X-direction (γ = 0◦ ) and in 30◦ from the X-direction in the X-
Y-plane (γ = 30◦ ), where “Forward” is in +X/+Y-direction and “Backward” in
Fig. 5: Position accuracy of the CaRo Printer where “Forward” is in +X/+Y-

direction and “Backward” in -X/-Y-direction.
-X/-Y-direction. Therefore, a laser doppler vibrometer (0FV-303 from Polytec)

was used to measure a reference point at the top of the platform. The relative
position errors are shown in Fig. 5. Linear position errors can be seen, which
increase in the upper region of the working space.
A position error of less than 3 % at S1 is similar to the simulation in Fig. 2.
This effect is bigger in S2 (around 10 %). This is caused by moderate tilting
of the platform ,which is caused by mechanical tolerances. Accuracy in terms
of repeatability has in all points errors below 0.5 %. For printing applications
these errors are well permissible, especially for architectural parts, and can be
improved by increasing the platform weight and the accuracy of the mechanics.
After measuring the position error once, the accuracy can be optimized by means
of compensation lists. In contrast to rigid-architecture machines the accuracies
of the CaRo printer is worse, but would far cheaper solutions for many printing
applications on constructions zones.
Printing tests were carried out with supplied PLA material. The printing
results with linear interpolated paths have caused poorer surface and geometri-
cal qualities as opposed to B-spline interpolation. The smoother transitions and
the associated lower jerk have significantly improved the process. One reference
workpiece, a double helix shell, is shown in Fig. 6, in which the difference be-
388 A. Pott et al.
Fig. 6: Double helix shell reference workpiece. Linear interpolated trajectories

(right), B-Spline interpolation (left).
tween the interpolation types can be seen. This cylinder has two 0.8 mm wide
peripheral tracks and a layer thickness of 0.7 mm. The B-Spline interpolated
printed workpiece has less than 1 % geometrical length errors.
During the exhibition time, from October 2017 to May 2018, the cable robot
was running about eight hours a day, six days a week. During this time, no main-
tenance work was carried out except for calibration procedures. On average, the
visitors actuated the platform movements 113 times a day. The process data sent
to the cloud were analyzed and evaluated regularly. For long-time running evalu-
ation of the behavior, the lag distances (eΘi = ||Θs,i − Θi ||) of the winches were
considered, since the platform position was not measured directly (see Fig. 7).
Here, the platform moves in a commanded feed rate of 20 000 mm min−1 . Beside
some short-term outliers, the lag distances are less than 1.5 mm in all measure-
ments which were taken and processed on five different days. Furthermore, it can
be seen that the mechanical behavior has not significantly changed over time.
An outlook towards avoidance of tilting movements, mechanical differentials
for the cable pairs, which compensate the cable forces, will be further investi-
gated. Nevertheless, the lag distances are remarkable in highly dynamic motions.
During the printing process positional accuracy is significantly higher due to the
low feed rates.
5 Conclusions
In this paper, we introduced the suspended cable-driven parallel robot CaRo

printer designed for additive manufacturing. Experimental tests with the sus-
pended robot show encouraging results in terms of accuracy that meet the re-
Fig. 7: Lag distances measured during the execution of one representative G-code
cycle on different days.
quirements of 3D printing i.e., placing the fiber layers accurately. In addition,

long-term running robustness of the mechanics and of the control system of the
CaRo printer was proven during an exhibition time of six months. The CaRo
printer is understood as a scale prototype of potentially huge printing systems
in construction, where objects of several meters shall be printed with materials,
such as concrete.
Acknowledgment
The authors would like to thank the German Research Foundation (DFG) for
financial support of the project within the Transregional Collaborative Research
Centre (SFB/TRR) 141 “Biological Design and Integrative Structures” and par-
tial support by the DFG under Germany’s Excellence Strategy in Simulation
Technology (EXC 310/1) and in IntCDC (EXC 2120/1 – 390831618) at the
University of Stuttgart.
References
[1] Albus, J. S., Bostelman, R. V., and Dagalakis, N. G., 1993. “The NIST
ROBOCRANE”. Journal of Robotic Systems, 10(5), pp. 709–724.
[2] Khoshnevis, B., 2004. “Automated construction by contour crafting—related
robotics and information technologies”. Automation in Construction, 13(1),
pp. 5–19.
[3] Li, H., Sun, J., Pan, G., and Yang, Q., 2018. “Preliminary running and perfor-
mance test of the huge cable robot of FAST telescope”. In Cable-driven Parallel
Robots, Vol. 53. pp. 402–414.
390 A. Pott et al.
[4] Lamaury, J., and Gouttefarde, M., 2013. “Control of a large redundantly actuated
cable-suspended parallel robot”. In IEEE International Conference on Robotics
and Automation, 2013, pp. 4659–4664.
[5] Pott, A., Mütherich, H., Kraus, W., Schmidt, V., Miermeister, P., and Verl, A.,
2013. “IPAnema: A family of Cable-Driven Parallel Robots for Industrial Ap-
plications”. In Cable-Driven Parallel Robots, Mechanisms and Machine Science,
Springer, pp. 119–134.
[6] Tempel, P., Schnelle, F., Pott, A., and Eberhard, P., 2015. “Design and Pro-
gramming for Cable-Driven Parallel Robots in the German Pavilion at the EXPO
2015”. Machines, 3(3), pp. 223–241.
[7] Miermeister, P., Lächele, M., Boss, R., Masone, C., Schenk, C., Tesch, J., Kerger,
M., Teufel, H., Pott, A., and Bulthoff, H. H., 2016. “The CableRobot Simulator:
Large Scale Motion Platform Based on Cable Robot Technology”. In IEEE/RSJ
International Conference on Intelligent Robots and Systems (IROS), pp. 3024–
3029.
[8] Bosscher, P., Williams II, R. L., Bryson, L. S., and Castro-Lacouture, D., 2007.
“Cable-suspended robotic contour crafting system”. Automation in Construction,
17(1), pp. 45–55.
[9] Bosscher, P. M., and Williams II, R. L., 2009. Apparatus and Method Associated
with Cable Robot System.
[10] Izard, J.-B., Dubor, A., Hervé, P.-E., Cabay, E., Culla, D., Rodriguez, M., and
Barrado, M., 2017. “Large-scale 3D printing with cable-driven parallel robots”.
Construction Robotics, 7(1), p. 27.
[11] Izard, J.-B.Barnett, E., and Gosselin, Clément, 2017. “Large-scale 3D printing
with a cable-suspended robot”. Additive Manufacturing, 7(1), p. 27.
[12] Vu, D. S., Barnett, E., Zaccarin, A. M., and Gosselin, C., 2018. “On the design of a
three-DOF cable-suspended parallel robot based on a parallelogram arrangement
of the cables”. In Cable-driven Parallel Robots, Vol. 53. pp. 319–330.
[13] Pott, A., Meyer, C., and Verl, A., 2010. “Large-scale assembly of solar power
plants with parallel cable robots”. In International Symposium on Robotics (ISR)
and German Conference on Robotics (ROBOTIK), pp. 1–6.
[14] Izard, J.-B., Gouttefarde, M., Michelin, M., Tempier, O., and Baradat, C., 2013.
“A Reconfigurable Robot for Cable-Driven Parallel Robotic Research and Indus-
trial Scenario Proofing”. In Cable-Driven Parallel Robots, Mechanisms and Ma-
chine Science, Springer, pp. 135–148.
[15] Verhoeven, R., 2004. “Analysis of the Workspace of Tendon-based Stewart Plat-
forms”. PhD thesis, University of Duisburg-Essen, Duisburg, Germany.
[16] Peng, L., and Hongwei, M., 2016. “On the stability for a cable-driven parallel
robot while considering the cable sag effects”. In 13th International Conference
on Ubiquitous Robots and Ambient Intelligence (URAI), pp. 538–543.
[17] Pott, A., 2008. “Forward Kinematics and Workspace Determination of a Wire
Robot for Industrial Applications”. In Advances in Robot Kinematics (ARK),
Springer, pp. 451–458.
[18] Pott, A., 2013. “An improved Force Distribution Algorithm for Over-Constrained
Cable-Driven Parallel Robots”. In Computational Kinematics. Springer, pp. 139–
146.
[19] Perreault, S., Cardou, P., Gosselin, C., and Otis, M. J.-D., 2010. “Geometric
determination of the interference-free constant-orientation workspace of parallel
cable-driven mechanisms”. ASME Journal of Mechanisms and Robotics, 2(3).
A Dual Joystick-Trackball Interface for Accurate
and Time-Efficient Teleoperation of
Cable-Driven Parallel Robots within Large
Workspaces
Kwun Wang Ng1 , Robert Mahony2 , and Darwin Lau1

1
The Chinese University of Hong Kong, Sha Tin, Hong Kong,
kwunwang [email protected], [email protected]
2
Australian National University, Canberra, Australia,
[email protected]
Abstract. There is an increasing number of applications (e.g. building

construction and sport camera systems) for cable-driven parallel robots
(CDPRs) due to its large workspace feature. Although the autonomous
operations of the CDPRs is well studied, human command of the robots
(teleoperation) is still very crucial to provide the intelligent decision for
complicated tasks and dynamic environments. However, existing teleop-
eration schemes struggle to provide both fast and accurate commands of
CDPRs that typically operate within large workspaces. Thus, we devel-
oped a novel dual joystick-trackball interface for the teleoperation of the
CDPRs. The simulation and the hardware tests demonstrate the poten-
tial of the dual interface to maximize the accuracy and time-efficiency of
the teleoperation of the CDPRs.
1 Introduction
Cable-driven parallel robots (CDPRs) are well-known as a type of parallel mech-
anism where an end-effector is actuated by multiple cables in parallel. There
are many attractive advantages over conventional serial and parallel manipula-
tors, such as high payload to weight ratio, simple in structure, potentially large
workspace and high reconfigurability. One important feature of CDPRs is that
the cables can only be pulled in tension and cannot apply pushing forces. This
unilateral positive force constraint results in challenges in the analysis [1, 2, 3]
and control [4, 5, 6] that are different from the serial and parallel manipulators.
A key advantage of CDPRs is its large potential workspace, which is useful
for different applications including sports stadium camera systems [7], build-
ing construction [8] and transportation of heavy loads [9]. Although the fully
autonomous operation of CDPRs within a large workspace has been already
well-established, there is still a strong need for the human involvement in com-
manding and operating such robots. For example, the professional judgement
from a camera operator is necessary in the control of the stadium camera move-
ment during a sport match. This is because humans are strongly capable of

392 K. W. Ng et al.
making cognitive and intelligent decisions to deal with different complex scenar-
ios in a fast and adaptive manner.
Teleoperation is a method that directly involves humans in the control of
robots, and allows them to remotely operate the robots (slaves) through teleop-
eration devices (masters). One of the most common master devices is a joystick
controller. In principle, the master devices receive the position/force inputs from
human user, and execute the desired actions on the slave robots. Compared with
fully autonomous robots, teleoperation allows the human user skills, such as de-
cision making and fine manipulation, to augment the advantages of the robot
for more complex tasks in dynamic environments. Teleoperation has been widely
used in traditional robots, such as underwater robots [10], search and rescue [11],
and medical applications [12]. However, teleoperation has not been well studied
for CDPRs. Primary challenges include achieving both fast and accurate motion
in large workspaces, and the presence of cable force constraints.
In traditional teleoperation, there primarily exists two types of mapping
schemes between the master and the slave robots: Position-to-Position (P2P)
and Position-to-Velocity (P2V) mappings (as depicted in Fig. 1). In the P2P
scheme, the position of the master device is mapped to the position of the slave
robot. Alternatively, the P2V scheme maps the position of the master device to
the slave robot’s velocity (both speed and direction). Compared with the P2V
scheme, the main advantage of P2P is its higher positional accuracy because the
position of the slave robots can be directly servo-ed. However, the P2V scheme
is more favorable for robots that have to operate over a large or even unlimited
workspace as commanding the slave robot’s velocity directly avoids the limited
and finite positional range of P2P devices. Moreover, the P2V scheme allows
the operators to control the slave robots more efficiently at higher speeds within
a large workspace. In summary, the P2V scheme has higher time efficiency at
travelling over large workspaces at the sacrifice of positional accuracy.
Although there has been a lack of teleoperation interfaces for CDPRs, many
works on large workspace teleoperation have been studied for traditional robots,
such as mobile robots, using both P2V and P2P schemes. For the P2V scheme,
impedance and admittance controllers [13, 14, 15] are commonly employed for
users to smoothly control the robot’s velocity to achieve time-efficient teleoper-
ation but does not address the issue of less accurate position control. To address
this issue, the scrolling-based approach [16] and the workspace drifting approach
[17] were proposed and based on the P2P mapping by resetting the workspace
mapping between the master devices with limited workspace and the mobile
robots. Nonetheless, these P2P schemes result in low time-efficiency on the dis-
continuous position mapping due to the workspace resetting. Thus, the existing
works struggle to achieve both high accuracy and high time-efficiency in a single
system. Furthermore, teleoperation interfaces for CDPRs should take the effects
of the positive force constraint and workspace limitations into consideration.
In this paper, a novel dual joystick-trackball teleoperation interface for the
CDPRs to operate accurately and efficiently over large workspaces is introduced.
This interface is novel through the combination of two haptic input devices: (1)
A Dual Joystick-Trackball Interface for Accurate and Time-Efficient… 393
a P2V configured joystick controller, and; (2) a novel P2P configured trackball
controller. Detecting the user’s force applied onto the trackball, it serves to reg-
ulate the speed and fine positioning of the slave robot. The advantages of both
schemes, including high accuracy and time-efficiency, can be achieved concur-
rently through this dual interface approach. Through integrating the interface
and CDPRs, it is shown that human users can accurately servo the end-effector
to a desired position in a spatial CDPR. Furthermore, a workspace analysis on
CDPRs is used to generate constraints to prevent the teleoperated robot from
leaving the workspace, and this constraint is also fed back to the user through
the haptic devices. The potential of the proposed teleoperation interface for the
CDPRs with larger workspace and higher speeds is also investigated through
simulation results.
Master Device Slave Robot
P2V
Communication q
or
Platform
P2P
ξ
Operator v
Fig. 1: Teleoperation of CDPRs through P2P or P2V mapping scheme
2 Background of CDPRs
A spatial CDPR (SCDPR) with 6-DoFs (degrees of freedom) is a common type

of CDPRs used for operations within a large workspace. Consider a SCDPR
with m cables (m > 6). The cartesian coordinates of the SCDPR is q = [x y z α
β γ]T , where x, y and z represent the position of the end-effector (translation),
and α, β and γ are the orientation of the end-effector. Moreover, the cable forces
have to be bounded by positive minimum and maximum values, fmin > 0 and
fmax > 0, respectively. Based on [18, 19, 20], the equation of motion (EoM) of
the SCDPR and the cable force constraints are given by
M (q)q̈ + V(q, q̇) + G(q) = −LT (q)f (1)
0 fmin f fmax (2)

where M(q) ∈ R6×6 , V(q, q̇) ∈ R6 and G(q) ∈ R6 represent the inertial matrix,
centrifugal and coriolis forces, and gravity forces, respectively. The generalized
wrenches are calculated by the cable forces f ∈ Rm and the cable jacobian matrix
L(q) ∈ Rm×6 .
394 K. W. Ng et al.
3 Master Devices of the Dual Interface

The proposed dual joystick-trackball interface is composed of two individual
master devices to be simultaneously operated: (1) a 3-DoFs translation joystick,
and; (2) a 2-DoFs rotational trackball, as shown in Fig. 2. In the dual interface,
the P2V configured joystick commands the velocity of the robot to achieve time-
efficient motion over large workspace distances. The addition of the trackball
serves three purposes: (1) providing accurate finer position control; (2) providing
haptic feedback on the robot’s velocity, and; (3) acting as a damping injection
tool to reduce the speed of the CDPR when required.
(a) P2P rotational trackball (b) P2V joystick
Fig. 2: Dual interface teleoperation devices
3.1 P2V Configured Joystick Controller

In teleoperation, one of the common controllers suitable for P2V configured
joysticks is admittance control. In the admittance control scheme, the user’s
applied force on the joystick is sensed and mapped to the velocity command of
the CDPR. The position set-point ξ ∈ R3 of the joystick is determined directly
by the linear velocity feedback v ∈ R3 of the CDPR. The user can perceive
the motion of the CDPR through the position change of the end-effector of the
joystick. A high gain position controller is implemented to measure the applied
force on the rigid end-effector of the joystick. Through the admittance controller,
the motion range of the CDPR could be large and even unlimited since the
motion command is decoupled from the positional range of the joystick, and
only depends on the measured force.
The admittance control law between the CDPR and the joystick is
v m = km f m (3)
ξ = ks v (4)
where km , ks are the scaling factors, vm ∈ R3 is the velocity command and
fm ∈ R3 is the user force input.
3.2 Novel P2P Configured Trackball Controller

Compared with the existing master devices, a continuous rotation trackball has
no limit on the workspace for controlling the robot to travel unlimited distance
in two dimensions. Thus, the trackball is suitable for the positioning of the
CDPRs within a large workspace. To control the trackball, the motors inside
the trackball are controlled by an admittance controller with high torque and
high bandwidth actuation. In the admittance framework of the trackball, the
current measurements on the motors are used to estimate the user’s applied
torque τ b ∈ R2 on the trackball. The estimated torques in 2-DoF are mapped
to linear velocity commands vb ∈ R3 of the CDPRs on the x and y directions.
The operator can servo the horizontal planar motion of the CDPRs by rolling
the trackball. The mapping law between the torque and velocity is
⎡ ⎤
0 1
vb = μ1 ⎣ −1 0 ⎦ τ b (5)
0 0
where μ1 is the scaling factor. The magnitude of the scaling factor affects how
fine the motion of the CDPRs could be controlled from the trackball. A smaller
scaling factor results in slower velocity and smaller movement for human force
inputs at each time step. Otherwise, sudden large movements may occur and
lead to being harder for the human users to servo the position accurately.
Moreover, the trackball is not only an input device and can also provide
haptic feedback to the human users. The velocity of the CDPRs is directly
reflected on the trackball through the following relationship:
0 −1 0

ω = μ2 v (6)
1 0 0
The operator can intuitively perceive the displacements of the CDPRs in x and
y directions by applying a pressing force on the top of the trackball. This is
because the pitch and roll of the trackball is directly reflecting the displacement
of the planar motion of the CDPRs by integrating both sides of (6). The second
scaling factor μ2 dominates the perception of the motion of the CDPRs. In order
to get the noticeable velocity perception on the CDPRs, μ2 should be sufficiently
large. By resisting the rotating trackball, the operator can decrease the speed of
the trackball and consequently also the CDPR.
4 System Integration of a Dual Interface and CDPRs

This section illustrates the system integration of the dual interface and CDPRs.
The corresponding system architecture is depicted in Fig. 3.
4.1 Inputs Blending for the Dual Master Interface

One challenge for the formulation of the dual interface is how to merge two inputs
of both the P2V joystick and the P2P trackball. In this system, the velocity is
396 K. W. Ng et al.
ξ
vm
Joystick
fm Admittance 1−α +
vd Workspaces l_ v
CDPR
! controller vb and Kinematics
Trackball
τb α
Fig. 3: Architecture of the teleoperation system to command CDPRs
the common domain to blend two individual inputs to result in the reference
CDPR trajectory through the blending policy (7).
During teleoperation, the force of the joystick and the torque of trackball
applied by the operator are measured at every time instant. Through the admit-
tance controller, two different velocity inputs according to the above control laws
in Sec. 3 can be measured. In the dual interface, there are two modes: (1) joystick
only (highly time-efficient mode), and; (2) trackball only (highly accurate mode).
Instead of discrete mode switching, a linear blending policy (7) is implemented
to combine vm and vb by a blending value α(τ b ) ∈ [0, 1]. In (7), the magnitude
of the blending value determines which of the input values including vm and vb
to dominate the desired velocity vd of the CDPR.
vd = (1 − α)vm + αvb (7)
1
α(τ b ) = (8)
1 + e−kb ( τ b −s)
In practice, the CDPR is mainly operated in highly time-efficient mode from

one target to another, meaning that the joystick is mainly used in such instances.
When the CDPR approaches the desired target, the user would apply a torque
on the ball in order to position the robot more accurately. Higher applied torque
τb means higher intention α to switch from highly time-efficient mode to highly
accurate mode. The intention from the user is recognized by (8), where kb ∈ R
and s ∈ R represent the growth rate and the offset. Through this blending policy,
the control authority on the CDPR could continuously slide between the P2V
joystick and the P2P trackball. Finally, the entire spectrum shown in Fig. 4 can
be achieved by this framework to strike a balance between time-efficiency and
accuracy in a smooth and fast manner.
There are two reasons why a logistic function (8) to determine the blending
value is chosen. First, the logistic function increases slowly for large increase on
the applied torque near the origin. This creates a buffer that allows the operators
to lightly press on the trackball and apply a torque to perceive the motion
of the CDPR without significantly affecting the joystick command. Moreover,
the logistic function provides a bounded growth on the blending value to avoid
unbounded large velocity commands on the CDPR.
vd
α
High time-efficiency Low time-efficiency
Low accuracy High accuracy
Fig. 4: Schematic of Eq. 7 representing relationship among accuracy, time-

efficiency and the blending value
4.2 Workspaces and Control of the CDPRs within Teleoperation
Consideration of the workspace of the CDPRs In order to guarantee a

valid motion on the CDPRs, the feasible workspaces of the CDPRs have to be
calculated for the teleoperation. A workspace boundary determination method
[21] can compute a workspace based on different criteria, such as wrench clo-
sure or wrench feasibility, and ensure the continuity of the workspace inside a
bounding box depicted in Fig. 5. Due to the continuity, the end-effector can be
teleoperated in arbitrary speeds and directions based on the user’s command
inside the bounding box. In this method, an estimated boundary for the trans-
lational workspace with a fixed orientation is constructed by k triangular planes
{F1 , .., Fk }. To integrate the desired velocity with the workspace, when the end-
effector reaches the triangular planes, and the direction of the desired velocity
is not pointing towards the workspace ( q ∈ Fi and vdT n̂i > 0), the desired ve-
locity is projected to the located plane Fi . This projection can eliminate the
velocity component along the normal direction n̂i ∈ R3 of the boundary to the
end-effector from leaving the feasible workspace. The corresponding formulation
of the projection is given by

projFi (vd ) if q ∈ Fi and vdT n̂i > 0, i=1,...,k
vc = (9)
vd otherwise
where vc is the desired velocity after the workspace is considered.
Control of the CDPRs In the controller of the CDPR, inverse kinematic is

performed to calculate the cable length commands based on the relationship
(10), where l̇ is the change of the cable lengths. The desired motion from the
user can be performed by controlling the cable lengths. In practice, a offset
based on impedance control and the cable stiffness is added on the cable length
commands to generate the tensions in cables. Finally, the velocity feedback v of
the CDPR is reflected on both master devices through the admittance controller,
and provides the perceptions of the boundary.
398 K. W. Ng et al.
l̇ = L(q)vc (10)
Workspace Fi
î
n
vc
vd
Fig. 5: Workspace boundary for a CDPR (Left: the projection of the desired
velocity; Right: the bounding box)
5 Results
In this section, the results of the simulation on the Portable Cable-Driven Robot
(PoCaBot) are presented to evaluate the feasibility of the proposed dual interface
to achieve accurate and time-efficient teleoperation. A set of experiment results
will be presented to show the advantages of the additional trackball.
5.1 Experimental Setup

PoCaBot is a spatial cable-driven robot actuated by 8 cables, as shown in Fig. 6a.
The experiments and analysis of the PoCaBot were conducted on the simulation
platform, Cable-robot Analysis and Simulation Platform for Research (CASPR)
[22]. The dual interface includes the customized P2P configured trackball and
the modified Novint Falcon to emulate the P2V configured joystick. The way to
operate the dual interface is shown in Fig. 6b. Two views shown in Fig. 7 are
provided for the user to see and know where the robot is. In the broad view, it
is preferable to use the joystick to roughly control the moving direction of the
robot. The accurate position control can be achieved for the user by observing
the position of the robot in the narrow view.
In practice, the range of the magnitude of vd is dominated by the velocity
of the P2V joystick vm . The maximum velocity of the joystick is limited. In the
experiments, the user was instructed to control the end-effector of the PoCaBot
to reach a target position on the xy-plane in three control methods: (1) P2V
configured joystick with 0.1m/s for the maximum vd ; (2) P2V configured joystick
with 0.4m/s, and; (3) dual interface with 0.4m/s. For each method, the user was
required to conduct 20 trials to control the robot to approach the target point.
A error bound is set at the target. Whenever the end-effector stays at the error
bound after a time period (stopping time), the travelling time from the starting
point to the zone is recorded. The shorter travelling time and low variance among
different trials mean high efficiency and accuracy.
(a) Visualization of the PoCaBot (b) Operation of the dual inter-

face
Fig. 6: Dual interface teleoperation devices
Fig. 7: The simulation environment. Left: The broad view of the PoCaBot in
xy-plane; Right: The narrow view of the target zone (Grey area represents the
target zone.)
5.2 Simulation and Results
The comparison results of the three schemes are based on 1 mm error bound and
3 seconds for the stopping time, and depicted in Fig. 8. In Fig. 8, the diamond
markers represent that the robot stay at the error bound after stopping time. In
the operation in lower speed, the P2V joystick has higher positional resolution.
Less sudden movements of the robot with lower speed occurred in the narrow
view of the target point. It is easier for the user to control the robot on the narrow
view. The small variance, shown in Fig. 8a, on P2V configured joystick with the
maximum velocity 0.1m/s supports this claim. Compared with the maximum
velocity 0.4m/s, it leads to longer travelling time due to low travelling speed. On
the other hand, it is obvious that the proposed dual interface provides accurate
400 K. W. Ng et al.
control in high speed by comparing the results of P2V configured joystick and
the dual interface with the same maximum velocity 0.4m/s, shown in Fig. 8b
and Fig. 8c. From the comparison of the three methods, shown in Fig. 8d, it is
proven that the user can obtain an accurate and time-efficient control on CDPRs
through the proposed dual interface.
During teleoperation, the most effective way to use the dual interface from
the user’s experience is separated into three stages. The first stage is to use
the joystick to drive the robot as fast as the robot can. Secondly, whenever the
robot is near the target, the trackball is pressed to damp the motion. Finally,
the trackball is rolled by the user to finely control the robot to the target zone.
Moreover, it is found that the haptic feedback can provide the perception of the
workspace boundary. Perceiving the workspace boundary, the user can quickly
react to the boundary, and decide on-task corrective action to leave the boundary
during the operation.
P2V configured joystick (maximum velocity = 0.1m/s) P2V configured joystick (maximum velocity = 0.4m/s)
1.8 1.8
1.6 1.6
1.4 1.4
1.2 1.2
Error norm (m)
Error norm (m)
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55
Time (s) Time (s)
(a) (b)
Dual Interface (maximum velocity = 0.4m/s) Comparison of time between control methods
1.8 55
1.6 50
1.4 45
1.2 40
Error norm (m)
1 35
Time (s)
0.8 30
0.6 25
0.4 20
0.2 15
0 10
-0.2 5
0 5 10 15 20 25 30 35 40 45 50 55 P2V (v = 0.1) P2V (v = 0.4) Dual Interface (v = 0.4)
Time (s) Control scheme
(c) (d)
Fig. 8: The experiment results of the three control methods: (a) P2V configured
joystick with 0.1m/s; (b) P2V configured joystick with 0.4m/s; and (c) dual
interface with 0.4m/s; (d) Comparison of the three methods

A novel dual joystick-trackball interface for the teleoperation of the CDPRs with
a large workspace is proposed in this paper. The proposed dual interface aims
to address the lack of teleoperation interfaces for the CDPRs within a large
workspace, and fill the needs of accurate and time-efficient teleoperation for the
CDPRs. The simulation results verified that the performance of the the dual
interface is capable of achieving highly accurate and time-efficient teleoperation
of the CDPR. Future work will look into further incorporating the force and
velocity constraints into the teleoperation, and performing hardware validation
of higher speed systems over large workspaces.
References
[1] Lau, D., Oetomo, D., and Halgamuge, S. K. “Generalized modeling of multi-
link cable-driven manipulators with arbitrary routing using the cable-routing
matrix”. In: IEEE Transactions on Robotics 29.5 (2013), pp. 1102–1113.
[2] Gouttefarde, M. and Gosselin, C. M. “Analysis of the wrench-closure workspace
of planar parallel cable-driven mechanisms”. In: IEEE Transactions on Robotics
22.3 (2006), pp. 434–445.
[3] Pham, C. B., Yeo, S. H., Yang, G., Kurbanhusen, M. S., and Chen, I.-M. “Force-
closure workspace analysis of cable-driven parallel mechanisms”. In: Mechanism
and Machine Theory 41.1 (2006), pp. 53–69.
[4] Eden, J., Tan, Y., Lau, D., and Oetomo, D. “On the positive output control-
lability of linear time invariant systems”. In: Automatica 71 (2016), pp. 202–
209.
[5] Oh, S.-R. and Agrawal, S. K. “Cable suspended planar robots with redundant
cables: Controllers with positive tensions”. In: IEEE Transactions on Robotics
21.3 (2005), pp. 457–465.
[6] Zi, B., Duan, B., Du, J., and Bao, H. “Dynamic modeling and active control of
a cable-suspended parallel robot”. In: Mechatronics 18.1 (2008), pp. 1–12.
[7] Skycam, Advanced technology.@ONLINE. Sept. 2018. url: http://www.skycam.
tv/.
[8] Wu, Y., Cheng, H. H., Fingrut, A., Crolla, K., Yam, Y., and Lau, D. “CU-brick
cable-driven robot for automated construction of complex brick structures: From
simulation to hardware realisation”. In: Simulation, Modeling, and Programming
for Autonomous Robots (SIMPAR), 2018 IEEE International Conference on.
IEEE. 2018, pp. 166–173.
[9] Tempel, P., Herve, P.-E., Tempier, O., Gouttefarde, M., and Pott, A. “Esti-
mating inertial parameters of suspended cable-driven parallel robots—Use case
on CoGiRo”. In: Robotics and Automation (ICRA), 2017 IEEE International
Conference on. IEEE. 2017, pp. 6093–6098.
[10] Khatib, O., Yeh, X., Brantner, G., Soe, B., Kim, B., Ganguly, S., Stuart, H.,
Wang, S., Cutkosky, M., Edsinger, A., et al. “Ocean one: A robotic avatar
for oceanic discovery”. In: IEEE Robotics & Automation Magazine 23.4 (2016),
pp. 20–29.
402 K. W. Ng et al.
12 REFERENCES
[11] Marques, C., Cristóvao, J., Alvito, P., Lima, P., Frazao, J., Ribeiro, I., and Ven-
tura, R. “A search and rescue robot with tele-operated tether docking system”.
In: Industrial Robot: An International Journal 34.4 (2007), pp. 332–338.
[12] Taylor, R. H., Menciassi, A., Fichtinger, G., Fiorini, P., and Dario, P. “Medical
robotics and computer-integrated surgery”. In: Springer handbook of robotics.
Springer, 2016, pp. 1657–1684.
[13] Schill, F., Hou, X., Mahony, R., et al. “Admittance mode framework for haptic
teleoperation of hovering vehicles with unlimited workspace”. In: Proc. Aust.
Conf. Robot. Autom.(ACRA). 2010, pp. 1–7.
[14] Hou, X. and Mahony, R. “An intuitive multimodal haptic interface for tele-
operation of aerial robots”. In: Robotics and Automation (ICRA), 2014 IEEE
International Conference on. IEEE. 2014, pp. 838–845.
[15] Diolaiti, N. and Melchiorri, C. “Teleoperation of a mobile robot through hap-
tic feedback”. In: Haptic Virtual Environments and Their Applications, IEEE
International Workshop 2002 HAVE. IEEE. 2002, pp. 67–72.
[16] Conti, F. and Khatib, O. “Spanning large workspaces using small haptic de-
vices”. In: First Joint Eurohaptics Conference and Symposium on Haptic Inter-
faces for Virtual Environment and Teleoperator Systems. World Haptics Con-
ference. IEEE. 2005, pp. 183–188.
[17] Rüesch, A., Mersha, A. Y., Stramigioli, S., and Carloni, R. “Kinetic scrolling-
based position mapping for haptic teleoperation of unmanned aerial vehicles”.
In: Robotics and Automation (ICRA), 2012 IEEE International Conference on.
IEEE. 2012, pp. 3116–3121.
[18] Eden, J., Lau, D., Tan, Y., and Oetomo, D. “Available acceleration set for the
study of motion capabilities for cable-driven robots”. In: Mechanism and Ma-
chine Theory 105 (2016), pp. 320–336.
[19] Lau, D., Eden, J., Halgamuge, S. K., and Oetomo, D. “Cable function analysis
for the musculoskeletal static workspace of a human shoulder”. In: Cable-Driven
Parallel Robots. Springer, 2015, pp. 263–274.
[20] Lau, D., Bhalerao, K., Oetomo, D., and Halgamuge, S. K. “On the task spe-
cific evaluation and optimisation of cable-driven manipulators”. In: Advances in
reconfigurable mechanisms and robots I. Springer, 2012, pp. 707–716.
[21] Pott, A. “Efficient Computation of the Workspace Boundary, Its Properties and
Derivatives for Cable-Driven Parallel Robots”. In: Computational Kinematics.
Springer, 2018, pp. 190–197.
[22] Lau, D., Eden, J., Tan, Y., and Oetomo, D. “CASPR: A comprehensive cable-
robot analysis and simulation platform for the research of cable-driven parallel
robots”. In: Intelligent Robots and Systems (IROS), 2016 IEEE/RSJ Interna-
tional Conference on. IEEE. 2016, pp. 3004–3011.
Active Vibration Damping of a Cable-Driven Parallel
Manipulator Using a Multirotor System
Yue Sun1‫خ‬, Matthew Newman1, Arthur Zygielbaum2, Benjamin Terry 1

1
Department of Mechanical and Materials Engineering, University of Nebraska-Lincoln,
Lincoln, USA
2School of Natural Resources, University of Nebraska-Lincoln
Lincoln, USA
[email protected],[email protected],[email protected],
[email protected]
Abstract. A cable-driven parallel manipulator (CDPM) is a robotic manipulator

designed to control the position and/or orientation of its end effector within the
system’s workspace by use of actuated cables. Some configurations of cable-
driven systems are inherently low in stiffness and so the end effector is prone to
vibration under disturbances. In this paper, we study the feasibility of implement-
ing a multirotor system to actively damp the translational vibration of CDPM end
effectors. In a simulated experiment, the system attenuates vibration time by 80%
for a 30 kg mass with a vibration frequency of 1 Hz and an undamped amplitude
of 15 cm.
Keywords: cable-driven parallel manipulator, vibration damping, multirotor

system.
1 Introduction
A cable-driven parallel manipulator (CDPM) is one of the best options for precise po-
sitioning applications where large workspace, quick motions, heavy loads, and cost ef-
ficiency are required. Recently these systems have been applied to field phenotyping,
which is a process requiring automated measurement of the growth traits of plants to
help increase agricultural productivity [1]. Compared to conventional field phenotyping
strategies, cable-driven systems demonstrate promising advantages with minimal infra-
structure, automatic control, repeatable positioning, and scalability to a larger range of
field dimensions and shapes.
A drawback of this type of system, however, is its limited ability to reject excessive
vibration at the end effector due to the low effective stiffness of long, sagging cables
[2]. In agricultural phenotypic studies, there are several sensors and cameras that re-
quire prolonged exposure times, and so platform vibrations can compromise the sensor
readings, which further increase the settling time of the end effector and decrease the
efficiency of the system.

404 Y. Sun et al.
There is a wide range of work related to vibration damping for CDPMs. In general,
damping methodology currently fits within three categories: 1) addition of extra cables
to increase stiffness, 2) active control of cable tension, and 3) active damping using
inertia. The addition of cables can be used to passively damp the vibrations of CDPMs.
For instance, Duan et al. added three downward acting cables to increase tension and
therefore rigidity in the cable system, serving as the passive vibration damper to help
reduce vibrations of their CDPM [3]. However, this passive strategy can be insufficient
for applications that have very high stability requirements or when system geometry
does not allow for downward acting cables. In the case of the phenotyping system, the
downward acting cables would impinge on the plant foliage in the field, some of which
grows to heights of 5 m.
Internal forces in CDPMs contribute to an increase in the structure stiffness [4], thus,
the vibration of the end effector can be alleviated by controlling the internal forces in
the cables. Kawamura et al. used bias tension and nonlinear springs to increase the
internal forces in cables to minimize the vibrations of an ultra-highspeed robot, the
FALCON-7 [5]. A drawback of this approach is that the performance of actively con-
trolling the internal forces depends on precise dynamic control of the system, which is
difficult to achieve on large CDPMs.
Active vibration damping using inertia refers to the use of mechanisms that imple-
ment principles of momentum conservation within the system to counteract vibration
and improve end effector stability. Shao et al. used this approach by integrating a Stew-
art platform into the terminal of the FAST system to reduce the vibrations and increase
the terminal positioning accuracy [6]. Weber et al. used reaction wheels to actively
damp the rotational vibration of a lightweight (2kg) CDPM [7]. Lesellier et al. inte-
grated rotary arms into a CDPM to actively damp vibrations [8]. Shortcomings of this
method are that inertial vibration damping strategies are heavy and require expensive
and precise hardware, careful calibration, and sophisticated controls.
This research adds another active vibration damping strategy to the repertoire of vi-
bration damping for CDPMs. Motivated by the recent achievements of multirotor sys-
tems (also known as quadcopters or drones) in applications from data collection, to
parcel delivery, to crop surveillance [9]–[11], we proposed to integrate a multirotor
system into a CDPMs to help increase the stability of the end effector. Integrating an
autonomous multirotor system with a CDPM can take advantage of the benefits of both
systems and enhance data collection. In such a hybrid system, the multirotor is used
only for stabilization while cables are used for support and locomotion. As a result, the
hybrid system has longer data collecting time, better stability under disturbance, and
larger payload capacity of the end effector. To our knowledge, this work is the first to
apply a multirotor system to CDPMs for vibration damping.
The contribution of this work is the feasibility evaluation of improving the stability
of the large cable-driven parallel manipulators by using a multirotor system. In our
previous work [12], we derived the static analysis and dimensional optimization of a
field phenotyping platform (FPP), which has been constructed at the Agriculture Re-
search and Development Center (ARDC) of the University of Nebraska-Lincoln
(UNL), as shown in Fig. 1. This paper extends that work by developing a multirotor
Active Vibration Damping of a Cable-Driven Parallel Manipulator… 405
system which generate reaction forces by interacting with the air around the end effec-
tor to actively damp the vibrations. This technique could lead to a simpler, low-cost,
robust method of stabilizing a suspended CDPM end effector.
(a)
(b) (c) (d)
Fig. 1. The field phenotyping platform (FPP) with main components. (a) Overview of the FPP
system at Nebraska, USA. (b) The end effector of the FPP system (without the multirotor system).
(c) Sensors on-board the end effector. (d) Close view of pole with winch house located at its base.
2 System Requirements
FPP is an eight-cable field phenotyping system that covers a 0.4 hectare (1.0 acre) field.
There are four 25-meter high poles, one located at each corner of the rectangle field
(Fig. 1). Two cables are fed through pulleys from the top of each pole, one connected
to the top of the end effector and the other one fixed to the bottom of the end effector,
as shown in Fig. 1(b). With such cable configurations, the four top cables enable the
end effector to achieve the desired three-dimensional translational motion while the
four redundant bottom cables significantly improve the end effector stability to rota-
tional vibrations. The system, however, still has low stability in translational vibration
due to the inevitable low stiffness of the cable supports, especially in the vertical direc-
tion. Thus, we are aiming to increase the stability of the system to vertical translational
vibration by using a multirotor system.
Based on our observation of the FPP system during operation, the vertical transla-
tional vibration frequency is approximately 1 to 2 Hz depending on the location of the
end effector. The observed maximum vertical translational vibration amplitude is about
15 cm. In this research, we set 1 Hz as the target vibration frequency to attenuate, and
406 Y. Sun et al.
we evaluated the vibration damping performance of the multirotor system at this fre-
quency. Future work will focus on controlling vibration within the full range of ex-
pected frequencies from 1 to 2 Hz.
At the most general level, the functional requirement of the phenotyping system
is to achieve accurate and timely end effector sensor readings. This general requirement
can be broken down into the following design parameters for the multirotor vibration
damping system.
1) Positional accuracy: Due to the large footprint of a typical growing plot, the end
effector must be located within 20 cm of the target position. The positional accuracy of
the FPP at steady state is ±1 cm, thus the maximum peak-to-peak value allowed from
vibration must be less than 18 cm.
2) Blur: Blur is caused by the relative motion of the camera with respect to the im-
aging target during exposure. For images to be accurate, the peak-to-peak amplitude of
vibration must be less than 2.5 cm if the frequency of vibration is 1 Hz in order to
prevent unacceptable blur [13]. We chose a target maximum peak-to-peak amplitude
of vibration of 1 cm in case frequencies are up to 2 Hz. Note that the blur design pa-
rameter dominates the positional accuracy design parameter.
3) Settling Time: We define settling time (Ts) as the time interval between the start
of vibration and the moment when the peak-to-peak vibration amplitude is equal to or
less than 1 cm (as specified by the blur design parameter). During daily operation, the
end effector visits 360 plots in the field and operates 6~8 hours. On average, the settling
time, before data collection can begin after stopping at a plot, is about 20 seconds. Thus,
about 2 hours, or between 25% and 33% of the operation time is spent settling. Quicker
settling time will result in more time for data collection. Our target is to reduce Ts to 5
seconds or less.
4) Thrust Force: The multirotor system needs to provide sufficient thrust force to
damp the vibration. More specifics about necessary thrust force estimation are pre-
sented in Section 3.1.
5) Weight: For this proof-of-principle device, we target a total end effector weight
of 30 kg.
6) Power: Although we are using batteries for this proof-of-principle device, we will
assume that power can be delivered via a tethered cable in a formal implementation of
this system. This is a reasonable assumption for a CDPM of this type.
3 System Modelling and Design
With the system requirements in mind, we now describe how we address these require-
ments through estimating necessary thrust force for vibration damping and the mechan-
ical, electrical, and control design of the multirotor system.
3.1 Thrust force estimation
To find the necessary thrust force for vibration damping, we created a simplified model
of the end effector with the multirotor system in MATLAB, then used the actuator force
sensing capability provided by Simscape Multibody to estimate the thrust based on a

position feedback control scheme.
(a) (b)
Fig. 2. (a) The simplified mass-spring-damper model of the eight-cable FPP system. (b) Sim-
scape model of the simplified spring-mass-damper system with position control.
Since we are focused only on the vertical translational vibration damping, the eight-
cable FPP system can be simplified as a mass-spring-damper system that only has trans-
lational vibration along the vertical axis, as shown in Fig. 2(a). The multirotor system
and the end effector are simplified as a point mass M, while K and C are the global
vertical stiffness and damping coefficient of the system, respectively. We modelled this
simplified system in MATLAB, as showed in Fig. 2(b). The vertical translational mo-
tion is represented by a prismatic joint, where the stiffness coefficient K and the damp-
ing properties C can also be specified. In this model, we assign C with a negligible
value due to the small damping characteristics of cable-driven systems. As a result, the
damped natural frequency ‫ݓ‬ௗ of the system will be very close to the natural frequency
‫ݓ‬௡ of the system. Then, the stiffness coefficient K can be estimated by following (1),
as the natural frequency ‫ݓ‬௡ equals to the target frequency (1 Hz) and M is 30 kg.
಼
‫ݓ‬௡ ൌ ටಾ (1)
Using these values, K is determined to be 1183 N/m.

After we determined the coefficients, a PID controller was added to the model to
adjust the actuator force based on a position feedback control scheme. The input to the
controller is the position error of the point mass while the output of the controller is the
actuating force effecting on the prismatic joint. We set the initial displacement of the
point mass to be 15 cm, which is the observed maximum vertical translational displace-
ment of the FPP system. Because the maximum thrust force of a commercial hobby
quadcopter is around 600 N, we add a saturation block that has a band range of [-600,
0] to limit the force vector from the PID controller. The amplitudes of the point mass
408 Y. Sun et al.
with PID controller on and off are compared in Fig. 3, and the sensed actuator force is
presented in the figure as well. From the figure, we can conclude that the point mass
can be stabilized within 0.5 seconds with a maximum of 400 N actuator force. Thus,
the thrust force of the multirotor system needs to be 400 N or larger to ensure sufficient
thrust force for vibration damping.
Fig. 3. Sensed actuator force corresponding to initial condition of 15 cm displacement.
3.2 Mechanical Design

Similar to the design of a popular quadcopter, our multirotor system consists of four
identical arm-motor-propeller (AMP) sets. Each AMP consists of five components: the
propeller, the brushless motor, the electrical speed controller (ESC), the arm, and the
arm mounter, as shown in Fig. 4. The number of the AMPs can be varied based on
different applications. In this paper, we selected four AMP sets which turn the multi-
rotor system into a conventional quadrotor system.
Fig. 4. An arm-motor-propeller (AMP) set.
3.3 Electrical Design

The overall electrical design of the multirotor system is presented in Fig. 5, which
consists of two parts: the remote controller and the multirotor controller. The quad-
rotor controller can be turned on and off through the remote controller. Communication
between these two parts was accomplished through the NRF24L01L wireless module
which works at 2.4 GHz and has a range of 1 Km.
The electronics of the multirotor controller were designed to execute vibration damp-
ing during data acquisition. This was accomplished by monitoring the acceleration of
the end effector and controlling the speed of the motors. The acceleration of the end
effector was measured by a BNO055 IMU with a sampling rate of 50 Hz. To ensure
sufficient output thrust, we selected T-motor MN801S brushless motors with carbon
fiber propellers (0.66 m diameter). The maximum output thrust of the system is 10 kg
with a 12s lipo battery. Thus, a maximum of 40 kg thrust force can be obtained from
our multirotor system. Each motor was controlled by a FLAME 60A HV electronic
speed controller (ESC) with a peak current of 60 amps. Two LiPo 6s batteries were
connected in series to power the system.
Fig. 5. Electrical design diagram of the multirotor system.
3.4 Controller Design

Because the thrust force is unidirectional, the multirotor system needs to be actuated
when the end effector is moving upward and off when it moves downward. This uni-
lateral throttle control strategy is illustrated in Fig. 6 and drives the design of the con-
troller.
Fig. 6. Unilateral throttle control strategy illustration.

410 Y. Sun et al.
The dynamics of the multirotor system model is based on our previous work [14],
with the simplification that the system has constrained pitch, yaw, and roll angels. Ver-
tical acceleration from the IMU was the controlled system state and thrust was modu-
lated to drive that state to zero using a PD control scheme. We avoid the integral term
because it slows the control system response and steady state error in the vertical direc-
tion can be compensated by the cable lengths.
To achieve unilateral throttle control, we compared the current reading ( Az) and the
last reading (PreAz) from IMU to determine the direction of oscillation. The system is
moving upward when the difference between the last and current acceleration readings
is positive and so the thrust is activated only when this is the case. The control flow
diagram is shown in Fig. 7.
Fig. 7. Control scheme of the multirotor system for vibration damping.
4 Indoor Study of the Multirotor System
This section discusses the indoor testbed that was built to evaluate the translation vi-
bration damping performance of the multirotor system.
4.1 Indoor Testbed

The indoor testbed to evaluate vibration damping performance is shown in Fig. 8. It
consists of four subsystems: 1) parallel spring unit, 2) linear track, 3) multirotor system,
and 4) testbed frame.
The parallel spring unit was designed to have approximately the same stiffness (1183
N/m) that was calculated from (1) by using off-the-shelf springs. It consists of ten iden-
tical extension springs, with each spring having a stiffness of 112 N/cm and maximum
extendable length of 40 cm. The total stiffness of the system is 1120 N/m. The linear
track allows the multirotor system to have linear vertical motion only. We added extra
standard weights to the multirotor system to achieve the target weight of 30 Kg.
To introduce initial displacements, we used a cable-pulley mechanism as shown in
Fig. 8 (a). To collect real time vibration amplitudes of the multirotor system, a distance
measure/recording unit was designed and attached to the bottom plate of the spring unit,
as shown in Fig. 8 (b). The real time relative distance between the end effector and
ground was measured by a Lidar V3 sensor at a sampling rate of 50 Hz. These data
were logged into an SD card through an Arduino SD shield. The translational vibration
frequency of the testbed is about 1 Hz, which is our target value.
(a) (b)
Fig. 8. Indoor testbed. (a) Overall view of the indoor testbed. (b) Partial bottom view of the in-
door testbed.
4.2 Controller Evaluation

Before we evaluated the vibration damping performance of the complete multirotor
system, we used a beam load cell to verify the timing performance of the controller,
which includes the thrust force control based on the acceleration of the system and the
unilateral throttle control.
The experimental setup is shown in Fig. 9. Two plates are connected with the load
cell, and a motor-propeller set was mounted on the top plate while the bottom plate was
fixed on to the ground. The overall data architecture between this setup and the indoor
testbed is shown in Fig. 10. The load cell readings and the acceleration profile were
synchronized by using a second BNO055 IMU attached to the indoor testbed (Fig. 8)
to measure its acceleration.
Fig. 9. Testbed for measuring thrust force. (a) Overall view of the testbed with main compo-
nents. (b) The connection between the load cell and the motor-propeller set.
412 Y. Sun et al.
Fig. 10. Data architecture between the indoor testbed and the controller evaluation setup.
The readings from the load cell and the corresponding acceleration profile of the
testbed is shown in Fig. 11. As seen in this figure, there is small lag between the thrust
and acceleration profiles which is a result of both the dynamic response of the large
propeller and the known 122 ms response time of load cell. We time-shifted the thrust
force profile to the left 122 ms to compensate for the load cell delay in order to demon-
strate the actual thrust profile that the testbed experiences. As seen in Fig. 11, the thrust
force as expected to the acceleration of the system. One artifact to note from this figure
is the bimodal nature of the thrust force during decreasing acceleration which is due to
the proportional nature of the control.
Fig. 11. Thrust force profile based on the acceleration of the testbed.
4.3 Vibration Damping Performance Evaluation

We selected two initial displacement values (10 cm and 15 cm) to assess the perfor-
mance of the multirotor system. These two displacements are within the elastic dis-
placement range of the springs. Fig. 12 displays the comparisons with the multirotor
system on and off under these initial conditions. The effectiveness of the active vibra-
tion damping, by using the multirotor system is summarized in Table. 1, in terms of
settling time, maximum position overshot, and root mean square (RMS) error. Since
the equilibrium point of the system changed once the multirotor system was on, we
used the RMS error to describe the relative spread of the vibration amplitude around
the equilibrium position.
Overall, the results present the evidence that our approach improves the stability of
large scale CDPMs. As expected, the time to damp the vibration becomes longer as the
initial displacement increases; however, with the multirotor system, the settling time
was reduced by about 80%, and the spread of the vibration amplitude around the equi-
librium position was minimized by about 70%.
(a) (b)
Fig. 12. Comparison with the multirotor system on and off with different initial displacement:
(a) 10 cm. (b) 15 cm.
Table 1. Indoor test evaluation table

Disp. Settling Time (s) Max. Position Overshot (cm) RMS error
(cm) Free Active Attenu. Free Active Attenu. Free Active Attenu.
10 19.8 3.5 82.3% 8 4 50% 3.8 1.2 68.4%
15 27.3 5.4 80.2% 13 6 53.8% 4.7 1.2 74.4%
A video of the experiment with controller on and off can be found in this link:
https://unl.box.com/s/hdruu5wthumcp2p971kdz8pc7xq6fkys
5 Conclusion
This paper presented and evaluated a hybrid multirotor system for actively damping the
translational vibration of a large scale CDPM. The dynamics of the multirotor system
were derived, and a simple control strategy for active vibration damping is proposed.
An indoor testbed was designed to evaluate the vibration damping performance of the
multirotor system. The results from indoor experiments verify the potential of the mul-
tirotor system for vibration damping of larger scale CDPMs.
References
[1] N. Kirchgessner et al., “The ETH field phenotyping platform FIP: a cable-sus-
pended multi-sensor system,” Funct. Plant Biol., vol. 44, no. 1, pp. 154–168, Jan.
2017.
[2] J. Du, W. Ding, and H. Bao, “Cable Vibration Analysis for Large Workspace
Cable-Driven Parallel Manipulators,” in Cable-Driven Parallel Robots, T.
Bruckmann and A. Pott, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg,
2013, pp. 437–449.
414 Y. Sun et al.
[3] B. Y. Duan, Y. Y. Qiu, F. S. Zhang, and B. Zi, “On design and experiment of the
feed cable-suspended structure for super antenna,” Mechatronics, vol. 19, no. 4,
pp. 503–509, Jun. 2009.
[4] Y. Suilu, W. Zhao, L. Qi, and C. Yixin, “Stiffness analysis of a wire-driven par-
allel manipulator,” in 2012 IEEE International Conference on Computer Science
and Automation Engineering (CSAE), 2012, vol. 3, pp. 31–34.
[5] S. Kawamura, W. Choe, S. Tanaka, and S. R. Pandian, “Development of an ul-
trahigh speed robot FALCON using wire drive system,” in Proceedings of 1995
IEEE International Conference on Robotics and Automation, Nagoya, Japan,
1995, vol. 1, pp. 215–220.
[6] Z.-F. Shao, X. Tang, L.-P. Wang, and X. Chen, “Dynamic modeling and wind
vibration control of the feed support system in FAST,” Nonlinear Dyn., vol. 67,
no. 2, pp. 965–985, Jan. 2012.
[7] X. Weber, L. Cuvillon, and J. Gangloff, “Active vibration canceling of a cable-
driven parallel robot in modal space,” in 2015 IEEE International Conference on
Robotics and Automation (ICRA), 2015, pp. 1599–1604.
[8] M. Lesellier, L. Cuvillon, J. Gangloff, and M. Gouttefarde, “An Active Stabilizer
for Cable-Driven Parallel Robot Vibration Damping,” in 2018 IEEE/RSJ Inter-
national Conference on Intelligent Robots and Systems (IROS), Madrid, Spain,
2018, pp. 5063–5070.
[9] C. C. Murray and A. G. Chu, “The flying sidekick traveling salesman problem:
Optimization of drone-assisted parcel delivery,” Transp. Res. Part C Emerg.
Technol., vol. 54, pp. 86–109, May 2015.
[10] Y. Sun, A. Plowcha, M. Nail, S. Elbaum, B. Terry, and C. Detweiler, “Unmanned
Aerial Auger for Underground Sensor Installation,” in 2018 IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems (IROS), 2018, pp. 1374–
1381.
[11] D. Anthony and C. Detweiler, “UAV Localization in Row Crops,” J. Field Ro-
bot., vol. 34, no. 7, pp. 1275–1296, Oct. 2017.
[12] M. Newman, A. Zygielbaum, and B. Terry, “Static Analysis and Dimensional
Optimization of a Cable-Driven Parallel Robot,” in Cable-Driven Parallel Ro-
bots, 2018, pp. 152–166.
[13] I. Salafian, “Development of the End-Effector of a Cable-Driven Parallel Manip-
ulator for Automated Crop Sensing,” Mech. Mater. Eng. -- Diss. Theses Stud.
Res., Aug. 2017.
[14] M. Newman, “DESIGN AND EXPERIMENTATION OF CABLE-DRIVEN
PLATFORM STABILIZATION AND CONTROL SYSTEMS,” Mech. Mater.
Eng. -- Diss. Theses Stud. Res., Aug. 2017.
5HSURGXFWLRQRI/RQJ3HULRG*URXQG0RWLRQ
E\&DEOH'ULYHQ(DUWKTXDNH6LPXODWRU%DVHGRQ
&RPSXWHG7RUTXH0HWKRG
'DLVXNH0DWVXXUD7DLVKX8HNL<XVXNH6XJDKDUD0LQRUX<RVKLGD
DQG<XNLR7DNHGD

7RN\R,QVWLWXWHRI7HFKQRORJ\2RND\DPD0HJXURNX7RN\R-DSDQ

+DNXVDQ&RUS-7RZHU1LNNRFKR)XFKXVKL7RN\R-DSDQ
[email protected]
$EVWUDFW,QRUGHUWRUHSURGXFHVHLVPLFZDYHVFODVVLILHGLQORQJSHULRGJURXQG
PRWLRQE\GLVDVWHUSUHYHQWLRQHGXFDWLRQIDFLOLW\DQHDUWKTXDNHVLPXODWRUEDVHG
RQDFDEOHGULYHQSDUDOOHOPHFKDQLVPKDVEHHQFRQVWUXFWHG,QWKLVSDSHUDIHHG
IRUZDUGEDVHGFRQWUROVFKHPHZDVLPSOHPHQWHGLQWKHFDEOHGULYHQHDUWKTXDNH
VLPXODWRUIRUDFKLHYLQJKLJKDFFHOHUDWLRQZLWKIUHTXHQWUHYHUVHURWDWLRQUHTXLUHG
WR UHSURGXFH DFWXDOO\REVHUYHG VHLVPLF ZDYHV 7KH GHULYDWLRQ RI WKH G\QDPLF
PRGHOZDVGRQHLQWZRVWHSVWKHILUVWVWHSLVFDOFXODWLQJFDEOHWHQVLRQVWRDFKLHYH
D PRYLQJ SODWIRUP¶V WDUJHW DFFHOHUDWLRQ DQG WKH VHFRQG LV IRU HDFK VSRROHU¶V
WRUTXHFDOFXODWLRQWRDFKLHYHWKDWWHQVLRQ)RUDIHHGIRUZDUGWRUTXHFDOFXODWLRQ
XVLQJWKHVHFRQGLQYHUVHG\QDPLFPRGHORIFDEOHVSRROHUDQH[SHULPHQWDOLGHQ
WLILFDWLRQVFKHPHRIPRGHOSDUDPHWHUVZKLFKWDNHVDUDSLGFKDQJHRIWRUTXHYH
ORFLW\UHODWLRQVKLSRIVSRROHUDWWKHPRPHQWRIUHYHUVHURWDWLRQLQWRDFFRXQWZHUH
ILJXUHGRXW%\XVLQJWKHLPSOHPHQWHGFRQWUROVFKHPHZLWKLGHQWLILHGPRGHOSD
UDPHWHUV FDSDELOLW\ RI WKH GHYHORSHG HDUWKTXDNH VLPXODWRU IRU UHSURGXFLQJ DQ
DFWXDOO\REVHUYHGW\SLFDOORQJSHULRGJURXQGPRWLRQZDVGHPRQVWUDWHGE\PR
WLRQFRQWUROH[SHULPHQWV
.H\ZRUGV&DEOH'ULYHQ3DUDOOHO0HFKDQLVP(DUWKTXDNH6LPXODWRU6HLVPLF:DYH
5HSURGXFWLRQ&RPSXWHG7RUTXH0HWKRG)HHGIRUZDUG7RUTXH&RQWURO
,QWURGXFWLRQ
(DUWKTXDNHVNHHSRFFXUULQJJOREDOO\DQGXQH[SHFWHGO\,Q-DSDQIRUH[DPSOHWKHUHDUH
WKUHHVLJQLILFDQWHYHQWVZLWKLQWKHODVWWHQ\HDUV7RKRNXHDUWKTXDNH PDJQLWXGH
0Z .XPDPRWRHDUWKTXDNH 0Z DQG+RNNDLGR(DVWHUQ,EXULHDUWK
TXDNH 0Z $OWKRXJKWKHRFFXUUHQFHRIHDUWKTXDNHVFDQQRWEHSUHYHQWHGWKHUHLV
D SRVVLELOLW\ IRU SHRSOH DW WKH VLWH IRU WDNLQJ TXLFN DQG HIIHFWLYH UHDFWLRQ WR SUHYHQW
VHULRXVGDPDJHVZKHQWKH\KDYHKDGSURSHUSUDFWLFHGULOOVLQDGYDQFH(DUWKTXDNHVLP
XODWRUVDUHKHOSIXOIRUPDNLQJLWSRVVLEOHWRUHSURGXFHUHDOVHLVPLFZDYHVREVHUYHGLQ
WKHSDVWHYHQWVRUHYHQWKHLUDUUDQJHGYHUVLRQXQGHUFRQVLGHULQJWKHHIIHFWRIVHLVPLF

416 D. Matsuura et al.
SURSDJDWLRQLQWKHVRLODQGVWUXFWXUDOUHVSRQVHVRIEXLOGLQJVDQGVRRQLQVDIHHQYLURQ
PHQW:KHQVXFKVLPXODWRUVFDQEHRSHUDWHGLQPDQ\SODFHVWLPHWRWLPHSHRSOHFDQ
KDYHFKDQFHWRJHWXVHWRUHDOHDUWKTXDNHVDQGWRLPDJLQHKRZWKH\VKRXOGUHDFWZKHQ
DQDFWXDOHYHQWKDSSHQV)RUWKLVSXUSRVHHDUWKTXDNHVLPXODWRUYHKLFOHVFDOOHG.LVKLQ
VKD >@ KDYH EHHQ ZLGHO\ XVHG 7KHVH YHKLFOHV DUH JRRG IRU WUDQVSRUWDWLRQ EXW KDYH
OLPLWHG FDSDELOLWLHV LQ UHSURGXFLQJ VHLVPLF ZDYHV GXH WR WKH GHVLJQ RI WKHLU VKDNLQJ
WDEOHVPRXQWHGRQFRPPHUFLDOWUXFNV
5HJDUGLQJWKHFKDUDFWHULVWLFVRIVHLVPLFZDYHVWREHUHSURGXFHGLPSRUWDQFHRIORQJ
SHULRGJURXQGPRWLRQLVUHFHQWO\KLJKOLJKWHG>@7KLVNLQGRIVHLVPLFZDYHVRFFXUZLWK
YHU\ODUJHHDUWKTXDNHVDQGLQFOXGHSHULRGLFPRWLRQLQRQHWRWHQVHFRQGVRUORQJHUWKDW
SURSDJDWHJURXQGVXUIDFHLQODUJHGLVWDQFHVDQGH[FLWHORZIUHTXHQF\VWUXFWXUDOYLEUD
WLRQVRQPXOWLVWRU\EXLOGLQJVWKDWFDXVHVHULRXVGDPDJHVWRSHRSOHRUREMHFWVLQVLGHWKH
EXLOGLQJV,QRUGHUWRUHSURGXFHWKLVNLQGRIVHLVPLFZDYHKLJKDFFHOHUDWLRQDWWKHEH
JLQQLQJ 3ZDYH DQG KLJK YHORFLW\ DQG ODUJH GLVSODFHPHQW LQ WKH IROORZLQJ 6ZDYH
VKRXOGEHJHQHUDWHG$PDVVLYHVLPXODWRUFDOOHG('HIHQFH>@KDVEHHQFRQVWUXFWHG
IRU UHSURGXFLQJ YDULRXV VHLVPLF ZDYHV ZLWK KLJK SUHFLVLRQ RULJLQDOO\ WR UHYHDO WKH
PHFKDQLVPRIEXLOGLQJV¶GDPDJHVDIWHUWKH*UHDW+DQVKLQHDUWKTXDNH DQGKDV
EHHQUHSDLUHGIRUEHLQJFDSDEOHRIUHSURGXFLQJORQJSHULRGJURXQGPRWLRQDIWHU
7RKRNXHDUWKTXDNH7KLVVLPXODWRUKDVDPîPPRYLQJSODWIRUPZKLFKLVGULYHQ
E\ K\GUDXOLF DFWXDWRUV DQG WKHUHIRUH ODFN LQ WHUP RI WUDQVSRUWDELOLW\ -LVKLQ7KH
9XWRQ>@LVDQRWKHUGLIIHUHQWNLQGVLPXODWRU,WLVDVLQJOHVHDWHGYHKLFOHZLWK98721
FUDZOHUVXQGHUQHDWKGHYHORSHGE\+LURVH>@WRUHSURGXFHVHLVPLFZDYHVLQGLPHQ
VLRQVZLWKLQODUJHWUDYHOLQJGLVWDQFHDQGLVH[FHOOHQWLQWHUPRIWUDQVSRUWDELOLW\$VD
GUDZEDFNLWLVSURQHWRWLSSLQJEHFDXVHRIDVPDOOFRQWDFWDUHDRIWKHGULYLQJV\VWHP
DJDLQVWWKHXVHU¶VERG\VLWWLQJGRZQRQLW
,QVKRUWHDFKNLQGRIHDUWKTXDNHVLPXODWRULVDGYDQWDJHRXVLQWHUPRIVRPHRIWKHWKUHH
UHTXLUHGFKDUDFWHULVWLFV WUDQVSRUWDELOLW\KLJKDFFHOHUDWLRQDQGODUJHWUDYHOLQJGLVWDQFH
WREHDGHVLUDEOHGLVDVWHUSUHYHQWLRQHGXFDWLRQIDFLOLW\DJDLQVWORQJSHULRGJURXQGPR
WLRQHDUWKTXDNHVEXWLVQRWVDWLVI\LQJDOORIWKHPVLPXOWDQHRXVO\)RUIXOILOOLQJWKHWKUHH
UHTXLUHPHQWV DQ HDUWKTXDNH VLPXODWRU EDVHG RQ FDEOHGULYHQ SDUDOOHO PHFKDQLVP
&'30 VKRZQLQ)LJDQG)LJKDVEHHQFRQVWUXFWHG>@%\XVLQJFDEOHVODUJH
ZRUNVSDFHDQGORZLQHUWLDRIDPRYLQJSODWIRUPWKDWPDNHWKHVLPXODWRUEHLQJFDSDEOH
RISURGXFLQJYDULRXVW\SHVRIUHDOLVWLFVHLVPLFZDYHVLQFOXGLQJKLJKDFFHOHUDWLRQVFDQ
EHDFKLHYHG,QDGGLWLRQLWRIIHUVHDVHRIDVVHPEO\DQGGLVDVVHPEO\DQGKLJKWUDQVSRUW
DELOLW\,QSUHYLRXVZRUNVLQYHVWLJDWLRQRIFDEOHDUUDQJHPHQWWRHVWDEOLVKILQHVWDELOLW\
DQGGHVLJQRIZKROHVHWXSZHUHFRQGXFWHGDQGPRWLRQFRQWURORIDIDEULFDWHGSURWRW\SH
VKRZQ LQ )LJ D KDV EHHQ FDUULHG RXW +RZHYHU DELOLW\ RI WKH GULYLQJ V\VWHP IRU
DFKLHYLQJKLJKDFFHOHUDWLRQKDGDSUREOHPGXHWRDQDV\PPHWULFDQGIHHGEDFNEDVHG
GULYLQJVFKHPHDQGUHVXOWLQJORZDFFHSWDEOHFDEOHWHQVLRQV
,Q FDVH RI &'30 DFKLHYHPHQW RI VHLVPLF ZDYH LQFOXGLQJ KLJK DFFHOHUDWLRQ ZKLOH
GULYLQJDKHDY\PRYLQJPDVVLVVLJQLILFDQWO\GLIILFXOWVLQFHVRPHFDEOHVLQDQDQWDJR
QLVWLFGULYLQJV\VWHPHDVLO\KDYHQHJDWLYHWHQVLRQVHVSHFLDOO\DWWKHPRPHQWRIUHYHUVH
WXUQLQVXFKDFDVH7RRYHUFRPHWKLVSUREOHPWDUJHWWHQVLRQ WRUTXH WUDMHFWRU\RIHDFK
DFWXDWRUZLWKVDIHW\SUHORDGVKRXOGEHSUHFDOFXODWHGDQGIHGDVIHHGIRUZDUGLQSXWWRD
FRQWURO VFKHPH 5HJDUGLQJ WKH PRWLRQ FRQWURO RI &'30V D IHHGIRUZDUG FDEOH
Reproduction of Long-Period Ground Motion by Cable Driven Earthquake… 417
0RYLQJSODWIRUP
f 3$ f
3 3
r
0RWRU 0RWRU
3% r mg
f f
h z y
J 0RWRU
0RWRU R x
X xyJ
3 Z
Y

3
h DLUEHDULQJV
2 X
6WDWLFIUDPH
)LJ$VFKHPDWLFLOOXVWUDWLRQRIWKHHDUWKTXDNHVLPXODWRUEDVHVRQ&'30

WHQVLRQFRQWUROWHFKQLTXHFDOOHG&RPSXWHG7RUTXH0HWKRG>@LVZLGHO\XVHG:LWK
WKLVPHWKRG:LOOLDPVHWDWSHUIRUPHGDG\QDPLFVVLPXODWLRQRISODQDU&'30VKDYLQJ
WKUHHDQGIRXUVSRROHUVDQGSURSRVHGWKDWORRVHQLQJRIZLUHFDQEHDYRLGHG>@$IODNL
\DQHWDO>@SHUIRUPHGFRPSXWHUVLPXODWLRQRIPRWLRQFRQWURORIVXVSHQGHGW\SHVSD
WLDOPDQLSXODWRUVZLWKWKUHHVSRROHUVDQGSURSRVHGWKHFDSDELOLW\IRUSUHFLVHWUDMHFWRU\
WUDFNLQJ$OSHWDO>@LPSOHPHQWHGWKHPHWKRGLQWKHLUFRQWUROOHURIVXVSHQGHGPD
QLSXODWRUKDYLQJVDPHVWUXFWXUHZLWK$IODNL\DQ¶VPRGHOIRUKRXVHNHHSLQJRSHUDWLRQ
DVVLVWV\VWHPDQGVXFFHVVIXOO\GHPRQVWUDWHGPRWLRQFRQWUROH[SHULPHQWV)URPWKHGLV
FXVVLRQVLQDERYHSULRUOLWHUDWXUHVLWFDQEHVDLGWKDW&RPSXWHG7RUTXH0HWKRGVXS
SRVHGWREHHIIHFWLYHWRDFKLHYHILQHPRWLRQWUDFNLQJRIWKHFDEOHGULYHQHDUWKTXDNH
VLPXODWRUEXWWKHSRVVLELOLW\IRUWUDFNLQJKLJKIUHTXHQF\WDUJHWWRUTXHWUDMHFWRU\KDV
QRWEHHQUHYHDOHG7KLVSDSHUWKXVGLVFXVVHVDQLPSOHPHQWDWLRQRIDFRQWUROVFKHPHRI
WKHFDEOHGULYHQHDUWKTXDNHVLPXODWRUHPSOR\LQJ&RPSXWHG7RUTXH0HWKRGWRDFKLHYH
ILQHWUDFNDELOLW\IRUDFFHOHUDWLRQ,QRUGHUWRFDOFXODWHWKHIHHGIRUZDUGWRUTXHRIHDFK
FDEOHVSRROHUE\XVLQJDQLQYHUVHG\QDPLFPRGHOH[SHULPHQWDOSDUDPHWHULGHQWLILFD
WLRQVFKHPHZKLFKWDNHVDUDSLGFKDQJHRIWRUTXHYHORFLW\UHODWLRQVKLSRIVSRROHUDWWKH
PRPHQWRIUHYHUVHURWDWLRQLQWRDFFRXQWLVWKHQILJXUHGRXWDQGFDUULHGRXWZLWKWKH
GHYHORSHGHDUWKTXDNHVLPXODWRU%\XVLQJWKHHVWDEOLVKHGFRQWUROVFKHPHZLWKWKHH[
SHULPHQWDOO\ LGHQWLILHG PRGHO SDUDPHWHUV FDSDELOLW\ RI WKH HDUWKTXDNH VLPXODWRU IRU
UHSURGXFLQJDW\SLFDOORQJSHULRGJURXQGPRWLRQLVILQDOO\SUHVHQWHG
&RPSRVLWLRQRIWKHHDUWKTXDNHVLPXODWRU
7KHHDUWKTXDNHVLPXODWRULVFRPSRVHGRID&'30ZLWKIRXUFDEOHVFRQQHFWLQJWKHPRY
LQJSODWIRUPRIDSSUR[LPDWHO\NJLQFOXGLQJDNJGXPP\ORDGDQGDVXUURXQGLQJ
VWDWLFIUDPHRIP[PZLGHWRHVWDEOLVKDSODQDUPRWLRQFRQWURORQDKRUL]RQWDO
SODQHDQGIRXUDLUEHDULQJVWRVXSSRUWWKHSODWIRUPZLWKORZIULFWLRQ(DFKFDEOHLVGULYHQ
E\DVSRROPHFKDQLVPVKRZQLQ)LJ E 7KLVPHFKDQLVPGULYHVWKHFDEOHE\DVSLUDO
SXOOH\DWWDFKHGRQD:$&VHUYRPRWRUZLWKSODQHWDU\UHGXFHU
&DEOH
VSRRO

D 2XWORRNRIHQWLUHVHWXS E &ORVHXSRIDVSRROPHFKDQLVP
)LJ)DEULFDWHGSURWRW\SH

,QDGGLWLRQDQRWKHUJXLGHSXOOH\PRXQWHGRQDEDOOVFUHZKDYLQJDVDPHSLWFKZLWKWKH
VSLUDOSXOOH\LVSODFHGLQSDUDOOHODQGGULYHQV\QFKURQRXVO\ZLWKWKHVSLUDOSXOOH\WRDYRLG
LUUHJXODUZLQGLQJ$FTXLVLWLRQRIDQJXODUGLVSODFHPHQWDQGGULYLQJFXUUHQWRIIRXUHOHFWULF
PRWRUVYLD$'FRQYHUWHUDQGSXOVHFRXQWHUDQGUHIUHVKPHQWRIFRPPDQGVLJQDOIURP
'$FRQYHUWHUDUHGRQHLQPVLQWHUYDOUHJXODWHGE\DKDUGZDUHWLPHULQWHUUXSWSURJUDP
PLQJUXQQLQJRQD06:LQGRZVRSHUDWLQJV\VWHP
0RWLRQFRQWUROEDVHGRQWKH&RPSXWHG7RUTXH0HWKRG
,QYHUVHG\QDPLFV
,QWKHDQDO\VLVVFKHPHWKHHDUWKTXDNHVLPXODWRULVFRQVLGHUHGDVDSODQDUPHFKDQLVP
WKDWIRXUFDEOHVEHWZHHQ3i DQG3$DQG%DUHFRQQHFWLQJWKHVWDWLFIUDPHDQGWKH
PRYLQJSODWIRUP6LQFHIRXUDLUEHDULQJVXQGHUWKHPRYLQJSODWIRUPDFKLHYHVYHU\ORZ
IULFWLRQEHWZHHQWKHSODWIRUPDQGWKHJURXQGWKHIULFWLRQZDVQHJOHFWHGDQGRQO\WKH
LQHUWLDOIRUFHDQGPRPHQWRIWKHSODWIRUPDUHWDNHQLQWRDFFRXQW7KHPDVVDQGHODVWLF
LW\RIWKHIRXUZLUHVDUHDOVRLJQRUHGVLQFHWKH\DUHPDGHRIDFRPSRVLWHILEHUFDOOHG
'\QHHPDZKLFKDFKLHYHVERWKOLJKWZHLJKWDQGKLJKVWUHQJWK'XHWRDERYHDVVXPS
WLRQV3DQG3$%DUHDVVXPHGWREHFRQQHFWHGE\VWUDLJKWOLQHVORFDWHGRQDFRP
PRQKRUL]RQWDOSODQH,QDGGLWLRQWRWKHDERYHWKHUHLVDUHPDLQLQJIDFWRUWRDFKLHYHD
ILQHWUDFNLQJFDSDELOLW\DJDLQVWKLJKGHVLUHGDFFHOHUDWLRQV6LQFHWKHVSRROPHFKDQLVP
VKRZQLQ)LJ E LVFRPSRVHGRIPDQ\WUDQVPLVVLRQHOHPHQWVLQDGGLWLRQWRDPRWRU
ZLWKDUHGXFWLRQJHDULWVG\QDPLFFKDUDFWHULVWLFVVKRXOGEHWDNHQLQWRDFFRXQW,QWKLV
FKDSWHUDQLQYHUVHG\QDPLFVDQDO\VLVZLOOEHSHUIRUPHGWRUHYHDOLW
$VVKRZQLQ)LJDVWDWLFFRRUGLQDWHIUDPHIL[HGRQWKHJURXQGȭଵ DQGDPRYLQJ
FRRUGLQDWH IUDPH IL[HG RQ WKH PRYLQJ SODWIRUPȭଶ DUH GHILQHG 7KH RULJLQ RIȭଶ LV
FRLQFLGHZLWKWKHFHQWHURIPDVVRIWKHSODWIRUP3RVLWLRQDQGRULHQWDWLRQRIWKHSODW
IRUPDUHGHILQHGDVࢄ ൌ ሾ‫ߛ ݕ ݔ‬ሿ୘ DQGLWVPDVVDQGPRPHQWRILQHUWLDDURXQGYHU
WLFDO z D[LVDUHDOVRGHILQHGDVmDQGIzUHVSHFWLYHO\5HJDUGLQJHDFKiWK i
୘
FDEOHDQXQLWYHFWRU࢛௜ ൌ ൣ‫ݑ‬௜௫ ǡ ‫ݑ‬௜௬ ǡ ‫ݑ‬௜௭ ൧ GHQRWLQJLWVGLUHFWLRQDPSOLWXGHRIWHQVLRQ
fiDQGFRQQHFWLQJSRLQWRQWKHSODWIRUPriZKLFKLVGHILQHGLQȭଶ FRRUGLQDWHIUDPHDUH
UHVSHFWLYHO\GHILQHG$URWDWLRQPDWUL[RGHVFULEHVWKHFRRUGLQDWHWUDQVIRUPDWLRQIURP
ȭଶ WRȭଵ %\XVLQJWKHVHYDULDEOHVDQGPDWUL[WKHHTXDWLRQRIPRWLRQRIWKHPRYLQJ
SODWIRUPLVREWDLQHGDVIROORZLQJ
ࡹࢄሷ ൌ ‫ࡲۯ‬ǡ
ۗ
‫ݑ‬ ‫ݑ‬ଵ௬ ሼሺࡾ࢘ଵ ሻ ൈ ࢛ଵ ሽఊ ୘ ۖ
݉ Ͳ Ͳ ‫ ۍ‬ଵ௫ ‫ۖ ې‬
‫ݑ‬
‫ ێ‬ଶ௫ ‫ݑ‬ ଶ௬ ሼሺࡾ࢘ଶ ሻ ൈ ࢛ଶ ሽఊ ‫ۑ‬
ࡹ ൌ ൥ Ͳ ݉ Ͳ ൩ 㸪‫ ۯ‬ൌ ‫ێ‬ ǡ
‫ݑ‬ ଷ௫ ‫ݑ‬ ଷ௬ ሼሺࡾ࢘ଷ ሻ ൈ ࢛ଷ ሽఊ ‫ۘ ۑ‬
Ͳ Ͳ ‫ܫ‬௭ ‫ێ‬ ‫ۑ‬
‫ݑۏ‬ସ௫ ‫ݑ‬ସ௬ ሼሺࡾ࢘ସ ሻ ൈ ࢛ସ ሽఊ ‫ۖ ے‬ ۖ
ࢄ ൌ ሾ‫ߛ ݕ ݔ‬ሿ் 㸪ࡲ ൌ ሾ݂ଵ ݂ଶ ݂ଷ ݂ସ ሿ୘ Ǥ ۙ
,QWKHDERYHHTXDWLRQQRWDWLRQ³ሼࢻ ൈ ࢼሽஓ ´GHQRWHVDQH[WUDFWLRQRIWKHWKLUGURZRIWKH
FURVVSURGXFW+HUHOHWXVWKLQNVROYLQJ(TDERXWWKHFDEOHWHQVLRQ3RWWHWDOSUR
SRVHGDFORVHGIRUPVROXWLRQRIIRUFHGLVWULEXWLRQIRUUHDOWLPHIRUFHFRQWURORI'R)
&'30>@,QWKLVVROXWLRQ DFTXLVLWLRQRISRVLWLYHFDEOHWHQVLRQVLVJXDUDQWHHG,Q
DGGLWLRQWKHVROXWLRQLVFRQWLQXRXVZLWKUHVSHFWWRDPRYLQJWUDMHFWRU\DQGWKHUHIRUH
GLIIHUHQWLDEOH)RUDSSO\LQJWKLVVROXWLRQWRRXU'R)&'30(TLVUHZULWWHQDV
ࡲ ൌ ‫ܣ‬ା ‫ࢄܯ‬ሷ ൅ ሺ‫ ܫ‬െ ‫ܣ‬ା ‫ܣ‬ሻࡲ୫ ǡ ࡲ୫ ൌ ሾ݂୫ଵ ݂୫ଶ ݂୫ଷ ݂୫ସ ሿ୘ ǡ
ZKHUHVXSHUVFULSWLRQ³´GHQRWHVDSVHXGRLQYHUVHRIDPDWUL[7KLVHTXDWLRQOHDGVWR
SRVLWLYHDQGFRQWLQXXPFDEOHWHQVLRQVZKLOHWKHSODWIRUPDFKLHYHVDGHVLUHGPRWLRQ
ZKHQFPDYHFWRURIPHDQWHQVLRQVRIWKHIRXUFDEOHVKDVVXIILFLHQWO\ODUJHYDOXH
1H[WUHODWLRQVKLSEHWZHHQFDEOHWHQVLRQFDQGPRWRUWRUTXHĲLVFDOFXODWHG,QWKHVSRRO
PHFKDQLVPDPRWRUZLWKDUHGXFHUGULYHVDVSLUDOSXOOH\,QDGGLWLRQDEDOOVFUHZKDY
LQJDQH[DFWO\VDPHSLWFKWRWKHVSLUDOSXOOH\LVGULYHQE\DWLPLQJEHOWWRGULYHDPRYLQJ
JXLGHSXOOH\V\QFKURQRXVO\$OWKRXJKWKH&'30LVSODQDUWKHJUDYLW\DIIHFWLQJRQWKH
VSRROPHFKDQLVPFDQQRWEHLJQRUHGVLQFHWKHD[LVRIWKHEDOOVFUHZLVLQSDUDOOHOZLWK
WKH JUDYLWDWLRQDO D[LV :KHQ D YLVFRXV IULFWLRQ FRXORPE IULFWLRQ DQG WKH JUDYLW\ DUH
DIIHFWLQJWRWKHGULYLQJV\VWHPHVVHQWLDOPRWRU¶VWRUTXHWRDFKLHYHWDUJHWPRWLRQLV
࣎୧୬ ൌ ‫ܬ‬୫ ࣂሷ୫ ൅ ‫ࣂܦ‬ሶ୫ ൅ ࢌୡ ൫ࣂሶ୫ ൯ ൅ ࢀ௚ ൅ ߦ‫ݎ‬ୱ ࡲ
ZKHUHJPDfFTgșPȟDQGrVDUHPRPHQWRILQHUWLDYLVFRXVGDPSLQJFRHIILFLHQW
FRXORPEIULFWLRQJUDYLW\FRPSHQVDWLRQWHUPPRWRU¶VDQJXODUGLVSODFHPHQWUHGXFWLRQ
UDWLRDQGUDGLXVRIWKHVSLUDOSXOOH\UHVSHFWLYHO\,QDUHJLRQWKDWFDQVDWLVI\WKH9HFWRU
&ORVXUHVHFRQGGHULYDWLYHRIWKHVWDWHYDULDEOHVRIWKHV\VWHPܺሷFDQEHGHVFULEHGE\
XVLQJD-DFRELDQPDWUL[JDQGPRWRU¶VYHORFLW\DQGDFFHOHUDWLRQDV
ࢄሶ ൌ ࡭ା୘ ࢒ሶ ൌ ‫ࣂܬ‬ሶ୫ ǡ ࢄሷ ൌ ‫ࣂܬ‬ሷ୫ ൅ ‫ࣂܬ‬
ሶ ሶ ୫ ǡ
‫ ܬ‬ൌ ߦ‫ݎ‬௦ ‫ܣ‬ା୘ ǡ࢒ ൌ ሾ݈ଵ ݈ଶ ݈ଷ ݈ସ ሿࢀ ǡቑ
ࣂ୫ ൌ ሾߠଵ ߠଶ ߠଷ ߠସ ሿࢀ Ǥ
%\XVLQJ(TV DVLPXOWDQHRXVOLQHDUHTXDWLRQFDQEHREWDLQHGDV
࣎୧୬ ൌ ሺ‫ܬ‬୫ ൅ ߦ‫ݎ‬௦ ‫ܣ‬ା ‫ܬܯ‬ሻࣂሷ௠ ൅ ൫‫ ܦ‬൅ ߦ‫ݎ‬ୱ ‫ܣ‬ା ‫ܬܯ‬൯ሶ ࣂሶ୫

൅ࢌୡ ൫ࣂሶ୫ ൯ ൅ ࢀ௚ ൅ ߦ‫ݎ‬௦ ሺ‫ ܫ‬െ ‫ܣ‬ା ‫ܣ‬ሻࡲ୫ Ǥ
7KH&RPSXWHG7RUTXH0HWKRG
%DVHG RQ WKH DERYH LQYHUVH G\QDPLFV DQDO\VLV D FRQWURO VFKHPH WR REWDLQ UHTXLUHG
RXWSXWWRUTXHRIPRWRUVIRUWUDFNLQJWKHPRYLQJSODWIRUPRQWRGHVLUHGWUDMHFWRU\LVGH
VFULEHG$EORFNGLDJUDPRIWKHFRQWUROVFKHPHLVVKRZQLQ)LJ(TXDWLRQGHQRWHV
WKH UHODWLRQVKLS EHWZHHQ D PRWRU¶VRXWSXW GLVSODFHPHQW DQG WRUTXH DQG WKXV FDQEH
FRQVLGHUHG DV D IXQGDPHQWDO HTXDWLRQ WR FDOFXODWH WKH UHTXLUHG PRWRU¶V WRUTXH IURP
JLYHQșPZKLFKFDQEHXVHGDVDIHHGIRUZDUGLQSXWRIWKH&RPSXWHG7RUTXH0HWKRG
)LUVWGHVLUHGWUDMHFWRU\RIWKHSODWIRUP࢞ୢ ǡ ࢞ሶ ୢ ǡ ࢞ሷ ୢ ǡDUHFRQYHUWHGWRFRUUHVSRQGLQJPR
WRU¶VDQJXODUSDUDPHWHUVࣂ୫ ǡ ࣂሶ୫ ǡ ࣂሷ୫ 6LQFHPRWRU¶VFXUUHQWDQJXODUGLVSODFHPHQWࣂ
DQGYHORFLW\ࣂሶFDQEHREWDLQHGYLDVHQVRU\LQIRUPDWLRQFRPPDQGDQJOHࣂୡ YHORFLW\
ࣂሶୡ DFFHOHUDWLRQࣂሷୡ DUHFDOFXODWHGE\WKHIROORZLQJHTXDWLRQV
ࣂ௖ ൌ ࣂ
ࣂሶ௖ ൌ ࣂሶ ቑ
ࣂሷ௖ ൌ ࣂሷௗ ൅ ࡷ୴ ൫ࣂሶௗ െ ࣂሶ൯ ൅ ࡷ୮ ሺࣂௗ െ ࣂሻ
ZKHUH KY DQG KS DUH IHHGEDFN JDLQV %\ VXEVWLWXWLQJ WKHVH FRPPDQG SDUDPHWHUV
ࣂୡ ǡ ࣂሶୡ ǡ ࣂሷୡ LQWR(TLQVWHDGRIࣂ୫ ǡ ࣂሶ୫ ǡ ࣂሷ୫ FRPPDQGWRUTXHIRUWKHPRWRUĲLQFDQ
EHREWDLQHG
([SHULPHQWDOLGHQWLILFDWLRQRIPRGHOSDUDPHWHUV
0RGHOSDUDPHWHUVLQ(TZKLFKDUHUHTXLUHGWRFDUU\LQJRXWWKHFRQWUROVFKHPHYLV
FRXVGDPSLQJDPRPHQWRILQHUWLDOJPFRXORPEIULFWLRQfFDQGJUDYLWDWLRQDOORDGTg
DUHH[SHULPHQWDOO\LGHQWLILHG:KHQDVSRROHULVGULYHQZLWKRXWWKHFDEOHDUHODWLRQVKLS
EHWZHHQWKHPRWRU¶VRXWSXWWRUTXHDQGUHVXOWLQJPRWLRQFDQEHZULWWHQDV
߬୫୭୲ ൌ ‫ܬ‬୫ ߠሷ୫ ൅ ‫ߠܦ‬ሶ୫ ൅ ݂ୡ ൫ߠሶ୫ ൯ ൅ ܶ௚
:KHQWKHPRWRU¶VWRUTXHLVFRQVWDQWDWHUPLQDOYHORFLW\DIWHUDWUDQVLHQWUHVSRQVHKDV
EHHQFRQYHUJHGZLOOEHFRPHFRQVWDQWZKLFKLVGHWHUPLQHGE\WKHYLVFRXVIULFWLRQ$
SORWRIDFWXDOPHDVXUHPHQWGDWDLVVKRZQLQ)LJ D ,QWKHILJXUHILWWHGOLQHVRQWRWKH
DFTXLUHGGDWDLQQHJDWLYHDQGSRVLWLYHDQJXODUYHORFLW\UHJLRQVDUHVXSHUSRVHG6LQFH
WKHUHLVDUDSLGFKDQJHRIWKHPRWRUWRUTXHZKHQDQJXODUYHORFLW\FKDQJHVIURPSRVLWLYH
WRQHJDWLYHRUYLFHYHUVDWKHDYHUDJHGLQFOLQDWLRQRIWKHVHWZROLQHVDQGWZRFURVVLQJ
SRLQWVZLWKDYHUWLFDOOLQHDWșP DDQGfFZHUHLGHQWLILHG0RPHQWRILQHUWLDJPDQG
JUDYLWDWLRQDOWHUPTgFDQEHLGHQWLILHGE\DPHWKRGSURSRVHGE\7VXUXWDHWDO>@,Q
WKLVPHWKRGDPRWRULVGULYHQE\YHORFLW\FRQWUROZKLOHWKHWDUJHWYHORFLW\LVFKDQJLQJ
DVVKRZQLQ)LJ E $IWHUUHFRUGLQJDPHDVXUHPHQWVLJQDORIPRWRUWRUTXHLQWHJUDWHG
WRUTXHLQWKHH[SHULPHQWDQGWKHRUHWLFDOPRGHODUHREWDLQHGDQGLGHQWLILFDWLRQRIPR
PHQWRILQHUWLDFDQEHGRQHDV‫ܬ‬୫ ൌ ൫‫߬ ׬‬ୣ୶୮ Ȁ ‫߬ ׬‬୫୭ୢୣ୪ ൯‫ܬ‬୫ ԢZKHUH‫ܬ‬୫ ԢLVDQLQLWLDOJXHVV
RILQHUWLDIRUWKHG\QDPLFPRGHO7KHLQWHJUDWLRQRIWRUTXHZDVGRQHZLWKLQWZRVHF
WLRQVLQEHWZHHQaDQGbDQGaDQGb7KHVHVHFWLRQVDUHFKRVHQVRWKDW DQJXODU
GLVSODFHPHQWVZKHQt aDQGbDUHHTXDO GXUDWLRQRISRVLWLYHDQGQHJDWLYH
YHORFLW\ UHJLRQ DUH HTXDO DQG YHORFLW\ SURILOH LQ ab DQG ab DUH V\PPHWULF
7KDQNWRWKHVHFRQGLWLRQVGLVWXUEDQFHRIGDPSLQJFRXORPEIULFWLRQDQGFRQVWDQWH[
WHUQDOORDGFDQEHFDQFHOOHGRXWWKURXJKWKHLQWHJUDWLRQ$FTXLUHGYDOXHVIURPWKHH[
SHULPHQWDOLGHQWLILFDWLRQDUHVKRZQLQ7DEOH7KHYDOXHRIJPZKLFKLVWKHILQDOO\
REWDLQHGRQHLQWKHLGHQWLILFDWLRQZDV ή ଶ DJDLQVW ή ଶ FDWDORJDQG
&$'EDVHGYDOXH6LQFHWKLVNLQGRIH[SHULPHQWDOUHVXOWWHQGWREHELJJHUWKDQVLPXOD
WLRQUHVXOWWKHREWDLQHGUHVXOWFDQEHUHOLDEOH

xG șG șF

șF ĲLQ j
șG xt
xG
Kv șF
șP s șP
xG șG
Kp

)LJ$EORFNGLDJUDPRIWKHFRQWUROVFKHPHEDVHGRQWKH&RPSXWHG7RUTXH0HWKRG

PP@

7DUJHWYHORFLW\ V>UDGV@

>1䞉PP@
0RWRUWRUTXH
P

7DUJHWYHORFLW\
UHI
PRW
0RWRUWRUTXH

P

$QJXODUYHORFLW\ >UDGV@ 7LPH>V@

D &RQVWDQWWRUTXHGULYLQJ E ,WHUDWLYHGULYLQJZLWKYHORFLW\FRQWURO
)LJ$FTXLUHGGDWDLQWKHSDUDPHWHULGHQWLILFDWLRQH[SHULPHQW

7DEOH5HVXOWRIWKHH[SHULPHQWDOLGHQWLILFDWLRQRIPRGHOSDUDPHWHUV
Viscos damping ‫ܦ‬ ሾ㺃ȀሺȀሻሿ ͲǤ͸Ͷ
Coulomb friction ݂௖ ሾ㺃ሿ ͳͶͷ
Gravitational term ܶ௚ ሾ㺃ሿ 11
Moment of inertia ‫ܬ‬௠ ሾ ή ଶ ሿ 410

0RWLRQFRQWUROH[SHULPHQW
(YDOXDWLRQRIWKHWZRGLIIHUHQWFRQWUROVFKHPHVZLWKVLQXVRLGDOLQSXW
)LJXUHSORWVDPHDVXUHGGLVSODFHPHQWRIWKHPRYLQJSODWIRUP E\237275$.
1RUWKHUQ'LJLWDO,QFGHWDLOHGVSHFLILFDWLRQLVVKRZQLQ7DEOH ZKHQWKHV\VWHPLV
GULYHQE\WZRGLIIHUHQWFRQWUROVFKHPHVRXUSUHYLRXVRQHH[SODLQHGLQ>@DQGQHZO\
SURSRVHGRQHLQ)LJ)RUWKHQHZVFKHPHWKHPHDQWHQVLRQ HDFKHOHPHQWRIFP
ZDV VHW WR 1 DQG IHHGEDFN JDLQV IRU SRVLWLRQ FRQWURO ZHUH VHW WR KS DQG
KY UHVSHFWLYHO\,QERWKFDVHGHVLUHGWUDMHFWRU\LVJLYHQDV
‫ݔ‬ ʹͲͲ ሺʹߨ ή ͲǤͷ‫ݐ‬ሻ ሺʹߨ ή ͲǤͲͷ‫ݐ‬ሻ
൥‫ݕ‬൩ ൌ ൥ Ͳ ൩
ߛ Ͳ
LQZKLFKWKHxGLUHFWLRQDOGLVSODFHPHQWZLWKLQേPPLVJLYHQDVDVLQXVRLGDOZDYH
RI+]ZKRVHDPSOLWXGHLVPRGXODWHGE\ORZHUIUHTXHQF\ +] HQYHORSH,Q)LJ
LW LV REYLRXV WKDW WKHRXWSXW WUDMHFWRU\ RI WKHQHZ FRQWURO VFKHPH DJUHHVZLWK WKH
GHVLUHGWUDMHFWRU\WKDQWKHSUHYLRXVRQHHVSHFLDOO\LQRXWRIPRYLQJD[LV y GLUHFWLRQ
&DEOHWHQVLRQVZLWKWKHQHZFRQWUROVFKHPHDUHDOVRSORWWHGLQ)LJ,WFDQEHVHHQ
WKDWDOOFDEOHWHQVLRQVDUHPDLQWDLQHGLQWKHSRVLWLYHUHJLRQ

'HVLUHG[ 7DEOH6SHFLILFDWLRQRIWKHRSWLFDOGLVSODFH
'LVSODFHPHQW[>PP@
0HDVXUHG[ 2OGFRQWUROOHU
0HDVXUHG[ 1HZFRQWUROOHU
PHQWPHDVXUHPHQWDSSDUDWXV

0RGHOQXPEHU 237275$.

0HDVXUHPHQW x

DFFXUDF\>PP@ yz

6DPSOLQJUDWH>+]@

7LPH>V@
D xD[LVGLVSODFHPHQW

'LVSODFHPHQW\>PP@
:LUH$
:LUH%
:LUH&
:LUH'

7LPH>V@
E yD[LVGLVSODFHPHQW
)LJ0HDVXUHGFDEOHWHQVLRQZLWKVLQXVRLGDO
)LJ0HDVXUHGRXWSXWGLVSODFHPHQWRIWKH LQSXWZKLOHPRWLRQFRQWUROLVXVLQJ
PRYLQJSODWIRUPZLWKVLQXVRLGDOLQSXW WKHQHZO\SURSRVHGFRQWUROVFKHPH

&DSDELOLW\RIWKHV\VWHPLQUHSURGXFLQJDFWXDOO\REVHUYHGVHLVPLFZDYH
,QRUGHUWRGHPRQVWUDWHWKHFDSDELOLW\RIWKHGHYHORSHGHDUWKTXDNHVLPXODWRULQUHSUR
GXFLQJDFWXDOO\REVHUYHGVHLVPLFZDYHVPRWLRQFRQWUROH[SHULPHQWZDVFDUULHGRXW
7KHGHVLUHGSRVLWLRQWUDMHFWRU\RIWKHPRYLQJSODWIRUPZDVDFTXLUHGE\QXPHULFDOLQ
WHJUDWLRQRIDQRULJLQDODFFHOHUDWLRQWUDMHFWRU\REVHUYHGDW2IXQDWRFLW\LQ,ZDWHSUH
IHFWXUHGXULQJWKH7RKRNXHDUWKTXDNHDQGVDPHFPKSDQGKYZLWKWKHSUHYLRXV
H[SHULPHQWZHUHXVHG0HDVXUHGRXWSXWWUDMHFWRU\RIWKHSODWIRUPSRVLWLRQDQGFDEOH
WHQVLRQVDUHSORWWHGLQ)LJVDQGUHVSHFWLYHO\,WFDQEHVDLGWKDWWKHGHVLUHGWUDMHF
WRU\ZDVILQHO\WUDFNHGHVSHFLDOO\LQWKHIRUPHUKDOISHULRGLQZKLFKKLJKDFFHOHUDWLRQ
RILQLWLDO3ZDYHLVLQFOXGHG,QWKHODWWHUKDOIVRPHWUDFNLQJHUURUEHJDQWRVKRZXS
DIWHUWKHPHDQFDEOHWHQVLRQVXGGHQO\GHFUHDVHGIURP1WR1DSSUR[LPDWHO\
7KHORVVRIFDEOHWHQVLRQLVVXSSRVHGWREHRFFXUUHGGXHWRDORRVHQHVVRIWKHFDEOHDQG
SXOOH\GXULQJWKHFDEOHLVLQLWLDOO\ZLQGHGXS7KLVORRVHQHVVLVVDYHGZKHQWKHWUDYHOLQJ
GLVWDQFHLVVPDOOGXULQJ3ZDYHWKHQVXGGHQO\UHOHDVHGZKHQWKHWUDYHOLQJGLVWDQFH
LQFUHDVHVDFFRUGLQJWRWKHEHJLQQLQJRI6ZDYH7KLVPXVWEHVROYHGE\LQFUHDVLQJWKH
WLJKWQHVVRIWKHFDEOHDQGSXOOH\EHIRUHVWDUWLQJWKHVHLVPLFZDYHUHSURGXFWLRQ
'LVSODFHPHQW[>PP@

ZLUH$
'LVSODFHPHQW\>PP@
ZLUH%
:LUHWHQVLRQ>1@
ZLUH&
ZLUH'

7LPH>V@
)LJ0HDVXUHGSODWIRUPSRVLWLRQ )LJ0HDVXUHGFDEOHWHQVLRQ
&RQFOXVLRQ
,QWKLVSDSHUDIHHGIRUZDUGEDVHGFRQWUROVFKHPHZDVLPSOHPHQWHGLQDFDEOHGULYHQ
HDUWKTXDNHVLPXODWRUWRDFKLHYHILQHWUDFNDELOLW\RIWDUJHWWUDMHFWRU\RIDPRYLQJSODW
IRUPLQFOXGLQJKLJKIUHTXHQF\DQGKLJKDFFHOHUDWLRQVKDNLQJWKDWDUHQHFHVVDU\IRUUH
SURGXFLQJDFWXDOO\PHDVXUHGVHLVPLFZDYHVRIORQJSHULRGJURXQGPRWLRQEXWGLIILFXOW
WRDFKLHYHVLQFHIUHTXHQWUHYHUVHURWDWLRQVRIPRWRUVDUHUHTXLUHGZKLOHPDLQWDLQLQJDOO
FDEOHV¶SRVLWLYHWHQVLRQV2EWDLQHGUHVXOWVDUHVXPPDUL]HGDVIROORZLQJ
'\QDPLF PRGHOV RI WKH HQWLUH FDEOH GULYHQ SDUDOOHO PHFKDQLVP DQG HDFK FDEOH
VSRROHUKDYHEHHQGHULYHG(VVHQWLDOPRGHOSDUDPHWHUVIRUWKHIHHGIRUZDUGWRUTXH
FDOFXODWLRQZHUHH[SHULPHQWDOO\LGHQWLILHGZKLOHFDQFHOOLQJWKHHIIHFWRIGLVWXUE
DQFHIDFWRUVE\XVLQJLQWHJUDWLRQRIPRWRUWRUTXHDQGWDNLQJWKHUDSLGFKDQJHRI
PRWRUWRUTXHDWWKHPRPHQWRIUHYHUVHWXUQ2QHWKRVHYDOXHVZKLFKZDVILQDOO\
REWDLQHGWKURXJKWKHSURFHVVPRPHQWRILQHUWLDZDVFORVHWRDFDWDORJDQG&$'
EDVHGYDOXH%DVHGRQWKHREWDLQHGUHVXOWDFRQWUROVFKHPHEDVHGRQ&RPSXWHG
7RUTXH0HWKRGKDVEHHQLPSOHPHQWHG
0RWLRQFRQWUROH[SHULPHQWVSHUIRUPHGGHVLUHGWUDMHFWRU\WUDFNLQJEDVHGRQDVL
QXVRLGDOZDYHDQGDFWXDOO\REVHUYHGVHLVPLFZDYH)URPWKHIRUPHUH[SHULPHQW
LPSURYHPHQWRIWUDFNDELOLW\DJDLQVWKLJKDFFHOHUDWLRQDQGODUJHWUDYHOLQJGLVWDQFH
ZDVFRQILUPHG)URPWKHODWWHUH[SHULPHQWFDSDELOLW\RIWKHGHYHORSHGVLPXODWRU
LQ UHSURGXFLQJ ORQJSHULRG VHLVPLF PRWLRQ ZDV GHPRQVWUDWHG 7KH VLPXODWRU
FRXOGWUDFNWKHGHVLUHGWUDMHFWRU\LQERWKLQLWLDO3ZDYHZLWKKLJKDFFHOHUDWLRQDQG
IROORZLQJ6ZDYHZLWKODUJHWUDYHOLQJGLVWDQFH
Acknowledgement.
7KLV ZRUN ZDV SDUWLDOO\ VXSSRUWHG E\ -.$ DQG LWV SURPRWLRQDO IXQGV IURP
$8725$&( 0 7KH DXWKRUV JUHDWO\ WKDQN 3URI 6KLJHR +LURVH &R
IRXQGHUDQG&(2RI+L%RWFRUS-DSDQ DQG3URI.HLVXNH$ULNDZD .DQDJDZD,QVW
RI7HFK-DSDQ IRUWKHLUYDOXDEOHDGYLFHVRQWKHGHVLJQRIFDEOHGULYHPHFKDQLVP
5HIHUHQFHV
<RVKLR0DVDNL 'XUDWLRQRI³IHHOLQJRIEHLQJVKDNHQ´DV$VVHVVHG8VLQJDQ(DUWK

TXDNH6LPXODWRU9HKLFOH(TXLOLEULXP5HVHDUFKYRO
+.RNHWVXDQG+0L\DNH $6HLVPRORJLFDO2YHUYLHZRI/RQJ3HULRG*URXQG0R
WLRQ-RXUQDORI6HLVPRORJ\9RO,VVXHSS
.2KWDQL12JDZD77DND\DPDDQG+6KLEDWD &RQVWUXFWLRQRI('()(16(
'IXOOVFDOHHDUWKTXDNHWHVWLQJIDFLOLW\ 7KLUWHHQWK:RUOG&RQIHUHQFHRQ(DUWKTXDNH(Q
JLQHHULQJ
5RK6H*RQ<7DJXFKL<1LVKLGD5<DPDJXFKL<)XNXGD6.XURGD0<RVKLGD
)XNXVKLPD()DQG6+LURVH 'HYHORSPHQWRIWKHSRUWDEOHJURXQGPRWLRQVLPX
ODWRURIDQHDUWKTXDNH,(((,QWHUQDWLRQDO&RQIHUHQFHRQ,QWHOOLJHQW5RERWVDQG6\VWHPV

6+LURVH6$PDQR 7KH98721+LJKSD\ORDGKLJKHIILFLHQF\KRORQRPLFRPQL
GLUHFWLRQDOYHKLFOH,Q3URFHHGLQJVRIWKH6L[WK6\PSRVLXPRQ5RERWLFV5HVHDUFK
'0DWVXXUD6,VKLGD0$NUDPLQ(%.oNWDEDN<6XJDKDUD67DQDND1)XNXZD
0<RVKLGD<7DNHGD &RQFHSWXDO'HVLJQRID&DEOH'ULYHQ3DUDOOHO0HFKDQLVP
IRU3ODQDU(DUWKTXDNH6LPXODWLRQ7KHVW&,60,)7R006\PSRVLXPRQ5RERW'HVLJQ
'\QDPLFVDQG&RQWURO 520$16< 6SULQJHUSS
5&3DXO0RGHOOLQJ7UDMHFWRU\&DOFXODWLRQDQG6HUYRLQJRID&RPSXWHU&RQWUROOHG$UP
7HFKQLFDO5HSRUW$,06WDQIRUG8QLYHUVLW\$UWLILFLDO,QWHOOLJHQFH/DERUDWRU\
5REHUW/:LOOLDPV,,3DROR*DOOLQD 7UDQVODWLRQDO3ODQDU&DEOH'LUHFW'ULYHQ5R
ERWV-RXUQDORI,QWHOOLJHQWDQG5RERWLF6\VWHPV9RO,VVXHSS±
$$IODNL\DQ+%D\DQL070DVRXOHK &RPSXWHGWRUTXHFRQWURORIDFDEOHVXV
SHQGHG SDUDOOHO URERW 3URF RI UG 56, ,QWHU &RQI RQ 5RERWLFV DQG 0HFKDWURQLFV
,&520 SS
$%$OS6.$JUDZDO &DEOH6XVSHQGHG5RERWV)HHGEDFN&RQWUROOHUVZLWK3RV
LWLYH,QSXWV3URFRIWKH$PHULFDQ&RQWURO&RQISS
$3RWW7%UXFNPDQQ/0LNHOVRQV &ORVHGIRUP)RUFH'LVWULEXWLRQIRU3DUDOOHO
:LUH5RERWV&RPSXWDWLRQDO.LQHPDWLFV6SULQJHUSS
. 7VXUXWD 6 )XWDPL + 1DNDPXUD DQG 7 0XUDNDPL 7KH ,QHUWLD ,GHQWLILFDWLRQ
0HWKRG(OLPLQDWLQJWKH,QIOXHQFHVRI9LVFRXV)ULFWLRQ&RXORPE)ULFWLRQDQG&RQVWDQW'LV
WXUEDQFH7RUTXHLQ0RWLRQ6\VWHPV-RIWKH-DSDQ6RFLHW\IRU3UHFLVLRQ(QJLQHHULQJ9RO
,VVXHSS ,Q-DSDQHVH

Author Index
A Gonzalez-Rodríguez, Antonio, 13
Ament, Christoph, 367 Gouttefarde, Marc, 61, 173, 221, 345
Ananthasuresh, G.K., 3 Gueners, Damien, 307
Arena, Andrea, 319
Arsenault, Marc, 185 H
Asl, Hamed Jabbari, 209 Hamann, Marcus, 367
Harada, Takashi, 23, 35
B Heidel, Robin, 281
Bandyopadhyay, Sandipan, 161 Hirosato, Koki, 23, 35
Birglen, Lionel, 61
Boumann, Roland, 269 I
Bruckmann, Tobias, 269, 281 Idá, Edoardo, 333
Izadbakhsh, Alireza, 209
C
Cardou, Philippe, 109, 345 K
Caro, Stéphane, 73, 109, 257 Kamali, Kaveh, 345
Carricato, Marco, 333
Castillo-García, Fernando J., 13 L
Chanal, Hélène, 307 Lau, Darwin, 391
Chaumette, François, 73 Lee, Dongwon, 295
Chedli Bouzgarrou, B., 307 Lemmen, Patrik, 281
Chemori, Ahmed, 221 Lesellier, M., 173
Lessanibahri, Saman, 109
D Li, Haisheng, 149
Desrosiers, Christian, 345 Li, Hui, 245
Li, Mingzhe, 245
F Long, Philip, 257
Fabritius, Marc, 137
Flügge, Wilko, 233 M
Froitzheim, Pascal, 233 Mahony, Robert, 391
Fuchs, Normen, 233 Marquez-Gamez, David, 257
Martin, Christoph, 137
G Matsuura, Daisuke, 415
García-Vanegas, Jorge A., 13 Merlet, Jean-Pierre, 47, 333
Gattulli, Vincenzo, 319

https://doi.org/10.1007/978-3-030-20751-9
426 Author Index
N Tang, Lewei, 99
Narikiyo, Tatsuo, 209 Tang, Xiaoqiang, 99
Newman, Matthew, 403 Tempel, Philipp, 121, 295, 379
Ng, Kwun Wang, 391 Terry, Benjamin, 403
Notash, Leila, 197 Trautwein, Felix, 295
Nüsse, Pauline Marie, 367 Tremblay, Nicolas, 345
O U
Otis, Martin J.-D., 345 Ueki, Taishu, 415
Ottaviano, Erika, 13, 319
V
P Verl, Alexander, 121, 379
Pedemonte, Nicolò, 73 Vikranth Reddy, M., 3
Peng, Fazhong, 149
Potenza, Francesco, 319 W
Pott, Andreas, 121, 137, 295, 379 Wang, Liping, 149
Praneet, N.C., 3 Winter, David, 367
Woernle, Christoph, 233
R Wu, Li, 99
Radojicic, Jelena, 85 Wulle, Frederik, 379
Rasheed, Tahir, 257 Wu, Xiaoyu, 99
Reichenbach, Thomas, 121
Rodríguez-Rosa, David, 13 X
Roos, Adolfo Suarez, 73 Xu, Wenfu, 357
Rubio-Gómez, Guillermo, 13
Y
S Yoshida, Minoru, 415
Santos, João Cavalcanti, 221 Yuan, Han, 357
Shahi, Ambuj, 161
Shao, Zhufeng, 149 Z
Stoltmann, Michael, 233 Zake, Zane, 73
Sugahara, Yusuke, 415 Zhang, Yongqing, 357
Sun, Yue, 403 Zhang, Zhaokun, 149
Surdilovic, Dragoljub, 85 Zygielbaum, Arthur, 403
T
Takeda, Yukio, 415

Cabledriven Parallel Robots 2019

Uploaded by

Copyright:

Available Formats

Cabledriven Parallel Robots 2019

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cabledriven Parallel Robots 2019

Uploaded by

Copyright:

Available Formats

Mechanisms and Machine Science 74

More information about this series at http://www.springer.com/series/8779

Cable-Driven Parallel Robots

ISSN 2211-0984 ISSN 2211-0992 (electronic)

S. Agrawal, Columbia University

Stuttgart, Germany Andreas Pott

Part II Kinematics and Static

Practical Stability of Under-Constrained Cable-Suspended

Part III Workspace

Part V Motion Planning

Part VI Advanced Cable Modeling

Part VII Calibration and Identiﬁcation

Part VIII Applications

M. Vikranth Reddy1, N. C. Praneet2, and G. K. Ananthasuresh3

Keywords: Planar cable-driven robot, orientations.

© Springer Nature Switzerland AG 2019 3

2 Limited orientability of the moving platform in cable-

2.1 Three-cable planar robot

Fig. 1. Model of the three-cable planar robot

rank (L) rank (L_aug)

Fig. 3. Orientations feasible at (2, 2)

2.2 Four-cable robot

3 Complete orientability in cable robots

3.1 An extra cable to enhance orientation

3.2 Re-calculated tensions for orienting in a given location

Fig. 7. A Force of [0 5.59 0]T is acting on the triangular platform.

Component/ Part Specifications

Aluminum (Outer diameter = 42.60mm , width = 32mm,

4.1 Three-cable robot with complete orientability

Guillermo Rubio-Gómez1, David Rodríguez-Rosa1, Jorge A. García-Vanegas2, Anto-

Abstract. This work presents Chain-Driven Parallel Robots replacing cables by

Keywords: Chain-Driven Robot, Parallel Robot, Industrial Applications.

© Springer Nature Switzerland AG 2019 13

2 Chain-Driven Robot vs Cable-Driven Robot

As introduced in Section 1, chain-driven robots present several important advantages

Fig. 1. Drum vs. Sprocket

x Cable plasticity limitation: When parallel robots are commanded by means of

x Minimum tension of chains for ensuring an accurate positioning (sagged

Fig. 2. Counterweight for decreasing needed torque.

3.1 Prototype design

Fig. 3. Scheme of 2DOF planar chain-driven robot

3.2 Experimental platform

The experimental parameters are summarized in Table 1.

Table 1. Parameters of experimental platform

The values of end-effector mass, ݉௘ , and counterweights, ݉ଵ and ݉ଶ will be con-

3.3 Tension Analysis

3.4 Workspace Analysis

ȁ߬௜ ȁ ൌ ȁ‫ ݎ‬൉ ሺܶ௜ െ ݉௜ ݃ሻȁ ൏ ߬௠௔௫ (2)

being ߬௜ the torque exerted by motor ݅.

Finally, Fig. 8 represents the maximum rectangular workspace which is 0.810 m ൈ

Fig. 8. Maximum rectangular workspace

4.1 Experiments setup

Fig. 9. Hardware architecture of the robot

4.2 First movements

Fig. 2. First experiments setup.

Takashi Harada1 and Koki Hirosato2

Abstract. Non-slipping conditions of endless-cable driven parallel robot

Keywords: endless-Cable, friction drive, the Euler-Eytelwein’s formula,

© Springer Nature Switzerland AG 2019 23

2 Endless-cable driven parallel robot

Fig. 3. Orientations feasible at (2, 2)