3806 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 5, NO.

3, JULY 2020

Nanometer Precision With a Planar Parallel

Continuum Robot
Benjamin Mauzé , Redwan Dahmouche , Guillaume J. Laurent , Antoine N. André ,
Patrick Rougeot, Patrick Sandoz, and Cédric Clévy

Abstract—In many cases, soft and continuum robots represent an designs [8], [9], actuation systems [10] and materials [11], classi-
interesting alternative to articulated robots because they have the cal robots have seen improved in miniaturization capability [12],
advantages of miniaturization capability, safer interactions with [13], lightweight designs, adaptability [14], payload-to-robot
humans and often simpler fabricating and integration. However,
these benefits are usually considered to arise at the expense of weight ratios [15], length-to-width ratios [16] and others char-
accuracy and precision because of the soft or flexible limbs. This acteristics [17].
paper demonstrates that, with a proper design, a planar parallel However, a property of soft and continuum robots that is
continuum robot is capable of great precision. Indeed, the proposed currently missing is precision. Indeed, very little work has been
3-Degrees-of-Freedom planar parallel continuum robot exhibits done in this field. For instance, one of the rare studies found a pre-
a precision of 9.13 nm in position and 1.2 µrad in orientation.
In addition, the novel robotic design leverages the effect of the cision of about 1 mm for a serial continuum robot [18]. In addi-
actuators’ defects, making the robot more precise than its own tion, Orekhov et al. [19] proposed a surgical Parallel Continuum
actuators. Finally, the workspace of the proposed robot (62.3 mm2 , Manipulator with a large workspace and a precision of 0.88 mm
0.6452 rad) is significantly larger than most compliant mechanisms, in position and 1.96 deg in rotation. Black et al. [20] studied the
which is particularly interesting when both very high precision and
pros and cons of Parallel Continuum Robots (PCR) and pointed
relatively large displacements are required.
out that an analysis of their performance was still needed.
Index Terms—Soft robot applications, micro/nano robots, Atuzarra et al. [21], by analyzing the characteristics of PCR,
parallel robots. announced that they may achieve better precision than their
I. INTRODUCTION serial counterparts. However, no evaluation was made of their
potential precision compared to other technologies such as clas-
FTER decades of development, industrial robots use today
A very mature technologies. Yet, they use bearings and gear-
boxes that limit their precision and accuracy due to backlashes,
sical parallel robots. The current conjecture, from the industrial
and academic practices, leads to think that articulated robots are
better suitable for precision manipulation [22].
flexibilities and frictions. Parallel robots leverage some of these The question that arises then is, “Is it possible to combine
drawbacks thanks to more rigid structures but still basically the advantages of the different technologies to obtain a high
experience the same types of defects [1]. precision over a large workspace?”
In applications such as micro and nano assembly, auto- In this paper, simulated and experimental results show that
mated biological cell manipulation, X-ray lithography, and oth- parallel continuum robots are relevant candidates for high pre-
ers, where high-grade precision positioning is required, such cision manipulation. Indeed, their mechanical structure does
mechanical components should be avoided [2], [3]. Flexure not have mechanical joints. Moreover, the limbs’ bending can
hinge-based compliant structures actuated by piezoelectric, elec- be large, with appropriate materials allowing relatively large
tromagnetic, electrostatic or electrothermal actuators are thus displacement range compared to compliant mechanisms. This
preferred [4]–[7]. However, this type of mechanism has very paper shows that a 3-Degrees-of-Freedom (3-DoF) planar par-
limited workspace in position as well as in rotation [5]. allel continuum robot can reach nanometer precision along a
Soft and continuum robots have interesting properties over relatively large workspace, ranking it among the best solutions
articulated and flexure-based robots. Thanks to innovative for high-grade precision applications.
The next section introduces the 3-DoF PCR structure whose
Manuscript received October 15, 2019; accepted March 4, 2020. Date of model is provided in Section III. The design and prototype
publication March 20, 2020; date of current version April 13, 2020. This letter
was recommended for publication by Associate Editor O. Ozcan and Editor C. development is proposed in Section IV. Precision performances
Laschi upon evaluation of the reviewers’ comments. This work was supported in of the robot are investigated first by simulation in Section V and
part by ISITE-BFC Contract ANR-15-IDEX-03, in part by Equipex ROBOTEX then experimentally in Section VI.
(ANR-10-EQPX-44-01), in part by Région Bourgogne Franche-Comté, and
in part by EIPHI Graduate School (ANR-17-EURE-0002). (Corresponding
author: Redwan Dahmouche.)
The authors are with FEMTO-ST Institute, University Bourgogne II. 3-DOF PARALLEL CONTINUUM ROBOT
Franche-Comté, UMR CNRS, 6174 Besançon, France (e-mail: benjamin.
[email protected]; [email protected]; guillaume.laurent@ This paper aims to demonstrate that parallel continuum robots
femto-st.fr; [email protected]; [email protected]; patrick.
[email protected]; [email protected]).
(PCR) can achieve high positioning precision (comparable with
Digital Object Identifier 10.1109/LRA.2020.2982360 flexure hinge-based compliant stages) while having longer travel
2377-3766 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 MAUZÉ et al.: NANOMETER PRECISION WITH A PLANAR PARALLEL CONTINUUM ROBOT 3807

Fig. 2. Rod model within the (O, x, y) plane with frames and boundary
conditions.

actuator line passes through the center of the circle; for

α = π/2, the actuator line is tangent to the circle).

III. MODELING
In this section, we detail the model of the proposed PCR
structure that will be used to simulate its precision.

A. Frames and Transformations

Fig. 1. Model of the new planar Parallel Continuum Robot.
For the purpose of analysis, we attach a work reference frame
(OW , xW , yW , zW ) to the fixed base with its origin located at
the center of the circumscribed circle of triangle A1 A2 A3 when
the actuators are in their reference positions and with the zW axis
ranges. For that purpose, a planar PCR, inspired by the well-
perpendicular to the base. A mobile frame (OP , xP , yP , zP ) is
known 3-PRR parallel robot, was designed.
attached to the moving platform with its origin located at the
The regular 3-PRR mechanism [23] is composed of three
center G of the triangle B1 B2 B3 . Finally, mobile frames are
planar kinematic chains, each with a prismatic joint and two
attached to the proximal ends and to the distal ends of each rod
revolute joints, all connecting the fixed base to the moving
(Fig. 1).
platform. The direction of the three chains are star-shaped with
These frames allow us to define the following transformations:
120-degree angles. This mechanism allows positioning of the r The transformations W TA from the work frame to the
i
platform along three-degrees-of-freedom, x, y, and θ in the plane
rod’s proximal end, which are each a function of the
according to the prismatic joint values q1 , q2 , and q3 .
prismatic articulate variable qi each;
The proposed design of the planar PCR is also composed r The rod transformations Ai TB that define the relative
i
of three kinematic chains but links and joints are replaced by
position and orientation between the proximal and the
slender rods that provide large and continuous deformations.
distal end of each rod, depending on the external forces
Each rod is connected to a precision linear actuator and to the
and moments applied to it;
rigid moving platform. The motion of the platform is restricted r The transformations Bi TP from each rod’s distal end to
to the 3-DoF planar displacements.
the mobile frame, which are constant since the platform is
The rods are clamped on their proximal ends, denoted Ai , to
rigid;
the linear stages in line with the direction of their motions. The r The transformation W TP that gives the position and the
three stages are arranged at 120-degree angles such that A1 A2 A3
orientation of the mobile platform with regards to the work
is an equilateral triangle when the actuators are in their reference
frame.
positions. The clamping points of the distal ends of the rods, Bi ,
to the mobile platform also form an equilateral triangle with
vertices B1 , B2 and B3 . B. Rods Modeling
As illustrated in Fig. 1, only four parameters are required to 1) Kirchhoff Rod Equations: The rods are slender elements,
describe the manipulator geometry: straight in their free-stress configuration. However, to avoid
r l is the length at the free-stress configuration of the rods singular configurations, the rods are constrained in the home
(the rods are straight before assembling); configuration (see Fig. 2). The characteristic dimension of their
r rP is the radius of the circle defined by the three distal ends cross-section (diameter for cylindrical rod) is more than 100
of the rods B1 , B2 , B3 ; times smaller than the rod’s length. Thus, shear and elongation
r rA is the radius of the circle defined by the three proximal can be neglected. Considering those assumptions, the Cosserat
ends A1 , A2 , A3 of the rods when the actuators are in their rod model can be simplified to the Kirchhoff model.
reference positions; In this article, we consider the planar case of this model.
r α corresponds to the angle between the prismatic joint axis We suppose that the movement of each rod is planar inside the
and the direction of the center of the circle (for α = 0, the (O, x, y) plane. All variables are illustrated in Fig. 2.

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 3808 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 5, NO. 3, JULY 2020

A scalar parameter s represents the curvilinear abscissa over

the entire rod, s ∈ [0, l] where l is the free-stress length of a rod.
The shape of a rod along its arc length is defined by the cross-
section centroid position p(s) = [ x(s)
y(s)
] represented in the frame
attached to its proximal end Ai and R(s) = [ cos θ(s) −sin θ(s)
sin θ(s) cos θ(s)
]
the rotation matrix between the frame attached to Ai and the one
attached to p(s).
The evolution of position v(s) ∈ R2 and orientation uz (s) ∈
R along the arc length, related to the material strain, can be
expressed in the local frame as:

dp(s)
R(s)T = v(s)
ds
  Fig. 3. Prototype of the proposed parallel continuum robot.
dR(s) 0 −uz (s)
R(s)T = (1)
ds uz (s) 0

(s) platform are the following:

The internal force n(s) = [ nnxy (s) ] and moment m(s) are de-
fined with respect to the arc length. We assume that no distributed W
TBi = W TP · P TBi (4)
external forces or moments are applied on the rod. The nonlinear
equations of the static equilibrium of Kirchhoff rod yield: Using both boundaries conditions, the ith rod transformation
Ai
TBi is then given by:
dn(s)
=0 Ai
TBi = Ai TW · W TBi (5)
ds
dm(s) dp(s) 4) Static Equilibrium Conditions: The static equilibrium
+ × n(s) = 0 (2)
ds ds conditions of the platform, needed in order to determine the
From all previously defined equations, the system of differ- position and orientation of the platform, yield:
ential equations is the following: 3
⎧ dx(s) ⎫ [ni (L)] − f P = 0
⎪ ⎧ ⎫
⎪
⎪ ds ⎪ ⎪ ⎪ cos(θ(s))
⎪ dy(s) ⎪ ⎪ i=1
⎪ ⎪ ⎪
⎪ ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ sin(θ(s)) ⎪
⎪ 3
⎪
⎪ ds ⎪ ⎪
⎪ ⎪ ⎪
⎪
⎨ dθ(s) ⎬ ⎨ m
+ u (0) ⎬ [pBi × ni (L) + mi ] − pP × f P − mP = 0 (6)
z
ds
= EI
(3) i=1
⎪ dnx (s) ⎪
⎪ ⎪ ⎪ ⎪ 0 ⎪
⎪
⎪
⎪ ds ⎪
⎪ ⎪ ⎪ ⎪
⎪
⎪
⎪ dny (s) ⎪
⎪ ⎪
⎪ 0 ⎪
⎪ where pBi and pP are the positions in the work frame of Bi and
⎪
⎪ ⎪
⎪ ⎪
⎩ ⎪
⎭
⎪
⎩ dm(s) ⎪
ds
⎭ nx sin(θ(s)) − ny cos(θ(s)) P , f P and mP are respectively the external forces and moment
ds applied on the platform and ni and mi the ones applied by the
rods expressed in the work frame.
Where E is the Young modulus of the rod material, uz (0) is the
initial orientation of the rod, which is zero in our case, and I is
the second area moment of the rod cross-section. IV. PCR DESIGN
This system of equations describes the shape of the rod and the This section details the different parts of the PCR and its
internal forces and moments. The rod shape corresponds to the assembly. Fig. 3 shows the developed prototype.
transformation between the point Ai and Bi where i ∈ {1, 2, 3} The mobile platform is a silicon wafer of radius rP =
is the limb number. 51.85 mm (2 inches) on which three fiber holders were glued
2) Proximal Boundary Conditions: The proximal boundary 120 degrees apart. This choice of platform was motivated by a
condition of the ith rod corresponds to the position of the point potential application of wafer positioning.
Ai . The transformation W TAi from the work frame to the rod’s Underneath the platform, a 50 mm diameter vacuum
proximal end depends on the joint coordinate qi . When qi = 0, preloaded air-bearing S205001, from the IBS company, main-
the joint is in the reference position. In this paper, the considered tains the platforms at a stable height (±5 μm). The air-bearing is
joint is prismatic, thus the transformation W TAi follows a mounted on a manual linear stage, Newport M-DS25-Z, which
translation in the direction of the joint. One can notice that this allows the level of the wafer to be adjusted.
transformation can be calculated for other kinds of actuators. Fiber holders are aluminum pieces drilled on one of their
3) Distal Boundary Conditions: The distal ends of the rods side in order to insert the flexible rods. The rods are 125 μm
are coupled through their connection with the platform. Rigid- diameter single-core optical fibers, which were stripped of the
body conditions between the distal ends of the rods and the plastic part. Their Young modulus E is 73 GPa. Their length at

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 MAUZÉ et al.: NANOMETER PRECISION WITH A PLANAR PARALLEL CONTINUUM ROBOT 3809

TABLE I
UNKNOWNS, INPUTS AND OUTPUTS FOR SOLVING THE TWO KINEMATIC
MODELS. THE INDEX i IS THE NUMBER OF THE LIMBS WHICH IS
EQUAL TO 1, 2 AND 3

two elements, two manual linear translation stages, Newport

M-SDS-40, allow setting the distance rA at 85 mm. Those
manual stages allow the assembly of the prototype. They adjust
the initial actuators’ position to fix the corresponding holders,
reduce the initial stresses inside the fiber and, with the manual
stage underneath the air-bearing, adjust the planarity of the robot.
The footprint of the prototype is less than an equilateral
triangle with side of 235 mm.

V. SIMULATED PRECISION EVALUATION

In this section, the simulated precision along the workspace
of the PCR prototype is evaluated by Monte Carlo simulations.

A. Software Implementation
The quasi-static behavior of the previously defined PCR is
simulated using two algorithms. Fig. 4a and Fig. 4b represent
respectively the two routines used to solve the two kinematic
models, the Forward Kinematic Model (FKM) and the Inverse
Kinematic Model (IKM). Both models are numerically solved
with a shooting method.
Table I points out the unknowns, inputs and outputs for both
algorithms. pP = [ xyPP ] and θP correspond respectively to the
position and the orientation of the mobile platform in the work
frame. We use the fourth-order Runge-Kutta method in order
to integrate spatially the system of equations. The optimization
process is performed using the built-in function lsqnonlin
from Matlab software. The residue is the vector composed of the
Fig. 4. Flowchart of the shooting method used to numerically solve the bound- static equilibrium equations (6) and the rigid-body conditions (4)
ary conditions problem for (a) the Forward Kinematic Model (FKM) and (b) the between the platform and the distal end of the rod.
Inverse Kinematic Model (IKM). For both cases, the spatial integration uses the The gravity is compensated by the air-bearing so we assume
fourth-order Runge-Kutta (RK4) method. “Res.” in the flowchart represents
the optimization residues. that no external forces or moments are applied to the platform
when the platform is in equilibrium. By considering as null
the external forces and moments, respectively f P and mP , the
equations (3) and (6) are simplified.
their free-stress configuration is 30 mm. They are clamped in
line with the translation axis of the actuator.
B. Workspace Simulation
The robot limbs are actuated by stick-slip piezo-electric lin-
ear actuators SmarAct slc-1730-s-hv whose resolution is about The flexibility of the PCR’s elements allows them to have a
1 nm. Their precision in closed-loop is 18.02 nm for a displace- relatively large workspace. In order to demonstrate the preci-
ment of 100 μm. Those actuators are a good trade-off between sion of the proposed structure along its motion range without
size, travel range and resolution. Moreover, their compatibility breaking the limbs, the safe workspace must be estimated by
with high vacuum opens the possibility to integrate them into a simulation first.
scanning electron microscope thanks to a more compact design The workspace simulation is performed using the IKM
in which the air-bearing would not be required. (Fig. 4b). It has been computed following the method proposed
Below the actuators, a manual rotation stage Newport M- by Merlet et al. [24]. Stable and reachable positions are com-
RS40 allows setting the angle α at 15 degrees. Below those puted using the IKM considering fixed platform’s orientations.

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 3810 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 5, NO. 3, JULY 2020

Fig. 5. Simulated workspace in position for a constant orientation of the plat-

form (initial orientation) and for maximal stresses in the limbs below 200 MPa.
The workspace is sampled by steps of 250 μm. The colors correspond to the
maximal values of the stresses inside the three legs. The four black points P1 ,
P2 , P3 and P4 are the positions where the precision will be quantified.

The position workspace for the initial platform’s orientation is Fig. 6. Results of the propagation of uncertainty from the actuators to the
robot’s end-effector based on Monte Carlo simulations. The considered displace-
described and illustrated in Fig. 5. ment is a translation of 100 μm along q1 from its initial position. (a) Platform
At each position, the stress inside the legs is computed and the coordinates in the work frame. (b) Distance deviation (distances to the mean
maximal value is reported. The maximal stress variations inside position). (c) Angle deviation (differences between angles and their mean).
the robot’s legs are illustrated by the color of the corresponding
position. To avoid breaking the robot’s limbs, the workspace is
limited to the area in which the maximal stress is below 200 MPa. The orientation precision is defined as the standard deviation
This value was chosen by considering the safest elastic limit of stda with :
the optic fiber, which varies between 200 MPa and 700 MPa for n   2
standard optical fiber without the acrilate protective coating [25]. j=1 θP j − n1 ni=1 θP i
The area of this safe workspace is equal to 62.3 mm2 . Despite stda = (8)
n−1
the fact that this workspace does not exploit the full actuators
stroke, it is far larger than the best XYθ flexure hinge-based Using the previously defined model and algorithms, the sim-
compliant mechanisms reported (16 mm2 ) [6]. Fig. 5 shows the ulated precision of the robot is quantified by taking into account
safe workspace and the four points P1 , P2 , P3 and P4 that were the actuation uncertainty.
considered to quantify the precision of the robot in different 2) Monte Carlo Method: This precision is simulated using
positions. Due to the symmetries of the structure, these four the Monte Carlo method for the propagation of uncertainty dis-
points seem to be representative of the precision over the robot’s tributions. This method was introduced by the Joint Committee
safe workspace. for Guides in Metrology (JCGM) [26].
One considers the uncertainties of translation in the direc-
tion of the actuator displacement. Those uncertainties follow a
C. Monte-Carlo Evaluation of the Precision normal distribution with a standard deviation of 18.02 nm for
1) Precision Definition: The precision of the robot is defined a displacement of 100 μm. This value was extracted from the
by its position and orientation precision. The position precision manufacturer’s documents.
is defined as the standard deviation of the distance to the mean The principle of the method is the following. One considered n
position of the platform, stdl . values of q1 sorting from a normal distribution with a mean value
is 100 μm and a standard deviation of 18.02 nm, which reflect
n
¯l = 1 lj
respectively the desired position and precision of the actuator.
n Those joint coordinates are considered as inputs to the FKM to
j=1
calculate the position and the orientation of the platform as out-
2 2 puts. The value n was set to 30 to be statistically representative.
1 n 1 n
lj = xP j − x P i + yP j − yP i Using this method, the precision for the previously considered
n i=1 n i=1
points were estimated by simulation.
n 3) Simulation Results: Fig. 6 presents the results of the
− ¯l)2
j=1 (lj
stdl = (7) propagation of uncertainty from the actuators to the platform
n−1 considering the displacement along q1 from its initial position

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 MAUZÉ et al.: NANOMETER PRECISION WITH A PLANAR PARALLEL CONTINUUM ROBOT 3811

TABLE II
SIMULATED AND EXPERIMENTAL POSITION AND ORIENTATION PRECISION FOR
DIFFERENT CONSIDERED CASES. THE FIRST FOUR CASES CORRESPOND TO
THE FOUR POINTS CHOSEN IN THE WORKSPACE AND THE LAST CASE
CORRESPONDS TO THE ADDITION OF THE UNCERTAINTIES OF THE THREE
ACTUATORS AT THE INITIAL POSITION (P1 )

(P1 in Fig. 5). Fig. 6(a) shows the platform position where
we can see the direction of uncertainty. Fig. 6(b) shows the
distance deviation to the mean position (lj in equation (7)).
The standard deviation of those distances corresponds to the Fig. 7. Experimental setup showing the measurement system and the PCR
simulated position precision of the robot. Fig. 6(c) shows the prototype attached to an anti-vibration table. Pattern used with a Fast Fourier
angle deviation. Transform phase-based algorithm. This algorithm provides position and orien-
tation with a resolution of 0.5 nm and 1 μrad. [27].
The angle standard deviation corresponds to the simulated
orientation precision.
Additional simulations have been performed for the chosen
points and considering the uncertainties of the three actuators. Mitutoyo 10× zoom. The camera is configured with an exposure
The results of the simulations are reported in Table II. In the last time of 150 ms and takes 6.5 pictures per second of a calibrated
case, a displacement of 100 μm is simulated for each actuator pattern (Fig. 7) glued on the platform. All images are recorded
moving together from the robot’s initial position. after a warm-up cycle of the camera of at least one hour.
One can notice that the precision of the robot’s platform is The pattern was manufactured by etching a chromium layer
better than the actuators precision. One explanation for this result onto a transparent glass wafer. The periodicity between features
is that, depending on their configurations, most parallel robots is 9 μm along the two directions. The pattern was fabricated by
kinematics exhibit the property of amplifying or reducing the direct laser writing with an off-the-shelf instrument (Heidelberg
actuators motions. In the last case, the robot kinematics may DWL200) whose position and displacements are continuously
leverage the actuators positioning uncertainties. Although, the controlled by a HeNe laser interferometer.
presented parallel continuum robots has an additional property A phase-based method is used in order to get an accurate po-
in which the rods act like springs that absorb the actuators sition and orientation measurement of the platform. It measures
motions. Thus, a fraction of the actuators work is not transmitted the phase of the image in the Fourier domains in order to get a
to the platform as motion but stored into the rods as a potential sub-nanometric resolution. This algorithm together with a binary
energy (rods bending). This property, which can be found only in code provides the absolute pose of the pattern with a resolution
deformable robots, may contribute in enhancing the platform’s of 0.5 nm in both directions and 1 μrad for the orientation. More
precision against the actuators defects. details on this measuring method can be found in [27]–[30].
The simulated values obtained in this section were checked
against the ones obtained experimentally. B. Preliminary Experiments
A vision-based measurement system can be disturbed by en-
VI. EXPERIMENTAL PRECISION EVALUATION vironmental factors. Quantifying their impact enables defining a
In this section, the experimental set-up is described and the threshold below which the precision will not be distinguishable
robot’s precision is quantified experimentally. from the environment noise.
Three different cases are considered. The first one repre-
sents the quantification of the measurement noise, a.e. before
A. Measurement System powering the air-bearing or the actuators. The second one is
The measurement system can be seen in Fig. 7. All the the uncertainty change due to the air-bearing. The last one
components are placed on an anti-vibration table to reduce corresponds to the normal functioning conditions when the robot
the vibrations transmitted through the ground. The motion of is holding a stable position, for instance, its initial position.
the robot, shown in Fig. 3 and presented in section IV, is captured One hundred images are recorded for about one minute. The
by a vision measurement system that is composed of a camera vision algorithm is used to obtain the position and the orientation
IDS UI3880CP-C-HR-R2, an Edmund microscope tube and a of the platform. Distances to the mean (lj in equation (7)) and the

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 3812 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 5, NO. 3, JULY 2020

Fig. 8. (a) Quantification of the positioning measurement uncertainty and

(b) the orientation measurement uncertainty when the robot is holding a position.
The points’ size depends on the image acquisition time.

differences between angles and their mean are computed. The

standard deviations of those values quantify the measurement
uncertainty threshold.
For the three cases detailed above, the uncertainty mea- Fig. 9. (a) Platform positions during the experimental precision quantification.
surements were similar. Consequently, only the measurement (b) Distances to the mean for the platform positions. (c) Differences between
angles and their mean. Standard deviations stdl , stda correspond respectively
uncertainty when the robot is holding a position is illustrated in to position and orientation precision.
Fig. 8. The size of the points depends on the acquisition time
to show the position drift over the images sequence. Positioning
and orientation standard deviations are respectively 8.2 nm and The precision obtained with the proposed PCR (9.13 nm
2.6 μrad. and 1.2 μrad) is in the same order of magnitude as classical
To reduce this uncertainty, three consecutive measurements compliant XY θ mechanisms [6] that are specifically designed
are averaged to get the platform pose. The position standard for precise applications while the workspace of the proposed
deviation drops to 6.6 nm taking three images into account while PCR is significantly larger.
the angle standard deviation is only 1.85 μrad.
VII. CONCLUSION
C. Experimental Evaluation of the Precision In this paper, we proposed a new XY θ parallel continuum
The precision of the robot is quantified using repeated motions robot. The proposed structure has a safe workspace of 62.3 mm2
measured by the vision-based reference system. The actuators’ and 0.6452 rad (39.97 deg) and is able to experimentally reach a
speed is set to 20 μm.s−1 , which is enough for micro- or nano- precision of 9.13 nm in position and 1.2 μrad in orientation. The
applications where high dynamic is not required. The same presented results show that a planar parallel continuum robot can
positions and displacements used in the simulations have been outperforms conventional compliant mechanisms by achieving
considered. Those displacements are repeated 30 times to be a very high precision within a large workspace. Consequently,
statistically representative. At each position, three images are planar parallel continuum robots appear to be promising candi-
taken in order to reduce the measurement noise. dates for high-grade precision applications, which expends the
Once the images are recorded, the vision processing algorithm potential application scope of PCRs. Future work will investigate
provides the positions and orientations of the platform. Fig. 9 the accuracy assessment of the proposed PCR and how the
shows the experimental results for the first experimental case model‘s parameters influence the robot’s performances.
(P1 ). The maximal values of the distances to the mean for the
position and orientation are respectively 40.93 nm and 4.2 μrad. REFERENCES
The standard deviations, corresponding to the position and ori-
[1] C. E. Bryson and D. C. Rucker, “Toward parallel continuum manipulators,”
entation precision, are 9.13 nm and 1.2 μrad. Table II shows in Proc. IEEE Int. Conf. Robot. Autom., May 2014, pp. 778–785.
the experimental precision results for the five cases discussed [2] B. Brazey, R. Dahmouche, J.-A. Seon, and M. Gauthier, “Experimental
previously. One can notice that the measured deviations are validation of in-hand planar orientation and translation in microscale,”
Intell. Service Robot., vol. 9, no. 2, pp. 101–112, Apr. 2016.
extremely low, even below the estimated ones. Moreover, all [3] J.-A. Seon, R. Dahmouche, and M. Gauthier, “Enhance in-hand dexterous
precision results are at the same level as the measurement micromanipulation by exploiting adhesion forces,” IEEE Trans. Robot.,
uncertainties, and we cannot conclude whether the observed vol. 34, no. 1, pp. 113–125, Feb. 2018.
[4] P. R. Ouyang, R. C. Tjiptoprodjo, W. J. Zhang, and G. S. Yang, “Micro-
deviations are mainly due to positioning errors of actuators or motion devices technology: The state of arts review,” Int. J. Adv. Manuf.
to environmental noise. Technol., vol. 38, pp. 463–478, 2008.

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.
 MAUZÉ et al.: NANOMETER PRECISION WITH A PLANAR PARALLEL CONTINUUM ROBOT 3813

[5] B. Ding, Y. Li, X. Xiao, Y. Tang, and B. Li, “Design and analysis of a [19] A. L. Orekhov, C. B. Black, J. Till, S. Chung, and D. C. Rucker,
3-DOF planar micromanipulation stage with large rotational displacement “Analysis and validation of a teleoperated surgical parallel continuum
for micromanipulation system,” Mech. Sci., vol. 8, no. 1, pp. 117–126, manipulator,” IEEE Robot. Autom. Lett., vol. 1, no. 2, pp. 828–835,
May 2017. Jul. 2016.
[6] Z. Wu and Q. Xu, “Survey on recent designs of compliant micro-/nano- [20] C. B. Black, J. Till, and D. C. Rucker, “Parallel continuum robots: Mod-
positioning stages,” Actuators, vol. 7, no. 1, Feb. 2018. [Online]. Available: eling, analysis, and actuation-based force sensing,” IEEE Trans. Robot.,
https:doi.org.10.3390/act7010005 vol. 34, no. 1, pp. 29–47, Feb. 2018.
[7] L. Zhang and P. Yan, “Design of a parallel XYθ micro-manipulating [21] O. Altuzarra, D. Caballero, Q. Zhang, and F. J. Campa, “Kine-
system with large stroke,” in Proc. Chin. Control Decis. Conf., May 2016, matic characteristics of parallel continuum mechanisms,” in Proc. Ad-
pp. 4775–4780. vances in Robot Kinematics, vol. 8. Berlin, Germany: Springer, 2019,
[8] D. Rus and M. T. Tolley, “Design, fabrication and control of soft robots,” pp. 293–301.
Nature, vol. 521, no. 7553, pp. 467–475, May 2015. [22] Y. Sidyganov, V. Smelik, K. Semenov, D. Kostromin, and A. Medyakov,
[9] D. Trivedi, C. D. Rahn, W. M. Kier, and I. D. Walker, “Soft robotics: “The study of the positioning of a flexible manipulator,” J. Appl. Eng. Sci.,
Biological inspiration, state of the art, and future research,” Appl. Bionics vol. 16, no. 4, pp. 480–479, 2018.
Biomech., vol. 5, no. 3, pp. 99–117, Dec. 2008. [23] J.-P. Merlet, “Direct kinematics of planar parallel manipulators,” in Proc.
[10] C. Ahn, X. Liang, and S. Cai, “Bioinspired design of light-powered IEEE Int. Conf. Robot. Autom., 1996, vol. 4, pp. 3744–3749.
crawling, squeezing, and jumping untethered soft robot,” Adv. Mater. [24] D. Oetomo, D. Daney, B. Shirinzadeh, and J.-P. Merlet, “Certified
Technol., vol. 4, Jun. 2019, Art. no. 1900185. workspace analysis of 3RRR planar parallel flexure mechanism,” in Proc.
[11] J. Paik, “Soft robot design methodology for ‘push-button’ manufacturing,” Int. Conf. Robot. Autom., May 2008, pp. 3838–3843.
Nature Rev. Mater., vol. 3, no. 6, pp. 81–83, Jun. 2018. [25] P. Antunes, F. Domingues, M. Granada, and P. Andr, “Mechanical prop-
[12] J. Burgner-Kahrs, D. C. Rucker, and H. Choset, “Continuum robots for erties of optical fibers,” in Selected Topics on Optical Fiber Technology,
medical applications: A survey,” IEEE Trans. Robot., vol. 31, no. 6, M. Yasin, Ed., Rijeka, Croatia: InTech, Feb. 2012.
pp. 1261–1280, Dec. 2015. [26] Joint Committee for Guides in Metrology (JCGM) 101:2008, Supple-
[13] W. Haouas, R. Dahmouche, J. Agnus, N. L. Fort-Piat, and G. J. ment to the Guide to the expression of Uncertainty in Measurement
Laurent, “New integrated silicon-PDMS process for compliant micro- (GUM).
mechanisms,” J. Micromech. Microeng., vol. 27, no. 12, Oct. 2017, Art. [27] A. N. Andre, P. Sandoz, B. Mauze, M. Jacquot, and G. J. Laurent, “Sensing
no. 127001. one nanometer over ten centimeters: A micro-encoded target for visual
[14] J. Hughes, U. Culha, F. Giardina, F. Guenther, A. Rosendo, and F. Iida, in-plane position measurement,” IEEE/ASME Trans. Mechatronics, to be
“Soft manipulators and grippers: A review,” Front. Robot. AI, vol. 3, published.
Nov. 2016. [Online]. Available: https:doi.org.10.3389/frobt.2016.00069 [28] N. Tan, C. Clévy, G. J. Laurent, P. Sandoz, and N. Chaillet, “Accuracy
[15] S. Voisembert, N. Mechbal, A. Riwan, and A. Aoussat, “Design of a novel quantification and improvement of serial micropositioning robots for
long-range inflatable robotic arm: Manufacturing and numerical evaluation in-plane motions,” IEEE Trans. Robot., vol. 31, no. 6, pp. 1497–1507,
of the joints and actuation,” J. Mechanisms Robot., vol. 5, no. 4, Nov. 2013, Dec. 2015.
Art. no. 045001. [29] V. Guelpa, G. J. Laurent, P. Sandoz, J. Galeano Zea, and C. Clévy,
[16] C. Girerd, K. Rabenorosoa, and P. Renaud, Combining Tube Design and “Subpixelic measurement of large 1D displacements: Principle, process-
Simple Kinematic Strategy for Follow-the-Leader Deployment of Concen- ing algorithms, performances and software,” Sensors, vol. 14, no. 3,
tric Tube Robots. Berlin, Germany: Springer, 2018, pp. 23–31. pp. 5056–5073, 2014.
[17] M. Runciman, A. Darzi, and G. P. Mylonas, “Soft robotics in minimally [30] V. Guelpa, P. Sandoz, M. A. Vergara, C. Clévy, N. Le Fort-Piat, and G.
invasive surgery,” Soft Robot., vol. 6, pp. 423–443, Mar. 2019. J. Laurent, “2D visual micro-position measurement based on intertwined
[18] H.-J. Kim, A. Kawamura, Y. Nishioka, and S. Kawamura, “Mechanical twin-scale patterns,” Sensors and Actuators A: Phys., vol. 248, pp. 272–
design and control of inflatable robotic arms for high positioning accuracy,” 280, 2016.
Adv. Robot., vol. 32, no. 2, pp. 89–104, Jan. 2018.

Authorized licensed use limited to: Rajamangala Univ of Technology Thanyaburi provided by UniNet. Downloaded on July 31,2024 at 08:06:00 UTC from IEEE Xplore. Restrictions apply.