robotics

Article
A Robot Arm Design Optimization Method by Using
a Kinematic Redundancy Resolution Technique
Omar W. Maaroof 1, * , Mehmet İsmet Can Dede 2 and Levent Aydin 3
1 Mechatronics Engineering Department, University of Mosul, Mosul 41002, Iraq
2 Department of Mechanical Engineering, İzmir Institute of Technology, Gülbahçe Mahallesi, Urla,
Izmir 35430, Turkey; [email protected]
3 Department of Mechanical Engineering, İzmir Katip Çelebi University, A.O.S.B. Mahallesi, Çiğli,
Izmir 35620, Turkey; [email protected]
* Correspondence: [email protected]

Abstract: Redundancy resolution techniques have been widely used for the control of kinematically
redundant robots. In this work, one of the redundancy resolution techniques is employed in the
mechanical design optimization of a robot arm. Although the robot arm is non-redundant, the
proposed method modifies robot arm kinematics by adding virtual joints to make the robot arm
kinematically redundant. In the proposed method, a suitable objective function is selected to optimize
the robot arm’s kinematic parameters by enhancing one or more performance indices. Then the robot
arm’s end-effector is fixed at critical positions while the redundancy resolution algorithm moves
its joints including the virtual joints because of the self-motion of a redundant robot. Hence, the
optimum values of the virtual joints are determined, and the design of the robot arm is modified
accordingly. An advantage of this method is the visualization of the changes in the manipulator’s
structure during the optimization process. In this work, as a case study, a passive robotic arm that



is used in a surgical robot system is considered and the task is defined as the determination of


the optimum base location and the first link’s length. The results indicate the effectiveness of the
Citation: Maaroof, O.W.; Dede,
proposed method.
M.İ.C.; Aydin, L. A Robot Arm
Design Optimization Method by
Keywords: design optimization; redundancy resolution; robot mechanism design; optimization
Using a Kinematic Redundancy
Resolution Technique. Robotics 2022,
techniques; surgical robots
11, 1. https://doi.org/10.3390/
robotics11010001

Academic Editor: Alessandro

1. Introduction
Di Nuovo
Optimization methods have been employed in a wide range of areas from economical
Received: 15 November 2021 sciences to design processes in engineering applications. All optimization techniques de-
Accepted: 19 December 2021 pend on the numerical and/or algorithmic approach. With the improvement and availabil-
Published: 22 December 2021 ity of powerful computers, many techniques for optimization studies are presented. Such
Publisher’s Note: MDPI stays neutral methods can be listed as genetic algorithms (GA) [1], Ant Colony Optimization (ACO) [2],
with regard to jurisdictional claims in and Particle Swarm Optimization (PSO) method [3]. These methods are categorized as
published maps and institutional affil- modern and nontraditional optimization methods. However, optimization techniques
iations. have rewards and drawbacks. Various modifications to improve these techniques have
been a focus of many studies. The readers are directed to related resources on methods of
optimization and comparative studies such as the study in [4] as well as the review study of
the seven stochastic optimization methods that are preferred in optimization of industrial
Copyright: © 2021 by the authors. designs [5].
Licensee MDPI, Basel, Switzerland. All systems that require an optimum design are inherently redundant. In robotics,
This article is an open access article
among the other possible redundancy of components, kinematic redundancy has been an
distributed under the terms and
attractive research area since kinematically redundant robot arms may be used to perform
conditions of the Creative Commons
additional task/s while performing their main tasks. This is due to the infinite number of
Attribution (CC BY) license (https://
solutions received for the inverse kinematics analysis of a redundant robot resulting in an
creativecommons.org/licenses/by/
infinite number of configurations of the robot for the same end-effector pose. Consequently,
4.0/).

Robotics 2022, 11, 1. https://doi.org/10.3390/robotics11010001 https://www.mdpi.com/journal/robotics

Robotics 2022, 11, 1 2 of 21

the motions of the links of a robot that are not affecting the motion of the end-effector are
named “self-motion” by Nakamura [6]. The algorithms that are developed to regulate
the self-motion of the redundant manipulators for a specific aim are called redundancy
resolution techniques.
In this study, a design optimization method is formulated specifically for robot ma-
nipulators by using Denavit–Hartenberg parameters. The proposed optimization method
adopts a redundancy resolution technique for a kinematically non-redundant manipulator.
In this new method, an originally non-redundant manipulator is modified to be a redun-
dant manipulator by including virtual joints. Hence, the employment of a redundancy
resolution technique is possible for this modified redundant manipulator. The type and
location of the added virtual joints are selected by the designer for selectively optimizing
the specific parameters of the robot manipulator. The current approach immerses the robot
mechanism designer into the optimization process by providing the convergence steps as
design parameters continuously change to their optimal values. Hence, the optimization
process becomes more intuitive to the robot mechanism designer than the optimization
methods involving randomness. This intuitive approach is our motivation in carrying out
this study and the main novelty of this work.
The structural optimization process is carried out, considering the design constraints,
to validate the applicability of this new technique. Design optimization procedures fol-
lowed in the design of industrial robot manipulators [7] and the design of haptic devices [8]
are utilized in this present work in terms of analyzing the requirements, stating the prob-
lem, assigning design constraints, and nominating objective functions. Consequently,
performance indices such as manipulability and condition number can be utilized to eval-
uate kinematic and/or dynamic performances of manipulators [9]. Specifically, in this
present work, the objective function that is used in redundancy resolution via null-space
optimization is derived by using the modified form of these two indices.
The case scenario selected for this work is the passive arm of a surgical robotic system
called NeuRoboScope [10] as shown in Figure 1. In this system, the passive arm is required
to be backdriven by the surgeon to the designated locations of the surgical workspace
with minimal effort. Therefore, its performance measures related to both kinematic and
dynamic manipulability are studied in this paper as objective functions to test the proposed
optimization technique. In the next sections, after a brief review of redundancy resolution
techniques, the mechanical design optimization technique is described and the passive
arm mechanism of the NeuRoboScope system is introduced. Related to this specific case
scenario, the design constraints that are used in modifying the problem as the optimization
of a two degree-of-freedom (DoF) planar manipulator are explained. This modification also
facilitated the understanding and verification of the method described in this paper since
the two-DoF planar manipulator is an extensively studied manipulator in the literature.
The optimization procedure for this case scenario is explained along with the modifications
of the well-known performance indices that are used in this study. Finally, simulations that
are carried out to determine the optimal design of the mechanical structure are presented
and discussions are given on the obtained results.
Robotics
Robotics2022,
2022,10,
11,x 1FOR PEER REVIEW 3 of
3 of2221

Thesurgical
Figure1.1.The
Figure surgicalrobotic
roboticsystem
systemfor
forminimal
minimalinvasive
invasivepituitary
pituitarytumor
tumorsurgery:
surgery:NeuRoboScope.
NeuRoboScope.

2. A Brief Review of Redundancy Resolution Techniques

2. A Brief Review of Redundancy Resolution Techniques
A variety of redundancy resolution methods have been introduced in the literature
suchAas variety of redundancy
the Jacobian pseudo-inverseresolution methods
method, weightedhave pseudo-inverse
been introduced in the literature
method, and singu-
such as the Jacobian pseudo-inverse method, weighted pseudo-inverse
larity robustness method (damped least-squares DLS). All those redundancy resolution method, and sin-
gularity robustness method
methods are grouped as Jacobian-based. (damped least-squares DLS). All those redundancy resolution
methods By areaddinggrouped as Jacobian-based.
constraints in the form of additional tasks to redundancy resolution, the
infinite number of solutions the
By adding constraints in form of additional
is narrowed down to atasks to redundancysolution.
specific/bounded resolution, This theis
infinite number of solutions is narrowed down to a specific/bounded
exactly equivalent to establishing design constraints in design optimization techniques. solution. This is ex-To
actly equivalent to establishing design constraints in design optimization
incorporate an additional task, which is usually called the subtask, to the resolution process techniques. To
incorporate
of kinematic anredundancy,
additional task, which is usually
a null-space-based called the
method can subtask,
be applied. to the resolution
In this method, pro- the
cess of kinematic
gradient redundancy,
of a differentiable a null-space-based
objective method can
function is projected in thebe null-space
applied. Inofthis themethod,
Jacobian
the gradient
matrix so that of ita does
differentiable
not affectobjective
the main function
task. Here, is the
projected
main taskin the null-space
is the task thatofisthe Ja-
usually
cobian
assignedmatrix so tracking
as the that it does notend-effector’s
of the affect the main task. Here,
motion the main task is the task that
trajectory.
is usually
Anotherassignedmethod as theof tracking
redundancy of the end-effector’s
resolution is themotion trajectory.
decomposition method which de-
Another method of redundancy resolution is
composes joint-space variables into two groups (two minor Jacobian matrices)the decomposition method which as theyde- are
composes
related tojoint-space
the main task variables
and theinto two groups
additional task.(two minor Jacobian
Afterward, constraint matrices)
objective asequality
they areis
related
utilized toastheanmain
implicit task and thetoadditional
function reduce thetask. Afterward,
gradient constraint
of optimization objective
objective equality
function [11].
isThis
utilized
method as anhas implicit functionof
the attribute to eliminating
reduce the gradient of optimization
the unnecessary intensiveobjective functionof
computation
[11]. This methodwhich
pseudo-inverse has the attributethe
increases ofefficiency
eliminating the unnecessary
of calculation time. intensive computation
In the task augmentation
of pseudo-inverse which increases null-space-based
the efficiency of method, the Jacobian
calculation time. matrix is extended
by the
In theaddition of an auxiliary
task augmentation task [12] to result
null-space-based in a square
method, augmented
the Jacobian Jacobian
matrix matrix.
is extended
byInthe
thisaddition
method,ofthe anpseudo
auxiliary inverse
task [12]is not
to to be used
result [13] andaugmented
in a square the kinematic solution
Jacobian is no
matrix.
Inlonger redundant.
this method, the pseudo inverse is not to be used [13] and the kinematic solution is no
longerMulti-task
redundant. priority is another null-space-based method [14,15]. In this method, other
thanMulti-task
the Jacobian prioritymatrix related null-space-based
is another to the main task,method for each additional
[14,15]. In thissubtask,
method,another
other
Jacobian matrix exists. The self-motion of the first subtask
than the Jacobian matrix related to the main task, for each additional subtask, another is projected to the null-space
of the main
Jacobian matrixtask’s Jacobian
exists. matrix. Theofmotion
The self-motion the firstofsubtask
the second subtaskto
is projected is the
projected
null-spaceinto of the
null-space
the main task’s of the first subtask’s
Jacobian matrix. The Jacobian
motionmatrix.
of theIn the same
second subtaskmeans, other lower-order
is projected into the
priority subtasks
null-space of the first can be embedded
subtask’s in thematrix.
Jacobian earlier subtask
In the same that has higher
means, priority.
other lower-order
priorityAmong
subtasks possible
can besecondary
embeddedtasks, in thethe extrasubtask
earlier DoFs have thatbeen used for
has higher obstacle avoid-
priority.
anceAmong
in [16],possible
mechanical joint-limit avoidance in [17], minimization
secondary tasks, the extra DoFs have been used for obstacle of joint velocities
avoid- and
accelerations in [18], and reducing interaction forces in physical
ance in [16], mechanical joint-limit avoidance in [17], minimization of joint velocities and human-robot interaction
accelerations in [18], and reducing interaction forces in physical human-robot interaction
Robotics 2022, 10, x FOR PEER REVIEW 4 of 22
Robotics 2022, 11, 1 4 of 21

in [19]. In [20], the manipulability measure was used, and dynamic manipulability was
introduced
in [19]. In in [21].
[20], theMost methods formeasure
manipulability resolving wasredundancy
used, and in manipulation
dynamic involve de-
manipulability was
fining an objective function to satisfy specific additional tasks. In accordance,
introduced in [21]. Most methods for resolving redundancy in manipulation involve in the pro-
defin-
posed
ing anoptimization technique
objective function presented
to satisfy in additional
specific this paper,tasks.
thoseInobjective functions
accordance, in theare used
proposed
asoptimization
potential performance
technique indices to assign
presented in thisdesign
paper,constraints in thefunctions
those objective optimization problems
are used as po-
for structural
tential design of
performance manipulators.
indices to assign design constraints in the optimization problems for
structural design of manipulators.
3. Description of the New Mechanical Design Optimization Method for Robot Manip-
3. Description of the New Mechanical Design Optimization Method for
ulators
Robot Manipulators
To resolve the redundancy in robot manipulators, first, the manipulator has to be
To resolve
kinematically the redundancy
redundant in robot
with respect to themanipulators,
requirementsfirst,of thethe manipulator
task. That is thehasDoFtoofbe
kinematically redundant with respect to the requirements of the task.
the manipulator 𝑛 should be higher than the DoF needed for the task 𝑚. In the proposed That is the DoF of
the manipulator n should be higher than the DoF needed for the task m. In the proposed
optimization method, if 𝑛 = 𝑚, by including 𝑝 number of virtual joints, the modified
optimization method, if n = m, by including p number of virtual joints, the modified robot
robot arm has 𝑛 + 𝑝 DoF and becomes a redundant one. The additional virtual joint var-
arm has n + p DoF and becomes a redundant one. The additional virtual joint variables
iables represent the design parameters of a manipulator. In the structural synthesis of a
represent the design parameters of a manipulator. In the structural synthesis of a robot
robot arm, design parameter/s can be any Denavit–Hartenberg (DH) [22] parameter/s
arm, design parameter/s can be any Denavit–Hartenberg (DH) [22] parameter/s shown in
shown in Figure 2 other than the joint variable. Hence, (1) for link 𝑘 that is connected to
Figure 2 other than the joint variable. Hence, (1) for link k that is connected to link k − 1 via
link 𝑘 − 1 via a revolute joint, the possible design parameters are the effective link length
a revolute joint, the possible design parameters are the effective link length a , the twist
𝑎 , the twist angle 𝛼 and the relative offset between the two links defined kalong the
angle α and the relative offset between the two links defined along the revolute joint’s axis
revolute kjoint’s axis of rotation 𝑑 (2) for link 𝑘 that is connected to link 𝑘 − 1 via a
of rotation dk (2) for link k that is connected to link k − 1 via a prismatic joint, the possible
prismatic joint, the possible design parameters are the effective link length 𝑎 , the twist
design parameters are the effective link length ak , the twist angle αk and the relative rotation
angle
between𝛼 and the links
the two relative rotation
defined aboutbetween the two
the prismatic links
joint’s defined
motion axisabout
passing thefrom
prismatic
the DH
joint’s motion axis passing from the DH joint center 𝜃
joint center θk . When a virtual joint is included in the manipulator’s mechanismincluded
. When a virtual joint is to change
inthe
thedesign
manipulator’s
parameter/s,mechanism to changeof
the self-motion the
thedesign
robotparameter/s,
arm becomesthe self-motion ofofthe
a consequence the
robot arm becomes a consequence of
change of the selected design parameter/s. the change of the selected design parameter/s.

Possibledesign
Figure2.2.Possible
Figure designparameters
parametersdefined
definedwith
withrespect
respecttotoa amodified
modifiedDH
DHconvention.
convention.

AsAsa anext
next step,
step, a suitable
a suitable objective
objective function
function to minimized
to be be minimized or maximized
or maximized should
should be
be selected to optimize the selected design parameters, which are, in this case,
selected to optimize the selected design parameters, which are, in this case, the added the added
virtual joints. The objective functions are generally selected to represent a performance
virtual joints. The objective functions are generally selected to represent a performance
index defined for robot manipulators such as manipulability or condition number. However,
index defined for robot manipulators such as manipulability or condition number. How-
the designer can also formulate an objective function based on the requirements of the
ever, the designer can also formulate an objective function based on the requirements of
manipulator’s task.
the manipulator’s task.
The aim in this method is to calculate the optimal values of the virtual joints by
The aim in this method is to calculate the optimal values of the virtual joints by max-
maximizing or minimizing the objective function via a redundancy resolution algorithm
imizing or minimizing the objective function via a redundancy resolution algorithm and
and thus, selecting optimal design parameters. However, for the redundancy resolution
thus, selecting optimal design parameters. However, for the redundancy resolution algo-
algorithm to control the self-motion of the manipulator, joints should move at a finite
rithm to control the self-motion of the manipulator, joints should move at a finite rate.
rate. This means enough time should be provided during the optimization so that the self-
Robotics 2022, 11, 1 5 of 21

motion of the manipulator is completed, and the optimal values of the design parameters
are obtained. To ensure that there is enough time for self-motion of the manipulator
to be completed, the end-effector of the manipulator is fixed to a specific pose and the
convergence of the parameters is observed. This specific pose may be selected as a critical
or most visited pose of the manipulator. In this way it is possible to observe the changes
in the design parameters as the optimization process is carried out. Consequently, the
designer can gain insight into the effect of the design parameters on the performance of
the manipulator.

4. Optimization Methodology Described for the Passive Arm of the NeuRoboScope

System
In this section, the robot manipulator’s mechanism that is selected as a case study is
introduced. Later, the structural synthesis optimization of this robot arm mechanism is
explained by defining the specific optimization procedure applied in this case scenario.

4.1. Mechanism of the Robotic Manipulator and the Description of the Case Scenario
The manipulator that is considered for the case scenario is designed as a passive arm
of the NeuRoboScope surgical system (Figure 1). The NeuRoboScope system is designated
to work alongside the surgeon assisting him/her by handling the camera system, the
endoscope, throughout the surgery. The passive arm carries an active arm mounted on its
last link and the endoscope is attached to the active arm.
The passive arm has six revolute joints that are not actuated. The surgeon is expected
to backdrive the passive arm to locate the active arm at some desired poses during the
surgery. Accordingly, the joints of the passive arm are equipped with brake systems to
maintain their angular positions when desired by the surgeon.
The passive arm’s kinematic architecture is shown in Figure 3. The first two revolute
joints are for the planar motion on the horizontal plane. Third and fourth revolute joints
have axes parallel to the horizontal plane and they are interrelated to each other with
a parallelogram loop so that θ3 + θ4 = 2π. The other three joints compose the wrist
mechanism that is responsible for adjusting the orientation of the active arm’s base located
Robotics 2022, 10, x FOR PEER REVIEW 6 of 22
at the last link of the passive arm. In Figure 3, MP is identified as the manipulation point at
which the ease of manipulation of the passive arm is designated to be calculated.

Figure 3.
Figure 3. Kinematic scheme of
Kinematic scheme of the
the passive
passive robot
robot arm.
arm.

The parameters
DH position of MP is assigned
of the as the
passive arm aremaneuvering
provided in point
Table which is essential
1. In this table, a1 for being
length is
a critical
kept as toconsideration for optimization.
be designed (TBD) on purpose since in the case study, this is the link length that
( ) ( ) ( ) ( ) ( )
𝑝⃗ = 𝑎 𝑢⃗ + 𝑑 𝑢⃗ + 𝑎 𝑢⃗ + 𝑎 𝑢⃗ + 𝑎 𝑢⃗ (1)
In Equation (1), 𝑑 , 𝑎 , 𝑎 , and 𝑎 are assigned as fixed parameters. The only varia-
ble is the design parameter that is selected for this optimization study which is the effec-
Robotics 2022, 11, 1 6 of 21

is selected to be optimized for this manipulator. The eighth row of this table indicates the
coordinate frame selected for the active arm’s motion definition.

Table 1. DH parameters of the passive arm.

k dk θk ak αk
1 0 θ1 TBD 0
2 d2 θ2 a2 −π/2
3 0 θ3 a3 0
4 0 θ4 a4 π/2
5 d5 θ5 0 −π/2
6 0 θ6 a6 −π/2
7 d7 θ7 0 π/2
8 0 δ8 0 −π/2

The position of MP is assigned as the maneuvering point which is essential for being a
critical consideration for optimization.

→ → (1) → (1) → (2) → (3) → (4)

p MP = a1 u 1 + d2 u 3 + a2 u 1 + a3 u 1 + a4 u 1 (1)

In Equation (1), d2 , a2 , a3 , and a4 are assigned as fixed parameters. The only variable
is the design parameter that is selected for this optimization study which is the effective
link length of the first link, a1 .
The joint variables are θ1 , θ2 , θ3 , and θ4 , three of which are independent parameters so
that the position in the Cartesian space is to be considered as moving on a horizontal plane
as related to θ1 and θ2 . Apart from that, θ3 is related to the vertical motion and it is selected
as a constant value as explained previously. In this way, the passive arm is reduced into a
planar revolute-revolute (RR) manipulator. Consequently, the length of the new second
link is calculated as a2∗ = a2 + a4 + a3 cos θ3 .
The position of the MP is assigned as a constraint for design. Concerning that this
position on the horizontal plane and the base of the passive arm is fixed at a specific
point on the surgery table, two possible configurations can be calculated as elbow-up and
elbow-down. Any one of the two solutions can be selected to find the initial position for
each of θ1 and θ2 during the optimization study. The inverse kinematics solutions are not
presented here since it is trivial for a planar two DoF revolute-jointed arm.
For verifying and testing the objective function on the actual passive arm, the original
Jacobian matrix is a Ĵ ∈ <2×2 matrix for the planar arm with two DoF. However, for
optimization study, virtual joints are added to the original passive arm and the Jacobian
matrix is modified as Ĵ2×4 ∈ <2×4 and Ĵ2×3 ∈ <2×3 . The case with the two virtual joints
( Ĵ2×4 ) includes a prismatic joint acting along the y-axis and its positive direction motion
→ (0)
is defined along − u 2 axis by a virtual joint parameter Y0 . The other virtual joint in the
two-virtual joint case ( Ĵ2×4 ) and the single-virtual joint case ( Ĵ2×3 ) is the prismatic joint
included to change the effective link length, a1 . The dimension of the Jacobian matrix is
related to the DoF of the workspace and DoF of the original or modified passive arm. In
Equations (2)–(4), Jacobian matrices with two virtual joints, one virtual joint, and no virtual
joints are presented, respectively.
 
0 c1 − a1 s1 − ( a2 + a4 + a3 c3)s12 −( a2 + a4 + a3 c3)s12
Ĵ2×4 = (2)
−1 s1 a1 c1 + ( a2 + a4 + a3 c3)c12 ( a2 + a4 + a3 c3)c12
 
c1 − a1 s1 − ( a2 + a4 + a3 c3)s12 −( a2 + a4 + a3 c3)s12
Ĵ2×3 = (3)
s1 a1 c1 + ( a2 + a4 + a3 c3)c12 ( a2 + a4 + a3 c3)c12
 
− a1 s1 − ( a2 + a4 + a3 c3)s12 −( a2 + a4 + a3 c3)s12
Ĵ2×2 = (4)
a1 c1 + ( a2 + a4 + a3 c3)c12 ( a2 + a4 + a3 c3)c12
 𝑐1 −𝑎 𝑠1 − (𝑎 + 𝑎 + 𝑎 𝑐3)𝑠12 −(𝑎 + 𝑎 + 𝑎 𝑐3)𝑠12
𝐽 × = (3)
𝑠1 𝑎 𝑐1 + (𝑎 + 𝑎 + 𝑎 𝑐3)𝑐12 (𝑎 + 𝑎 + 𝑎 𝑐3)𝑐12

Robotics 2022, 11, 1 7 of 21

−𝑎 𝑠1 − (𝑎 + 𝑎 + 𝑎 𝑐3)𝑠12 −(𝑎 + 𝑎 + 𝑎 𝑐3)𝑠12
𝐽 × = (4)
𝑎 𝑐1 + (𝑎 + 𝑎 + 𝑎 𝑐3)𝑐12 (𝑎 + 𝑎 + 𝑎 𝑐3)𝑐12

InInthethe equations
equations above,
above, 𝑠𝑘 =
sk = sin θk 𝑠𝑖𝑛
and𝜃ck = cos𝑐𝑘
and θk =
for𝑐𝑜𝑠
k =𝜃 1, 2,
for 𝑘 =s12
3, and 1,2,3
= ,sin
and 𝑠12
( θ1 + θ2 )=
𝑠𝑖𝑛(𝜃 + 𝜃 ) and 𝑐12
and c12 = cos(θ1 + θ2 ). = 𝑐𝑜𝑠(𝜃 + 𝜃 ).

4.2.Design
4.2. DesignOptimization
OptimizationConstraints
Constraints
NeuRoboScopesystem
NeuRoboScope systemis is designed
designed to to
be be used
used in ainminimally
a minimally invasive
invasive pituitary
pituitary tumortu-
mor surgery.
surgery. In thisIn this surgery,
surgery, the natural
the natural openings openings
throughthrough theare
the nostrils nostrils
used are used surgical
to insert to insert
surgical
tools. tools.the
Hence, Hence, thearm
passive passive arm of NeuRoboScope
of NeuRoboScope system issystem is back-driven
back-driven manually manually
by the
by the surgeon
surgeon while thewhile the surgeon
surgeon places
places the the endoscope
endoscope in and outin and outnostril.
of the of the Concerning
nostril. Concern-
this
ing thisuse
special special
of theuse of thearm,
passive passive arm,
design design constraints
constraints are defined arebelow.
defined below.
1.1. Thesurgeon
The surgeoncancaninsert
insertthetheendoscope
endoscopefrom fromeither
eithernostril.
nostril.
2.2. The
Theendoscope
endoscopeand andthethe
active robot
active arm arm
robot should not interfere
should with the
not interfere surgeon’s
with hands,
the surgeon’s
and theyand
hands, should
they not block
should notthe surgeon’s
block view of the
the surgeon’s monitor,
view see Figure
of the monitor, see4.Figure 4.
3.3. The
The passive
passive arm
arm should
should locate
locate the
theactive
activearm arminside
insidethethesurgery
surgeryworkspace
workspace by byap-
ap-
proaching from behind the patient’s
proaching from behind the patient’s head. head.
4.4. The
Thepassive
passivearm
armshould
shouldbe befixed
fixedtotothe
thesurgery
surgerytable.
table.
5.5. Physical
Physicaldimensions
dimensions of of the
the links
links should
should not not bebe large,
large, and
andtheytheyshould
should not
not be
beheavy,
heavy,
but
butthey
theyshould
shouldbe berigid
rigidenough
enoughto tocompose
composean aninertial
inertialframe
frameforforthe
theactive
activearm
armwhen
when
their
theirbrakes
brakesatatthe
thejoints
jointsof ofthe
thepassive
passivearm armare areactivated.
activated.
6.6. There
Thereshould
shouldbebenonoactuators
actuatorson onthe
thejoints
jointsof ofthe
thepassive
passiverobot
robotarm.
arm.
7.7. When
When the passive arm’s brakes are released, the surgeon shouldbe
the passive arm’s brakes are released, the surgeon should beable
abletotomove
movethe
the
endoscope freely while the endoscope is still attached to the
endoscope freely while the endoscope is still attached to the active robot arm.active robot arm.

Figure4.4.The
Figure Thesurgery
surgeryroom
roomsetting
settingwith
withthe
themonitor,
monitor,the
theNeuRoboScope
NeuRoboScopesystem
systemand
andthe
thesurgeons.
surgeons.

AAstudy
studycan
canbe becarried
carriedoutoutfor
forall
allof
ofthe
theDHDHparameters,
parameters,excluding
excludingthethejoint
jointvariables,
variables,
of the passive arm. However, in this study, the passive arm’s motion
of the passive arm. However, in this study, the passive arm’s motion on the horizontal on the horizontal
planeisisconsidered
plane considered to to facilitate
facilitate thethe demonstration
demonstration of optimization
of the the optimization method.
method. Accord-
Accordingly,
ingly,
from from
this thison,
point point
the on, the kinematics
kinematics of the passive
of the passive arm arearm are reduced
reduced to a planar
to a 2-DoF 2-DoF planar
robot
robot
arm armtwo
with with two revolute
revolute joints. joints.
DesignDesign parameters
parameters that
that are are considered
considered are
are the the effective
effective link
link length
length of the of thelink
first first link
and theand the ground
ground frame’sframe’s origin,iswhich
origin, which is the location
the location that the
that the passive
passive
arm armon
is fixed is the
fixed on thetable.
surgery surgery table.

4.3. Optimization through Mechanical Redundancy

Depending on the previously defined design constraints, the requirements for the
passive arm are set for the optimization procedure by considering the necessities and
conditions of the surgery:
 4.3. Optimization through Mechanical Redundancy
Depending on the previously defined design constraints, the requirements for the
passive arm are set for the optimization procedure by considering the necessities and con-
ditions of the surgery:
Robotics 2022, 11, 1 8 of 21
1. The surgeon should have minimal effort when he/she intends to push the active robot
arm in or away from the surgery zone.
2. The parallelogram loop in the passive robot arm is utilized with no modifications
1. The surgeon should have minimal effort when he/she intends to push the active robot
since it is designed with counter-spring for gravity compensation. This linkage is re-
arm in or away from the surgery zone.
sponsible for providing vertical motion of the base of the active arm.
2. The parallelogram loop in the passive robot arm is utilized with no modifications
3. The optimization is related only to the ease of manipulation only on the horizontal
since it is designed with counter-spring for gravity compensation. This linkage is
plane.
responsible for providing vertical motion of the base of the active arm.
4. The fixing point position on the y-axis (with respect to reference-frame in Figure 5) is
3. The optimization is related only to the ease of manipulation only on the horizontal plane.
selected as a possible design parameter, which is related to another design parameter
4. The fixing point position on the y-axis (with respect to reference-frame in Figure 5) is
that is the first link’s length.
selected as a possible design parameter, which is related to another design parameter
5. MP position is fixed at the coordinate (−20, −30) cm (which can be considered as an
that is the first link’s length.
average position of workspace required by the surgeon) relative to the reference
5. MP position is fixed at the coordinate (−20, −30) cm (which can be considered as
frame
an in Figure
average 5. of workspace required by the surgeon) relative to the reference
position
6. frame in Figure 5. length of the first link should be limited depending on its manu-
The effective link
6. facturability,
The finallength
effective link weightofandthe allowed
first linkcompliant
should be displacements due on
limited depending to loads.
its manufac-
7. The linear density of the first link is taken as follows: mass/length
turability, final weight and allowed compliant displacements due to loads. = 1 kg/m.
8.
7. The linear
The third joint
density of the 𝜃firstinlink
variable Figure 3 is fixed
is taken at −30°
as follows: which is the= condition
mass/length 1 kg/m. when
8. the endoscope is located just above the patient’s nostrils.
◦
The third joint variable θ in Figure 3 is fixed at −30 which is the condition when the
3
Since the objective
endoscope is locatedofjust
theabove
optimization is to nostrils.
the patient’s ease the manipulation at MP, which is
denoted in Figure 3, forward and inverse kinematics,
Since the objective of the optimization is to ease theandmanipulation
the Jacobian matrix calculations
at MP, which is de-
are presented for MP. Hence, these calculations are used for the proposed optimization
noted in Figure 3, forward and inverse kinematics, and the Jacobian matrix calculations are
method. for MP. Hence, these calculations are used for the proposed optimization method.
presented

Figure5.5.Coordinate
Coordinatesystem
systemfixed
fixed → (0)
Figure onon
thethe surgery
surgery table
table (where
(where the the
unitunit vector
vector alongalong the x-axis
the x-axis is u 1 is
( ) ( )
𝑢⃗ and the unit vector along the y-axis
→(0is) 𝑢⃗ ).
and the unit vector along the y-axis is u 2 ).

5. Implementation
5. Implementation of of the
the New
New Optimization
Optimization Strategy
Strategy
When there
When thereisiskinematic
kinematicredundancy,
redundancy,therethereisisa afree
free motion
motion ofof
thethe mechanism
mechanism even
even if
if the
the end-effector
end-effector is fixed.
is fixed. ThisThis so-called
so-called self-motion
self-motion happenshappens
in theinnull
thespace
null space
of the of the Ja-
Jacobian
cobian matrix.
matrix. In the implementation
In the implementation of the proposed
of the proposed optimization
optimization strategy,strategy, this property
this property is used.
is used. However, to use this property, the non-redundant
However, to use this property, the non-redundant passive arm must be modifiedpassive arm must be modified
to have
to have
more more joints.
joints.
For the
For the case
case with
with twotwo virtual
virtual joints,
joints, the DoF of the RR planar arm is increased by 2.
Consequently,a a4-DoF
Consequently, 4-DoF PRPR
PRPR planar
planar armarm operating
operating for afor a 2-DoF
2-DoF planar
planar task istask is formed.
formed. Since
the objective is to find optimum design parameters, these new joint variables (Y0 and a1 )
are included in the control of the self-motion of the resultant redundant arm.
Although a designer is free to choose any redundant manipulator controller, a previ-
ously designed controller for redundant robot manipulators is utilized for this optimization
task [18]. However, only the kinematic part of this controller is used in this work. This
controller is used to control both the main task in task-space, earlier defined by the MP
Robotics 2022, 11, 1 9 of 21

point’s position x in horizontal plane, and a desired subtask by adjusting the joints’ motion
in the null space.
An error term is defined as e = x d − x as the tracking error, and x d is defined as
the desired position/trajectory in task-space. The designed controller is presented in
Equation (5) where K̂v and K̂ p are diagonal constant feedback gain matrices related to the
proportional-derivative (PD) controller of the main task.
..

.. . . . ..
+
q = Ĵ x d + K̂v e + K̂ p e − Ĵ q + β N (5)

The column of joint variables in position level are denoted as q ∈ <n , where n, from
now on, is the summation of actual and virtual joints in this approach. We consider that
a vector function z(.) ∈ <n is calculated as a gradient of optimization objective function
f (q) for a specific optimization objective function (which may be time-dependent function,
including design constraints, etc.), and joint velocities in the null space are required to track
the projection of z onto the null space of Ĵ. Since În − Ĵ + Ĵ projects vectors onto the null
space of Ĵ, this can be formulated in an error signal calculation as presented in Equation (6),
which converges to zero. Here, Ĵ + represents the pseudo-inverse of Ĵ and În ∈ <n×n is the
identity matrix.
. .
e N = În − Ĵ + Ĵ z − β N


(6)
Assuming the manipulator does not go through a singularity condition, it is needed to
.. ..
design β N to obtain the desired result for the subtask objective. β N is determined as;
..
.
 . 
.
β N = În − Ĵ + Ĵ z − Ĵ + Ĵ Ĵ + + Ĵ + Ĵz + K̂ N e N (7)

In Equation (7), K̂ N is a diagonal positive definite feedback matrix. This designed

control law guarantees that the error will be bounded and converged to zero [18]. After
the description of the controller to be used in the proposed optimization method, the
performance indices that are used to formulate the objective function are explained in the
next subsections. It should be noted that instead of the performance metrics mentioned
below, the objective function can be developed by considering other performance metrics
and/or their variations.

5.1. Manipulability Ellipsoid and Singular Value Decomposition (SVD)

Manipulability of a selected point in a mechanical linkage can be represented as a
scalar value related to the area/volume of velocity ellipse/ellipsoid calculated at this point.
It is first developed in [20] and introduced as a performance index. Since the motion
of any point on a mechanical linkage can be related to the motion of the joints by the
Jacobian matrix, the scalar representation of the manipulability index Mp is provided in
the following equation. q
Mp = det( Ĵ Ĵ T ) (8)
In addition to the manipulability index shown in Equation (8), another manipulability
measure in Equation (9) was also formulated by Paul and Stevenson [23] as follows

Mp = |det( Ĵ )|. (9)

The objective set when using the manipulability index is to maximize this value via
changing the positions of joints by staying inside the null-space when MP is fixed at the
desired position. During this motion, the effective link length of the first link (a1 ) and the
position of the fixing point of the passive arm (Y0 ) will be changing to reach the optimum
value in this optimization.
Robotics 2022, 11, 1 10 of 21

Singular value decomposition on the matrix Ĵ Ĵ T is used to solve for singular values
of Ĵ, which also represents the semi-axes of manipulability ellipse during the simulation.
The rank of Ĵ represents the number of singular values, which is two in this case be-
cause in the formulation of the objective function, the Jacobian matrix is related to the
original manipulator.

5.2. The Modified Condition Number

Another way to represent force/motion relation of a point in the workspace of a
mechanism is done by a scalar number called condition number [24]. Manipulability
represents ease of manipulation of the end-effector at a certain location of the workspace
and the condition number relates the length of the maximum and minimum axes of
manipulability ellipse achieved along with different directions at that specific location of
the end-effector. The condition number is represented by the ratio of the maximum singular
value to the minimum singular value of the Jacobian matrix, which represents the radii of
the manipulability ellipse. The distance between any point on the ellipse and its center was
defined in [25] as the velocity transmission ratio on one specific direction of motion of the
end-effector. In an ideal case, the performance index is 1, which means at that specific point,
the velocity transmission ratio is the same in all directions and a circle will be representing
the manipulability.
Force condition number and velocity condition number are calculated in similar ways.
Salisbury and Craig [24] introduced force condition number as the amplification of the
relative force error at the task-space to the relative torque error at the joint-space. While
Merlet [26] described a velocity condition number using relative motion error in joint-space
and relative motion error in the task-space.
In this work, the difference between the maximum and minimum singular values of
the Jacobian matrix (σmax and σmin , respectively) is used instead of the condition number
and it will be referred as the modified condition number from this point on. The objective
function shown in Equation (10) is used in the optimization procedure to be minimized to
a minimum value or if possible zero value.

Cn = σmax − σmin (10)

5.3. Generalized Inertia Matrix

To include a dynamic orientated design constraint for the design of the passive arm,
the dynamic model of the arm is studied via finding its dynamic equation of motion and the
generalized inertia matrix. This is essential to find the dominant part of design parameters
in the relation between the forces displayed at the end-effector and the consequent motion
of the manipulator. In this work, since the end-effector is moved at a slow rate, Coriolis
and centripetal forces, and the viscous frictional forces are neglected. Consequently, the
remaining part of the dynamic equation is shown below.
..
τ = M̂ q (11)

In Equation (11), τ is the column of actuator torques/forces acting on the joints and M̂
is the generalized inertia matrix. To represent Equation (11) in the task-space, where the
interaction is taking place, the task-space forces are mapped to the joint space forces by
using the Jacobian matrix as follows:

τ = Ĵ T F. (12)

Here, in Equation (12), F is the external forces acting at the end-effector. For the
. .. ..
slow-motion of the end-effector, where q → 0 , we consider q = Ĵ −1 x. Subsequently,
..
Ĵ T F = M̂ Ĵ −1 x. (13)
Robotics 2022, 11, 1 11 of 21

As a result of this expression in Equation (13), mapped generalized inertia matrix

(mGI M) M̂G is defined in the task-space as shown in Equation (14).

Ĵ −T M̂ Ĵ −1 = M̂G (14)

The dynamic performance of the passive arm can be represented by the ellipse (in the
case of the 2-DoF manipulator) which can be plotted for the eigenvalues and eigenvectors
of M̂G matrix [27]. On the other hand, the dynamic manipulability (the determinant of
Ĵ M̂−1 ) as defined in [21] by Yoshikawa relates the loads at actuators to the acceleration
output at end-effector as shown in Equation (15).
..
x = Ĵ M̂−1 τ (15)

For the design optimization study of the passive arm, the objective is to minimize the
resistance shown to the operator during he/she backdrives the arm, which can be termed
as the mechanical impedance [28] of the passive arm. By doing so, the end-effector can be
moved freely inside its workspace with minimum force reflected to the user. This can be
achieved by either minimizing the determinant of the numerator part of M̂G in Equation (14)
hence the generalized inertia matrix ImN (previously defined as M̂), and/or maximizing
the determinant of denominator part of M̂G , which corresponds to the manipulability
index ImD since det( Ĵ T Ĵ ) = det( Ĵ Ĵ T ). In this work, the mapped generalized inertia matrix
M̂G and its abovementioned numerator (ImN) and denominator parts (ImD) are used as
indicators of the dynamic performance measure while selecting the objective function.

6. Simulation Tests and Results

Two simulation tests are conducted to verify the presented approach. Two design
parameters are used for both tests. The Jacobian matrix developed for the MP of the
actual passive arm (RR manipulator version) is used to determine the desired objective
function that is related to the modified condition number. All tests are carried out in Matlab
Simulink setting the fixed step calculation frequency to 100 Hz with an ODE3 solver. The
fixed position of MP is selected as a design constraint on the horizontal planar workspace
at −20, −30 cm for this particular scenario of the presented case study, which is determined
relative to the frame described in Figure 5. In all the tests, the initial value of the first link
is chosen as a1 = 30 cm and the initial value of fixing position (first joint’s axis location)
along the y-axis is selected to be Y0 = 0 cm. The other parameters of the passive arm are
assigned as fixed parameters.

6.1. Simulation Test with the Modified Condition Number Performance Index
In this test, the modified condition number and the manipulability index is calculated
to visualize the effect of the optimization. Both the fixing point of the manipulator along
the y-axis and the first link’s length are selected as design parameters.
By using the modified condition number, which is presented in Equation (10), as the
only performance index in forming the objective function f (q) = −Cn, singular values
are forced to be equal during the optimization procedure due to the minimization of the
objective function. As a result, the manipulability ellipse is forced to be a circle and the
robot arm moves into an isotropic pose as can be seen in Figure 6.
This result is the same result that is presented and discussed in [29]. The obtained
results are exactly as expected for the isotropic pose. The √ optimal length of the first
link came out to be a1 = 38.47 cm which is equal to a1 = a2 2. The fixing point position
converged to a position at Y0 = −11.57 cm. It is observed in Figure 7 that the manipulability
index decreases while the modified condition number index increases which is making
maximum and minimum singular values to be equal.
Robotics 2022,
Robotics 2022, 11,
10, 1x FOR PEER REVIEW 1212of
of 22
21

Figure 6. Change of the robot arm structure and the manipulability ellipse (printed in blue color)
during the optimization routine by using the modified condition number.

This result is the same result that is presented and discussed in [29]. The obtained
results are exactly as expected for the isotropic pose. The optimal length of the first link
came out to be 𝑎 = 38.47 cm which is equal to 𝑎 = 𝑎 √2. The fixing point position con-
verged to a position at 𝑌 = −11.57 cm. It is observed in Figure 7 that the manipulability
index decreases
Figure 6.
Figure 6. Change of
Change while
of the the modified
the robot
robot arm condition
arm structure
structure and thenumber
and the index ellipse
manipulability
manipulability increases
ellipse which
(printed
(printed is making
in blue
in blue color)
color)
maximum
during and minimum
the optimization singular
routine values
by using to be equal.
the modified condition
condition number.
number.

This result is the same result that is presented and discussed in [29]. The obtained
results are exactly as expected for the isotropic pose. The optimal length of the first link
came out to be 𝑎 = 38.47 cm which is equal to 𝑎 = 𝑎 √2. The fixing point position con-
verged to a position at 𝑌 = −11.57 cm. It is observed in Figure 7 that the manipulability
index decreases while the modified condition number index increases which is making
maximum and minimum singular values to be equal.

Figure7.
Figure 7. Variation
Variationof
ofthe
the manipulability
manipulabilityindex
index and
and singular
singular values
values during
during the
the optimization
optimization routine
routine
by using the modified condition number.
by using the modified condition number.

As another consideration of the optimization process, the variation of inertial proper-

ties of the manipulator is investigated. In Figure 8a, it can be noticed that the determinant
of the mapped generalized inertia matrix (mGIM) is increased because of this optimization.
This is due to the increase in the first link length and decrease in overall manipulability.
In addition,
Figure the determinant
7. Variation of the generalized
of the manipulability inertia values
index and singular matrixduring
(the numerator of mGIM)
the optimization is
routine
increased as can be seen in Figure
by using the modified condition number.8b, which is represented by ImN in this figure. ImD rep-
resents manipulability (the denominator of mGIM). As the objective function is minimized,
 erties of the manipulator is investigated. In Figure 8a, it can be noticed that the determi-
nant of the mapped generalized inertia matrix (mGIM) is increased because of this opti-
mization. This is due to the increase in the first link length and decrease in overall manip-
ulability. In addition, the determinant of the generalized inertia matrix (the numerator of
Robotics 2022, 11, 1 mGIM) is increased as can be seen in Figure 8b, which is represented by ImN in this13fig- of 21
ure. ImD represents manipulability (the denominator of mGIM). As the objective func-
tion is minimized, the modified condition number becomes zero since in this work, it is
described as the
the modified difference
condition between
number the maximum
becomes zero sinceand minimum
in this work, it singular value
is described as of
thethe
dif-
manipulability matrix.
ference between the maximum and minimum singular value of the manipulability matrix.

Figure
Figure8.8.The optimization
The procedure
optimization with
procedure the modified
with condition
the modified number:
condition (a) Variation
number: of the of
(a) Variation gen-
the
eralized inertia matrix, (b) variation of components of the objective function, (c) variation in the
generalized inertia matrix, (b) variation of components of the objective function, (c) variation in the
values
valuesofofdesign
designparameters.
parameters.

During
Duringthe
theoptimization
optimizationprocess,
process, design parameters
parameters couldcouldconverge
convergetotoconstant
constantvalues.
val-
ues. As shown in Figure 8c, the length of the first link is increased due to
As shown in Figure 8c, the length of the first link increased due to this optimization,this optimization,
which
whichleads
leadstotoa adecrease
decrease inin
manipulability
manipulability and as as
and a result increase
a result in the
increase determinant
in the of
determinant
𝑚𝐺𝐼𝑀.
of mGIThis
M. This
can be considered
can as a drawback
be considered but the
as a drawback but objective in using
the objective the modified
in using con-
the modified
condition
dition number
number is to is to result
result in aninisotropic
an isotropic
posepose in terms
in terms of manipulability
of manipulability index.
index.

6.2.Simulation
6.2. SimulationTest
Testwith
withthe
theModified
ModifiedCondition
ConditionNumber
NumberPerformance
PerformanceIndex
Indexand
andGeneralized
Generalized
Inertia Matrix
Inertia Matrix
In this final test, modified condition number and the numerator part of the mapped
In this final test, modified condition number and the numerator part of the mapped
generalized inertia matrix, which corresponds to the generalized inertia matrix, are used
generalized inertia matrix, which corresponds to the generalized inertia matrix, are used
with selected weights of w = 1 and w = 2 in forming the objective function, respec-
with selected weights of 𝑤 1= 1 and 𝑤 2= 2 in forming the objective function, respec-
tively. In this way, the objective function is modified to have both the effect of mod-
tively. In this way, the objective function is modified to have both the effect of modified
ified condition number and the determinant of the generalized inertia matrix ImN as
condition number and the determinant of the generalized inertia matrix 𝐼𝑚𝑁 as 𝑓(𝑞̅) =
f (q) = −w1 Cn − w2 ImN. As a result, shorter length for the first link is obtained and ma-
−𝑤 𝐶 − 𝑤 𝐼𝑚𝑁. As a result, shorter length for the first link is obtained and manipulabil-
nipulability ellipse is reshaped as can be noticed in Figure 9. In Figures 6 and 9: (1) the
ity ellipse is reshaped as can be noticed in Figure 9.
red line indicates the first link, and the black line indicates the second link, (2) the joint
centers for the first and second joint are indicated with red and black circles, respectively
(3) during the optimization process, the link colors are drawn darker as the links move
from their initial states to their final states.
 Robotics2022,
Robotics
Robotics 2022,11,
2022, 10,1xxFOR
10, FORPEER
PEERREVIEW
REVIEW 14
14 of 21
14 of 22
22

Figure
Figure9.
Figure 9.Variation
9. Variationof
Variation ofthe
of therobot
the robotarm
robot armstructure
arm structureand
structure andthe
and themanipulability
the manipulabilityellipse
manipulability ellipse(printed
ellipse (printedin
(printed inblue
in bluecolor)
blue color)
color)
during
during the
the optimization
optimization routine
routine by
by using
using thethe modified
modified condition
condition number
number and and
the the generalized
generalized
during the optimization routine by using the modified condition number and the generalized inertia inertia
matrix.matrix.
inertia
matrix.

InInthis
In thisoptimization
this optimizationprocedure,
optimization procedure,due
procedure, dueto
due tothe
to thedecrease
the decreasein
decrease inmanipulability,
in manipulability,which
manipulability, whichisis
which isshown
shown
shown
in Figure and the decrease in the determinant of generalized
in Figure 10, and the decrease in the determinant of generalized inertia matrix 𝐼𝑚𝑁, which
in Figure 10, and the decrease in the determinant of generalized inertia
inertia matrix 𝐼𝑚𝑁,
matrix ImN,
which
which
is shown
is is
shown in shown
in Figure in
Figure 11b,Figure
11b, the 11b, the
the determinantdeterminant
determinant of 𝑚𝐺𝐼𝑀, as
of 𝑚𝐺𝐼𝑀, of mGI M,
as shown
shown inas shown
in Figure in
Figure 11a, Figure
11a, is 11a,
is increased
increasedis
increased
more relative
more more
relative to relative
to the to
the results the results
results presented
presented inpresented
in Section in
Section 6.1. Section
6.1. The 6.1.
The positiveThe positive
positive result
result of result
of this of this
this optimiza-
optimiza-
optimization
tion is obtainingis obtaining
a smaller a smaller
link lengthlinkforlength
the for
first the
link, first
which link,
is
tion is obtaining a smaller link length for the first link, which is indicated in Figure 11c. which
indicated is indicated
in Figure in
11c.
Figure 11c.
Nevertheless, Nevertheless,
the weights the
usedweights
for the used for
influence theofinfluence
the of
modified
Nevertheless, the weights used for the influence of the modified condition number and the modified
condition condition
number and
number
the inertia
inertiaandmatrix
the inertia thematrix
on the on the
objective objective
function function
is critical.
critical. Theischoice
critical.of The
choice thesechoice
weights of these
could
the matrix on objective function is The of these weights could
weights could decrease
decrease manipulability
manipulability in manipulability
in one
one direction in
direction to one direction
to aa value
value near to a
near zero,value
zero, whichnear
which wouldzero, which
would minimize would
minimize the the
decrease
minimize the backdrivability
backdrivability of the
the manipulatorof the manipulator
manipulator in that
that direction.in that direction.
direction.
backdrivability of in

Figure10.
Figure
Figure 10.Variation
10. Variationofof
Variation ofthe
the
the manipulability
manipulability
manipulability index
index
index and
and
and singular
singular values
values
singular during
during
values the
thethe
during optimization
optimization pro-
process
optimization pro-
cess
cess
by byusing
by
using using themodified
the modified
the modified condition
condition
condition numbernumber
number and
and
and the theinertia
the
inertiainertia matrix.
matrix.
matrix.
Robotics 2022, 11, 1 15 of 21
Robotics 2022, 10, x FOR PEER REVIEW 15 of 22

Figure11.
Figure 11. The
Theoptimization
optimizationprocedure
procedurewith modified
with condition
modified number
condition and inertia
number matrix: matrix:
and inertia (a) Var-
iation
(a) of theofgeneralized
Variation inertiainertia
the generalized matrix, (b) variation
matrix, of components
(b) variation of the of
of components objective function,
the objective (c)
func-
variation
tion, in the values
(c) variation in the of design
values parameters.
of design parameters.

7.7.Validation
Validationofofthe
theProposed
ProposedOptimum
OptimumDesign
DesignApproach
Approach
Toease
To easethethecomparison
comparisonofofour ouroptimization
optimizationapproach,
approach,we wenamed
namedour ouroptimization
optimization
approach
approachasasOptimization
Optimization viavia
Redundancy
Redundancy Resolution
Resolution (ORR).
(ORR).To show
To showthe accuracy
the accuracy of theof
proposed
the proposed optimization
optimizationapproach, we compared
approach, we comparedthe results obtained
the results by stochastic
obtained optimiza-
by stochastic op-
tion algorithms
timization Simulated
algorithms Annealing
Simulated (SA) algorithm,
Annealing DifferentialDifferential
(SA) algorithm, Evolution (DE) algorithm,
Evolution (DE)
Nelder–Mead
algorithm, Nelder–Mead(NM) algorithm, (NM) Random
algorithm, Search
Random (RS)Search
algorithm, based on four
(RS) algorithm, based different
on four
phenomenological search approaches, and that of ORR for
different phenomenological search approaches, and that of ORR for the mechanism the mechanism design problem de-
we
signdefined.
problem Here,we the basic Here,
defined. rationale
the for choosing
basic rationale these
for algorithms
choosing these for validation
algorithms is for
(i) they
vali-
have
dationproven theirhave
is (i) they reliability
proven in their
the solution of mathematical
reliability in the solutionoptimization
of mathematical problems, which
optimization
have been used
problems, which in have
manybeen different
used disciplines, (ii) because
in many different they have
disciplines, differentthey
(ii) because phenomeno-
have dif-
logical
ferent bases, a solution alternative
phenomenological that an alternative
bases, a solution algorithm might that anmiss can bemight
algorithm compensated
miss canby be
this way. At this stage, two different optimization scenarios
compensated by this way. At this stage, two different optimization scenarios were defined: First, to were
find thede-
results
fined: that
First,minimize
to find the only the Cthat
results n parameter,
minimizeand onlyin the
this𝐶context, optimize
parameter, andthe parameters
in this context,
Yoptimize
, θ , θ and a
0 1 2 the parameters 1 under the given nonlinear equality and linear inequality
𝑌 , 𝜃 , 𝜃 and 𝑎 under the given nonlinear equality and linear constraints; The
second
inequalitywas constraints;
to define theThe importance
second was levels
to define Cn and
of the the ImN parameters
importance with𝐶theand
levels of the weight
𝐼𝑚𝑁
values of w and w , to define a new objective function
parameters with the weight values of 𝑤 and 𝑤 , to define a new objective func-
1 2 (w 1 C n + w 2 ImN) and to use the
constraints used in the first scenario. Weights are selected as w
tion(𝑤 𝐶 + 𝑤 𝐼𝑚𝑁) and to use the constraints used in the first scenario. Weights are se-
1 = 1 and w 2 = 2 in forming
the objective
lected as 𝑤 function,
= 1 and 𝑤 respectively.
= 2 in formingThe unit for the distances
the objective function,isrespectively.
m and the unit The forunitthe
for
angular positions is rad in description of the two optimization scenarios.
the distances is m and the unit for the angular positions is rad in description of the two
Find design
optimization variables: Y0 , θ1 , θ2 , a1
scenarios.
Scenario 1:
Find design variables: 𝑌 , 𝜃 , 𝜃 , 𝑎
To minimize the objective function: Cn (Y0 , θ1 , θ2 , a1 )
Scenario 1:
Scenario 2:
To minimize the objective function: 𝐶 (𝑌 , 𝜃 , 𝜃 , 𝑎 )
To minimize the objective function: w1 Cn (Y0 , θ1 , θ2 , a1 ) + w2 ImN (Y0 , θ1 , θ2 , a1 )
Scenario 2:
To minimize the objective function: 𝑤 𝐶 (𝑌 , 𝜃 , 𝜃 , 𝑎 ) + 𝑤 𝐼𝑚𝑁(𝑌 , 𝜃 , 𝜃 , 𝑎 )
Robotics 2022, 11, 1 16 of 21

Subjected to: −1 ≤ Y0 ≤ 0, 0 < a1 < 0.5

a2 = 0.350;
a3 = 0.216;
a4 = 0.05;
θ3 = −30π/180;
X0 = 0;
a2∗ = a2 + a4 + a3 cos θ3 ;
[ X MP , YMP ] = [ − 0.2, −0.3]
X MP = a1 cos(θ1 ) + a2∗ cos(θ1 + θ2 );
YMP = Y0 + a1 sin(θ1 ) + a2∗ sin(θ1 + θ2 );
where
Cn (Y0 , θ1 , θ2 , a1 ) = 12 ((( a21 + ( a41 + 1.088a31 cos(θ2 ) + 0.592a21 cos2 (θ2 ) + 0.161a1 cos(θ2 ) + 0.022)0.5
+0.544a1 cos(θ2 ) + 0.148)
−(( a21 − ( a41 + 1.088a31 cos(θ2 ) + 0.592a21 cos2 (θ2 ) + 0.161a1 cos(θ2 ) + 0.022)0.5 )
+0.544a1 cos(θ2 ) + 0.148)0.5 )0.5 )2
ImN (Y0 , θ1 , θ2 , a1 ) = 0.172707a21 + 0.100746a31 − 0.0740175a21 cos2 (θ2 )
In the Appendix A, summary information about the working logic of the optimization
algorithms we use for validation is given. For more detailed information, you can refer to
the relevant reference [5].
For the optimization problems solved for the scenarios, the algorithm options given in
Table 2 are used.

Table 2. Corresponding options for the optimization algorithms DE, NM, RS, and SA.

Options DE NM RS SA
Crossover
0.5 - - -
fractions
Random Seed 1 5/10 0 2
Scaling factor 0.6 - - -
Tolerance 0.001 0.001 0.001 0.001
Contact ratio - 0.5 - -
Expand ratio - 2.0 - -
Reflect ratio - 1.0 - -
Shrink ratio - 0.5 - -
Level iterations - - - 50
Perturbation
- - - 1.0
scale
Penalty Function - - Automatic -
Search Points - - 2 -
Method - - Interior Point -

The obtained results using the four well-established optimization algorithm’s results
are tabulated along with the result obtained by the proposed ORR algorithm. Due to the
“RandomSeed” dependency, which is one of the advantages of the NM algorithm, it is
possible to produce alternative results when two different choices (5 and 10) are made for
this problem. Therefore, the results for two version of NM are also presented. Tables 3 and 4
present the result obtained for Scenarios 1 and 2, respectively.
In Table 3, the solution obtained in DE algorithm is the positive solution alternative
for the RR manipulator. Therefore, the obtained optimization result in DE is identical to the
ones obtained via SA, NM2, RS and the proposed ORR algorithm. The result obtained via
NM1 algorithm is different than the other ones. However, when the constraint is narrowed
to −0.4 ≤ Y0 ≤ 0, this algorithm also gives the same result with the others.
Robotics 2022, 11, 1 17 of 21

Table 3. The optimization results for Scenario 1 obtained via DE, NM, RS, SA and ORR algorithms.

Optimization Algorithm Objective Function Y0 (cm) θ1 (rad) θ2 (rad) a1 (cm)

SA 2.21008 × 10−14 −11.56 −1.611 −2.356 38.48
DE 1.21496 × 10−13 −11.56 3.079 2.356 38.48
NM1 8.39033 × 10−15 −48.44 1.611 2.356 38.48
NM2 2.21008 × 10−14 −11.56 −1.611 −2.356 38.48
RS 2.21008 × 10−14 −11.56 −1.611 −2.356 38.48
ORR 5.03038 × 10−9 −11.57 −1.611 −2.356 38.47

Table 4. The optimization results for Scenario 2 obtained via DE, NM, RS, SA and ORR algorithms.

Optimization Algorithm Objective Function Y0 (cm) θ1 (rad) θ2 (rad) a1 (cm)

SA 0.0363314 −30.00 −1.998 −2.408 28.50
DE 0.0363314 −30.00 1.998 2.408 28.50
NM1 0.0363314 −29.997 −1.998 −2.408 28.49
NM2 0.0363314 −29.994 1.998 2.408 28.50
RS 0.0363314 −30.00 1.998 2.408 28.50
ORR 0.036349 −26.57 −1.835 −2.398 28.55

When the results in Table 4 are considered, although the proposed ORR approaches to
the optimization results of the other well-established algorithms, there is still a considerable
difference for the Y0 value. As ORR is developed based on a control algorithm using
redundancy resolution, the convergence to the optimal result can take some time due to
the characteristics of the robot mechanism’s controller. The controller can be optimized
to converge to the end result faster. To test this, the same controller is used for extended
simulation durations. The obtained optimization results for Scenario 2 are presented in
Table 5.

Table 5. The optimization results for Scenario 2 obtained via ORR algorithm for different simulation
durations.

Method—Duration Objective Function Y0 (cm) θ1 (rad) θ2 (rad) a1 (cm)

ORR—200 s 0.0363496 −26.57 −1.835 −2.398 28.55
ORR—300 s 0.0363404 −27.22 −1.864 −2.402 28.54
ORR—400 s 0.0363369 −27.63 −1.883 −2.404 28.53
ORR—500 s 0.0363351 −27.92 −1.897 −2.405 28.52
ORR—800 s 0.0363329 −28.48 −1.923 −2.406 28.51
ORR—1600 s 0.0363317 −29.22 −1.959 −2.408 28.50

It is observed in Table 5 that the design parameters converge to their optimal values
as the simulation duration is extended. This convergence trend is expected since in an
actual robot control case, it takes a certain amount of time for the robot to move to the
desired location.

8. Discussion and Conclusions

A new approach of mechanical optimization of robot manipulators through mechanical
redundancy and the use of redundancy resolution algorithm designed for the control of
redundant robots is presented in this paper. The advantages of this approach with respect
to the other design optimization methods are (1) the possibility of visualization of the
robot’s configuration variations during the optimization, (2) selectively optimizing specific
structural parameters of the robot manipulator, and (3) during the optimization process,
the design parameters are continuously changed until they approach near the optimal
results and no randomness exists during the optimization. In this way, the designer can
perceive the effects of the selected performance index and the selected structural parameter
for optimization.
 optimal results and no randomness exists during the optimization. In this way, the de-
signer can perceive the effects of the selected performance index and the selected struc-
Robotics 2022, 11, 1
tural parameter for optimization. 18 of 21
Figure 12 shows the Pareto chart for two performance indices to be minimized (the
modified condition number, 𝐶 , and the determinant of the generalized inertia matrix,
𝐼𝑚𝑁,) by using the selected design parameters within their design limits of 𝑌 = −0.3 m →
Figure 12 shows the Pareto chart for two performance indices to be minimized (the
0 m and 𝑎 = 0 m → 0.4 m with the discrete step size of 0.01 m. Within these ranges, a
modified condition number, Cn , and the determinant of the generalized inertia matrix, ImN),
Pareto set of 961 individual solutions are evaluated for the corresponding performance
by using the selected design parameters within their design limits of Y0 = −0.3 m → 0 m and
indices which are presented with “x” mark on the figure. The distribution of the Pareto
a1 = 0 m → 0.4 m with the discrete step size of 0.01 m. Within these ranges, a Pareto set of
set shows the opposing requirements of the two performance indices. In this constrained
961 individual solutions are evaluated for the corresponding performance indices which
multi-objective optimization problem, the procedure described in test B is selected for ob-
are presented with “x” mark on the figure. The distribution of the Pareto set shows the
serving the new optimization method’s performance via the Pareto set. The execution of
opposing requirements of the two performance indices. In this constrained multi-objective
the proposed optimization method is printed on the figure which initiates from the point
optimization problem, the procedure described in test B is selected for observing the new
marked with the red circle and follows the blue line to terminate at the red plus mark. The
optimization method’s performance via the Pareto set. The execution of the proposed
result of the proposed optimization method with two performance indices indicates that
optimization method is printed on the figure which initiates from the point marked with
its final result on the Pareto Front is located at the lower curve in this Pareto set. This
the red circle and follows the blue line to terminate at the red plus mark. The result of
demonstrates
the proposedthat the proposed
optimization optimization
method with twomethod guarantees
performance the final
indices solution
indicates thatwill be
its final
located at the Pareto Front without setting a stopping criterion as in genetic optimization
result on the Pareto Front is located at the lower curve in this Pareto set. This demonstrates
methods.
that the However, the exact location
proposed optimization methodon the lower curve
guarantees the depends on thewill
final solution weighting val-at
be located
ues
the(𝑤Pareto 𝑤 ) which
and Front cansetting
without be selected depending
a stopping on the
criterion asdesired performance
in genetic of the
optimization robot
methods.
manipulator.
However, the exact location on the lower curve depends on the weighting values (w and 1
w2 ) which can be selected depending on the desired performance of the robot manipulator.

Figure
Figure12.
12.Pareto
Paretoset
setand
andthe
theinitiation-termination
initiation-terminationpoints
pointsofofthe
theoptimization
optimization procedure
procedurefor
fortest
testB B
for
forminimizing
minimizing the
themodified
modified condition
condition number,𝐶C,nand
number, , andthe
thedeterminant
determinantofofthe
thegeneralized
generalizedinertia
inertia
matrix,𝐼𝑚𝑁.
matrix, ImN.

InInthis
thisapproach,
approach,optimum
optimumsolutions
solutionsfor forvarious
variousdesign
designparameters
parameterscan canbebeobtained
obtained
bybyincluding
includingthese
thesedesign
designparameters
parametersasasvirtual
virtualjoints
jointsofof
a virtually constructed
a virtually constructedredundant
redundant
robot.
robot.These
Thesevariables
variablesare
are adjusted
adjusted in the null-space
null-space of ofthe
theJacobian
Jacobianmatrix
matrixthrough
through re-
redun-
dundancy resolution
dancy resolution techniques
techniques so that
so that it will
it will not affect
not affect the main
the main task ortask or design
design con-
constraint/s.
However,
straint/s. the manipulation
However, directly affects
the manipulation theaffects
directly selected thesubtask,
selectedwhich is thewhich
subtask, optimization
is the
procedure’s procedure’s
optimization objective function.
objectiveThus, the design
function. Thus,parameters are optimized
the design parameters areaccording
optimizedto
the selected
according objective
to the selectedfunction that
objective includes
function theincludes
that selectedthe performance indices of manipula-
selected performance indices
tors. A flowchart representing the implementation of this new technique is presented in
Figure 13.
Robotics 2022, 10, x FOR PEER REVIEW 19 of 22

Robotics 2022, 11, 1 19 of 21

of manipulators. A flowchart representing the implementation of this new technique is
presented in Figure 13.

Figure
Figure13.
13.Flowchart
Flowchartfor
forthe
theimplementation
implementationofofthe
thenew
newdesign
designoptimization
optimizationdesignated
designatedfor
forrobot
robot
manipulators.
manipulators.

The
Theimplementation
implementationprocedure
procedureof ofthis
thisnew
newtechnique
techniqueisisinvestigated
investigatedby byusing
usingaacase
case
scenariofor
scenario fordetermining
determining the
the fixing point of the base
base platform
platform and
and the
thefirst
firstlink’s
link’slength
lengthofofa
asurgical
surgicalrobotic
roboticsystem’s
system’spassive
passivearm.
arm.The
Themain
mainconcern
concernforforthe
theoptimization
optimizationisisincreasing
increas-
the backdrivability of this passive arm. The presented results
ing the backdrivability of this passive arm. The presented results with with various
variousnumbers
numbers of
design
of designparameters
parametersand andvarious
varioususes
usesofof performance
performance indices
indices in thethe objective
objective functions
functions
verifythe
verify theapplicability
applicabilityof
ofthis
thisnew
newmethod
methodfor forthe
themechanical
mechanicaldesign
designandandoptimization
optimizationof of
robotmanipulators.
robot manipulators.The Thevalidation
validationtest
testresults
resultsagainst
againstthe
thewell-established
well-establishedoptimization
optimization
algorithmsindicate
algorithms indicatethat
thatthe
theproposed
proposedoptimization
optimizationalgorithm
algorithmORR ORRcan canconverge
convergeto toglobal
global
optimal results for various robot mechanism design objectives.
optimal results for various robot mechanism design objectives.
Oneshortcoming
One shortcomingof ofthe
theproposed
proposedoptimization
optimizationalgorithm
algorithmisisthetheduration
durationofofthetheopti-
opti-
mization process. This shortcoming can be addressed by selecting suitable gains
mization process. This shortcoming can be addressed by selecting suitable gains for faster for faster
motionof
motion ofthe
themanipulator
manipulatorand andthus,
thus,convergence
convergenceto tothe
theoptimal
optimal results.
results.

AuthorContributions:
Author Conceptualization,O.W.M.;
Contributions:Conceptualization, O.W.M.;methodology,
methodology,O.W.M.
O.W.M.andandM.İ.C.D.;
M.İ.C.D.;software,
software,
O.W.M.and
O.W.M. and L.A.;
L.A.;validation,
validation,O.W.M.
O.W.M.andandL.A.;
L.A.;formal
formalanalysis,
analysis,O.W.M.;
O.W.M.;investigation,
investigation,O.W.M.;
O.W.M.;
resources, O.W.M.; data curation, O.W.M. and L.A.; writing—original draft preparation,
resources, O.W.M.; data curation, O.W.M. and L.A.; writing—original draft preparation, O.W.M. O.W.M.
and M. İ.C.D.; writing—review and editing, M. İ.C.D. and L.A.; visualization, O.W.M.
and M.İ.C.D.; writing—review and editing, M.İ.C.D. and L.A.; visualization, O.W.M. and M.İ.C.D.; and M. İ.C.D.;
supervision, M. İ.C.D.; project administration, M. İ.C.D.; funding acquisition, M. İ.C.D.
supervision, M.İ.C.D.; project administration, M.İ.C.D.; funding acquisition, M.İ.C.D. All authors All authors
haveread
have readand
andagreed
agreedtotothe
thepublished
publishedversion
versionofofthe
themanuscript.
manuscript.
Thiswork
Funding:This
Funding: work is
is supported
supported by
by The
The Scientific
Scientificand
andTechnological
TechnologicalResearch
ResearchCouncil
Councilof of
Turkey via
Turkey
grant
via number
grant 115E726.
number 115E726.
InstitutionalReview
Institutional ReviewBoard
BoardStatement: Notapplicable.
Statement:Not applicable.
InformedConsent
Informed ConsentStatement: Notapplicable.
Statement:Not applicable.
Authorsthank
Acknowledgments:Authors
Acknowledgments: thankthe
theNeuRoboScope
NeuRoboScopeproject
projectteam
teamfor
fortheir
theirefforts
effortsin
indeveloping
developing
thesurgical
the surgicalsystem.
system.
Conflictsof
Conflicts ofInterest: Theauthors
Interest: The authorsdeclare
declare no
no conflict
conflict of
of interest.
interest.
Robotics 2022, 11, 1 20 of 21

Appendix A
Nelder–Mead algorithm:
NM is designed firstly for unconstrained optimization problems. Essential steps of the
traditional algorithm are Ordering, Centroid, and Transformation. Since our mathematical
optimization problem includes nonlinear equality constraints a hybrid form including
conventional NM, conjugate gradient, and principle axis methods has been selected. To
implement the hybrid approach, NMaximize command as a postprocessor embedded in
Mathematica software is performed.
Simulated Annealing algorithm:
A combination of traditional SA and the principle axis algorithm is selected because of
inherent nonlinearity of the objective functions and constraints for the presented problem.
The main stages of the algorithm are [5]
(1) Introduce an initial guess zi .
(2) Generate next point, zi+1 , in the neighboring point of zi .
(3) The main goal of step 2 is to obtain smaller radius of the neighborhood for each iteration.
(4) If g(zbest ) = g(zi+n ), zi+n replaces zbest and z.
(5) Boltzmann’s probability distribution function is used to measure the distance between
these two points
where
zi : The starting point,
zi+1 : The next iteration point in the algorithm steps,
g(zbest ): The fitness function value for zbest ,
g(zi+n ): The fitness function value for zi+n ,
zi+n : The maximum number of iteration (last) points in the algorithm steps
zbest : The best point found so far in the algorithm steps
Differential Evolution algorithm:
DE is a widely used stochastic optimization method and has the stages crossover,
population size, and scaling factor to produce the generations. In this problem, we first
converted the constrained optimization problem into the canonical form. To ensure conver-
gence, we use an augmented Lagrangian merit function. Mathematica implementation of
the DE algorithm follows the procedures:
(1) Introduce a population of h points.
(2) Produce randomly generated population points.
(3) Use the real scaling factor rsF and select Cross-Probability value in the interval [0, 1].
(4) Compare the difference between the two most recently generated points.
where
h: The number of population points
rsF: The real scaling factor and it scales applied to the difference vector in creating
a mate
Random Search algorithm:
RS algorithm is the simplest method utilized for both discrete and continuous opti-
mization problems. However, for our optimization problem a combination of conventional
RS and Levenberg–Marquardt algorithms is performed at the final step. This process is
important in the evaluation of convergence and provides the merging quality of selected
starting points to the local minimum. The main steps of the algorithm are:
(1) Enter the start parameter.
(2) Create working group point Vk+1 .
(3) Update Xk+1 , Qk+1 for Vk+1
(4) Compare the difference between the two most recently generated points.
where
Vk+1 : A collection of candidate points
Qk+1 : Algorithm parameters
Xk+1 : The current iterate
Robotics 2022, 11, 1 21 of 21

References
1. Holland, J.H. Genetic Algorithms and the Optimal Allocation of Trials. SIAM J. Comput. 1973, 2, 88–105. [CrossRef]
2. Colorni, A.; Dorigo, M.; Maniezzo, V. Distributed Optimization by Ant Colonies. In Toward a Practice of Autonomous Systems,
Proceedings of the First European Conference on Artificial Life, Paris, France, 11–13 December 1991; Varela, F.J., Bourgine, P., Eds.;
Elsevier: Paris, France, 1992; pp. 134–142.
3. Kennedy, J.; Eberhart, R. Particle Swarm Optimization (PSO). In Proceedings of the IEEE International Conference on Neural
Networks, Perth, Australia, 27 November–1 December 1995; pp. 1942–1948. [CrossRef]
4. Ab Wahab, M.N.; Nefti-Meziani, S.; Atyabi, A. A Comprehensive Review of Swarm Optimization Algorithms. PLoS ONE 2015,
10, e0122827. [CrossRef]
5. Erten, H.I.; Deveci, H.A.; Artem, H.S. Stochastic Optimization Methods. In Designing Engineering Structures Using Stochastic
Optimization Methods; Aydin, L., Artem, H.S., Oterkus, S., Eds.; CRC Press: Boca Raton, FL, USA, 2020; pp. 10–23. [CrossRef]
6. Nakamura, Y. Advanced Robotics: Redundancy and Optimization, 1st ed.; Addison-Wesley Longman Publishing Co., Inc.: Boston,
MA, USA, 1990.
7. Briot, S.; Goldsztejn, A. Topology Optimization of Industrial Robots: Application to a Five-Bar Mechanism. Mech. Mach. Theory
2018, 120, 30–56. [CrossRef]
8. Vulliez, M.; Zeghloul, S.; Khatib, O. Design Strategy and Issues of the Delthaptic, a New 6-DOF Parallel Haptic Device. Mech.
Mach. Theory 2018, 128, 395–411. [CrossRef]
9. Iqbal, H.; Aized, T. Workspace Analysis and Optimization of 4-Links of an 8-DOF Haptic Master Device. Rob. Auton. Syst. 2014,
62, 1220–1227. [CrossRef]
10. Dede, M.İ.C.; Kiper, G.; Ayav, T.; Özdemirel, B.; Tatlıcıoğlu, E.; Hanalioglu, S.; Işıkay, İ.; Berker, M. Human–Robot Interfaces of
the NeuRoboScope: A Minimally Invasive Endoscopic Pituitary Tumor Surgery Robotic Assistance System. J. Med. Device 2021,
15, 011106. [CrossRef]
11. De Luca, A.; Oriolo, G. The Reduced Gradient Method for Solving Redundancy in Robot Arms. In IFAC 11th Triennial World
Gongress; Elsevier: Tallinn, Estonia, 1990; Volume 23, pp. 133–138. [CrossRef]
12. Peng, Z.X.; Adachi, N. Compliant Motion Control of Kinematically Redundant Manipulators. IEEE Trans. Robot. Autom. 1993, 9,
831–836. [CrossRef]
13. Seraji, H. Configuration Control of Redundant Manipulators: Theory and Implementation. IEEE Trans. Robot. Autom. 1989, 5,
472–490. [CrossRef]
14. Nakamura, Y.; Hanafusa, H.; Yoshikawa, T. Task-Priority Based Redundancy Control of Robot Manipulators. Int. J. Rob. Res. 1987,
6, 3–15. [CrossRef]
15. Uzunolu, E.; Tatliciolu, E.; Dede, M.C. A Multi-Priority Controller for Industrial Macro-Micro Manipulation. Robotica 2020, 39,
217–232. [CrossRef]
16. Dede, M.İ.C.; Maaroof, O.W.; Tatlicioğlu, E. A New Objective Function for Obstacle Avoidance by Redundant Service Robot Arms.
Int. J. Adv. Robot. Syst. 2016, 13, 48. [CrossRef]
17. Tatlicioğlu, E.; Braganza, D.; Burg, T.C.; Dawson, D.M. Adaptive Control of Redundant Robot Manipulators with Sub-Task
Objectives. Robotica 2009, 27, 873–881. [CrossRef]
18. Maaroof, O.W.; Gezgin, E.; Dede, M.İ.C. General Subtask Controller for Redundant Robot Manipulators. In Proceedings of the
2012 12th International Conference on Control, Automation and Systems, Jeju, Korea, 17–21 October 2012; IEEE: Piscataway, NJ,
USA, 2012; pp. 1352–1357.
19. Maaroof, O.W.; Dede, M.İ.C. Physical Human-Robot Interaction: Increasing Safety by Robot Arm’s Posture Optimization. In
Proceedings of the ROMANSY 21—Robot Design, Dynamics and Control. ROMANSY21 2016. CISM International Centre for
Mechanical Sciences (Courses and Lectures), Udine, Italy, 20–23 June 2016; Parenti-Castelli, V., Schiehlen, W., Eds.; Springer:
Cham, Switzerland, 2016; Volume 569, pp. 329–337. [CrossRef]
20. Yoshikawa, T. Manipulability and Redundancy Control of Robotic Mechanisms. In Proceedings of the IEEE International
Conference on Robotics and Automation, St. Louis, MO, USA, 25–28 March 1985; IEEE: Piscataway, NJ, USA, 1985; pp. 1004–1009.
[CrossRef]
21. Yoshikawa, T. Dynamic Manipulability of Robot Manipulators. Trans. Soc. Instrum. Control Eng. 1985, 21, 970–975. [CrossRef]
22. Denavit, J.; Hartenberg, R.S. A Kinematic Notation for Lower Pair Mechanisms Based on Matrices. ASME J. Appl. Mech. 1955, 22,
215–221. [CrossRef]
23. Paul, R.P.; Stevenson, C.N. Kinematics of Robot Wrists. Int. J. Rob. Res. 1983, 2, 31–38. [CrossRef]
24. Salisbury, J.K.; Craig, J.J. Articulated Hands: Force Control and Kinematic Issues. Int. J. Rob. Res. 1982, 1, 4–17. [CrossRef]
25. Chiu, S.L. Task Compatibility of Manipulator Postures. Int. J. Rob. Res. 1988, 7, 13–21. [CrossRef]
26. Merlet, J.-P. Parallel Robots, 2nd ed.; Springer Science & Business Media: Berlin/Heidelberg, Germany, 2006.
27. Asada, H. A Geometrical Representation of Manipulator Dynamics and Its Application to Arm Design. ASME. J. Dyn. Sys. Meas.
Control 1983, 105, 131–142. [CrossRef]
28. Colgate, J.E.; Brown, J.M. Factors Affecting the Z-Width of a Haptic Display. In Proceedings of the IEEE International Conference
on Robotics and Automation, San Diego, CA, USA, 8–13 May 1994; IEEE: Piscataway, NJ, USA, 1994; pp. 3205–3210. [CrossRef]
29. Yoshikawa, T. Foundations of Robotics: Analysis and Control; MIT Press: Cambridge, MA, USA, 1990.