Mechanism and Machine Theory 99 (2016) 1–18

Contents lists available at ScienceDirect

Mechanism and Machine Theory

journal homepage: www.elsevier.com/locate/mechmt

Optimum kinematic design of a planar cable-driven parallel

robot with wrench-closure gait trajectory
Ghasem Abbasnejad, Jungwon Yoon ⁎, Hosu Lee
School of Mechanical Engineering and ReCAPT, Gyeongsang National University, Jinju, 660-701, South Korea

a r t i c l e i n f o a b s t r a c t

Article history: We report the design of a planar cable-driven parallel robot with four cables. The design
Received 2 July 2015 can generate a wrench-closure trajectory of the lower limb, shank, with applications in gait
Received in revised form 20 December 2015 rehabilitation. Using such a design, any external wrench on the target limb can be balanced
Accepted 21 December 2015
using cables for all poses of the limb near to the trajectory in the gait cycle. We calculate the
Available online 18 January 2016
largest wrench-closure circular zone centered at an arbitrary point of the trajectory for a
given range of orientations around a reference orientation of the limb. Taking the area of
Keywords: such zones into account for a set of points on a given trajectory, we optimize the robot
Gait rehabilitation
kinematics with fixed cable attachment points. However, static evaluation of the robot in the
Parallel robot
trajectory indicates that, in some part of the trajectory, a general external wrench cannot be
Cable-driven
Planar balanced. Therefore, a reconfigurable design of the robot is investigated in which the cable
Trajectory attachment points on the base can move with respect to the motion of the limb in its tra-
Wrench-closure jectory. The area of wrench-closure zones in the trajectory can be increased using different
actuation schemes, which are obtained and compared. Finally, a redundant reconfigurable
robot with an optimum wrench-closure gait trajectory is proposed.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Gait training or gait rehabilitation is the process of learning how to walk and is applied to patients who suffer from gait dis-
orders. Physical therapists (or physiotherapists) generally aim to train their patients by encouraging them to practice normal gait
patterns. This is typically achieved by applying the torque required by a patient at the hip and knee joints during walking. Because
of the importance of proper rehabilitation, several robotic systems have been developed to aid in this therapy [1–7].
These systems are commonly classified as either exoskeleton based or end-effector based. Exoskeleton-based robotic systems
attach and apply loads to several segments to derive the target limbs. They would require a precise design of the device geometry
that often has to be adapted on the patient. Setting-up such device for a particular patient, especially if the device has many seg-
ments, may take a significant amount of time. Also, the limb-exoskeleton coupled systems are generally overconstrained which
impose unnatuaral motion and loads at joints. Investigating muscle activation patterns with and without the aid of an exoskeleton
reveals significant differences when a healthy subjects walks on a treadmill [8,9]. Comparing to the exoskeleton, end-effector sys-
tems attach to a single interface (i.e., the end-effector). Movements of the end-effector also indirectly change the position of other
segments to which it is attached. By contrast to the exoskeleton system, in end-effector system the overall limb motion is less
constrained by external device and thus the muscle activation patterns are more natural and comparable to those with treadmill
walking [8].

⁎ Corresponding author.
E-mail addresses: ghasem.abbasnejad@gmail.com (G. Abbasnejad), jwyoon@gnu.ac.kr (J. Yoon), eldoon@nate.com (H. Lee).

http://dx.doi.org/10.1016/j.mechmachtheory.2015.12.009
0094-114X/© 2016 Elsevier Ltd. All rights reserved.
2 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

Rigid links are used with conventional robotic systems, which impose an additional mass on the system and require support.
This leads to some drawbacks, such as a need for more powerful actuators, and introduces safety concerns [10]. Using cables in
place of a rigid body system can overcome these limitations and hence improves the performance of the robotic system. In par-
ticular, and similar to the end-effector-based design, using cable-driven rehabilitation systems, the cables move the target limb in
its trajectory by directly applying forces to the limb, rather than by actuating the joints. In this manner, the advantage of the
similarity of muscle activation patterns to those of treadmill walking can be exploited when cables are used in rehabilitation.
Moreover, the use of cables in a robotic system provides additional advantages, including a larger workspace, reduced
manufacturing and maintenance costs, ease of assembly and disassembly, improved transportability, and superior modularity
and reconfigurability.
The use of cables in place of rigid bodies has been studied for lower-limb rehabilitation in little literature [10–14]. In [10], a
cable actuation system was developed to control the motion of each joint of a patient by acting on the lower limbs while the pa-
tient was lying down. In [11], a cable-driven locomotion trainer (CaLT) was developed for gait training. The design of the system
in which cables are attached to the legs around the ankles, provided compliant assistance, and encouraged active involvement of
the patient. However, the ability of the robot to provide balance as well as general planar wrench on the lower leg to fully control
it through all points of its trajectory during the gait cycle was not demonstrated. In [12], a cable-driven loading system is used in a
test rig for in vitro analysis of the knee joint behavior. In this system, a generic wrench can be applied to the tibia which is housed
in a ring as the end-effector. The arrangement of the cables and their connections to the end-effector are such that the wrench
provided to the platform can be practically fully decoupled. In [13], the design of a cable-driven active leg exoskeleton (C-
ALEX) for human gait training is presented. The exoskeleton uses four cables and three cuffs which are connected to the waist,
thigh, and shank of the wearer. All four cables are routed through the waist cuff. Two of these are attached to the thigh cuff,
and the other two are routed through the thigh cuff and attached to the shank cuff. These four cables actuate two degrees of free-
dom of the wearer's leg: the hip flexion/extension and the knee flexion/extension. In [14], design of a lightweight, reconfigurable
hybrid cable-actuated articulated multibody system in which multiple cables are attached from a ground-frame to various loca-
tions on the lower limbs is introduced. In this system, external torsion springs are attached to hip, knee, and ankle joints, respec-
tively, to guarantee the tension force in cables and increase the workspace of the system.
The problem of balancing the general wrench on an end-effector and ensuring full control is challenging with cable-driven
parallel robots (CDPRs), whereby cables are employed to control the end-effector postures. With CDPRs, if all f degrees of freedom
(DOFs) of the end-effector with respect to the base are controlled by n cables, this is referred to as fully-constrained [15–19]. The
minimum number of cables, n, that are necessary to fully control the output motion is equal to the number f. However, since ca-
bles may only exert tensile axial forces, a redundancy of control actions, n N f, is usually necessary, in order to guarantee full con-
trol is preserved for a generic loading condition [20]. Conversely, a CDPR is under-constrained if the ability of the robot to balance
an arbitrary external wrench on the end-effector is lost [21–23]. A CDPR is naturally under-constrained when n b f whereas when
n ≥ f, it operates as such when mechanical equilibrium would require a negative tension in some cables. Here we use a fully-
constrained CDPR to control and move the lower limb, including the shank, which is considered as an end-effector. The trajectory
is controlled using a number of cables attached to the limb. Consequently, the thigh is also guided through its trajectory by fol-
lowing the movement of the shank, and the foot of the other limb is guided by a moving treadmill. As regards the human walking
pattern in the sagittal plane, in general, the shank preserves three DOFs, which must be restrained completely. To fully control
these three DOFs of the end-effector, although the minimum number of three cables are necessary, mostly four cables are re-
quired because cables can only exert tensile axial forces. However, even in fully-constrained CDPRs, full control of all f DOFs of
an end-effector may be lost when the number of cables that effectively contribute to control is less than f. This may happen if
balance of a generic exerted wrench on the end-effector would require a negative tension in some cables, which cannot be pro-
vided as the cables can exert only tensile forces. Consequently, the relationship between a pose and the feasible wrenches at the
end-effector is a fundamental issue for CDPRs [18]. This problem has led to several definitions of the workspace of CDPRs
[18,19,24–26]. One definition of the workspace relationship between the poses and wrenches at the end-effector is wrench-
closure workspace (WCW), which provides a strong tool for the general design of CDPRs [18]. The WCW is the set of end-
effector poses for which any wrench imposed can be balanced by the tension force of the cables. A comprehensive method to
compute the WCW of 3-DOF planar parallel mechanisms was addressed in [18]. Using this method, a WCW with a constant ori-
entation of the end-effector can be computed exactly as a bounded subset of the plane whose boundary is composed of sections
of quadratic curves. Then the exact WCW can be obtained as the common areas computed for different orientations of the end-
effector. However, calculating the exact geometry of the WCW can be complicated and time-consuming, with information that is
superfluous for the purpose of CDPR design.
Instead, by ignoring small parts of the WCW, the approximate geometry and size can be calculated using a simple geometry,
e.g., a circle, to enable easy and rapid computation. This simplification has received much interest when the size of a WCW must
be calculated iteratively. Using a simple method, we calculate the wrench-closure circular zone around an arbitrary point within
the workspace of a CDPR. The property of this zone is that as long as the poses of the end-effector are within this zone, the robot
has a wrench-closure configuration for a given range of orientations of the end-effector. However, with a broad range of orienta-
tions of the end-effector, the size of this zone is often limited or even non-existent. Nevertheless, for gait rehabilitation, it is im-
portant that the robot be in wrench-closure configuration for a limited range of orientations of the shank around the desired
orientation when the poses of the shank are close to its trajectory. Mathematically, this can be interpreted as the existence of
wrench-closure zones for all poses of the shank in its trajectory. Clearly, enlargement of these zones yields a robot with a
wrench-closure configuration for the poses of the shank in a wider area around its trajectory, which may be the aim of the design.
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 3

Considering an increase in the size of the wrench-closure zones for a sufficiently large number of poses of the shank in its trajec-
tory as the objective function, an optimum design of a reconfigurable CDPR is investigated.
In contrast to traditional CDPRs in which the attachment points of their cables are fixed, in a reconfigurable CDPR, the geom-
etry structure of the CDPR can be reconfigured by changing its cable attachment points which results to a larger WCW.
Reconfigurable CDPRs have recently been studied in several literature [27–30]. In [27], the concept of adaptive cable-driven sys-
tems is introduced to formalize a new design methodology for such CDPRs. Based on a total or partial decoupling between cable
disposition and end-effector pose, this methodology allows to achieve a design solution which minimizes or maximizes some local
performance indices such as cable tension criteria and dexterity of the CDPR. In [28], the string-operated planar haptic interface
for arm rehabilitation (Sophia-3) is introduced as first application of reconfigurable CDPRs. Sophia-3 is a 2-DOF, fully-constrained
CDPR and consists of three cables which share the same attachment point on the end-effector. Two cables are reeled on fixed pul-
ley, whereas the third one is connected to a moving pulley that allows the robot to achieve excellent force capabilities, despite the
low number of cables. In [29], an analysis framework for reconfigurable CDPRs is presented. The proposed method investigates
improvement of tension distribution by solving reconfiguration of such systems according to a tension factor over the workspace.
In [30], a new type of large-dimension reconfigurable suspended CDPR is introduced as a means to hand large and heavy parts
across a wide workspace. A systematic procedure to solve the problem of finding optimum reconfiguration for the robot is pre-
sented. It is shown that the reconfigurability of the proposed CDPR offers better performances in term of workspace, flexibility,
and power consumption.
In this paper, to solve the optimization problem regarding the design of a CDPR for gait rehabilitation, particle swarm optimi-
zation (PSO) [31] is utilized as a global nonlinear optimization algorithm. PSO has been shown to be useful in a wide variety of
optimization problems. Advantages of PSO include ease of implementation and that there are few parameters to adjust. Example
of remarkable performance of PSO in robotics can be found in [32,33].

2. Circular wrench-closure zone

Fig. 1 shows a schematic diagram illustrating the basic concept of the use of a planar 4–4 CDPR in treadmill gait training,
i.e., the notation 4–4 denotes that four cables are fixed to the end-effector (shank) and driven using four motors. By considering
the planar motion of the shank in sagittal plane as a subset of its spatial motion, this system can potentially control all three DOFs
of the shank in this plane and balance any planar wrench exerted on it. However, this can be achieved only for poses of the shank
in which any planar external wrench on the shank can be balanced by applying tension to the cables. The relation between the
poses of the shank and balancing the external wrench in sagittal plane is investigated in this section which also hold when the
shank has small motion out of this plane.

2.1. Kinematic model of the 4–4 CDPR

Here the shank is considered as an end-effector separated from the leg, the movement of which in its trajectory on sagittal
plane is controlled by four cables. As shown in Fig. 2, the ith cable (where i = 1 … 4) is attached to the base at the point Ai =
[ai1, ai2]T, and to the end-effector at the point Bi = [bi1, bi2]T. For the sake of analysis, we consider a fixed Cartesian frame A,
where x and y are unit vectors along the coordinate axes attached to the base at point O, and a moving Cartesian frame B attached

Fig. 1. The basic concept of the use of cables in gait rehabilitation.

4 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

Fig. 2. 4–4 CDPR model of gait rehabilitation.

to the platform in G, where u and v are unit vectors along its axes. The cable length, ρi, is the distance between Ai and Bi at the
considered configuration. The position of the end-effector relative to the base frame is defined by the vector from O to G, and de-
noted x = [x, y]T, whereas its orientation is defined by the angle ϕ, which is the angle between x and u. Using these assumptions,
the following symbols are defined:


cosðϕÞ − sinðϕÞ
R¼
sinðϕÞ cosðϕÞ
ð1Þ
ai ¼ Ai −O; ri ¼ Bi −G ¼ Rbi
si ¼ Bi −Ai ¼ x þ Rbi −ai

where bi is the position vector of Bi in B. The platform is assumed to be acted upon by a force in the sagittal plane applied in G,
as well as a torque along the axis out of the plane, and is described by a planar wrench Le. The normalized Plücker vector of the
line associated with the ith cable is Li =ρi, where, in axis coordinates, Li ¼ ½si ; ri  si . The wrench exerted by the ith cable on the
platform is ðτi =ρi ÞLi, where τi is a positive scalar representing the intensity of the cable tensile force. If the wrench Le is balanced
by the sum of the cable wrenches, the following relation can be written in matrix form:
2 3
ðτ1 =ρ1 Þ
6 ðτ2 =ρ2 Þ 7
½L1 L2 L3 L4  6 7 −Le ¼ 0; ð2Þ
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} 4 ðτ3 =ρ3 Þ 5
M ðτ4 =ρ4 Þ
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
t

where M is a 3 × 4 matrix that depends on both the pose and orientation of the end-effector for a robot with a fixed geometry.
Using this model in a wrench-closure configuration of the robot, for any external wrench Le , there exists at least one vector
t ≥ 0 such that Mt ¼ Le. Here, for any vector q, q N 0, q ≥0 and q b 0 imply that all the components of q are greater than zero, greater
than or equal to zero, and smaller than zero, respectively. It has been shown [34–36] that the necessary condition for these con-
figurations is given by:

rankðMÞ ¼ 3 ð3Þ

which implies that the wrench matrix M is full rank. However, this condition is not sufficient for making sure the tensile force in
all the cables which should be considered necessarily because of the inherent feature of cables. For this, a sufficient condition is
required as

∃z∈ ker ðMÞ such that zN0: ð4Þ

 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 5

Now let z0i be defined as

iþ1
z0i ¼ ð−1Þ det ðMi Þ; i ¼ 1; 2⋯; 4; ð5Þ

where Mi is the submatrix of M obtained by deleting the ith column of M. It has been shown in [36] that when the necessary
condition, Eq. (3), is satisfied (rank(M) =3), the sufficient condition (4) is equivalent to:

z0 N0 or z0 b0
where ð6Þ
T
z0 ¼ ½z01 ; z01 ; z03 ; z04 

These relations greatly simplify testing the sufficient condition of the wrench-closure conditions are valid. However, our only
concern is the sign of each of the components z0i of the vector z0 defined by (5).

2.2. Wrench-closure boundary surface of a 4–4 CDPR

For a pose of the end-effector, the 4–4 CDPR shown in Fig. 2 is in wrench-closure configuration if and only if the necessary
condition in Eq.(3) is satisfied and all the functions z0i have the same sign. From this condition, the boundary of the wrench-
closure area can be defined as all the points in the workspace of the robot corresponding to the configurations with

∃i∈f1; 2; 3; 4g : z0i ¼ 0 and∀j∈f1; 2; 3; 4g; j≠i z0 j N0 or z0 j b0: ð7Þ

Thus, z0i = 0 describe the possible boundary of the wrench-closure area. For a fixed geometry, each z0i is an analytical function
involving powers of x ,y , cos (ϕ)x , y , cos (ϕ), and sin(ϕ). When the orientation of the end-effector is constant, ϕ =C, each of these
functions transforms to a set of quadratic functions in terms of the coordinates x and y of the platform reference point G.

Fig. 3. Example of surfaces which potentially contain wrench-closure boundary surface of a 4–4 CDPR.
6 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

Geometrically, these functions describe a set of curves in the x–y plane, which describe the possible boundary of the wrench-
closure area. As discussed in [18], these curves have close correspondence with the boundary of the constant orientation
singularity-free area of a 3RPR parallel mechanism. Similarly, in this mechanism, the boundary of the singularity-free area for a
constant orientation of the end-effector can be described by a quadratic function. In both cases, these curves are surfaces in x–
y–ϕ coordinates if the range of orientation of the end-effector is taken into account. Fig. 3 shows an example of these surfaces
which possibly contain the wrench-closure boundary of the CDPR shown in Fig. 1 with the geometric parameters listed in
Table 1. It is clear that the intersection of a plane that is parallel to the x–y plane with ϕ = C yields a set of curves as the boundary
of the constant orientation wrench-closure area.
As discussed in [37,32,38,39], with the help of similar singularity-free surfaces of parallel mechanisms, a zone with a certain
geometry inscribed by the singularity-free area can be calculated. For example, in [32], a method based on PSO was introduced
to compute the radius of a circle inscribed by the singularity-free workspace of a planar 3-DOF parallel mechanism. In the follow-
ing sections, we review the PSO algorithm and then describe the implementation of this procedure to compute a circular zone
inscribed by a WCW using wrench-closure boundary surfaces.

2.3. Overview of particle swarm optimization

PSO is a technique inspired by the social behavior of flocking animals, such as birds or ants [31]. A swarm of Np particles
is used to probe a D-dimensional search space. Each particle moves in the search space on the basis of the optimal position
encountered during the search by either the particle itself or by its neighbors. The optimal position is that which provides
the best fitness value. The algorithm consists of adapting the velocity and position of the particles iteratively, over NI itera-
tions, as follows:

h i h i
best best
Δwh ðk þ 1Þ ¼ ψΔwh ðkÞ þ C 1 R1 wh ðkÞ−wh ðkÞ þ C 2 R2 wG ðkÞ−wh ðkÞ ;
ð8Þ
wh ðk þ 1Þ ¼ wh ðkÞ þ Δwh ðk þ 1Þ:

where k is the iteration number, wh is the current position of the particle h, wbest
h is the best position achieved so far by the
particle, wbest
G is the best position globally attained by any particle in the swarm, Δwh is the increment of the particle posi-
tion, ψ is an inertial weighting, C1 and C2 are positive acceleration constants, and R1 and R2 are random numbers uniformly
distributed in the interval [0, 1]. The first term in Eq. (8) represents the weighting of the current velocity of the particle, the
second term expresses a trend toward a local optimum, and the third term describes a global search tending toward the op-
timal solution.

2.4. Circular wrench-closure zone calculation

The end-effector of the 4–4 CDPR is expected to follow a trajectory inside its workspace. By considering an arbitrary position of
the reference point G on the end-effctor, C0 = [x0,y0], the circular wrench-closure zone is defined as a circle centered in C0 with
the property that, as long as the end-effector reference point G lies inside it, any external wrench can be balanced by applying
tension to the cables for any orientation of the end-effector within the range of [ϕ0 − δϕ, ϕ0 + δϕ]. ϕ0 is the desired orientation
of the end-effector in gait cycle when the reference point G lies on C0 and δϕ is a constant tolerance of the end-effector orienta-
tion which can be achieved by preserving wrench-closure configuration of the robot. The center of the circle C0, the angles ϕ0 and
δϕ are called as the center position, the center orientation and orientation deviation, respectively. By locating a circle in the
wrench-closure area, one can make sure the limited movement of the end-effector from the point C0 in all directions while the
wrench-closure configuration is preserved. On the other hand, the circular zone provides a simple estimation of the exact
wrench-closure area as the common areas of the projection of the four boundary surfaces on x-y plane. As this exact area may
not be in a regular geometry, computation of its size can be difficult and challenging. Here a simple algorithm is described to lo-
cate this circle. This problem may be simplified to locating the maximum radius of the circular zone centered in C0 and inscribed
in the wrench-closure area. To this end, a point on the wrench-closure surfaces of a 4–4 CDPR with the end-effector orientation
range [ϕ0 − δϕ, ϕ0 + δϕ] defined in the previous section is found such that it has the smallest distance to the center. We search for
three variables x, y, and ϕ, which are constrained with one of 4 equations z0i =0 such that the square of the distance from [x,y] to

Table 1
Geometric parameters of the 4–4 CDPR with potential wrench-closure boundary surface
shown in Fig. 3.

i Ai Bi

1 [0.8, 1]T [0.07, 0.13]T

2 [0.8, 0]T [0.07, −0.13]T
3 [−0.8, 0]T [−0.07, −0.13]T
4 [−0.8, 1]T [−0.07, 0.13]T
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 7

the center C0 is minimal; i.e.,

2 2 2
min d min ¼ ðx−x0 Þ þ ðy−y0 Þ
While : ð9Þ
∃i∈f1; 2…4g : z0i ¼ 0

To solve this optimization problem, a PSO algorithm searches in a two-dimensional plane, e.g., the y-ϕ plane. As described in
Algorithm 1, these two variables are chosen randomly. Substitution of these variables in the constraint equations s then yields
quadratic polynomials in x.

Algorithm 1. A pseudo-code representation of the algorithm used to compute the inscribed circular zone in the WCW of the 4–4
CDPR.

On solving each of these polynomials, a set of solutions of the variable x is obtained. By evaluating the corresponding fitness
function in Eq. (9), the closest point on the wrench-closure surface to the central position can be calculated. This procedure is re-
peated as part of the PSO algorithm, and the initial random variables are updated to minimize the fitness function. Since this op-
timization problem is simple, the algorithm typically converge to the optimum solution in few iterations and less than a second
on a PC with a 3.4 GHz processor and 4GB of RAM.1 Accordingly, one may conclude that in an x–y–ϕ coordinate frame, a cylinder
with the axis parallel to the ϕ axis and centered on C0 with radius dmin obtained from Eq. (9) does not intersect with any of the
wrench-closure surfaces. Clearly, it is guaranteed that, as long as the platform reference point G lies inside the projected circle of
the cylinder on the x–y plane, the robot always remains in the wrench-closure configuration for any orientation of the end-
effector within the range of [ϕ0 − δϕ, ϕ0 + δϕ]. For the example of the geometric parameters listed in Table 1, a center position
[0.1625, 0.2555]Tm, a center orientation 0.0293rad, and an orientation deviation 0.1745rad, the algorithm results in the radius of
the zone dmin = 0.090464. To visualize this, Fig. 4 shows the corresponding cylinder with the wrench-closure surfaces.

2.5. Enlargement of the circular wrench-closure zone

So far, we have considered that for a given set of geometric parameters of a CDPR, a circular wrench-closure zone centered on
C0 may exist for a range of orientations of the end-effector, and its radius can be computed by solving an optimization problem.
However, this zone is often small, or even non-existent for some geometries of the robot. This underlies the importance of iden-
tifying a robot geometry that results in a large wrench-closure circular zone.

1
This workstation is used for all the computation in this paper.
 8
G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18
Fig. 4. (a) The cylindrical wrench-closure zone for the robot parameters listed in Table 1. (b) Top view of the wrench-closure zone (dmin =0.090464 m).
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 9

Algorithm 2. The pseudo-code of the algorithm for enlargement of circular wrench-closure zone.

It is clear that the fitness function in Eq. (9) depends on the geometric parameters of the robot consisting of the coordinates of the
attachment points for cables on the end-effector and base. These are parameters which are also considered as the first input of
Algorithm 1, to determine the radius of the wrench-closure zone. For different inputs of the algorithm, different radii of the wrench-
closure zone will result for the same C0, ϕ0 and δϕ of the end-effector. However, the optimal set of such parameters that results in the
largest circular wrench-closure zone, max(dmin), is of interest. As described in Algorithm 2, for given C0, ϕ0, and δϕ of the end-effector,
these parameters can be determined using PSO and Algorithm 1. First, a set of geometric parameters is randomly chosen, and their
corresponding fitness values dmin are computed using Algorithm 1. Then, these parameters are updated according to the PSO algorithm
toward maximization of dmin. To account for some design limitations, the following geometric constraints are also considered:

1Þ0:5≤a11 ¼ a21 ≤0:8;

2Þa22 þ 0:06 ≤a12 ≤1;
3Þ0≤a22 ≤0:94;
4Þ−0:5≤a31 ¼ a41 ≤−0:8;
5Þ0≤a32 ≤0:94;
6Þa32 þ 0:06 ≤a42 ≤1;
ð10Þ
7Þb11 ¼ b21 ¼ 0:07;
8Þb22 ≤b12 ≤0:13;
9Þ−0:13 ≤b22 ≤0:13;
10Þb31 ¼ b41 ¼ −0:07;
11Þ−0:13 ≤b32 ≤0:13;
12Þb32 ≤b42 ≤0:13

Fig. 5. A schematic diagram showing a model of a 4–4 CDPR with the parameters listed in Table 2 (dmin =0.32742 m).
10 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

As shown in Fig. 5, the constraints 1 and 4 are imposed because the cables are attached to vertical bars in front of and behind
the patient. From constraints 2, attachment point A1 is above attachment point A2, and a small gap is always preserved between
these two points so that the two motors can be positioned properly. This constraint in addition to the constraint 8, which is to
position attachment point B1 above attachment point B2, makes sure that the interference between the cables is avoided. This
is specially important for safe use of the robot for rehabilitation and interaction with patients. From constraints 6 and 12, the
same applies to the attachment points A3 and A4 and the other two cables. The remaining constraints consider the average perim-
eter and the length of the shank. We assume in this example that the shank is at least 0.26 m long, with a diameter of not more
than 0.14 m.
For the example of C0 = [0.1625,0.2555]Tm, ϕ0 = 0.0293rad, and δϕ = 0.1745rad given at the end of Section 2.4, the radius of
the wrench-closure zone for the parameters of the robot listed in Table 1 was computed as dmin = 0.090464 m previously. How-
ever, now Algorithm 2 is used to find a new set of geometric parameters with maximum radius of wrench-closure zone for the
same C0, ϕ0, and δϕ. Table 2 lists the results of the optimization, and Fig. 5 shows a schematic diagram of the corresponding
model. The radius of the wrench-closure zone is dmin = 0.32742 m for this optimal set of parameters, which is significantly larger
than the radius of the zone computed previously. The CPU time required for this algorithm to converge to the optimum param-
eters in this example is roughly 8.8641 min.

3. Optimum kinematic design for gait trajectory

In this section, we describe the design of a 4–4 CDPR that can follow the gait trajectory of the shank. In such a design, the
robot remains in wrench-closure configuration in each position, and the orientation of the shank near the desired trajectory. How-
ever, before investigating the problem, we must derive data describing the positions of the reference point G on the shank, as well
as the orientation of the shank in each part of the gait cycle.

3.1. Gait Analysis

The two main phases of the gait cycle are the stance phase and the swing phase. In the stance phase, the foot is on the ground,
which accounts for approximately 60% of the cycle. It begins with the heel strike and terminates when the foot leaves the ground.
The remaining 40% of the cycle is the swing phase, which begins as soon as the foot leaves the ground, whereby the subject ac-
celerates the leg forward. Termination of this phase begins with the slowing of the leg and stabilizing of the foot in preparation
for the subsequent heel strike.
In [40], this measurement in 3D was studied for all three lower extremities (i.e., the thigh, calf, and foot) via a set consisting of
a minimum number of markers placed on anatomical landmarks that could be reliably located. Fig. 2 shows a schematic diagram
of a model of the isolated shank with two markers indicated as M1 and M2. Fig. 6 shows the example of trajectory curves of two
components of these markers in sagittal plane. For this, normal gait cycle of the example of a male subject with the average phys-
ical dimensions listed in Table 3 is considered.
When the trajectories of two points on an object are known, the trajectory of a reference point on the object and its orienta-
tion can be easily computed in plane. Assuming that the reference point G is the mid point between M1 and M2 (see Fig. 2), the
position of the point G and the orientation of the shank in the desired gait trajectory can be determined as follows:

M 1x þ M 2x
Gx ¼
2
M 1y þ M 2y
Gy ¼ ð11Þ
2 
M 2y −M 1y
ϕ ¼ arctan
M 2x −M 1x

where the subscripts x and y imply coordinates in the x and y directions, respectively. For the example of the subject with the physical
dimensions listed in Table 3, the calculated position of the reference point G and the orientation of the shank in each part of the gait

Table 2
The optimal geometric parameters for a 4–4 CDPR with an enlarged circular wrench-closure zone
in 0% of the gait cycle.

i Ai Bi

1 [0.5667, 1]T [0.07, 0.0897]T

2 [0.5667, 0]T [0.07, 0.0897]T
3 [−0.8, 0]T [−0.07, −0.13]T
4 [−0.8, 0.06]T [−0.07, 0.1274]T
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 11

Fig. 6. The trajectories of markers M1 and M2 in the gait cycle.

trajectory is shown in Fig. 7. Assuming another subject with different physical dimensions, trajectory of the shank would be similar to
what is depicted in Fig. 7. As described in [41], the lengths of different body segments have approximately constant ratio of the body
height and, for the example of subject of a given height, these lengths can be computed as listed in Table 3. Thus, for a subject with
physical dimensions a% smaller or larger than those in Table 3, the trajectory of the point G in the direction of x and y can be estimated
as a deviation of a% from the trajectory shown in Fig. 7, whereas the orientation remains unchanged. For example, Fig. 8 shows the
deviation of the trajectory of the point G for a subject with physical dimensions 6% greater than those listed in Table 3. This 6% differ-
ence is similar to the difference of 0.1 m in the normal range of height. It follows that the maximum difference in the two trajectories in
the gait cycle is 0.0243 m.

3.2. Calculation of the circular wrench-closure zone with respect to the gait cycle

From Fig. 7, we can see that the shank undergoes a wide range of orientations through the gait cycle, i.e., [−1.0363, 0.0904]rad.
The initial computation for a robot with different arbitrarily chosen geometric parameters shows that a wrench-closure
workspace for this wide range of orientations typically does not exist, or is very small. In another words, it is very unlikely
to find a set of geometric parameters for a robot with the property that, for each position of reference point G on the shank
around the desired gait trajectory, the robot remains in a wrench-closure configuration for any of its orientations in the
range. However, as shown in Fig. 7, in each stage of the desired gait cycle, the shank has a specific orientation. Therefore,
we may conclude that investigating the wrench-closure configuration for the entire range of orientations of the shank in
the position of the shank reference point G near the desired trajectory is unnecessary. Rather, considering such a configura-
tion for the orientation of the shank with limited deviation from the desired position of each reference point G in a zone
around the desired trajectory is of greater interest. This problem can be redefined as calculating the wrench-closure zone
of a robot for a coordinate of the center position C0 as the position of G on the shank in its trajectory, and for a small
range of its orientation that deviates from the desired orientation. Assuming reference point C0 and reference orientation
ϕ0 in Algorithm 1 δϕ. For example, in the desired trajectory of the shank during the gait cycle, the position and orientation
at 0% is [0.1625, 0.2555] m and 0.0293rad. These are the same assumptions for C0 and ϕ0 for which the wrench-closure was
computed for δϕ = 0.1745rad, as shown in Fig. 4. We may therefore conclude that the same computation can be performed
for a pose of the shank in any other part of the gait cycle. More generally, the radius of the circular wrench-closure zones for

Table 3
The dimensions of the male subject with the gait trajectory shown in Fig. 2.

Body part Thigh Shank Foot

Length (m) 0.460 0.430 0.260

12 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

Fig. 7. The position and orientation of the shank through the gait trajectory.

an orientation deviation regarding a set of poses of the shank in a desired gait trajectory can be calculated using Algorithm
1, as well as the corresponding positions of point G and the orientations of the shank.
This calculation was performed for the set of geometric parameters of the robot listed in Table 2 for 50 sample poses of the
shank in a gait trajectory. Fig. 9 shows the radius of the wrench-closure zones in each of these poses. The previous calculation of
this set of robot parameters (see Subsection 2.5) corresponded to the maximum radius of the wrench-closure zone in 0% of the

Fig. 8. The trajectory of G for (a) a subject with the physical dimensions listed in Table 3, and (b) a subject with physical dimensions that are 6% larger.
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 13

Fig. 9. The radius of the circular wrench-closure zone for 50 sample poses of the shank in the gait cycle for the robot parameters listed in Table 2.

gait trajectory. Therefore, as expected, the radius of the wrench-closure zone shown in Fig. 9 is a maximum in 0% of the gait cycle.
However, this diagram shows a sharp decline after 30% of the gait cycle, and the wrench-closure zone becomes non-existent be-
tween 50% and 80%. It follows that in this stage of the gait cycle, the robot may fails to balance the wrench imposed on the shank,
and one or more cables may become slack.

3.3. Optimum kinematic design with wrench-closure configuration in the gait cycle

In the design of a CDPR, the radius of wrench-closure zones should be considered for all poses of the shank in its trajectory
through the gait cycle. The most important consideration in the design of the cable robot is to ensure that the robot maintains
full control of the shank in all of the configurations of the robot during the gait cycle. Computationally, this can be implemented
by defining a new fitness function in the Algorithm 2 to ensure that the radius of the wrench-closure zones is not less than a
certain constant value ε for a sufficiently large number of points in the trajectory. As described in Algorithm 3, the gait trajectory
is divided into NC sections, and the objective function, NNε, is defined as the maximum number of points among NC for which the
radius of the wrench-closure zone is larger than ε. In this manner, it is guaranteed that the robot never loses control of the shank
at a sufficient distance from the desired trajectory for a certain range of orientations of the shank. As discussed in Subsection 3.1,
the comparison of the trajectory of the reference point G shown in Fig. 8 reveals that a deviation of 0.1 m from the physical di-
mensions of the subject results in a maximum deviation of 0.0246 m in the trajectory. From this comparison, it can be realized
that although a design of the robot with the minimum radius of the wrench-closure zone equal or less than 0.0246 m can operate
properly for a subject with the physical dimensions reported in Table 3, it cannot preserve the wrench-closure configuration dur-
ing the gait cycle when the subject's physical dimensions deviate 0.1 m or more. Thus, for the robot to be applicable for subjects
with relatively wide range of physical dimensions, the minimum radius of the wrench-closure zone is required to be sufficiently
larger than 0.0246 m. According to this fact, the geometric parameters of the robot are investigated by Algorithm 3 with ε = 0.05,
NC = 50, and by taking into account the constraints in Eq. (10). Since in this algorithm for each particle the radius of wrench-
closure zones for NC number of points in the trajectory should be computed, the CPU time increase significantly. For this example,

Table 4
Optimum geometric parameters of a 4–4 CDPR with an enlarged circular wrench-closure zone in the en-
tire gait cycle.

i Ai Bi

1 [0.7134, 0.2092]T [0.07, 0.13]T

2 [0.7134, 0.1492]T [0.07, −0.13]T
3 [−0.6847, 0]T [−0.07, −0.0689]T
4 [−0.6847, 0.7560]T [−0.07, −0.0689]T
14 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

Fig. 10. A schematic diagram showing a model of a 4–4 CDPR with the parameters listed in Table 4.

roughly 206.3073 min is required for the Algorithm 3 to converge to the optimum set of parameters of the robot listed in Table 4.
Fig. 10 shows a schematic diagram of the model of the robot with this set of parameters.
Fig. 11 shows the corresponding radii of the wrench-closure zone in the gait cycle. This reveals that, although the wrench-
closure zone exists in a broader range of the gait cycle, the minimum radius of the wrench-closure zone around the desired tra-
jectory is not fulfilled in part of the trajectory. This diagram indicates that the robot may lose the ability to balance the wrench
imposed on the shank when it undergoes movement between 60% and 70% of the gait cycle. Performing the optimization proce-
dure with different adjustments to eradicate local optima and to find a new set of robot parameters with the desired performance
fails every time. Accordingly, the final recourse to find a kinematic design for a robot with the preferred wrench-closure trajectory
is to add redundancy. This is investigated in following section.

Fig. 11. The radius of the circular wrench-closure zone for 50 sample poses of the shank in the gait cycle for the robot parameters listed in Table 4.
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 15

Algorithm 3. The pseudo-code of the algorithm to ensure the wrench-closure configuration of the robot in whole the gait
trajectory.

4. Optimum kinematic design of a reconfigurable CDPR for gait trajectory

As discussed in the previous section, the design of a CDPR to control the motion of the leg for gait rehabilitation with a given
wrench-closure configuration near the desired trajectory is challenging, and the attempt to create such a design was unsuccessful.
However, in this section, the kinematic performance of the CDPR is improved by introducing reconfigurability. Here the design of
a reconfigurable CDPR inferred from the optimization results listed in Table 4 is considered and is shown schematically in Fig. 12.
Generally, in parallel manipulators, adding redundancy can result in several advantages such as avoiding most kinematic singular-
ities, enlarging workspace, as well as improving dexterity [42]. Similarly in reconfigurable CDPR, capability of the robot is im-
proved greatly because of increase in the number of redundancies [28–30]. This strategy can also improve the performance of
the CDPR for gait rehabilitation and avoid loss of control of the shank for poses near the desired trajectory.

Fig. 12. A schematic diagram showing a 4–4 reconfigurable CDPR.

16 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

Two attachment points of cables at A1 and A2 are fixed in the previous design, and these are now assumed to be driven by
actuator S1, so they are no longer fixed to the base and may move together. From the parameters listed in Table 4, these two at-
tachment points should be as close to each other as possible. However, the separation between them on the actuator S1 is limited
because we must ensure that the two motors in these points cannot interfere with each other. Similarly, the actuator S2 should
drive the attachment point A4, as listed in Table 4. These two actuators are chosen by trial and error and by considering all the
other possible actuators.
Although reconfigurability may improve the performance of a CDPR, it complicates planning and control because there are
more independent variables for a manipulator with greater redundancy. Here, a reconfigurable CDPR with actuators S1 and/or
S2 is compared while following the desired gait trajectory in terms of the ability to ensure a wrench-closure configuration.
Reconfigurability increases the number of inverse displacement solutions from a limited number of solutions to a locus of solu-
tions. Therefore, it is necessary to apply a method to select a solution from all available solutions. To solve this problem, a strategy
has been proposed [42] for an actuation scheme for a planar redundant manipulator. Using this strategy together with PSO opti-
mization, point-to-point path planning was achieved for a reconfigurable CDPR based on locally optimizing the radius of the
wrench-closure zone for each pose of the end-effector in its desired trajectory. To avoid large actuator velocities, the rate of
change of actuator displacements was constrained not to exceed a certain limit. For the actuation schemes obtained here, the
gait cycle was divided into NC, and for each increment of the gait cycle, the rate of change of actuators could not exceed
0.03 m. This method is illustrated in Algorithm 4.

Algorithm 4. The pseudo-code of the algorithm for path planning of reconfigurable CDPR.

This algorithm is used to investigate for the actuation schemes of the robot with various reconfiguration. In the first scenario,
we assume that only the actuator S1 is active, and that the second actuator is locked at S2 = 0.756 m. Then, the first actuator is
locked at S1 = 0.14 m, which is equivalent to having A12 = 0.14 m and A22 = 0.2 m. In the second scenario, both S1 and S2 are
active. In each case, this algorithm solve an optimization problem for NC points on the trajectory which can increase the CPU
time accordingly. In these examples for NC = 50 the CPU time was roughly 192.983 min.
Fig. 13 shows the radius of the wrench-closure zone for each point of the desired gait trajectory with different
reconfigurations. The average radius of the wrench-closure zone in the entire gait cycle improves when actuator S1 is active.

Fig. 13. The radius of the wrench-closure zone for each point of the gait trajectory with different reconfigurations.
 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18 17

Fig. 14. The actuation scheme for a robot with a single actuator S1.

Fig. 13 shows that when the second actuator S2 is relaxed, the radius of the wrench-closure zone does not increase significant-
ly. The wrench-closure radius for a robot with two actuators S1 and S2 is almost the same as that for a robot with only one ac-
tuator S1. Moreover, when the actuator S1 is fixed, with a single active actuator S2, the radius of the wrench-closure zone does
not change significantly compared with that for a robot with no reconfiguration. From this comparison, we may infer that,
when a robot has actuator S1, the wrench-closure configuration near the desired trajectory is sufficient. Fig. 14 shows the actuator
scheme for a robot with an actuator S1 throughout the entire gait cycle.

5. Conclusion

We have discussed the design of a 4–4 planar cable-driven parallel robot for gait rehabilitation. Cables may exert only tensile
axial forces; therefore, ensuring full control of a robot with generic loading conditions with no slack cables is challenging. This
problem was investigated for a 4–4 planar cable-driven parallel robot with applications in gait rehabilitation. The cables should
attach to the shank as end-effectors and balance the general load when the shank moves near the desired trajectory. The desired
trajectory of the shank was determined by investigating the motion of the lower extremities in the normal gait cycle. Using a
schematic model of the robot, a simple mathematical representation of a wrench-closure configuration in which the robot can bal-
ance a general load was provided. A method based on PSO was used to determine the radius of a circular zone in which the
wrench-closure configuration of the robot is guaranteed for a given range of orientations of the end-effector. Calculation of this
zone around an arbitrary point on the desired trajectory was described, and this computation was extended for a set of points
on the trajectory. Moreover, the size of these zones depends on the geometric parameters of the robot, and an algorithm was pro-
vided to determine the optimum geometric parameters with an enlarged wrench-closure zone. However, application of this op-
timization procedure to gait rehabilitation failed to result in optimum robot parameters with a sufficiently large wrench-closure
zone in the desired trajectory. To overcome this, we investigated design of a reconfigurable cable-driven parallel robot. The
wrench-closure zone in the desired trajectory was studied considering three different actuation scenarios. By comparing the out-
comes obtained for each case, we may conclude that, when a reconfiguration is introduced, the robot operates efficiently near to
the desired trajectory.

Acknowledgments

This work was supported by the National Research Foundation Korea (NRF) (2012-0009524 and 2014R1A2A1A11053989) and
Technology Development Project for Advancement of NIPA SW Convergence Technology and Dual Use Technology Program of
Civil and Military.

References

[1] J. Yoon, B. Novandy, Y. C., K. Park, A 6-dof gait rehabilitation robot with upper and lower limb connections that allows walking velocity updates on various ter-
rains, IEEE/ASME Trans. Mechatron. 15 (2) (2010) 201–215.
[2] G. Belforte, G. Eula, S. Appendino, S. Sirolli, Pneumatic interactive gait ehabilitation orthosis: design and preliminary testing, Proc. Inst. Mech. Eng. H 225 (2)
(2011) 158–169.
[3] P. Beyl, M. Damme, R. Ham, D. Lefeber, Design and control concepts of an exoskeleton for gait, Proceedings of the 2nd Biennial IEEE/RAS-EMBS International Con-
ference on Biomedical Robotics and Biomechatronics 2008, pp. 103–108.
18 G. Abbasnejad et al. / Mechanism and Machine Theory 99 (2016) 1–18

[4] G. Colombo, M. Joerg, R. Schreier, V. Dietz, Treadmill training of paraplegic patients using a robotic orthosis, J. Rehabil. Res. Dev. 37 (6) (2000) 693–700.
[5] M. Frey, G. Colombo, M. Vaglio, R. Bucher, M. Jorg, R. Riener, A novel mechatronic body weight support system, IEEE Trans. Neural Syst. Rehabil. Eng. 14 (3)
(2006) 311–321.
[6] F. Gazzani, A. Fadda, M. Torre, V. Macellari, A pneumatic system for body weight relief in gait rehabilitation, IEEE Trans. Neural Syst. Rehabil. Eng. 8 (4) (2000)
506–513.
[7] S. Hesse, D. Uhlenbrock, A mechanized gait trainer for restoration of gait, J. Rehabil. Res. Dev. 37 (6) (2000) 701–708.
[8] H. Schmidt, M. Volkmar, C. Werner, I. Helmich, F. Piorko, J. Kruger, S. Hesse, Muscle activation patterns of healthy subjects during floor walking and stair climbing
on an end-effector-based gait rehabilitation robot, IEEE 10th International Conference on Rehabilitation Robotics, ICORR 2007, pp. 1077–1084.
[9] J. Hidler, A. Wall, Alterations in muscle activation patterns during robotic-assisted walking, Clin. Biomech. 20 (2005) 184–193.
[10] K. Homma, O. Fukuda, J. Sugawara, Y. Nagata, M. Usuba, A wire-driven leg rehabilitation system: development of a 4-dof experimental system, Proceedings of the
2003 IEEWASME International Conference on Advanced Intelligent Mechatronics 2003, pp. 908–913.
[11] M. Wu, T. Hornby, J. Landry, H. Roth, B. Schmit, A cable-driven locomotor training system for restoration of gait in human sci, Gait Posture 33 (2011) 256–260.
[12] M. Forlani, N. Sancisi, M. Conconi, V. Parenti-Castelli, A new test rig for static and dynamic evaluation of knee motion based on a cable-driven parallel manipulator
loading system, Meccanica http://dx.doi.org/10.1007/s11012-015-0124-1.
[13] X. Jin, X. Cui, S. Agrawal, Design of a cable-driven active leg exoskeleton (c-alex) and gait training experiments with human subjects, 2015 IEEE International
Conference on Robotics and Automation (ICRA), 2015.
[14] A. Alamdari, V. Krovi, Robotic physical exercise and system (ropes): a cable-driven robotic rehabilitation system for lower-extremity motor therapy, Proceedings
of the ASME 2015 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE 2015, 2015.
[15] A. Ming, T. Higuchi, Study on multiple degree-of-freedom positioning mechanism using wires—Part 1: concept, design and control, Int. J. Jpn Soc. Precis. Eng. 28
(2) (1994) 131–138.
[16] P. Bosscher, A.T. Riechel, I. Ebert-Uphoff, Wrench-feasible workspace generation for cable-driven robots, IEEE Trans. Robot. 22 (5) (2006) 890–902.
[17] K. Kozak, Q. Zhou, J. Wang, Static analysis of cable-driven manipulators with non-negligible cable mass, IEEE Trans. Robot. 22 (3) (2006) 425–433.
[18] M. Gouttefarde, C.M. Gosselin, Analysis of the wrench-closure workspace of planar parallel cable-driven mechanisms, IEEE Trans. Robot. 22 (3) (2006) 434–445.
[19] M. Gouttefarde, D. Daney, J.-P. Merlet, Interval-analysis-based determination of the wrench-feasible workspace of parallel cable-driven robots, IEEE Trans. Robot.
27 (1) (2011) 1–13.
[20] M. Carricato, J.-P. Merlet, Direct geometrico-static problem of under-constrained cable-driven parallel robots with three cables, Proc. of the 2011 IEEE Int. Conf. on
Robotics and Automation, Shanghai, China 2011, pp. 3011–3017.
[21] G. Abbasnejad, M. Carricato, Direct geometrico-static problem of underconstrained cable-driven parallel robots with n cables, IEEE Trans. Robot. 31 (2) (2015)
468–478.
[22] A. Fattah, S.K. Agrawal, On the design of cable-suspended planar parallel robots, ASME J. Mech. Des. 127 (5) (2006) 1021–1028.
[23] G. Abbasnejad, M. Carricato, Real solutions of the direct geometrico-static problem of under-constrained cable-driven parallel robots with 3 cables: a numerical
investigation, Meccanica 47 (7) (2012) 1761–1773.
[24] I. Ebert-Uphoff, P. Voglewede, On the connections between cable-driven robots, parallel manipulators and grasping, IEEE International Conference on Robotics
and Automation, vol. 5 2004, pp. 4521–4526.
[25] J. Pusey, A. Fattah, S. Agrawal, E. Messina, Design and workspace analysis of a 6–6 cable-suspended parallel robot, Mech. Mach. Theory 39 (7) (2004) 761–778.
[26] G. Barrette, C.M. Gosselin, Determination of the dynamic workspace of cable-driven planar parallel mechanisms, J. Mech. Des. 127 (2) (2005) 242–248.
[27] G. Rosati, D. Zanotto, S. Agrawal, On the design of adaptive cable-driven systems, J. Mech. Robot. 3 (2).
[28] D. Zanotto, G. Rosati, S. Minto, A. Rossi, Sophia-3: a semiadaptive cable-driven rehabilitation device with a tilting working plane, IEEE Trans. Robot. 30 (4) (2014)
974–979.
[29] X. Zhou, S. Jun, V. Krovi, Tension distribution shaping via reconfigurable attachment in planar mobile cable robots, Robotica 32 (2014) 245–256.
[30] D. Nguyen, M. Gouttefarde, O. Company, F. Pierrot, On the analysis of large-dimension reconfigurable suspended cable-driven parallel robots, IEEE International
Conference on Robotics & Automation (ICRA) 2014, pp. 5728–5735.
[31] J. Kennedy, R. Eberhart, Particle swarm optimization, Proceedings of IEEE International Conference on Neural Networks, vol. 4 1995, pp. 1942–1948.
[32] G. Abbasnejad, H.M. Daniali, S.M. Kazemi, A new approach to determine the maximal singularity-free zone of 3-rpr planar parallel manipulator, Robotica 30 (6)
(2012) 1005–1012.
[33] G. Abbasnejad, H.M. Daniali, A. Fathi, Architecture optimization of 4pus + 1 ps parallel manipulator, Robotica 29 (5) (2011) 683–690.
[34] R.G. Roberts, T. Graham, T. Lippitt, On the inverse kinematics, statics, and fault tolerance of cable-suspended robots, J. Robot. Syst. 15 (10) (1998) 581–597.
[35] R. Verhoeven, Analysis of the Workspace of Tendon-Based Stewart PlatformsPh.D. thesis University of Duisburg-Essen, Germany, 2004.
[36] P. Gallina, G. Rosati, Manipulability of a planar wire driven haptic device, Mech. Mach. Theory 37 (2) (2002) 215–228.
[37] H. Li, C. Gosselin, M. Richard, Determination of maximal singularity-free zones in the workspace of planar three-degree-of-freedom parallel mechanisms, Mech.
Mach. Theory 41 (10) (2006) 1157–1167.
[38] A. Karimi, M. Masouleh, P. Cardou, Singularity-free workspace analysis of general 6-ups parallel mechanisms via convex optimization, Mech. Mach. Theory 80
(2014) 17–34.
[39] M. Kaloorazi, M. Masouleh, S. Caro, Determination of the maximal singularity-free workspace of 3-dof parallel mechanisms with a constructive geometric ap-
proach, Mech. Mach. Theory 84 (2015) 25–36.
[40] C. Vaughan, B. Davis, J. O'Connor, Dynamics of Human Gait, Kiboho Publishers, 1992.
[41] D.A. Winter, Biomechanics and Motor Control of Human Movement, fourth ed. John Wiley & Sons, 2009.
[42] I. Ebrahimi, J. Carretero, R. Boudreau, Kinematic analysis and path planning of a new kinematically redundant planar parallel manipulator, Robotica 26 (2008)
405–413.