2390 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 23, NO.

5, OCTOBER 2018

Research on Controllable Stiffness of

Redundant Cable-Driven Parallel Robots
Zhiwei Cui , Xiaoqiang Tang , Senhao Hou, and Haining Sun

Abstract—In this paper, to solve the problem of variable Cable-driven parallel robots (CDPRs) are a special class of
stiffness of cable-driven parallel robots (CDPR), a new static parallel robots in which their moving-platforms are driven by
stiffness analysis and cable tension distribution method are cables instead of rigid links. This class of robots has been ex-
proposed for studying the CDPRs’ controllable stiffness.
First, a three-dimensional Hessian matrix of the structure tensively researched and is highly favored because of its ex-
matrix to position differential is deduced by introducing a cellent movement performance, high load capacity, and large
line vector and differential transform, and a static stiffness workspace [1]–[8]. Furthermore, as the driving cables can pull
model is established for analyzing the relationship between but cannot push [9], [10], a CDPR is always a redundantly ac-
cable tension and the stiffness of the robots. Furthermore,
tuated system. As a redundant CDPR has an infinite number of
a calculation algorithm for cable tension polygons based on
Graham’s Scan is introduced that effectively obtains the ca- feasible cable tension solutions at a specific position, many stud-
ble tension feasible region (CTFR) of CDPRs. Next, a method ies have been conducted on optimization of the cable tensions
involving “the relationship between the external force and [11]–[15]. Among the different types of CDPRs, the redundant
the pose change value of the moving-platform” is proposed CDPR has more actuators than the moving-platform’s degrees
to measure the variation of system stiffness. The CDPR’s
of freedom (DOFs) and has a degree of redundancy (DOR)
controllable stiffness is also analyzed based on the CTFR
by considering the variation of each driving-cable’s tension given by r = m − n, where m and n are the number of ca-
and each component of the moving-platform’s pose. The bles and DOF, respectively. Hence, at least m = n + 1 cables
results of experimental and theoretical analyses verify the are needed to completely control an n-DOF CDPR. Note that
correctness and efficacy of the proposed method. In addi- if gravity plays the role of an additional virtual cable, n-DOF
tion, they show that the proposed method is computation-
CDPRs can be completely controlled by n cables [16]. In addi-
ally efficient and easily establishes the relationship between
cable tension and the CDPR’s controllable stiffness. tion, redundant CDPRs can adjust to stiffness through changing
the cable tensions [2], [5], which makes it possible to develop
Index Terms—Cable-driven robots, cable tension distri- variable-stiffness robots with stiffness adaptive adjustment and
bution, controllable stiffness, redundant systems.
solve the security problem in “human–robot interaction” for col-
I. INTRODUCTION laboration robots. Redundant actuators can improve the working
space, stiffness, and load capacity of CDPRs [2], but cause ca-
ITH the rapid development of robot technology and
W the continuous increase in social demand, human–robot
communion and collaborative operation will become an essen-
bles and the environment to be prone to collision or interference,
which makes CDPRs more complex and costly. Considering the
advantages, cost, and complexity of the redundant drive, six-
tial characteristic of next-generation robots. However, the ques- DOF CDPRs are driven by eight cables [17]–[20]. This paper is
tion of how to ensure the safety of persons and machines for focused on the characteristics of controllable stiffness of redun-
human–robot interaction is a difficult problem that has to be dantly actuated CDPRs intended for human–robot communion
solved concomitant with the development of collaboration-robot and collaborative operation. Experiments on a two-DOR eight-
technology. Therefore, it is particularly important to study the cable CDPR prototype, TCPR-8 (see Fig. 1) are presented.
controllable stiffness of the robot. The stiffness problem is an important issue that needs to
be solved in order to improve the performance of robots. It
Manuscript received December 19, 2017; revised March 28, 2018; is closely related to the dynamic characteristics and position-
accepted July 30, 2018. Date of publication August 8, 2018; date of
current version October 15, 2018. Recommended by Technical Editor
ing accuracy of robots. Consequently, numerous theoretical and
T. Seo. This work was supported in part by the National Natural Science practical studies have been conducted in this area. From analy-
Foundation of China under Grant 91648107 and Grant 51475252, and in sis of the spring equivalent model of the cable in consideration
part by Tsinghua University Initiative Scientific Research Program under
Grant 2014z22068. (Corresponding author: Xiaoqiang Tang.)
of the prestress, studies [21] have verified that insufficient tor-
The authors are with the State Key Laboratory of Tribology and In- sional stiffness is the main reason for the instability of robots
stitute of Manufacturing Engineering, Department of Mechanical Engi- and stated that this can be improved by increasing the prestress.
neeringand the Beijing Key Lab of Precision/Ultra-precision Manufactur-
ing Equipment and Control, Tsinghua University, Beijing 100084, China
Recent studies that considered the cable sag due to gravity and
(e-mail:,[email protected]; [email protected]; elastic deformation of cables [4], [22] analyzed the stiffness
[email protected]; [email protected]). of CDPRs and found that the self-weight and elastic deforma-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
tion of cables have a significant effect on the system stiffness,
Digital Object Identifier 10.1109/TMECH.2018.2864307 which can be enhanced through optimization of the structural

1083-4435 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Université de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 CUI et al.: RESEARCH ON CONTROLLABLE STIFFNESS OF REDUNDANT CABLE-DRIVEN PARALLEL ROBOTS 2391

Fig. 1. TCPR-8 CDPR prototype.

Fig. 2. Simplified structural model of CDPR.

parameters. Yeo and Azadi installed a variable-stiffness device

on the driving cables to expand the range of robot stiffness
The remainder of this paper is structured as follows:
changes. However, they did not establish the exact relationship
Section II reviews static analysis of redundant CDPRs.
between the system stiffness and the tension of the cables; thus,
Section III develops the static stiffness model of redundant CD-
the system stiffness could not be controlled effectively by alter-
PRs. Section IV gives a cable tension distribution method for
ing the cable tension [5], [23]. Hesselroth and Hennessey [24]
two-DOR CDPRs. Section V analyzes and controls the control-
showed that the z-direction stiffness of a double Stewart platform
lable stiffness of CDPRs. Section VI presents experimental ver-
system is about three times greater than that of a single Stewart
ification and discussion. Section VII analyzes the experimental
platform system by calculation and analysis with MATLAB.
results and concludes.
However, the double Stewart platform system significantly in-
creases the energy consumption and complexity of the system.
Most of the above studies tried to improve the position accuracy II. STATIC ANALYSIS
or load capacity of the robot by analyzing or enhancing the stiff- Static analysis of CDPRs is the basis for studying the stiff-
ness of the robots. In contrast, ensuring the safety of persons and ness problem. In this section, a general model of CDPRs is
machines in human–robot interaction while improving the ac- established and its position and statics analyzed.
curacy of the work and load capacity, and establishing the exact Using an m-cable n-DOF CDPR (m − n = 2) as the object
relationship between the system stiffness and the tension of the of research, the corresponding general model is as illustrated in
cables in order to determine the characteristics of controllable Fig. 2. Each cable is connected to the moving-platform at one
stiffness of CDPRs have not been studied to date. end and is pulled by a fixed winch at the other end. According
When a CDPR is employed in an unstructured environment to the cable catenary model [25], as the length and width of the
or human–robot interaction, and collision is usually inevitable, robots in this paper are both less than 2 m, the deformation of
it poses a security threat to people and products owing to the cable due to gravity can be ignored and a linear model is
its high mechanical stiffness and lack of flexibility. Thus, to utilized instead of the cable catenary model.
improve the flexibility of these robots and ensure the safety To clearly describe the pose of the moving-platform for CD-
of persons or products, robots with low stiffness, which can PRs, illustrated in Fig. 2, the base coordinate system O and
absorb and cushion the energy generated by collisions with- the moving coordinate system O of CDPRs are assigned, re-
out harming persons or products, and can also be applied to spectively, where Ai is the fix-exit-point of the cable and Bi
flexible jobs such as polishing and rehabilitation therapy, are is the connection point between the cable and the moving-
required. platform. By using the two coordinate systems, the position
Therefore, the characteristics of two-DOR CDPRs, and their of the moving-platform can be described via the position vec-
kinematics and statics are analyzed in this paper. To research the tor p = [xo  , yo  , zo  ]T (in meters) of the origin of O in O,
characteristics of controllable stiffness for CDPR, a static stiff- and the posture can be described through the angular displace-
ness model is established by deducing the controllable stiffness ment Θ = [θx , θy , θz ]T (in degrees) of O relative to O. Thus,
and inherent stiffness matrix. Then, a cable tension distribution x = [p; Θ] can denote the pose of the moving-platform in O.
method and the relationship between cable tension and system Position analysis of CDPRs is conducted to establish the rela-
stiffness are investigated by calculating the cable tension poly- tionship between the pose of the moving-platform and the cable
gon. System stiffness can be effectively controlled by adjusting length. According to the vector loop equation, the relationship
cable tension. This satisfies the static balance and avoids the between the moving-platform’s pose and the cable length is as
pseudodrag of cables. Finally, the correctness and validity of follows:
the method presented are verified by theoretical analysis and

two sets of experiments. li = ai − p − R · o bi (1)

Authorized licensed use limited to: Université de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 2392 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 23, NO. 5, OCTOBER 2018

where li denotes the cable vector from Bi to Ai , ai is the system structure and the moving-platform’s pose. It is called the

position vector of Ai in O, o bi is the position vector of Bi in inherent stiffness.
O , R represents the rotation matrix of O relative to O, and
i = 1, 2, . . . , m. A. Controllable Stiffness
When the target pose x is determined, the cable length li 
The controllable stiffness is the key to realizing stiffness adap-
and the cable unit vector ui = li /li  can be calculated via (1).
tive adjustment of robots. This section deduces the matrix of the
The relationship between the moving-platform’s pose and the
controllable stiffness by introducing a line vector and differen-
cable length is also determined.
tial transformation. From (6), K 1 is written as follows:
Using (1), the expression for the kinematic equation of CD-
PRs is given as follows: K 1 = −HT (7)
x = f (l) (2) where H = [H 1 , H 2 , . . . , H n ] ∈ Rn ×n ×m represents the
where f denotes a nonlinear vector equation. The relationship three-dimensional Hessian matrix, H j ∈ Rn ×m is the jth sub-
between the velocity vector of cables and the moving-platform’s array of H, and
velocity vector via the differentiation of (2) with respect to time ⎡ ∂ u1 ∂ um
⎤
∂ xj ··· ∂ xj
can be obtained as follows: ∂J
Hj = = ⎣ ∂ (R·o  b )×u 
∂ [(R·o bm )×um ]
⎦ . (8)
∂xj [ 1]
···
1
l̇ = −J T · ẋ (3) ∂ xj ∂ xj

where l̇ = [l̇1 , l̇2 , . . . , l̇m ]T is the velocity vector of the cables, To simplify the derivation process of H, the cable unit vector
J ∈ Rn ×m is the Jacobian matrix of the CDPRs at a pose, and ui is written with the direction cosine as follows:
ẋ = [ṗ, Θ̇] denotes the velocity vector of the moving-platform,
where ui = [cαi , cβi , cγi ]T (9)
 
u1 ··· um where αi , βi , and γi denote the angle of ui about the Cartesian
J (x) =   .
(R · o b1 ) × u1 · · · (R · o bm ) × um axis x, y, and z, respectively.
According to the principle of virtual work, the static equilib- (Note: To simplify the expression, we use cθ and sθ instead
rium equation of CDPRs can be written as follows: of cos θ and sin θ in this paper, respectively.)

From (8) and (9), the partial derivatives of ui and (R · o bi ) ×
JT + F = 0 (4) ui with respect to xj can be obtained as follows:
where T = [t1 , t2 , . . . , tm ]T , ti is the value of the ith cable ∂ui ∂ui ∂αi ∂ui ∂βi ∂ui ∂γi
tension, F = [f e , me ]T , f e and me represent the resultant = + + , (10)
∂xj ∂αi ∂xj ∂βi ∂xj ∂γi ∂xj
force and torque (including the moving-platform’s weight) of  
the external environment acting on the moving-platform (with ∂ (R · o bi ) × ui ∂(R · o bi )  ∂ui
= × ui + (R · o bi ) × .
units N and N·m, respectively). ∂xj ∂xj ∂xj
(11)
III. STATIC STIFFNESS MODEL
Because ui is a unit vector, (10) can be written as follows:
Studying the static stiffness of the robots is the basis for  T
researching variable-stiffness robots and is also the key to re- ∂ui cαi · sαi ∂αi
= −sαi , 0, ·
alizing collaboration robots. The antideformation ability of the ∂xj cγi ∂xj
moving-platform under the action of external force is one of the  T
criteria to measure the system stiffness. Therefore, the relation- cβi · sβi ∂βi
+ 0, −sβi , · . (12)
ship between the small external force δF acting on the moving- cγi ∂xj
platform and the small variation δx of the moving-platform’s Using (1), the following expression can be obtained:
pose can be obtained as follows:

li · ui = ai − p − R · o bi . (13)
δF = K · δx (5)
where K denotes the system stiffness matrix. Because R = ω × R, via the derivatives of (13) with respect
Using (4) and (5), the expression of the stiffness matrix is to time, we obtain the following:

given by l˙i · ui + li · u̇i = −ṗ − ω × (R · o bi )
 
∂F ∂J ∂T 
K= =− T +J = K 1 + K 2. (6) = −I R · o bi × ẋ = Gi ẋ (14)
∂x ∂x ∂x
As can be seen from (6), the system stiffness matrix consists of where Gi ∈ R3×6 denotes the algebraic matrix.
two parts: K 1 is related to the structure matrix transformation From (3), we obtain the following:
and cable tension. It can be controlled by altering the cable l̇i = −J Ti ẋ (15)
tension at a certain pose, and is called the controllable stiffness
(it is the main research focus in this paper). K 2 is related to the where J Ti is the ith row vector of J T .

Authorized licensed use limited to: Université de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 CUI et al.: RESEARCH ON CONTROLLABLE STIFFNESS OF REDUNDANT CABLE-DRIVEN PARALLEL ROBOTS 2393

Using (15), we obtain the following: Substituting (23) and (24) into (21), we obtain the following:
⎡ ⎤  
cαi · J Ti E1 · A1 Em · Am
⎢ ⎥ K 2 = J · diag ,··· , · JT . (25)
l˙i ui = − ⎣cβi · J T ⎦ ẋ = Qi ẋ (16) lo1 lom
i
cγi · Ji
T The stiffness matrix of controllable stiffness K 1 and inher-
ent stiffness K 2 can be calculated using the above equations,
where Qi ∈ R3×6 represents the algebraic matrix. respectively, and the system stiffness K can be obtained.
Substituting (16) into (14), we obtain the following:
 T IV. CABLE TENSION DISTRIBUTION
li · −α̇i · sαi , −β̇i · sβi , −γ̇i · sγi = [Gi − Qi ] ẋ. (17)
When the external load of the moving-platform is constant,
From (17), we obtain the following: effective adjustment of the driving-cable’s cable tension, while
 ∂ αi satisfying the system balance and avoiding cable slack or over-
∂ xj = − l i ·sα i [Gi − Qi ]1,j
1
load, is important for cable tension distribution and is also the
(18) basis of control system stiffness. From Section III-A, it is clear
∂ βi
∂ xj = − l i ·sβ
1
i
[Gi − Qi ]2,j
that the controllable stiffness of CDPRs can be adjusted by al-
where [Gi − Qi ]1,j and [Gi − Qi ]2,j are the elements of the tering the cable tensions. Therefore, a method for cable tension
first row, the jth column and the second row, the jth column of distribution is proposed in this paper. The proposed method first
the matrix [Gi − Qi ], respectively. calculates the feasible region of cable tension, and distributes

Through the derivatives of R · o bi with respect to time, we cable tension in that region. The method is computationally effi-
obtain the following: cient, and more easily establishes the relationship between cable
 
  
 tension and CDPRs’ controllable stiffness.
R · o bi = ω × R · o bi = M i ẋ (19) As this paper is dedicated to the problem of cable tension
distribution for two-DOR CDPRs, the n × m Jacobian matrix
where M i ∈ R3×6 is the algebraic matrix. J is nonsquare. When J has full rank, (4) is equivalent to the
Equation (19) can be rewritten as follows: following well-known equation:

∂ R · o bi T = J + (−F ) + N λ = tp + tn (26)
= M i,j (20)
∂xj where J + = J T (J J T )−1 is the pseudo-inverse matrix of ma-
where M i,j denotes the jth column vector of M i . trix J , N = null(J ) denotes a full-rank m × r matrix, the two
H can be obtained by substituting (12), (18), and (20) into columns of N form an orthonormal basis of the null-space of J ,
(10), (11), and (8). The stiffness matrix K 1 can be calculated λ = [λ1 λ2 ]T is an arbitrary two-dimensional (2-D) vector rep-
using (7) and the cable tension vector T . resenting one point, λ1 ∈ R, λ2 ∈ R, J + (−F ) represents the
minimum-norm solution of (4), and N λ is the homogeneous
B. Inherent Stiffness general solution of (4).

Although the inherent stiffness is not controlled by the cable A. Cable Tension Polygon
tension, it still has a significant impact on the system stiffness.
Therefore, the process of deducing the matrix of inherent stiff- The region surrounded by the cable tension polygon is called
ness is given in this section. Using (6), we obtain the following: the cable tension feasible region (CTFR), which is the basis for
studying CDPR’s cable tension distribution and control. The
∂T ∂l definition and calculation method for the cable tension polygon
K 2 = −J . (21)
∂l ∂x are given in this section.
The relationship between the cables’ deformation and the Let Σ ⊂ Rm be the r-dimensional affine space of the solutions
cables tension is written as follows: to (4); the expression can be written as follows:
Ei · Ai Σ = {T |J T + F = 0} . (27)
∂ti = · ∂li (22)
loi
Similarly, we can define Ω ⊂ Rm as the m-dimensional or-
where Ei , Ai , and loi are the elastic modulus, cross-sectional thotope of feasible cable tensions; the expression can be ob-
area, and static length of the cable, respectively. tained as follows:
From (22), we obtain the following:
  Ω = {T |ti ∈ [tm in , tm ax ], 1 ≤ i ≤ m} (28)
∂T E1 · A1 Em · Am
= diag ,··· , . (23) where tm in and tm ax are the minimum and maximum limit-
∂l lo1 lom
ing values of cable tensions, the intersection Λ = Σ ∩ Ω is a
Using (3), we obtain the following: 2-D convex polytope [26], which can be referred to as a cable
∂l tension polygon. Λ denotes the set of tension solutions T to
= −J T . (24) (4) satisfying the inequalities tm in ≤ ti ≤ tm ax , which can be
∂x

Authorized licensed use limited to: Université de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 2394 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 23, NO. 5, OCTOBER 2018

Fig. 3. TCPR-8 in static equilibrium at pose x = [0.1, 0.1, 0.8, 5, 5, 5]T ,

and its moving-platform with an external load F = [300, 300, 300,
10, 10, 10]T , tm in = 10 N, and tm a x = 1000 N. (a) Incomplete cable ten-
sion polygon (two cables define four half-plane intersections). (b) Com-
plete cable tension polygon.

defined via the following set of 2m linear inequalities:

tm in − tp ≤ N λ ≤ tm ax − tp . (29)

When the limiting values tm in and tm ax of cable tensions

are determined, each inequality defines a half-plane through
a boundary line with λ changes. Any n(1 < n ≤ m) cables
define 2n half-plane intersections that can form the CDPR’s
incomplete cable tension polygon ψ and the CDPR’s complete
cable tension polygon (or cable tension polygon for short) if
and only if m is equal to n. For example, the intersection of the
four half-planes defined by cables No. 1 and No. 2 can form
Fig. 4. Flowchart of the cable tension polygon algorithm.
an incomplete cable tension polygon of CDPRs—illustrated in
Fig. 3(a). The intersection of the 2m half-planes in (29) forms
the CDPR’s complete cable tension polygon Λ—illustrated in
Fig. 3(b). Obviously, Λ ⊆ ψ, and is only one.
Note: To facilitate expression, L = {Lij |i = 1, 2, . . . , m; polygon calculation algorithm was designed based on Graham’s
j = 1, 2} denotes the boundary line of cable tension (i rep- Scan method. The calculation principle is as follows: The cable
resents the ith cable, j = 1 represents the case where tm in is the tension polygon is calculated stepwise via cyclic calculation of
limiting value of cables tension, j = 2 denotes the case where the convex hull of the intersection points for the boundary line
tm ax is the limiting value) in this paper. The number of inter- of cable tension.
section points for the boundary line can be calculated via the The calculation process is shown in Fig. 4. The steps are as
equation C22m − m = 4C2m . The CDPR’s cables are not parallel follows.
to each other; otherwise, the CDPRs would appear singular. The Step 1: Obtain the point sequence A p0 , p1 , . . . , pn by cal-
CDPR is driven via m identical cables in this paper, therefore, culating the intersections of the boundary line of cable
the limiting values tm in and tm ax of each cable tension are the tension and remove the repeat point.
same. Step 2: Compute the lowest and leftmost point in A, which is
called base point p0 .
Step 3: Sort the pole angles of the points in the sequence
B. Cable Tension Polygon Calculation
of points p1 , p2 , . . . , pn relative to point p0 in as-
According to the characteristic of the cable tension polygon cending order. Correspondingly, update the sequence
and the boundary line of the cable tension, the cable tension of points as follows: A p 1 , p 2 , . . . , p n (sort the

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 CUI et al.: RESEARCH ON CONTROLLABLE STIFFNESS OF REDUNDANT CABLE-DRIVEN PARALLEL ROBOTS 2395

Fig. 6. Cable tension polygon.

Fig. 5. Points after sorting.

points with the same polar angle according to the

distance to p0 ), as shown in Fig. 5.
Step 4: Push successively p0 and p1 into stack S.
Step 5: Let i = 1.
Step 6: Assign i = i + 1.
Step 7: Let the current point p= pi .
Step 8: Obtain the directed line segment L by connecting the
top two elements ptop−1 and ptop of stack S.
Step 9: If p is on the right side of L, delete the top elements of
stack S, and then perform Step 8; otherwise, execute Fig. 7. Reality edge and virtual edge.
the next step.
Step 10: Push p into stack S.
Step 11: If p = pn , execute the next step; otherwise, execute
Step 6.
Step 12: If the connection of any two elements in stack S
contain a reality edge, then execute the next step;
otherwise, execute Step 14.
Step 13: If A contains points that do not satisfy the reality edge, Fig. 8. Sketch map for closed points’ sets. (a) Sets of closure points.
delete those points and execute Step 2; otherwise, (b) Sets of nonclosure points.
execute Step 15.
Step 14: If A is closure, then let A = A − S and execute
Step 2; otherwise, execute the next step. The cable tension polygon has the following nature.
Step 15: If A contains invalid points, delete those points un- Nature 1: Let pc = {pic |i = 1, 2, . . . , m} be the set of ca-
til A does not contain any invalid points; otherwise, ble tension polygon vertices, and m ≥ 3, where pjc and pkc (j =
output A. k;j, k = 1, 2, . . . , m) are any two elements in pc . Then
Step 16: End. 1) pjc is not adjacent to pkc is a necessary and sufficient
The definition of special terms used in the algorithm and the condition for the line segment pjc pkc to be the virtual edge
nature of the cable tension polygons are given below. of the cable tension polygon, as illustrated in Fig. 7;
Definition 1: Let p = {pi |i = 1, 2, . . . , n} be the set of inter- 2) pjc is adjacent to pkc is a necessary and sufficient condition
sections for the boundary line of cable tension, where pj and pk for the line segment pjc pkc to be the reality edge of the cable
(j = k; j, k = 1, 2, . . . , n) are any two elements in p. If ∃ l ∈ L tension polygon, as shown in Fig. 7.
meet the line segment pj pk collinear with l, then pj pk is called Definition 3: Let pr = {pir |i = 1, 2, . . . , n} be the set of in-
the reality edge of the cable tension polygon, as shown in Fig. 6. tersections of the boundary line of cable tension, and n ≥ 3,
If ∀ l ∈ L does not meet the line segment pj pk collinear with l, where Sr is pr ’s convex hull. Let pd = pr − Sr , if the set of
then pj pk is called the virtual edge of the cable tension polygon, points pd can form a closed area, then pr is called closure;
as illustrated in Fig. 6. otherwise, pr is called nonclosure, as shown in Fig. 8.
Definition 2: Let p = {pi |i = 1, 2, . . . , n} be the set of in- In a certain pose of CDPRs, the CTFR can be calculated
tersections for the boundary line of cable tension, and pc = effectively using the above algorithm, when the external load
{pic |i = 1, 2, . . . , m} be the set of cable tension polygon ver- of the moving-platform and the limiting values of cable tension
tices, where m ≥ 3. If ∀pk ∈ p, pk ∈ / pc , then pk is an invalid are constant. The algorithm establishes the foundation for the
point. Otherwise, if ∀pk ∈ p, pk ∈ pc , then pk is a valid point, cable tension optimization of CDPRs. The calculation example
as shown in Fig. 6. of CTFR is illustrated in Fig. 3(b).

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 2396 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 23, NO. 5, OCTOBER 2018

Fig. 10. Change curves of cable tension.

Fig. 9. Centroid line.

C. Centroid Line
The centroid C of the cable tension polygon is the geometric
center of CTFR and is the only point as far away from the
boundary as possible in CTFR [27], and is also the point for the
most smooth and stable of cable tension. Any straight line that
crosses the centroid divides the CTFR into two parts of equal
area.
The calculation method for the centroid and centroid line is as
follows. Let pc be the set of cable tension polygon vertices, and
pic (i = 1) are sorted in ascending order of the pole angles that pic
is relative to point p1c , where pic = [λi1 , λi2 ]T . The set of vertices
pc can be calculated using the algorithm given in Section IV-B.
The centroid C = [λc1 , λc2 ]T of the cable tension polygon is
directly given via the following mathematical formula: Fig. 11. Configuration of TCPR-8.
 n −1
λ c1 = 1
6A i= 1 (λ i 1 + λ (i+ 1) 1 )(λ i 1 λ (i+ 1) 2 − λ (i+ 1) 1 λ i 2 )
n −1
λ c2 = 1
6A i= 1 (λ i 2 + λ (i+ 1) 2 )(λ i 1 λ (i+ 1) 2 − λ (i+ 1) 1 λ i 2 ) as an example, the TCPR-8’s curves of cable tension can be
(30) obtained when λ 1 and λ 2 change along the centroid line seg-
where A is the area of the cable tension polygon, and its calcu- ment, as shown in Fig. 10.
lated formula is as follows: As can be seen from Fig. 10, as λ1 and λ2 increase, the driving-
n −1
1 cable’s tension of CDPRs increases gradually. It can also be
A= (λi1 λ(i+1)2 − λ(i+1)1 λi2 ). (31) concluded that the tension values of each cable is positively
2
i=1 correlated with the changing value of λ1 and λ2 . Therefore, the
As illustrated in Fig. 3(b), the yellow dot is the centroid C of driving-cable’s tension can be controlled when λ1 and λ2 change
the cable tension polygon Λ. At centroid C, the cable tension along the centroid straight line; thus, the CDPR’s stiffness can
vector T C = J + (−F ) + N λC = tp + N λC . be controlled.
The straight line that crosses the centroid can be obtained via
the following mathematical formula: V. CONTROLLABLE STIFFNESS ANALYSIS AND CONTROL
 To show the details of the proposed model and algorithm, and
λ2 = (tanα)(λ 1 − λ c1 ) + λ c2 , α ∈ (−π/2, π/2)
(32) demonstrate its correctness and effectiveness, the TCPR-8’s real
λ1 = λc1 , α = π/2 robot prototype is presented in this paper. Fig. 1 illustrates the
prototype of the studied six-DOF CDPR designed for simulated
where α is the angle between the centroid line and the λ1 -axis.
satellite launch. The cable layout and the dimensions of the
For example, let α = α1 + k ∗ μ, where initial value α1 =
CDPR’s fixed and moving-platforms are represented in Fig. 11
−5π/12, μ = π/12, and k is equal to 0, 1, . . . , 11. The straight
and Table I. Table II displays the mechanical properties of the
lines that cross the centroid can be given when the exter-
robot.
nal load F = [300, 300, 300, 10, 10, 10]T and the pose x =
[0.1, 0.1, 0.8, 5, 5, 5]T of the TCPR-8’s moving-platform are de-
termined, as shown in Fig. 9. A. Stiffness Measurement Method
As illustrated in Fig. 9, in the CTFR, the values of λ 1 and In order to effectively measure the variation of system stiff-
λ 2 gradually increase along the straight line that crosses the ness, a method involving “the relationship between the external
centroid from left to right when α ∈ (0, π/2). Taking α = π/4 force and the pose change value of the moving-platform” is

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 CUI et al.: RESEARCH ON CONTROLLABLE STIFFNESS OF REDUNDANT CABLE-DRIVEN PARALLEL ROBOTS 2397

TABLE I
TCPR-8’S DIMENSIONS (IN METERS)

TABLE II
TCPR-8’S MECHANICAL PROPERTIES ROPE CROSS-SECTIONAL DIAMETER
Fig. 12. Cable tension feasible region.

proposed and adopted in this paper. Obviously, under the same

external force, the greater the pose changes of the moving-
platform the smaller the system stiffness at a certain pose, and
vice versa.
From (5), we obtain the following:
Δx = K − · F (33)
where Δx = [Δxo  , Δyo  , Δzo  , Δθx , Δθy , Δθz ]T denotes the
pose’s change value, and K − = (K T K)−1 · K T is the gener-
alized inverse matrix of K.
According to the external load of the moving-platform, the
pose’s change value of the moving-platform at a certain pose can
be calculated via (33). Therefore, the CDPR’s system stiffness
can be analyzed by analyzing the relationship between the exter-
nal force and the pose’s change value of the moving-platform.

B. Stiffness Analysis and Control Fig. 13. Change surface of cable tension.
To more clearly describe the relationship between the CDPR’s
stiffness and cable tension, and effectively control the robot
stiffness, a numerical analysis utilizing the proposed model and
algorithm are conducted in this section. In order to more clearly
show the control effect for the CDPR’s stiffness via cable ten-
sion, the limiting value tm ax of the CDPR’s driving-cable is
assumed to be 1000 N (the other parameters remain unchanged)
for numerical analysis.
As illustrated in Fig. 12, under the action of external load F =
[300, 300, 300, 10, 10, 10]T , the CTFR can be obtained when the
TCPR-8’s moving-platform at pose x = [0.1, 0.1, 0.8, 5, 5, 5]T .
As λ1 and λ2 change, the change surface of cable tension for each
driving-cable can be obtained in CTFR, as shown in Fig. 13. At
this time, the surface can be formed as each component of the
moving-platform’s pose changes in CTFR, illustrated in Fig. 14.
Table III compares the maximum and minimum change values
of the pose’s components.
The results from Section IV-C and Fig. 13 indicate that as
λ1 and λ2 increase, each driving-cable’s tension increases grad-
Fig. 14. Change surfaces of pose component.
ually. It can also be concluded that the tension values of each

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 2398 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 23, NO. 5, OCTOBER 2018

TABLE III
COMPARISON OF THE MINIMUM AND MAXIMUM CHANGE VALUES OF THE
POSE’S COMPONENTS

cable are positively correlated with the change value of λ1 and

λ2 . Among them, except for Δyo  and Δθx , the maximum and
minimum values of each pose component correspond to points
A and C in CTFR, respectively, as illustrated in Figs. 12 and
14. This shows that the stiffness in these directions gradually Fig. 15. Change curves of pose. (a) Line displacement. (b) Angular
increases as the cable tension increases. The maximum and min- displacement.
imum values of Δθx correspond to points B and C in CTFR,
respectively, as shown in Figs. 12 and 14. Points A and B are very TABLE IV
close, and their coordinate components λ1 and λ2 are also small. POSE’S CHANGING VALUE FOR THE END OF THE π/4 CENTROID LINE
Therefore, the θx -direction also conforms to the conclusion that
the stiffness gradually increases as the cable tension increases.
In this pose, the x- and θz -directions are the directions of the
fastest increases for translational and rotational stiffness, which
increase by almost 125% and 180%, respectively. Therefore,
the proposed method can effectively control CDPRs’ system
stiffness as it significantly changes the system stiffness in this
paper.
The maximum and minimum values of Δyo  correspond to
points C and A in CTFR, respectively, as shown in Figs. 12 and
14. Δyo  increases with the increasing cable tension owing to
the asymmetric increase of the cable tension of the symmetry
mechanism, as illustrated in Fig. 13. The rate of increase of the
cable tension for the third, fourth, seventh, and eighth cables is
VI. EXPERIMENTAL VERIFICATION AND DISCUSSION
significantly more than that of the four cables in the opposite
direction. To verify the proposed model and algorithm, two sets of ex-
From (33), illustrated in Fig. 9, under the control of varying periments were conducted for the TCPR-8’s moving-platform
cable tension, the change curves of pose along the α = π/4 at pose x = [0.1, 0.1, 0.8, 0, 0, 0]T . In each experiment, using
centroid line can be obtained, as shown in Fig. 15. Table IV a Leica AT901-B laser tracker measurement system, with pre-
presents the pose’s changing value for the end of the centroid cision 0.001 mm, the change of the moving-platform’s pose
line. was recorded. To provide a constant external load of 200 N
The CDPR’s stiffness gradually increases, as the results from for the TCPR-8’s moving-platform, heavy objects were hung
Fig. 15 and Table IV indicate, when λ1 and λ2 change along on the platform. The value of the driving-cable’s tension was
the α = π/4 centroid line from left to right. The stiffness of recorded using a DJSX-44-100KG S-type pull pressure sensor.
the x- and θz -directions increase by almost 120% and 160%, Figs. 1 and 17 show the details of the measurement systems and
respectively. Therefore, the CDPR’s stiffness can be effectively hanging objects.
controlled along the centroid line in CTFR, according to the In the first experiment, the moving-platform’s external load
stiffness requirements. For example, the stiffness of the θz - was 200 N in the x-direction. In CTFR, illustrated in Fig. 18,
−−−−−−−→
direction is increased to 150%, and the change of Δθz is point with the change of λ1 and λ2 in the Am in Am ax -direction, the
pe , as shown in Fig. 15(b). λ1 , λ2 , and the values of each cable tension change curves of each of the TCPR-8’s driving cables
tension are presented in Table V. was obtained, where Am in and Am ax denote the maximum and
Based on CTFR, CDPR’s stiffness can also be effectively minimum values of Δxo  corresponding to the points in CTFR.
controlled along the inner equidistant lines of the cable tension As shown in Fig. 18, we obtain five points equidistant,Am in ,
polygon, as illustrated in Fig. 16. A1 , A2 , A3 , and Am ax , on line segment AB, and five sets of

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 CUI et al.: RESEARCH ON CONTROLLABLE STIFFNESS OF REDUNDANT CABLE-DRIVEN PARALLEL ROBOTS 2399

TABLE V
COMPARISON OF CABLE TENSION AND Δθz IN INITIAL POSITION AND p e (IN NEWTON AND rad × 10−2 )

Fig. 16. Inner equidistant curve.

Fig. 19. Comparison curves between experimental and theoretical

Δxo  .

Fig. 17. (a) Leica AT901-B laser tracker. (b) Hanging heavy objects.

Fig. 20. Change curves of cable tension: 2nd experiment.

Fig. 18. Change curves of cable tension: 1st experiment.

Fig. 21. Comparison curves between experimental and theoretical
Δy o  .

values for cable tension accordingly for controlling the TCPR- experimental and theoretical change curves of Δyo  were ob-
8’s moving-platform. Under the control of these five value sets tained, as shown in Fig. 21.
for cable tension, the values of Δxm in 1 2 3
o  , Δxo  , Δxo  , Δxo  , and Similar methods can be used to verify the remaining direc-
m ax
Δxo  were recorded accordingly via the laser tracker, and the tional stiffness. Note: As shown in (4), because gravity is a
experimental curves obtained by connecting the experimental part of the external force acting on the moving-platform, to en-
points with a line segment, as illustrated in Fig. 19. sure that the external load F is constant in the z-direction, it is
In the second experiment, the moving-platform’s external necessary to apply a load of F + 10 N (the mass of the moving-
load was 200 N in the y-direction. In CTFR, illustrated in Fig. 20, platform is 1 kg, as shown in Table II) in the z-direction.
−−−−−−−→
with the change of λ1 and λ2 in the Bm in Bm ax -direction, the The theoretical data were obtained via the proposed method.
tension change curves of each of the TCPR-8’s driving cables In order to clearly express the deviations between experimental
were obtained, where Bm in and Bm ax denote the maximum and data and theoretical data, they are compared in Table VI.
minimum values of Δyo  , corresponding to the points in CTFR. The experimental curves for Δxo  and Δyo  have good co-
Adopting a method similar to the first set of experiments, the herence to the theoretical curves, as shown in Figs. 19 and 21.

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 2400 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 23, NO. 5, OCTOBER 2018

TABLE VI [4] H. Yuan, E. Courteille, and D. Deblaise, “Static and dynamic stiff-
COMPARISON OF EXPERIMENTAL AND THEORETICAL DATA (IN MILLIMETERS) ness analyses of cable-driven parallel robots with non-negligible ca-
ble mass and elasticity,” Mech. Mach. Theory, vol. 85, pp. 64–81,
2015.
[5] S. H. Yeo, G. Yang, and W. B. Lim, “Design and analysis of cable-driven
manipulators with variable stiffness,” Mech. Mach. Theory, vol. 69, no. 6,
pp. 230–244, 2013.
[6] E. A. Brackbill, Y. Mao, S. K. Agrawal, M. Annapragada, and V. N.
Dubey, “Dynamics and control of a 4-dof wearable cable-driven upper
arm exoskeleton,” in Proc. IEEE Int. Conf. Robot. Autom., 2009, pp. 2308–
2313.
[7] S. Kawamura, “High-speed manipulation by using parallel wire-driven
Considering the results in Table VI, it is clear that the maximum robots,” Robotica, vol. 18, no. 1, pp. 13–21, 2000.
absolute error between the experimental and theoretical values [8] S. E. Landsberger, “Design and construction of a cable-controlled par-
allel link manipulator,” Master’s thesis, Massachusetts Inst. Technol.,
of xo  and yo  are 1.9 × 10−3 mm and 1.8 × 10−3 mm, respec- Cambridge, MA, USA, 1984, pp. 29–32.
tively, and their deviation relative to the theoretical values are [9] X. Jiang and C. Gosselin, “Dynamically feasible trajectories for three-
less than 1%. These results verify that the proposed method is DOF planar cable-suspended parallel robots,” in Proc. 38th Mechanisms
Robot. Conf., 2014, paper DETC2014-34419.
correct and effective. [10] D. Lau, J. Eden, D. Oetomo, and S. K. Halgamuge, “Musculoskele-
tal static workspace analysis of the human shoulder as a cable-driven
robot,” IEEE/ASME Trans. Mechatronics, vol. 20, no. 2, pp. 978–984,
VII. CONCLUSION Apr. 2015.
[11] M. Hassan and A. Khajepour, “Optimization of actuator forces in cable-
In this paper, a simplified structural model of two-DOR CD- based parallel manipulators using convex analysis,” IEEE Trans. Robot.,
PRs was established and static analysis conducted. The relation- vol. 24, no. 3, pp. 736–740, Jun. 2008.
[12] H. D. Taghirad and Y. B. Bedoustani, “An analytic-iterative redundancy
ship between the CDPR’s stiffness and cable tension was also resolution scheme for cable-driven redundant parallel manipulators,” IEEE
determined by establishing a static stiffness model. Further- Trans. Robot., vol. 27, no. 6, pp. 1137–1143, Dec. 2011.
more, a calculation algorithm for cable tension polygon based [13] M. Hassan and A. Khajepour, “Analysis of bounded cable tensions in
cable-actuated parallel manipulators,” IEEE Trans. Robot., vol. 27, no. 5,
on Graham’s Scan was presented, and the TCPR-8’s CTFR at pp. 891–900, Oct. 2011.
a pose was effectively obtained via the algorithm. The CDPR’s [14] B. L. Wen, H. Y. Song, and G. Yang, “Optimization of tension distribu-
system stiffness was also analyzed based on the CTFR, and it tion for cable-driven manipulators using tension-level index,” IEEE/ASME
Trans. Mechatronics, vol. 19, no. 2, pp. 676–683, Apr. 2014.
was found that distribution of the CDPR’s cable tension along [15] J. Lamaury and M. Gouttefarde, “A tension distribution method with
the centroid line of the CTFR effectively controlled the CDPR’s improved computational efficiency,” in Proc. 1st Int. Conf. Cable-Driven
controllable stiffness. The following conclusions can be drawn. Parallel Robots, 2013, pp. 71–85.
[16] X. Jiang and C. Gosselin, “Trajectory generation for three-degree-of-
1) To effectively control the CDPR’s stiffness, the CDPRs’ freedom cable-suspended parallel robots based on analytical integration
cable tension should be distributed in CTFR. This method of the dynamic equations,” J. Mechanisms Robot., vol. 8, no. 4, 2015, Art.
can avoid many ineffective iterative calculations, and its no. 041001.
[17] S. Tadokoro, Y. Murao, M. Hiller, and R. Murata, “A motion base with
calculation efficiency is high. 6-DOF by parallel cable drive architecture,” IEEE/ASME Trans. Mecha-
2) Analysis of the CDPR’s system stiffness in CTFR verified tronics, vol. 7, no. 2, pp. 115–123, 2002.
that the CDPR’s stiffness can be effectively controlled by [18] J. A. Seon, S. Park, S. Y. Ko, and J. O. Park, “Cable configuration analysis
to increase the rotational range of suspended 6-DOF cable driven parallel
controlling the cable tension. As the results from Section robots,” in Proc. IEEE Int. Conf. Control, Autom. Syst., 2017, pp. 1047–
V-B indicate, the x- and θz -directions are the directions 1052.
of the fastest increasing of translational and rotational [19] A. Nasr and S. A. A. Moosavian, “Multi-criteria design of 6-DoF fully-
constrained cable driven redundant parallel manipulator,” in Proc. IEEE
stiffness, which increase by almost 125% and 180%, re- Rsi Int. Conf. Robot. Mechatronics, 2016, pp. 001–006.
spectively. [20] R. Chellal, E. Laroche, and L. Cuvillon, “An H methodology for position
The results of theoretical and experimental analyses verified control of 6-DoF cable-driven parallel robots,” in Proc. Eur. Control Conf.,
2014, pp. 358–363.
the efficacy and correctness of the proposed method in the x- [21] S. Behzadipour and A. Khajepour, “Stiffness of cable-based parallel ma-
and y-directions. nipulators with application to stability analysis,” J. Mechanical Des.,
In the future, we will study the CDPR’s dynamic stiffness, vol. 128, no. 6, pp. 303–310, 2006.
[22] R. Yao, X. Tang, J. Wang, and P. Huang, “Dimensional optimization design
and control the CDPR’s vibration and the accuracy of the work of the four-cable-driven parallel manipulator in fast,” IEEE/ASME Trans.
done by optimizing the cable tension. Mechatronics, vol. 15, no. 6, pp. 932–941, Dec. 2010.
[23] M. Azadi and S. Behzadipour, “An application of parallel singularity in
variable stiffness elements,” in Proc. ASME Int. Des. Eng. Tech. Conf.
REFERENCES Comput. Inf. Eng. Conf., 2008, pp. 13–19.
[24] A. H. Hesselroth and M. P. Hennessey, “Analytical evaluation of the
[1] M. Carricato and J. P. Merlet, “Stability analysis of underconstrained double stewart platform tensile truss stiffness matrix,” J. Mechanisms
cable-driven parallel robots,” IEEE Trans. Robot., vol. 29, no. 1, pp. 288– Robot., vol. 6, no. 1, 2014, Art. no. 011003.
296, Feb. 2013. [25] R. Yao, “Study on tension characteristic and dimensional synthetic design
[2] H. Jamshidifar, A. Khajepour, B. Fidan, and M. Rushton, “Kinematically- of cable driven parallel manipulators with large span,” Tsinghua Univ.,
constrained redundant cable-driven parallel robots: Modeling, redundancy Beijing, China, 2010, pp. 15–19.
analysis and stiffness optimization,” IEEE/ASME Trans. Mechatronics, [26] G. M. Ziegler, Lectures on Polytopes (Graduate Texts Mathematics). New
vol. 22, no. 2, pp. 921–930, Apr. 2017. York, NY, USA: Springer, vol. 152, 1993, pp. 1–41.
[3] M. Carricato and J. P. Merlet, “Direct geometrico-static problem of under- [27] M. J. Kaiser and T. L. Morin, “Characterizing centers of convex bodies
constrained cable-driven parallel robots with three cables,” in Proc. IEEE via optimization,” J. Math. Anal. Appl., vol. 184, no. 3, pp. 533–559,
Int. Conf. Robot. Autom., 2013, vol. 31, pp. 3011–3017. 1994.

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.
 CUI et al.: RESEARCH ON CONTROLLABLE STIFFNESS OF REDUNDANT CABLE-DRIVEN PARALLEL ROBOTS 2401

Zhiwei Cui received the Master’s degree in me- Senhao Hou received the B.S. degree in me-
chanical engineering from Beihang University, chanical engineering from Beijing Institute of
Beijing, China, in 2016. He is currently work- Technology, Beijing, China, in 2016. He is cur-
ing toward the Ph.D. degree in mechanical engi- rently working toward the Ph.D. degree in me-
neering at Tsinghua University, Beijing. chanical engineering at Tsinghua University,
His research interests include the parallel ma- Beijing.
nipulators, and cable-driven robots. His current research interests include cable
parallel robots.

Xiaoqiang Tang received the Ph.D. degree in Haining Sun received the B.S. degree in me-
mechanical engineering from Tsinghua Univer- chanical engineering from Shandong University,
sity, Beijing, China, in 2001. He is currently a Jinan, China, in 2017. He is currently working to-
Professor with the Department of Mechanical ward the Ph.D. degree in mechanical engineer-
Engineering, Tsinghua University. His research ing at Tsinghua University, Beijing, China.
interests include parallel manipulators, robots, His current research interests include cable
and reconfigurable manufacturing technology. parallel robots.

Authorized licensed use limited to: Universit&eacute; de Montpellier - Sci et Tech. Downloaded on September 25,2024 at 09:26:52 UTC from IEEE Xplore. Restrictions apply.