Mechanism and Machine Theory 86 (2015) 211234

Contents lists available at ScienceDirect

Mechanism and Machine Theory

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

Study on Jacobian, singularity and kinematics sensitivity of the

FUM 3-PSP parallel manipulator
Amir Rezaei , Alireza Akbarzadeh
Mechanical Engineering Department, Center of Excellence on Soft Computing and Intelligent Information Processing, (SCIIP), Ferdowsi University of Mashhad, Mashhad, Iran

a r t i c l e

i n f o

Article history:
Received 19 April 2014
Received in revised form 10 November 2014
Accepted 12 November 2014
Available online 8 January 2015
Keywords:
Sensitivity analysis
Jacobian matrix
Singularity
Operational modes
Actuator accuracy
End-effector accuracy

a b s t r a c t
The Jacobian matrices of a robot are commonly utilized in determining its dynamic behavior,
workspace singular regions as well as manipulability and sensitivity analysis. This paper provides
a new general perspective in analyzing workspace, singularity and sensitivity analyses. This perspective can aid both end-users and robot designers to assess effect of desired end-effector accuracy on required actuator accuracy as well as effect of selected actuator accuracy on the resulting
end-effector accuracy. To do this, a 3-PSP parallel robot with specic architecture is considered.
Two operational modes called non-pure translational and coupled mixed type modes are considered and Jacobian matrices are obtained and the inverse and direct relations for velocity and acceleration are derived. A methodology to select the most practical operational modes is represented.
Next the three well-known singularities as well as the constraint singularity are investigated. Additionally, notion of inverse kinematics singularity (IKS) is dened and existing IKS loci in
workspace for the two operational modes are determined. Finally, for the two operational
modes, the direct and inverse sensitivity analyses are investigated by dening concept of direct
and inverse squared Jacobian matrices. In the direct sensitivity analysis, inuence of actuator errors on the position and orientation errors of the moving platform (MP) is determined. In the inverse sensitivity analysis, the allowable actuator error boundaries are determined with respect to
desired MP pose errors.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Researches on applications of Parallel Kinematic Machines (PKMs) continue to attract the attention of many industrial companies,
specically, limited-degrees of freedom (DOFs) PKMs [1,2]. PKMs offer many advantages such as, high positioning accuracy, high
static/dynamic inherent rigidity, low inertia, high nominal load-to-weight ratio and good dynamic performance. Additionally, the
PKMs with the lower DOFs have most of the inherent capabilities of the parallel robots and can be made with lower manufacturing
cost [13]. These advantages of PKMs give them the capability to be used in many high-speed precision industrial applications
[46]. Examples of these applications are, machine tools, orienting devices, ne positioning devices, fast assembly, cutting and
welding machines, ight simulators and medical devices [510]. As applications of robotics increase so do the need for their accuracy.
Accuracy evaluation is one of the basic analyses used in robotics eld. It is one of the key features to development of the machine tool
and robotics industries. The servo actuator resolutions, inaccuracy in the assembly and manufacturing processes, joints clearance and
compliance modules in the mechanism are the some of the more common sources of errors in robotics systems [8,11,12].
The sensitivity analysis is an important activity during the design process of robotics systems. The sensitivity analysis determines
inuence of variations in the geometric parameters and/or actuators of a robot on its performance [12]. This analysis can also be
Corresponding author. Tel.: +98 915 525 5471; fax: +98 511 876 3301.
E-mail addresses: [email protected] (A. Rezaei), [email protected] (A. Akbarzadeh).

http://dx.doi.org/10.1016/j.mechmachtheory.2014.11.009
0094-114X/ 2014 Elsevier Ltd. All rights reserved.

212

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

viewed as providing relation between geometrical and actuator errors on the pose accuracy of the robot. Having information on the
sensitivity of a robot can aid the robot designers in design, component selection and construction of the robot [13]. To do this, the dexterity and the manipulability indices can be used to evaluate the sensitivity of robot performance in terms of actuated joint variations
[12,1417].
To evaluate the robot sensitivity in its entire workspace, rst the relations between actuators and end-effector variations must be
obtained. To do this, analyses such as the kinematics, Jacobian matrices and workspace analyses are used. The kinematics, Jacobian and
singularity analyses of PKMs have been studied by many researchers [7,8,1832]. Merlet [7] revisit the concepts of Jacobian matrix,
manipulability and condition number for parallel robots as accuracy indices in view of optimal design. In [8], the sensitivity model
of a spindle platform subjected to structure parameters is analyzed and inuence of all parameters on the position and orientation
of the spindle platform are analytically estimated. Jin and Chen [33] identied three basic types of constraint errors and suggested
an approach to evaluate the effects of constraint errors on decoupling characteristics of a 6-DOF parallel robot with decoupled translation and rotation. In [34] the effects of joint clearance on the parallel robot accuracy are evaluated using an analytical form to predict
pose errors for a 3-UPU parallel robot. Cardou et al. dened maximum rotation and maximum point-displacement sensitivities which
provide upper bounds for the end-effector rotation and point-displacement sensitivity [35]. A comparison of the sensitivity for 3-DOF
planar parallel manipulators is investigated [13,36] and two aggregate sensitivity indices for position and orientation of the moving
platform are determined. Also, Saadatzi et al. [37] presented a robust geometric approach for computing the kinematic sensitivity performance index for the case of planar parallel mechanisms. They also explored the concept of the global kinematic sensitivity index.
Wu et al. [38] developed a new error model with considerations of both conguration errors and joint clearances for a 3-PPR planar
parallel manipulator. Using the model, the pose error estimation problem is formulated as an optimization problem and the upper
bounds and distributions of the pose errors are established.
In the present paper, the sensitivity analysis includes the inuence of actuator variations on pose accuracy of the robot. To do this, a
specic architecture of the 3-PSP parallel robot is selected and its Jacobian, singularity as well as the direct and inverse sensitivity analyses are investigated. Additionally, the concepts of direct and inverse squared Jacobian matrices are introduced. The presented concept in this paper can aid both end-users and robot designers to assess effect of desired end-effector accuracy on required actuator
accuracy as well as effect of selected actuators accuracy on the resulting end-effector precision.
This paper is organized as follows. In Sections 2 and 3, structure and motion variables of the robot are addressed. In Section 4, identication of degrees of freedom and operational modes is investigated. In Section 5, vector description and inverse position analysis
are completed. In Section 6, using auxiliary vectors, analytical expressions for direct and inverse velocity and acceleration relations
are derived as overall form. Additionally, these relations are obtained for two selected operational modes. In Section 7, using Jacobian
matrices, the three conventional types of singularities and constraint singularity are analyzed. Additionally, the concept of inverse kinematics singularity for each operational mode is investigated for the 3-PSP robot. Finally, in Section 8, the inverse and direct sensitivity analyses are investigated for two operational modes by dening concept of direct and inverse squared Jacobian matrices of the
two modes.

2. The 3-PSP parallel mechanism

In the present paper, a specic architecture of a 3-PSP parallel manipulator which is constructed in FUM Robotics Research Center is
considered. The 3-PSP robot is a fully parallel mechanism with 3-DOFs in space. The robot consists of a moving platform shaped like a
symmetric tripetalous star, two xed platforms and three identical legs. The moving star (MS) and the xed platforms are connected
using three identical parallel legs. Each leg makes a serial kinematic chain as PrismaticSphericalPrismatic, PSP. The rst P-joint in
each leg is the active joint while the S-joint and the second P-joint are the passive joints [9,31]. See Fig. 1(a).

w
T
v
{T}

b1

w
{T}

h
T

S1

{B}

s1

q1

a1

{B}
y

A1

(a)

{B}
y

(b)

Fig. 1. (a) The physical model of a 3-PSP parallel robot and (b) Vectors and coordinate frames.

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

213

3. Description of joints coordinates for the 3-PSP robot

The virtual prototype and physical models of a 3-PSP parallel manipulator are depicted in Fig. 1. In this gure, the general and local
coordinate frames for the 3-PSP robot are depicted. The general coordinate frames are dened as
Fixed coordinate frame B{x, y, z} is embedded at point O on top xed triangle A1A2A3
Moving coordinate frame T{u, v, w} is attached to center of the MS, at point T.
In this paper, a leading superscript represents the coordinate frame in which the vector is referenced. The essential vectors to solve
the kinematics modeling are dened as follows

Three position vectors Bai locate corners of the xed base, Ai, in {B}
Three position vectors Bqac
i specify length of linear rods (LR)
Three position vectors Tbi each connects the end-effector, point T, to the ith S-joint, Si
Position of the end-effector, point T, with respect to {B} is given by vectors Bt
Position of the tool tip, point P, with respect to {B} is given by vectors Bp
The position vector Th is a vector which connects point T to point P in {T}.

4. Identication of degrees of freedom and operational modes

Earlier parallel robots were designed mostly with 6-DOFs. However, there are many practical applications where 6-DOFs are not all
required. Thereafter, interest remains for parallel robots with lower than 6-DOFs resulting in lower manufacturing cost while keeping
most of the inherent capabilities of the parallel robots. The 3-DOFs spatial parallel robots can be classied into 3 categories with respect to the type of DOFs used by their moving platform. These categories are: (a) pure translational, e.g. the 3-UPU and 3-PRC
robot, (b) pure rotational, e.g. spherical robots and the Agile Eye mechanism and (c) pure and non-pure coupled mixed-types motion
(two translational and one rotational, T2R1-type, or two rotational and one translational, R2T1-type), e.g. the 3-PRS robot [31]. The
spatial 3-DOFs parallel robots with pure coupled mixed-type motions, their moving platform have pure R2T1 or T2R1 DOFs. Whereas,
the moving platform of spatial 3-DOFs parallel robots with non-pure coupled mixed-type motions have non-pure R2T1 or T2R1 DOFs.
This means that, the moving platform of robots with non-pure coupled mixed-type DOFs have parasitic or uncontrollable motions e.g.
the 3-PRS and 3-PSP robots. Also, Li et al. study a number of pure R2T1 type parallel robots [39].
In spatial non-pure coupled mixed-type, i.e., complex m-DOFs parallel robots (m b 6), the operational modes can be dened
through selection of independent DOFs of the moving platform. For a spatial parallel robot with m degrees of freedom any m combination from n = 6 motion variables of the moving platform can be selected as mode of operation for the robot. Consider a spatial complex 3-DOFs parallel robot with 3 translational and 3 rotational motion variables as XYZ. Then, each 3 combination from 6 motion
variables of the moving platform can be selected and represent the corresponding operational mode. The selected 3-DOFs are independent and are controllable. The remaining 3-DOFs are not independent, are uncontrollable and are commonly referred to as causing
parasitic motion. It is worth noting that for each operational mode a different method must be employed to evaluate the inverse kinematics, Jacobian matrices and reachable workspaces. The DOFs of the 3-PSP moving platform robot result in non-pure coupled
mixed-type motions. To obtain all the 3-PSP operational modes, m = 3 combination from n = 6 motion variables of the moving platform, XYZ, can be selected. Therefore, the number of conceivable operational modes for the 3-PSP robot can be expressed as

n
m

n!
6!

20:
m!  nm! 3!  63!

The potential operational modes for the 3-PSP complex 3-DOFs parallel robot are manually selected and illustrated in Table 1.
The Euler angle is the rotation angle of the moving platform about z-axis of the moving frame {T}. For many applications such as
welding, this rotation is not necessary and in many other applications such as drilling, this rotation is commonly applied by the rotating tool. Furthermore, the rotational variables and can be changed independent of the variable . Conversely, structure of the 3-PSP
actuators does not allows the MS to independently rotate about z-axis, , without effecting the Euler angles and . Therefore, it can
be stated that the rotation angles of the MS about x and y-axes, i.e., and , are more desirable than the rotation angle . Finally, consider the non-pure rotational mode (R3), . If the orientation of the MS is given, the set of 9 non-linear algebraic kinematics equations, 9 1(q) = 09 1, is a linear system of nine equations in nine unknowns. Then, the number of answers for this set of equation, to
solve the inverse kinematics (InvKin) is innite. Hence, the non-pure rotational mode cannot be considered as a feasible
Table 1
The potential operational modes for the 3-PSP parallel robot.

Operational mode

Non-pure translational
mode (T3)

Non-pure rotational
mode (R3)

Non-pure coupled mixed-type

mode (T1R2)

Non-pure coupled mixed-type

mode (T2R1)

XYZ

X, Y, Z
X, Y, Z
X, Y, Z

XY, XY, XY
XZ, XZ, XZ
YZ, YZ, YZ

214

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

operational mode. Consequently, the operational modes containing the Euler angle may be removed from the practical list of the
operational modes. This eliminates 9 choices from 20 possible operational modes, Table 1.
Next, consider workspace analysis of the 3-PSP robot presented in [31]. It is clear that the magnitude of variations in x and y are
signicantly smaller than the magnitude of variation in z. In other words, motion of MS along the z-axis can be considered as a principle DOF of the MS.
Therefore, referring to Table 1, excluding the XYZ and Z operational modes, all other modes containing the motion variable z are
assumed nonpractical and therefore eliminated. In the present paper, the non-pure coupled mixed-type mode Z is considered as a
principle operational mode for the 3-PSP robot. In this mode, all of the three MS motion variables, Z, and , have widely ranges
throughout the robot workspace. The non-pure translational XYZ operational mode may also nd practical positioning applications
and is therefore selected as the second principle operational mode. Note that the 3-PSP robot does not have any pure operational
modes. Therefore, when the phrases XYZ mode and Z mode are used, these mean non-pure translational XYZ mode and nonpure coupled mixed-type Z mode.
5. Vector description and inverse position analysis
The two operational modes, non-pure translational XYZ and coupled mixed type Z modes are considered for the inverse position and Jacobian analyses (See [31] for more details). Fig. 1(b) illustrates vectors and coordinate frames used for the InvKin problem
of the 3-PSP. Note that the vectors referenced in frame {B} are denoted by Bv, while vectors referenced in frame {T} are denoted by Tv.
The symmetrical structure of the 3-PSP helps arrive at three algebraically similar constraint equations. Consider Fig. 1(b). Three closed
vector-loop equations can be written as,
B

B ac

ai qi T R


T
B
bi h p

for i 1; 2; 3

where BTR is a rotation matrix to transfer a vector dened in {T} to {B}. The constraint equations, Eq. (2), consist of 3 vector equations
and are equal to a set of 9 non-linear algebraic equations as 9 1(q12 1) = 09 1. The general coordinates used for kinematics modeling of the robot without considering the passive S-joints motion variables are

q121

ac

unac

q91

q31

ac

ac

ac

q1 ; q2 ; q3 ; xp ; yp ; zp ; ; ; ; b1 ; b2 ; b3
|{z}
|{z}
ac:
Unac:

un ac
are vectors of actuated and un-actuated variables, respectively. As shown in Fig. 1(b), the position vectors
where qac
3 1 and q9 1
used to describe the kinematic conguration of the 3-PSP parallel robot can be expressed by

ai f a cosi
B

a sini

t f xT

b f bi cosi bi sini 0 g ; for i 2i1=3

0g ;
 i
T
0 0 qac
for i 1; 2; 3
i
T
B
T
T
T
zT g ;
p f xP yP zP g ;
h f0 0 hg :

B ac
qi

yT

To transfer a vector dened in {T} to {B}, the rotation matrix, BTR maybe used as
2

B
TR

R z; R y; Rx;

ux
4 uy
uz

vx
vy
vz

3 2
c c
wx
wy 5 4 s c
s
wz

3
s c c s s s s c s c
c c s s s c s s s c 5
c s
c c

where, c and s stand for cosine and sine, respectively. Therefore, to obtain a vector in {B}, B, form a dened vector in {T}, T, we can
use, B = BTRT. Depending on the selected DOFs, three of the six MS variables are chosen and the remaining nine variables are calculated using InvKin problem. For each selected operational mode, a different InvKin solution strategy is used to solve Eq. (2). The relationship between inputs and outputs of the InvKin for the operational modes may be stated as
XYZ mode:

Inputs: xp ; yp ; zp

Z mode:

Inputs: ; ; zp

and

and

n
o
 ac ac ac

q1 ; q2 ; q3 ; ; ; ; b1 ; b2 ; b3 f 1 xp ; yp ; zp
n
o
n
o
ac
ac
ac
q1 ; q2 ; q3 ; xp ; yp ; ; b1 ; b2 ; b3 f 2 zp ; ; :

Upon solving Eqs. (6) or (7), kinematic values of the robot necessary for Jacobian analysis are obtained. In [31] both analytical and
numerical methods to solve the InvKin for XYZ and Z operational modes are presented. To graphically show the dominance of the
Z variables compared with the remaining variables, InvKin algorithm for the XYZ and Z operational modes are used for two different trajectories of the end-effector. For both trajectories, it is assumed that tool length is set to h = 8 cm.

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

215

5.1. Trajectory #1: planar trajectory in non-pure translational XYZ mode

In this trajectory, the tool tip traces a T-shape planar trajectory in Z = 10 cm plane. The end-effector trajectory, inverse kinematic
results and values for parasitic motion of the moving platform (,,) are depicted in Fig. 2.
5.2. Trajectory #2: trajectory in non-pure coupled mixed-type mode Z
In the second case study, trajectory of the moving platform is designed for Z operational mode. Consequently the values of X, Y
and are the parasitic motion (see Fig. 3).
As can be seen from Figs. 2 and 3, the magnitude of overall changes in X, Y and are signicantly smaller than values of Z, and ,
respectively.
6. Jacobian analysis
In this section, the two operational modes are considered and an analytical method is to obtain the Jacobian matrices as well as the
velocity and acceleration relations of the 3-PSP parallel robot.
6.1. Velocity analysis
For the velocity analysis, ith limb of the 3-PSP parallel robot is considered and both sides of the kinematics constraint equations,
Eq. (2), is time differentiated to yield
ac ac
^  b  hv
^ i b i b
q i q
i
MS
i
MS
p

for i 1; 2; 3

^ represent linear rods, LRs, and passive prismatic joints velocity vectors, respectively. Furthermore, vectors and
^ i andbi b
where qi q
s
i
^ , both
vp denote angular velocity vector of the MS and Cartesian velocity vector of the tool tip, respectively. To eliminate the vectors bi b
i
^ b
^ ; for i 1; 2; 3 and j 2; 3; 1 which is perpendicular to the
^i b
sides of Eq. (8) are dot multiplied with a specic unit vector m
i
j
^ . By dot multiplying unit vector m
^
three unit vectors b
in
Eq.
(8)
and
then
simplifying, this vector equation can be written in matrix
i
i
form as


ac ac

ac

Jac q

JMS tMS

(a)

(b)

(c)

Fig. 2. (a) The end-effector trajectory, (b) the actuator values and (c) values for parasitic motions in XYZ mode.

216

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

(a)

(b)

(c)

Fig. 3. (a) The end-effector trajectory, (b) the actuator values and (c) values for parasitic motions in Z mode.

where



^iq
^ ac
;
Jac diag m
i
33

h
JMS m
^ iT

^ i T
bi h  m

i
36

for i 1; 2; 3

10



ac
ac
ac T
where q qac
represents the LRs velocity vector and tMP f vP MS gT represents twist vector of the MS. Note
q3
1  q2



T
v P vx vy vz
and MS x y z T are dened in base coordinate frame {B}. In view of Eq. (9), we can rewrite this
equation in familiar form as


ac

JI JM tMP

11

where JIJM = J1
ac JMS is a 3 6 non-square matrix called Inverse Jacobian Matrix of the 3-PSP parallel manipulator. This matrix has many
practical applications such as calculating the robot stiffness matrix [9]. As stated earlier, although the MS of the 3-PSP robot has 3-DOF,
yet the MS is a 6-DOF rigid body in space where its 3-DOF is dependent on the remaining 3-DOFs. Equation (11) relates all 6 velocities
of the MS rigid body to 3 actuated joints velocities. Furthermore, it will be useful to obtain a 6 6 inverse square Jacobian matrix that
maps the twist tMP of the end-effector to actuated and constraint joints velocities. In what follows, the Jacobian of constraints and 6
6 inverse square Jacobian matrix referred to as Full Inverse Jacobian matrix are obtained. To obtain the Jacobian of constraints, the auxiliary vectors are introduced to eliminate passive velocity vectors (See [31] for more detailed explanations). To nd the relationship
ac ac
^ q
^
^i b
^ ac
between MS and vP, both sides of Eq. (8) are dot multiplied with unit vectors e
i ; for i 1; 2; 3 to eliminate qi qi and
i
^
bi bi terms. The unit vector i is along R-joint i in ith passive S-joint and is perpendicular to planes containing the two vectors, qac
i
and bi (see Fig. 4). Hence


Jvp vp Js MS 031

12

where
h
i
^i T
Jvp e

33

; Js

h
T i
^i
bi h  e

33

for i 1; 2; 3:

13

Therefore, Eq. (12) can be re-written as

Jc tMP 031

14

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

217

Fig. 4. Denition of the ith unit vectors i.

where, Jc is a 3 6 matrix called Jacobian of constraints for the 3-PSP parallel manipulator and is dened as
 h T
Js e
^i

Jc Jvp


T i
^i
bi h  e

36

for i 1; 2; 3:

15

The ith row in the Jacobian of constraints matrix, represents a unit wrench of constraints imposed by the joints of the ith limb of the
robot [29]. Then, matrix Jc maps a virtual displacement vector of the MS, X, to constraints' virtual motion directions which are equal
to zero, 03 1.
ac
Combining Eqs. (11) and (14), yields the relationship between the linear actuated joints velocities, q , and twist of the MS, tMP,
as


ac
q
031


JFIJM tMP

JFIJM

JI JM
Jc


16
66

where JFIJM is 6 6 square matrix called Full Inverse Jacobian matrix. The primary advantage of Eq. (16) with respect to Eq. (11) is
invertibility of the matrix JFIJM. This matrix enables us to conveniently determine the direct velocity relation for the 3-PSP robot
using Eq. (16). Estimating actuator sensitivities is one of the main motivations for obtaining the direct velocity relations. As stated earlier, JIJM is not a square matrix. Therefore, the direct velocity relation cannot be obtained using Eq. (11). To remedy this, employing
Eq. (16) and rearranging yields


ac

tMP JD JM q :

17

This equation is called the direct velocity equation of the robot. Additionally, JDJM is a 6 3 non-square matrix called Overall Direct
Jacobian Matrix of the 3-PSP and is obtained by selecting the rst three columns of the J1
FIJM as
1

JD JM JFIJM 613 :

18

The motivation for pursuing an Inverse Square Jacobian Matrix which maps the actuator velocities to only independent end effector
velocities can be demonstrated in trajectory planning applications. In such applications, the independent MS velocities are specied
and actuator speeds are obtained. As stated earlier, depending on the desired operational mode, three of the six motion variables of
the MS are selected. Then it is also desirable to determine a new 3 3 Inverse Square Jacobian Matrix which maps the three desirable
independent MS speed components to the three actuators components. Using Eq. (17), the velocity relations for any arbitrary operational modes can be obtained. To obtain the 3 3 Direct Square Jacobian Matrix for any arbitrary operational modes, we can
write


ac

vresmode JD JMmode q31

19

where vres, called restricted twist vector, represents a velocity vector containing any arbitrary three of the six variables of the MS twist
vector, tMP. The restricted twist vector, vres, can be dened as the restriction of tMP on the selected MS DOFs of the robot [7]. For any
operational mode, we can write
2
3
JD JM row#i
JD JMmode 4 JD JM row# j 5
JD JM row#k 33

20

218

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

where JDJM mode is a 3 3 matrix called Direct Square Jacobian Matrix for selected operational mode. The indices i, j and k allow us to
ac
select arbitrary rows of matrix JDJM. Therefore, using Eq. (19), the relationship between vz, x, y and q31 in the non-pure coupled
mixed-type mode Z can be expressed as


ac

ac

vZ JD JM 353 q31 JD JMZ q31

21

where JDJM Z is a 3 3 matrix called Direct Coupled Mixed-type Jacobian Matrix. Similarly, the relationship between vx, vy, vz and
ac
q31 in the non-pure translational mode XYZ can be obtained from Eq. (19) as


ac

ac

vXYZ vp JD JM 133 q31 JD JMXYZ q31

22

where JDJM XYZ is a 3 3 matrix called Direct non-pure translational Jacobian Matrix. Equations (21) and (22) allow us to quantify the
inuence of the actuated joints variables errors on the end-effector positioning errors in the XYZ and Z directions. Consequently, to
obtain the inverse velocity relations for each selected operational mode, Eqs. (21) and (22) are used as


ac

23

ac

24

q31 JD JMZ vZ

q31 JD JMXYZ vXYZ

1
1
where JDJM
Z and JDJM XYZ are 3 3 matrices called Inverse Coupled Mixed-type Jacobian Matrix and Inverse non-pure translational
Jacobian Matrix. These matrices map only independent end-effector velocities to the actuator velocities. Similarly, Eqs. (23) and (24)
allow us to quantify the inuence of positioning errors in the Z and XYZ DOFs of the end-effector on the actuated joints errors. In
general, we may also obtain the overall velocity relation using different approaches. Using this approach, additional velocities for the
3-PSP robot parameters may be obtained. For example, consider MS branch length, bi. Then Eq. (8) can be utilized to obtain the overall
velocity relation in matrix form. To do this, the vector velocity relation, Eq. (8), is rewritten in matrix form as

Jac;overall q

ac


JOIJM

tMP
b

25

91

where JOIJM is the Overall Inverse Jacobian Matrix and we can write
2

Jac;overall

^ ac
q
1
4 031
031

031
^ ac
q
2
031

2
3
I
031
6 33
031 5 ; JOIJM 4 I33
^ ac
q
I33
3
93


T
^ ac
Note that q
i f 0 0 1 g and b

b1

b2

MS  bi h bi h  I33 MS

b3

oT

b1 h  I33
b2 h  I33
b3 h  I33

^
b
1
031
031

031
^
b
2

031

3
031
7
031 5 :
^
b3 99

26

. Additionally, we know that

for i 1; 2; 3

27

therefore
8 9
8 9
8 93
<1=
<0=
<0=
4 bi h  0
bi h  1
bi h  0 5
: ;
: ;
: ;
0
0
1
2

bi h  I33

for i 1; 2; 3:

28

The JOIJM is a 9 9 square matrix and its inverse can be easily calculated. Using Eq. (25) the overall direct velocity relation may be
obtained as


tMP
b

JODJM q

ac

JODJM JOIJM Jac;overall

29

91

where JODJM is a 9 3 matrix called the Overall Direct Jacobian Matrix. Additionally, similar to Eq. (17) and using Eq. (29), the direct
velocity relation may be obtained as


tMP JD JM q

ac

h
i
1
JD JM JOIJM Jac;overall

163

30

By using Eq. (30), the direct and inverse velocity relations for any arbitrary operational mode can be obtained similar to the steps
taken in deriving Eqs. (19)(22) and (23)(24), respectively.

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

219

6.2. Acceleration analysis

To obtain the acceleration relations of the 3-PSP robot, both sides of Eq. (8) are time differentiated as
^
^ ac ^
ac
q
i qi bi bi 2MS  bi bi MS  bi h MS  MS  bi h vP


for i 1; 2; 3:

31

ac
The values q
i and bi represent the ith actuated joint acceleration and ith passive prismatic joint linear acceleration, respectively.
Additionally, vectors MS and vP denote angular and linear acceleration vectors of the MS, respectively. We know that


a  b  c a:cba:bc and a  b b  a
a  b:c c:a  b c:b  I33 a b  c:a

32

^
b
^ i and using Eq. (32), then simplifying the equation as well
By considering Eq. (31) and eliminating b
i i vector using unit vectors m
as writing the three scalar equations in matrix form [1], yields


ac JMS tMP NMS m:

Jac q

33

Therefore,


ac JI JM tMP J1ac NMS J1ac m

q

34

 ac
T

T
ac

ac ac and tMP vP MS are the actuated joints and MS acceleration vectors, respectively. Additionally,
where
 q q1 Tq2 q3

T
vP vx vy vz and MS x y z represent linear and angular accelerations of the MS which are dened in base coordinate frame {B}, respectively. Finally, the matrices N and m are dened as





T 
^ m
^ i MS  bi hm
^i
2bi b
i


33

^ i  bi h31
; m MS  MS m

for i 1; 2; 3:

35

Equation (34) provides the relationship between angular and linear acceleration of the MS with linear acceleration of the three
actuators and is called inverse acceleration relation. To solve Eq. (34), rst the InvKin, Eq. (2), and next inverse velocity relations,
ac
Eq. (11), need to be calculated. Using Eq. (29), values for i and MS can be calculated as a function of q . Since Jac is a 3 3 square
matrix, the inverse acceleration relations can easily be obtained. However, since matrix JMS in Eq. (33) is a 3 6 non-square matrix the
direct acceleration relations cannot be obtained using Eq. (33). To remedy this and to obtain the direct acceleration relations in overall
form for each of the operational modes, a vector approach, similar to what stated for velocity analysis, will be used. To obtain the overall acceleration relation, Eq. (31) is rewritten in a matrix form as


(
ac JOIJM
Jac;overall q

tMP

NOIJM MS mOIJM

36

91

where
NOIJM

h

i
^ I
2bi b
i
33 MS  bi hI33


93

mOIJM MS  MS bi h91
for i 1; 2; 3:

37

Therefore, similar to Eq. (29), we can write

(

tMP

)
1
ac J1
JODJM q
OIJM NOIJM MS JOIJM mOIJM :

38

Equation (34) provides a relationship between linear acceleration of the three actuators with all acceleration parameters of the MS.
ac by a user and solving the InvKin, Eq. (2), as well as
This equation is called the Overall Direct Acceleration relation. Upon specifying q
calculating values for i and MS by velocity relations, Eq. (29), all acceleration parameters of the MS can be determined. When calculation of the MP is only required, then Eq. (38) can be limited by selecting its rst 6 rows as


ac ND JM
MS mD JM 61
tMP JD JM q
63

39

where
h
i
1
ND JM 63 JOIJM NOIJM

163

h
i
1
; mD JM 61 JOIJM mOIJM

161

40

220

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

Equation (39) is called the direct acceleration relation. Similar to derivation of velocity relations, the direct acceleration relations for
each selected operational mode will be obtained. In general form, we can write
ac
vresmode JD JMmode q
31 ND JMmode MS mD JMmode


41

where, for desired operational mode, vresmode represents a restricted acceleration vector containing any arbitrary three of the six variables of the MS acceleration vector, MP. Additionally, NDJM mode and mDJM mode are dene as
2

ND JMmode

3
ND JM row#i
4
N

;
D JM row# j 5
ND JM row#k 33

mD JMmode

3
mD JM row#i
4
mD JM row# j 5
mD JM row#k 33

42

and i, j and k represent the selected number of arbitrary rows of Eq. (41). Therefore, using Eq. (41), the relationship between vz ; x ; y
ac
and q
31 in the non-pure coupled mixed-type mode Z can be expressed as
ac
vZ JD JMZ q
31 ND JMZ MS mD JMZ :


43

31 in the non-pure translational mode XYZ can be obtained from Eq. (41)
Similarly, the relationship between vx ; vy ; vz and q


ac

as
ac
vXYZ JD JMXYZ q
31 ND JMXYZ MS mD JMXYZ :


44

by a user, solving the InvKin, Eq. (2), and calculating values for i and MS using velocity relations, Eq. (29), the
Upon specifying q
restricted acceleration vectors of MS for Z and XYZ modes can be calculated using Eqs. (43) and (44). It is clear that JDJM Z and
JDJM XYZ are square matrices. Therefore, to obtain the inverse acceleration relations for each selected operational mode, Eqs. (43) and
(44) can be conveniently used as
ac



1
1
ac
q
31 JD JMZ v Z JD JMZ ND JMZ MS mD JMZ

45



1
1
ac
q
31 JD JMXYZ v XYZ JD JMXYZ ND JMXYZ MS mD JMXYZ :

46

7. Singularity analysis
Singularities are undesirable situations in manipulator operation for both motion and force control. In singular conguration, the
mobile platform may instantaneously gain one or more unconstrained degrees of freedom. For example, in some singular congurations; the moving platform can have motion even if all actuated joints are locked. Additionally, in some parallel robot's congurations
the architecture singularities occur whenever the actuators cannot control the velocity of the moving platform. Also, the constraint
singularity phenomenon in limited-DOF parallel robot occurs when the robot's limbs lose their ability to constrain the intended motion of the moving platform [29]. For example, when the constraint singularity occurs in the 3-UPU parallel robot assembled for translation, the moving platform will gain rotational DOF [29].
In this section, the singularity analysis for the 3-PSP parallel robot is presented and the concept of Inverse Kinematics Singularity for
each operational mode using force analysis is investigated. Next, for both operational modes, numerical approach is utilized and contours of determinant for the square matrices JDJM Z and JDJM XYZ are plotted throughout the workspace.
Since, the 3-PSP parallel robot is a spatial limited-DOF manipulator and its moving platform has 3-DOFs in space, a 6 6 square
Jacobian matrix JFIJM called Full Inverse Jacobian matrix was derived in Eq. (16) which provides information about both architecture
and constraint singularities [29]. By inspection of Eq. (16), we can see that determinant of Jacobian matrix JFIJM does not become
zero throughout the workspace. Therefore, the inverse of full inverse Jacobian matrix, J1
FIJM, is always computable. In other words,
the matrix JFIJM is invertible throughout the workspace and therefore the Overall Direct Jacobian Matrix of the 3-PSP parallel robot,
JDJM, is always computable (see Eq. (18)).
Additionally, Eqs. (9) and (11) can be used in deriving the singularity conditions of the parallel manipulators. Since, the 3-PSP parallel robot is a spatial limited-DOF manipulator and its moving platform has 3-DOFs in space, therefore, the concept of rank of Jacobian
matrix is used to determine the singular congurations of the robot. As reported in [31], using Eqs. (9) and (11), the direct kinematic
singularity occurs whenever JMS or JIJM becomes singular i.e. Rank(JMS) b 3 but Jac is invertible i.e. det(Jac) 0. Theoretically, this type of
singularity occurs whenever two rows of matrix JMS or JIJM are linearly dependent. By inspection of Eq. (10), we can see that the struc^ i to be parallel. Therefore the second type of singularity also does
ture of the MS does not allow any two of three vectors bi h  m
not occur for the 3-PSP robot.
Similarly, using Eq. (14), the constraint singularity occurs whenever Jc becomes singular i.e. Rank(Jc) b 3. Theoretically,
this type of singularity occurs whenever two rows of matrix Jc are linearly dependent. By inspection of Eq. (15), we can see that vector i
is perpendicular to vector (bi h) i. Therefore the constraint singularity also does not occur for the 3-PSP robot.

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

221

7.1. Inverse kinematics singularity

As reported in [31], the inverse kinematic singularity in overall form occurs whenever Jac becomes singular, det(Jac) = 0, but
Rank(JMS) becomes 3 i.e. matrix JMS has full rank. In reality, because motions of the LRs are limited, this condition cannot occur for
the 3-PSP robot. When the 3 3 square Jacobian matrix for each operational mode is derived, the inverse kinematics singularity
can be evaluated using Direct Square Jacobian Matrix JDJM mode for the selected operational mode, Eq. (20). Therefore the concept
of Inverse Kinematics Singularity, IKS, for each operational mode is investigated for the 3-PSP robot. Using Eqs. (21) and (22), matrices
JDJM Z and JDJM XYZ are employed to obtain the IKS for Z and XYZ operational modes, respectively. Additionally, a numerical
approach is utilized and contours of det( JDJM Z) and det(JDJM XYZ) at a specic plane of the workspace, z = 20 cm, is generated
and shown in Fig. 5. For the non-pure translational mode XYZ, two cases, both with and without tool length, h, are considered. Results
are illustrated in Fig. 5(a) and (b).
As shown in Fig. 5(a), the IKS for the XYZ mode for when h = 0 occurs whenever all three LRs lengths are equal and therefore the
end-effector is positioned at X = Y = 0. Additionally, as shown in Fig. 5(b), the IKS for XYZ mode for when h = 0.08 m occurs whenever the end-effector is positioned in points E, F and G of the workspace. These points are located at circumference with radius r =
0.0645 at angles 60, 180 and 300. Furthermore, as can be seen from Fig. 5(c), there are no singular points throughout the robot's
workspace whenever the robot is operated in the Z mode. Therefore, the IKS for Z mode does not occur throughout the
workspace. Using Eqs. (17)(22), for the end-effector position X = Y = 0 and z = 0.2 m and h = 0, we have
2

JD JM

0
6
0
6
6 0:33
6
6
0
6
4 3:683
0

0
0
0:33
3:19
1:842
0

3
0
0 7
7
0:33 7
7;
3:19 7
7
1:842 5
0

JD JMXYZ

0
0
4 0
0
0:33 0:33

3
0
0 5;
0:33

2
JD JMZ

3
0:33
0:33
0:33
0
3:19 3:19 5 :
3:683 1:842 1:842

(a)

(b)

(c)

Fig. 5. The singularity contours at z = 0.2 m for (a) XYZ with h = 0 (b) XYZ with h = 8 cm and (c) Z modes.

222

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

(a)

(b)

Singular configuration

{T}
v

y
z

z
x

{B}
y

Fig. 6. The inverse kinematics singularity for (a) 2 link robot, (b) 3-PSP parallel robot using force as explanation.

Consider matrix entries for the JDJM. We can conclude that for this end-effector's position, for an instantaneous actuators motion,
ac
q , there are no instantaneous motions for vx, vy and z. Therefore, the direct Jacobian matrices for each operational mode containing
X, Y and are singular as shown in the above JDJM XYZ.
The IKS concept is illustrated in Fig. 6. The IKS for operational modes can be explained using force relations concept of
the robot. From the point of view of the force analysis, direct velocity relation, Eq. (17), and using principle of virtual work, we
can write


f ac JD JM W:

47

This matrix equation should be called Inverse Force Inversion. Where fac and W f f ext Mext gT are the vector of actuated joints'
forces and the vector of applied external wrench to the end-effector with respect to frame {B}, respectively. Using Eq. (47) as well as
matrix JDJM obtained for IKS conguration X = Y = 0 and z = 0.2 m in case h = 0, we can state that forces in x- and y-directions and
moment about z-axis applied to the end-effector are not experienced by motors of the 3-PSP. Also, as shown in Fig. 5(b), in points of E,
F and G the IKS for XYZ mode occurs when h = 0.08 m.
The Jacobian matrices JDJM XYZ for these positions are obtained in Table 2. Determinant of these matrices is near zero. It is clear
that rows 1 and 2 of these matrices are dependent. In other words, using Eq. (22), we can express that in these singular positions the
end-effector motions along x- and y-directions is linearly dependent.
8. Sensitivity analysis: effect of actuators error on the end-effector
One of the most important indices used to explain robot capabilities is sensitivity. Sensitivity studies the inuence of the actuator
errors, joint clearances, tolerances and structure parameters on the end-effector errors and is a fundamental study in the robot design
process. In this paper, the two operational modes are considered and effect of actuator errors on desired position and orientation accuracy of the MS are determined. Additionally, effect of required end-effector accuracy on the actuator accuracy, consequently robot
cost, can be determined. These two points of view can be represented in Fig. 7 shown below.
8.1. Direct position sensitivity and evaluation of the end-effector error boundaries
A robot designer is always tempted to minimize the end-effector errors by minimizing its actuator errors. However, generally as
actuator accuracy increases so does their cost. In the direct sensitivity analysis, we will study how assumed actuators errors will inuence the overall accuracy of the moving platform. Consider the direct velocity equations Eqs. (21) and (22). The relations to investigate the direct sensitivity of the robot with respect to actuators motions can be expressed as
ac

XZ JD JMZ q31

48

Table 2
Jacobians in singular points for XYZ mode when h = 0.08 m.
JDJM XYZ for point E
2
3
0:077 0:074 0:148
4 0:125 0:12 0:241 5
0:476
0:461 0:063

JDJM XYZ for point F

2
3
0:284 0:142 0:142
4
0
0
0 5
0:0625 0:47
0:47

JDJM XYZ for point G

2
3
0:077 0:148 0:074
4 0:125 0:241
0:12 5
0:476
0:063
0:461

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

223

Inverse sensitivity analysis (enduser point of view)

Specify desired endeffector
accuracy by enduser

Determine the allowable

actuators accuracy

Robots cost $

Direct sensitivity analysis (designer point of view)

Specify the actuators
accuracy by enduser

Determine the endeffector

bound of errors

Evaluation of the
endeffector precision

Fig. 7. The strategy for inverse and direct sensitivity analyses.

ac

XXYZ JD JMXYZ q31

49

where XZ f z gT and XXYZ f x y z gT denote the innitesimal displacement/rotation vectors of the

moving platform for the two modes, respectively. Using direct Jacobian matrices, these two equations will allow us to investigate
the inuence of actuator positional errors, qac
3 1, on the positional errors of the MS, XZ and XXYZ. Therefore, the two 3 3 Jacobian
matrices, JDJM Z and JDJM XYZ, can also be called direct sensitivity matrices of the robot. Furthermore, these matrices will allow us
to obtain boundaries for the end-effector pose error in terms of boundaries for admissible error of the actuators.
It is feasible to assume that the admissible actuators errors are bounded. Therefore, the admissible space of actuator errors can be
assumed to be a cube with boundaries as qac
i 0.001(m). Fig. 8 illustrates the geometrical interpretation of these mapping for the
3D case. It is obvious when the space for actuator errors is bounded in a regular cube, then the MS pose error spaces are also bounded
in regular hexahedron spaces (see Fig. 8).
In this subsection, the direct position sensitivities for 3-PSP parallel robot are investigated for two operational modes XYZ and Z
as well as the end-effector error boundaries are also evaluated. First, the admissible actuator errors are applied and the MS pose error
spaces are determined. Next, the maximum MS pose errors for X, Y, Z, and are evaluated throughout the workspaces. Algorithm 1
is used to calculate the direct position sensitivity for the 3-PSP robot.
Algorithm #1. The direct position sensitivity Calculation of the maximax end-effector errors

Note that the maximum value among the maximum end-effector errors, worst case, represents the end-effector precision.
8.1.1. Direct position sensitivity for XYZ mode
Tables 3 and 4 show the maximum errors for constraint singular points as well as several selected end-effector positions where
values for | JDJM XYZ| are equal. To do this, two different tool lengths, h = 0 and h = 0.08(m), are considered. Also, it is assumed
that the actuator errors are equal to: qac
i 0.001(m).
The highlighted rows in Tables 3 and 4 refer to positions where inverse kinematics singularity occurs. For the XYZ mode, the select
robot pose with zero and non-zero tool heights and spaces of end-effector error are shown in Fig. 9(a)(b) and (c)(d), respectively.
The IKS conguration and its corresponding end-effector error are shown in Fig. 9(a) and (c). Also note that the maximum error limitation in the Z direction is always between 103 (m) the same for all points in the workspace.
As shown in Fig. 9(b) and (d), the error in Z direction is approximately twice the X or Y direction. This was inherently expected as
the workspace of the robot in the XYZ mode reects the same pattern. Next consider Fig. 9(a) where the robot is in its IKS state. As

224

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

G
H

Fig. 8. The mapping between the actuator errors and the MS pose error spaces as mode-wise.

shown in Fig. 9(a), the errors in X and Y become nearly zero. This is in line with what was shown in Fig. 6(b) which states forces applied to end-effector in X and Y directions are not transferred to the actuators. Similarly, innitesimal motions on the actuators do not
cause any motion in the X and Y directions of the end-effector. Finally, consider Fig. 9(c), where tip of the robot end-effector with nonzero tool length is at its IKS conguration. As shown in Fig. 9(c), unlike Fig. 9(a), the error space becomes a plane where errors in X and
Y directions are dependent.

Table 3
The Maximum end-effector position errors in case h = 0 for non-pure translational XYZ mode.

Outputs of the direct position sensitivity for XYZ mode

Points

Inputs of InvKin
Max. x
104 m

Max. y
104 m

Max. z
104 m

Max. XY
norm of
error
104 m

Error
volume
xyz
104 m3

X (m)

X (m)

Z (m)

0.2

10

0.04

0.2

0.0319

4.15

3.06

10

4.43

2.55

0.0128

0.0221

0.2

0.0319

5.22

3.7

10

5.47

2.55

0.02

0.0346

0.2

0.0319

3.41

4.37

10

4.43

2.55

0.0255

0.2

0.0319

2.4

5.33

10

5.47

2.55

0.02

0.0346

0.2

0.0319

3.41

4.37

10

4.43

2.55

0.0128

0.0221

0.2

0.0319

5.22

3.7

10

5.47

2.55

JDJMXYZ

Table 4
The maximum end-effector position errors in case h = 0.08(m) for non-pure translational XYZ mode.

Outputs of the direct position sensitivity for XYZ mode

Inputs of InvKin

Error
volume
xyz
1010 m3

Max. y
104 m

Max. z
104 m

Max. XY
norm of
error
104 m

2.92

4.87

10

5.678

5.678

10

5.678

0.2

2.92

4.87

10

5.678

0.2

0.0319

1.7

7.5

10

7.55

2.55

0.0370

0.2

0.0319

3.34

5.79

10

6.687

2.55

0.0133

0.0231

0.2

0.0319

6.93

4.49

10

7.55

2.55

0.0428

0.2

0.0319

6.687

1.9

10

6.687

2.55

0.0133

0.0231

0.2

0.0319

6.93

4.49

10

7.55

2.55

0.0214

0.0370

0.2

0.0319

3.34

5.79

10

6.687

2.55

Points

Max. x
104 m

X (m)

X (m)

Z (m)

JDJMXYZ

0.0323

0.0559

0.2

0.0645

0.2

0.0323

0.0559

0.0267

0.0214

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

(a)

225

0.4

z(m)

0.3
P

0.2

0.1

0
-0.1568

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

(b)
0.4

z(m)

0.3

0.2

0.1

0
-0.1568

-0.0905

-0.1
0

y(m)

(c)

0
0.1
0.1
0.1568 0.181

x(m)

0.4
P

z(m)

0.3

0.2

0.1

0
-0.1568

-0.0905

-0.1
0

y(m)

(d)

0
0.1
0.1
0.1568 0.181

x(m)

0.4

0.3

z(m)

0.2

0.1

0
-0.1568

-0.0905

-0.1

y(m)

0.1
0.1568 0.181

0.1

x(m)

Fig. 9. Conguration of the robot and error space of the end-effector in XYZ mode for cases: (a) h = 0, x = 0, y = 0 and z = 0.2, (b) h = 0, x = 0.0128, y = 0.0221, z =
0.2, (c) h = 0.08(m), x = 0.0645, y = 0, z = 0.2 and (d) h = 0.08(m), x = 0.0214, y = 0.0370, z = 0.2(m).

Next consider Fig. 10, which shows the error volume for an arbitrary point in robot workspace. Then, for each point, we can calculate maximum error in X, Y and Z directions as well as magnitude of the maximum error that ts in this volume. To do this an algorithm 1 is developed and results are shown in Fig. 10.
Also, the maximax end-effector values of errors for XYZ operational mode are separately equal X = 5.37 104(m) and Y =
5.52 104(m) and Z = 10 104(m).

226

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

Fig. 10. The XYZ Workspace for plane z = 0.2(m) with h = 0 and the maximum errors along x-, y- and z-axis for the end-effector in entire workspace, magnitude of the
XY error and volume of end-effector error.

8.1.2. Direct position sensitivity for Z mode

Table 5 shows maximum errors for several selected MS poses where values for |JDJM Z| are equal. It is assumed that the actuator
errors are equal to: qac
i 0.001(m).
To better illustrate, Fig. 11 depicts error spaces of the MS pose for Z mode for a number of MS poses presented in Table 5.
Additionally, the maximum MS pose errors and the end-effector error boundaries are evaluated throughout the workspaces for
the XYZ operational mode. Fig. 12 depicts contours of the maximum errors along x, y and z axes for the end-effector for h = 0 in
plane z = 0.2(m).
Also, the maximax end-effector values of errors for Z operational mode are separately equal Z = 10 104(m) and
= 6.38 103(deg) and = 7.37 103(deg).
8.2. Inverse position sensitivity and design of the actuators accuracy
Generally, there is a direct correlation between a robot nancial cost and its accuracy requiring. This cost requires the end user to
carefully consider the needed process accuracy when selecting a robot. On the other hand, a robot designer needs to carefully consider
the required end-effector accuracy when selecting the required accuracy of the robot's actuators. This process may be presented by an
inverse sensitivity analysis. Using Eqs. (23) and (24), the inverse sensitivity relations can be derived. Subsequently, robot's resulting
actuators accuracy with respect to allowable (desired) MS pose errors for each operational mode using the inverse sensitivity relations
will be available. Therefore
ac

50

ac

51

q31 JD JMZ XZ
q31 JD JMXYZ XXYZ :
Table 5
The maximum end-effector position errors for the coupled-mixed type Z mode.
Inputs of InvKin

Outputs of the direct position sensitivity for Z mode

Points

Z (m)

(deg)

(deg)

|JDJM Z|

Max.
z 104 m

Max.
104 deg

Max.
104 deg

Max norm of
error 104 deg

Error volume
Z 104 m (deg)2

1
2
3
4

0.2
0.2
0.2
0.2

0
30
17.41
17.41

0
0
30.15
30.15

11.75
7.63
7.63
7.63

10
10
10
10

0.365
0.274
0.333
0.333

0.422
0.365
0.301
0.301

0.422
0.365
0.386
0.386

3.085
2
2
2

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

(a)

227

0.4

z(m)

0.3
P

0.2

0.1

0
-0.1568

-0.0905

-0.1

y(m)

0.1
0.1568 0.181

0.1

x(m)

(b)
0.4

z(m)

0.3
P

0.2

0.1

0
-0.1568

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

Fig. 11. Conguration of the robot and error space of the MS in Z mode for cases: (a) = = 0, z = 0.2(m) and (b) = 17.41 (deg), = 30.15 (deg), z = 0.2 (m).

The inuences of the MS pose errors in each operational mode, XZ and XXYZ, on the actuator positional errors, qac
3 1, can be
investigated using inverse Jacobian matrices. In Eqs. (50) and (51), inverse of the direct Jacobian matrices for Z and XYZ modes,
1
1
JDJM
Z and JDJM XYZ, can be re-stated as sensitivity matrices in inverse position sensitivity analysis of the robot. In other words,
1
matrices JDJM Z and J1
DJM XYZ can be used in robot design process when a desired MS pose error is specied and required actuator
error needs to be determined. Consequently, we can select an actuator mechanism that meets the desired MS accuracy. The allowable

Fig. 12. The Z Workspace for plane z = 0.2(m) and the maximum errors along z-axis and about x-, y-axis for the end-effector in entire workspace, magnitude of the
error and volume of end-effector error.

228

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

(m)

G
H

(m)
(m,rad,rad)

G
F

H
(m,rad,rad)

Fig. 13. The mapping between the actuator errors and the MS pose error spaces as mode-wise.

space of MS pose errors can be assumed to be bounded on a regular cube-shape space. These boundaries are determined by desired
design inputs for the robot. Fig. 13 illustrates the geometrical interpretation of inverse sensitivity mapping for the 3D case. Similar to
the direct sensitivity mapping, when a cubic space of allowable MS pose errors is considered, the corresponding actuator error space is
bounded in a regular hexahedron space (see Fig. 13).
Algorithm 2 presents the inverse position sensitivity for the 3-PSP robot. Note that minimum value among all three actuator maximum
errors, within the robot workspace, represents maximum allowable actuator errors which generally directly inuence the robot cost.
Algorithm #2. The inverse position sensitivity Calculation of the minimax actuators errors

8.2.1. Inverse position sensitivity for XYZ mode

The inverse position sensitivities for the 3-PSP parallel robot are next investigated for the two operational modes XYZ and
Z. In the inverse position problem, the desired space for the end-effector errors is specied as a cube with boundaries as
XXYZ 105(m) and the resulting error boundaries for the actuators are determined. These errors will dene precision of
the robot's end effector in the XYZ mode. The algorithm 2 shows steps taken for the inverse position sensitivity.
Tables 6 and 7 show the maximum errors for several actuator errors when the allowable end-effector error space,
XXYZ 105(m), as shown in Fig. 13, is performed to the robot. The allowable actuator error boundaries and values for
maximum allowable actuator error are evaluated throughout the workspaces for both operational modes.
To better illustrate, Fig. 14 depict error spaces of the MS pose for XYZ mode for a number of MS poses presented in Tables 6
and 7.
Additionally, the maximum allowable actuator error boundaries are evaluated throughout the workspaces for XYZ operational
mode. Fig. 15 depicts contours of the maximum allowable actuator error for h = 0 in special plane z = 0.2(m).

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

229

Table 6
The maximum actuators error in case h = 0 for selected points in the XYZ mode.

Outputs of the inverse position sensitivity for XYZ mode

Inputs of InvKin
JDJMXYZ

Max. q1
105 m

Max. qac
2
105 m

Max. qac
3
105 m

Error volume
ac
ac
13 m3
qac
1 q 2 q 3 10

0.2

infinity

infinity

infinity

infinity

0.2

0.0319

3.35

7.58

7.58

2.51

0.0221

0.2

0.0319

3.93

3.18

11

2.51

0.02

0.0346

0.2

0.0319

5.83

4.2

7.57

2.51

0.0255

0.2

0.0319

8.33

3.91

3.91

2.51

0.02

0.0346

0.2

0.0319

5.83

7.57

4.2

2.51

0.0128

0.0221

0.2

0.0319

3.93

11

3.18

2.51

Points

X (m)

X (m)

Z (m)

0.04

0.0128

ac

Table 7
The maximum actuators error in case h = 0.08(m) for selected points in the XYZ mode.

Inputs of InvKin
Points

Outputs of the inverse position sensitivity for XYZ mode

Max. qac
2
105 m

Max. qac
3
105 m

Error volume
ac
ac
13 m3
qac
1 q 2 q 3 10

JDJMXYZ

Max. qac
1
105 m

0.2

330

390

20.9

143

0.2

4.3

15960

159.60

9000

X (m)

X (m)

Z (m)

0.0323

0.0559

0.0645

0.0323

0.0559

0.2

330

20.9

390

143

0.0267

0.2

0.0319

8.79

6.34

6.34

2.51

0.0214

0.0370

0.2

0.0319

8.31

4.5

2.51

0.0133

0.0231

0.2

0.0319

6.93

11.55

5.95

2.51

0.0427

0.2

0.0319

3.56

6.68

6.68

2.51

0.0133

0.0231

0.2

0.0319

6.93

5.95

11.5

2.51

0.0214

0.0370

0.2

0.0319

4.5

8.31

2.51

(a)

(b)
0.4

(c)
0.4

0.4

0.3

0.3

0.3

0.1

0
-0.1568

z(m)

z(m)

z(m)

0.2

0.2

0.1

0.1

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

0
-0.1568

0.2

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

0
-0.1568

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

Fig. 14. Conguration of the robot and error space of the actuators in XYZ mode for cases: (a) h = 0, x = 0.0128, y = 0.0221, z = 0.2, (b) h = 0.08(m), x = 0.0645,
y = 0, z = 0.2 and (c) h = 0.08(m), x = 0.0214, y = 0.0370, z = 0.2(m).

230

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

Fig. 15. The maximum allowable actuator error for h = 0 in z = 0.2(m) plane for XYZ mode.

4
4
Also, the minimax of actuator error values for XYZ mode are equal: qac
(m) and qac
(m) and
1 = 0.306 10
2 = 0.306 10
4
qac
(m).
3 = 0.306 10

8.2.2. Inverse position sensitivity for Z mode

Consider the admissible space of the end-effector errors as a cube with boundaries of 105 XZ 105 in Z mode as schematically shown in Fig. 13. Note, for simplicity, same values of 105(m) and 105(rad) is used for the Z and variables. Therefore,
the resulting boundaries of the actuators error can be obtained for each position of the end-effector using Eq. (50). Table 8 shows maximum resulting actuator errors.
To better illustrate, Fig. 16 depicts error spaces of the actuators for Z mode for a number of MS pose which are presented in Table 8.
The maximum resulting actuator error boundaries are also evaluated throughout the Z workspaces. Fig. 17 depicts contours of
the maximum actuator error in z = 0.2(m) plane.
4
4
Also, the minimax of actuator error values for Z mode is equal: qac
(m) and qac
(m) and
1 = 0.113 10
2 = 0.113 10
4
ac
q3 = 0.113 10 (m).
9. Conclusion
This paper provides a new perspective in the singularity and sensitivity analyses of the 3-PSP parallel robot and clear algorithms provide for these analyses. The 3-PSP robot is a limited-DOF parallel robot and its moving platform has 3-DOFs in

Table 8
Maximum end-effector positions error for selected points in the Z mode.
Inputs of InvKin

Outputs of the inverse position sensitivity for Z mode

Points

Z (m)

(deg)

(deg)

|JDJM Z|

5
Max. qac
m
1 10

5
Max. qac
m
2 10

5
Max. qac
m
3 10

15
ac
ac
Error volume qac
m3
1 q2 q3 10

1
2
3
4

0.2
0.2
0.2
0.2

0
30
17.41
17.41

0
0
30.15
30.15

11.75
7.63
7.63
7.63

1.18
1.167
1.29
1.29

1.25
1.36
1.21
1.32

1.25
1.355
1.32
1.21

0.68
1.05
1.05
1.05

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

231

(a)
0.4

z(m)

0.3
P

0.2

0.1

0
-0.1568

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

(b)
0.4

z(m)

0.3
P

0.2

0.1

0
-0.1568

-0.0905

-0.1
0

y(m)

0
0.1
0.1
0.1568 0.181

x(m)

Fig. 16. Conguration of the robot and error space of the actuators in Z mode for cases: (a) = = 0, z = 0.2(m) and (b) = 17.41 (deg), = 30.15 (deg), z = 0.2 (m).

space. Therefore, three of the six motion variables of the moving platform can be selected as operational modes. A methodology
is represented to select the most practical operational modes and two modes called non-pure translational and coupled mixed
type modes are considered. Using this approach, the Jacobian matrices for each of the two modes are obtained. Additionally, the

Fig. 17. The maximum resulting actuator error for h = 0 in z = 0.2(m) plane for the Z mode.

232

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

inverse and direct velocity and acceleration relations are derived based on using the auxiliary vectors to eliminate passive velocity and acceleration vectors. Next a survey to identify singular congurations of the robot is completed and the three wellknown singularities as well as the constraint singularity are investigated. Then, for two operational modes, the notion of inverse
kinematics singularity is dened. We can conclude that it can be possible that the overall Jacobian matrices are not singular
throughout the workspace, but its sub-matrices for each operational mode can be singular. This reality for the 3-PSP robot is
proved as the inverse kinematic singularity in overall form cannot occur but the inverse kinematic singularity occurs for the
XYZ mode. To clear this, the concept of IKS for XYZ mode is explained using the inverse force relation. Also, the contours of
IKS loci for two operational modes in a specic plane of the robot workspace are generated. Finally, the direct and inverse sensitivity analyses are investigated for two operational modes by dening concept of direct and inverse squared Jacobian sensitivity matrices for the two operational modes. The direct sensitivity matrices allow us to obtain boundaries for the end-effector
pose errors for each operational mode in terms of the admissible error boundaries of the actuators. For detail explanations, an
algorithm is used to evaluate the maximum MS pose errors for X, Y, Z, and throughout the workspaces. Additionally, the inverse sensitivity matrices allow us to evaluate the allowable actuator error boundaries with respect to desired MS pose errors for
each operational mode. In other words, the inverse sensitivity matrices can be used in robot design process. To do this, a desired
MS pose errors are specied and the required actuator errors are determined. For detail explanations, an algorithm is used to
evaluate the maximum actuator errors for two operational modes throughout the workspaces.
The major contributions are highlighted below:
Providing a new general perspective and a systematic study in analyzing work space, singularity and sensitivity analyses.
Presenting a design process for sensitivity analysis of robot from both end-user and designer point of views.
Presenting a methodology to select the most practical operational modes and select two most practical operational modes called
non-pure translational and non-pure coupled mixed-type modes.
Obtaining Jacobian matrices and the inverse and direct relations for velocity and acceleration for both operational modes.
Investigating on all types of singularity and representing the notion of inverse kinematics singularity for the two operational
modes.
Presenting the direct and inverse sensitivity analyses for the two operational modes.
Nomenclature
B{x, y, z} The xed coordinate frame which is attached to point O in top of xed platform
T{u, v, w} The moving coordinate frame which is attached to point T in the center of MS
a
Distance between the center point O of xed triangle A1A2A3, and point Ai
h
Tool length
B
ai
The position vectors locate corners of the xed base, Ai in frame {B}, for i = 1, 2, 3
B ac
qi
The position vectors which specify length of each linear actuator, for i = 1, 2, 3
B
t
The position vector of the end-effector, point T, with respect to {B}
B
p
The position vector of the tool tip, point P, with respect to {B}
T
h
The position vector which connects point T to point P
T
bi
The position vectors connect the end-effector to the ith S-joint, Si in frame {T}, for i = 1, 2, 3
, ,
Rotational variables about the x-, y- and z-axes (Euler angles)
xT, yT, zT Translational variables of the MS center along the x-, y- and z-axes
xP, yP, zP Translational variables for the Tool tip along the x, y and z-axes
qac
Translational variables for the linear rods, LRs, for i = 1, 2, 3
i
bi
Translational variables for the MS branches, for i = 1, 2, 3
B
The rotation matrix to transfer a vector dened in {T} to {B}
TR
tMS
Twist vector of the moving platform f vP MS gT
ac
qi
The values of ith actuated joint rate, for i = 1, 2, 3
i
The values of ith passive prismatic translational
i = 1, 2, 3
 joint rate, for
T
vP
The Cartesian velocity vector for the tool tip, vx vy vz

T
MS
The angular velocity vector of the MS, x y z
ac
^i
The unit vectors along ith LRs, for i = 1, 2, 3
q
^
b
The unit vectors along ith branch of the MS, for i = 1, 2, 3
i
^i
The three unit vectors which are all perpendicular to the MS plane, for i = 1, 2, 3
m
n
oT
ac
ac
ac
ac
q
Vector of the actuated joints velocity, q1 q2 q3


JIJM
Jc
JFIJM

ac

The Inverse Jacobian Matrix which maps tMS to q

The Jacobian of constraints
n
oT
Full Inverse Jacobian matrix which maps tMS to q ac013


ac

JDJM
The Overall Direct Jacobian Matrix which maps q to tMS
JDJM mode Direct Square Jacobian Matrix for selected operational mode
vres mode Restricted twist vector of the MS for selected operational mode
vZ
The MS velocities for Z operational mode, {vz, x, y}T

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

233

vXYZ
The MS velocities for XYZ operational mode, {vx, vy, vz}T
ac
JDJM Z Direct Coupled Mixed-type Jacobian Matrix which maps q to vZ
ac
JDJM XYZ Direct non-pure translational Jacobian Matrix which maps q to vXYZ
JOIJM
Overall Inverse Jacobian Matrix
n
oT
ac
JODJM
Overall Direct Jacobian Matrix which maps q to tMS b
ac
q
The values of ith actuated joint acceleration
i

b
The values of ith passive prismatic joint linear acceleration
i
 ac
T
ac
ac
ac
1 q
q
Vector of the actuated joints acceleration, q
2 q3
T
vP
The Cartesian acceleration vector for the tool tip, vx vy vz

T
MS
The angular acceleration vector of the MS, x y z

T
MP
Vector of the MS acceleration, vP MS
vresmode Restricted acceleration vector of the MS for selected
 operational
T mode
vZ
The MS accelerations for Z operational mode, vz ; x ; y

T
vXYZ
The MS accelerations for XYZ operational mode, vx ; vy ; vz


References
[1] S.M. Varedi, H.M. Daniali, D.D. Ganji, Kinematics of an offset 3-UPU translational parallel manipulator by the homotopy continuation method, Nonlinear Anal.
Real World Appl. 10 (2009) 17671774.
[2] M. Ruggiu, Kinematics analysis of the CUR translational manipulator, Mech. Mach. Theory 43 (2008) 10871098.
[3] Jaime Gallardo-Alvarado, Jos Mara Rico-Martnez, Grsel Alici, Kinematics and singularity analyses of a 4-dof parallel manipulator using screw theory, Mech.
Mach. Theory 41 (9) (2006) 10481061.
[4] K.H. Harib, A.M.M. Sharif Ullah, A. Hammami, A hexapod-based machine tool with hybrid structure: kinematic analysis and trajectory planning, Int. J. Mach. Tools
Manuf. 47 (2007) 14261432.
[5] H.T. Liu, et al., Optimal design of the TriVariant robot to achieve a nearly axial symmetry of kinematic performance, Mech. Mach. Theory 42 (12) (2007)
16431652.
[6] Marc Arsenault, Roger Boudreau, Synthesis of planar parallel mechanisms while considering workspace, dexterity, stiffness and singularity avoidance, J. Mech.
Des. 128 (2006) 69.
[7] J.P. Merlet, Jacobian, manipulability, condition number, and accuracy of parallel robots, ASME J. Mech. Des. 128 (2006) 199.
[8] Kuang-Chao Fan, et al., Sensitivity analysis of the 3-PRS parallel kinematic spindle platform of a serial-parallel machine tool, Int. J. Mach. Tools Manuf. 43 (15)
(2003) 15611569.
[9] Amir Rezaei, Alireza Akbarzadeh, Mohammad-R. Akbarzadeh-T, An investigation on stiffness of a 3-PSP spatial parallel mechanism with flexible moving platform
using invariant form, Mech. Mach. Theory 51 (2012) 195216.
[10] Amir Rezaei, Alireza Akbarzadeh, Position and stiffness analysis of a new asymmetric 2PRRPPR parallel CNC machine, Adv. Robot. 27 (2) (2013) 133145.
[11] N. Binaud, P. Cardou, S. Caro, P. Wenger, The Kinematic Sensitivity of Robotic Manipulators to Joint Clearances, Proceedings of ASME Design Engineering Technical
Conferences, August 1518, 2010, Montreal, QC., Canada, 2010.
[12] M. Tannous, S. Caro, A. Goldsztejn, Sensitivity analysis of parallel manipulators using an interval linearization method, Mech. Mach. Theory 71 (2014)
93114.
[13] Nicolas Binaud, Stphane Caro, Philippe Wenger, Sensitivity comparison of planar parallel manipulators, Mech. Mach. Theory 45 (11) (2010) 14771490.
[14] Geoffrey Pond, Juan A. Carretero, Formulating Jacobian matrices for the dexterity analysis of parallel manipulators, Mech. Mach. Theory 41 (12) (2006)
15051519.
[15] A. Yu, I. Bonev, P. Zsombor-Murray, Geometric approach to the accuracy analysis of a class of 3-DOF planar parallel robots, Mech. Mach. Theory 43 (3) (2009)
364375.
[16] H.S. Kim, L.-W. Tsai, Design optimization of a Cartesian parallel manipulator, ASME J. Mech. Des. 125 (2003) 4351.
[17] S. Briot, I.A. Bonev, Accuracy analysis of 3-DOF planar parallel robots, Mech. Mach. Theory 43 (2007) 445458.
[18] Y. Li, Q. Xu, Kinematic analysis of a 3-PRS parallel manipulator, Robot. Comput. Integr. Manuf. 23 (2007) 395408.
[19] Doik Kim, Wankyun Chung, Analytic singularity equation and analysis of six-DOF parallel manipulators using local structurization method, Robot. Autom. IEEE
Trans. 15 (4) (1999) 612622.
[20] Y. Li, Q. Xu, Kinematics and stiffness analysis for a general 3-PRS spatial parallel mechanism, ROMANSY, 2004.
[21] G. Alc, B. Shirinzadeh, Loci of singular configurations of a 3-DOF spherical parallel manipulator, Robot. Auton. Syst. 48 (2004) 7791.
[22] X.J. Liu, J. Wang, G. Pritschow, Kinematics, singularity and workspace of planar 5R symmetrical parallel mechanisms, Mech. Mach. Theory 41 (2006)
145169.
[23] Yi Lu, et al., Kinematics/statics analysis of a novel 2SPS + PRRPR parallel manipulator, Mech. Mach. Theory 43 (9) (2008) 10991111.
[24] O. Altuzarra, O. Salgado, V. Petuya, A. Herna_ndez, Point-based Jacobian formulation for computational kinematics of manipulators, Mech. Mach. Theory 41
(2006) 14071423.
[25] N. Simaan, M. Shoham, Geometric interpretation of the derivatives of parallel robots' Jacobian matrix with application to stiffness control, ASME J. Mech. Des. 125
(2003) 33.
[26] F. Firmani, R.P. Podhorodeski, Singularity analysis of planar parallel manipulators based on forward kinematic solutions, Mech. Mach. Theory 44 (7) (2009)
13861399.
[27] J. Angeles, Guilin Yang, I-Ming Chen, Singularity analysis of three-legged, six-DOF platform manipulators with RRRS legs, Advanced Intelligent Mechatronics,
2001. Proceedings. 2001 IEEE/ASME International Conference on, IEEE, vol. 1, 2001.
[28] Alfonso Hernndez, et al., Transitions in the velocity pattern of lower mobility parallel manipulators, Mech. Mach. Theory 43 (6) (2008) 738753.
[29] Sameer A. Joshi, Lung-Wen Tsai, Jacobian analysis of limited-DOF parallel manipulators, J. Mech. Des. 124 (2002) 254.
[30] N. Simaan, M. Shoham, Geometric interpretation of the derivatives of parallel robots' Jacobian matrix with application to stiffness control, J. Mech. Des. 125
(2003) 33.
[31] Amir Rezaei, et al., Position, Jacobian and workspace analysis of a 3-PSP spatial parallel manipulator, Robot. Comput. Integr. Manuf. 29 (4) (2013)
158173.
[32] Hamid Reza Hassanzadeh, Mohammad-R. Akbarzadeh-T, Alireza Akbarzadeh, Amir Rezaei, An interval-valued fuzzy controller for complex dynamical systems
with application to a 3-PSP parallel robot, Fuzzy Sets Syst. 235 (4) (2014) 83100.
[33] Y. Jin, I. Chen, Effects of constraint errors on parallel manipulators with decoupled motion, Mech. Mach. Theory 41 (8) (2006) 912928.
[34] A.-H. Chebbi, Z. Affi, L. Romdhane, Prediction of the pose errors produced by joints clearance for a 3-UPU parallel robot, Mech. Mach. Theory 44 (9) (2009)
17681783.

234

A. Rezaei, A. Akbarzadeh / Mechanism and Machine Theory 86 (2015) 211234

[35] Philippe Cardou, Samuel Bouchard, Clment Gosselin, Kinematic-sensitivity indices for dimensionally nonhomogeneous Jacobian matrices, IEEE Trans. Robot. 26
(1) (2010) 166173.
[36] Stphane Caro, Nicolas Binaud, Philippe Wenger, Sensitivity analysis of 3-RPR planar parallel manipulators, J. Mech. Des. 131 (2009) 121005.
[37] Mohammad Hossein Saadatzi, et al., Geometric analysis of the kinematic sensitivity of planar parallel mechanisms, Trans. Can. Soc. Mech. Eng. 35 (4) (2011)
477490.
[38] G. Wu, S. Bai, J.A. Kepler, S. Caro, Error modeling and experimental validation of a planar 3-PPR parallel manipulator with joint clearances, ASME J. Mech. Robot. 4
(4) (2012) 041008-1041008-12.
[39] Q. Li, J.M. Herve, 1T2R parallel mechanisms without parasitic motion, IEEE Trans. Robot. 26 (3) (2010) 401410.