Mechanism and Machine Theory 46 (2011) 577592

Contents lists available at ScienceDirect

Mechanism and Machine Theory

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / m e c h m t

Workspace analysis of positioning discontinuities due to clearances in

parallel manipulators
Oscar Altuzarra a, Jokin Aginaga b,, Alfonso Hernndez a, Isidro Zabalza b
a
b

University of the Basque Country UPV/EHU, Faculty of Engineering in Bilbao, Department of Mechanical Engineering, Alameda de Urquijo s/n, 48013, Bilbao, Spain
Public University of Navarre, Campus Arrosada s/n, Pamplona 31006, Spain

a r t i c l e

i n f o

Article history:
Received 30 September 2009
Received in revised form 13 September 2010
Accepted 11 January 2011
Available online 3 February 2011
Keywords:
Clearance
Workspace
Dynamics
Kinetostatics
Positioning discontinuities

a b s t r a c t
Clearances at joints produce a loss of accuracy when positioning a mechanism. The end-effector
pose error due to clearances depends on the mechanism configuration, the magnitude of the
clearance itself and applied external wrenches. Sudden changes which occur in the actual
posture of the mechanism owing to a change of contact mode at joints can be detected in the
neighbourhood of some configuration. These sudden changes lead to positioning discontinuities on certain trajectories, or on the workspace. In this paper, a methodology for analysing
the location of the discontinuities by means of a dynamic or kinetostatic analysis is presented.
The advantages of choosing either the dynamic or the kinetostatic approach are analysed,
making use of the 5R planar mechanism.
2011 Elsevier Ltd. All rights reserved.

1. Introduction
With exact design dimensions, mechanisms with n degrees of freedom have what can be called a nominal conguration space.
However, manufacturing and assembly errors cause changes in this space. Despite this, if the mechanisms are not overconstrainedwe will restrict ourselves to these for the time beingthe conguration space maintains its dimension when ideal
joints are considered. In the conguration space of a mechanism with ideal joints, these changes can be handled with an
appropriate calibration.
However, when considering clearance joints, there are additional changes in the conguration space and calibration is not
applicable. Clearances in joints increase the dimension of the conguration space, but within a limited boundary. So for given xed
inputs, there is an uncertainty about exactly where the end effector is, it being located somewhere within a clearanceconguration space limited by the clearance values.
Mechanisms take up clearances in different ways depending on the posture, applied loads and their dynamics. This determines
the actual pose that the end-effector can take due to clearances. In addition, from one pose to another, there may be sudden
changes in the way mechanisms take up clearances producing discontinuities in the trajectory. The dependency on applied loads
and dynamics as well as the discontinuities make a corrective procedure analogous to calibration difcult.
Accordingly, it is necessary to analyse the effect of clearance and identify discontinuities and their cause. A deterministic
accuracy analysis has to be established so that the error in positioning due to the clearance can be found and we can identify the
situations where discontinuities occur. In these postures we will need an analysis of the transmission of forces through the
mechanism to work out what is the cause of the discontinuity. This can be achieved using a kinetostatic analysis as most authors
do, or we could include dynamic analysis as well. We will show that this type of discontinuity is usually due to sudden changes in

Corresponding author. Tel.: +34 948 16 92 86; fax: +34 948 16 90 99.
E-mail address: jokin.aginaga@unavarra.es (J. Aginaga).
0094-114X/$ see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechmachtheory.2011.01.005

578

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

the relative position between the two parts of a joint. These can be described as changes in the contact mode of a joint, which occur
when contact is lost. Such changes lead to a sudden change in the direction of joint forces and moments.
Several research groups have worked on the accuracy analysis of mechanisms to predict the actual end-effector position
considering clearances. Most published studies propose methods that can be included in the following groups:
Geometrical methods. Some authors have evaluated the end-effector position inuenced by clearances using the position
problem or relying solely on the geometric characteristics derived from a simple modelling of the clearances. For example, the
workspace of the end-effector due to clearances in revolute pairs for given inputs is obtained in [1]. In [2] and [3], the authors use
an equivalent link for the revolute pair and evaluate position deviation using position equations and introducing probabilistic
distributions. The DenavitHartenberg closed-loop equation is used in [4] to introduce probabilistic variations in the DH
parameters due to clearance and manufacturing errors, and an optimisation problem to produce a robust design is proposed.
Common drawbacks of these geometrical methods are that: they are based on the position problem which is nonlinear, they
evaluate maximal or most-probable nal positions but not the actual one, and their result is the same irrespective of the load and
motion of the mechanism.
Kinematic methods. Other authors assess the problem of nding the maximum error resulting from a set of nominal joint
clearances on the basis that, for given inputs, the slop of the mechanism due to clearances is so small that it can be evaluated
approximately with velocity analysis. An objective function is stated with the clearance freedoms dened for several models of
the joint clearances, and maximised or minimised as necessary. In [5], the authors produce a seminal work on this type of
method: it is based solely on geometrical and velocity analysis using screw coordinates, and it is applied to a spatial parallel
manipulator. More recently, in [6], an improved version of the aforementioned technique has been published, in which the
maximal pose error is found allowing also passive joint parameters to vary due to clearances, and overconstrained mechanisms
can be analysed. In [7], the inuence of each joint clearanceincluding several contact modeson the pose error is assessed
using a set of six kinetostatic analyses with independent virtual loads applied to the end-effector, and subsequently making use
of the principle of virtual work to obtain the total error. This way the maximisation can be very much simplied by considering
the effects of each joint and at every contact mode separately. Later, in [8], a rst step toward a deterministic method was made
by including actual loads through a new kinetostatic analysis. On the one hand, the use of kinematic models in these methods has
made the accuracy analysis linear, and more complex contact models of the joints have been added. Also, maximisation analysis
has been simplied and overconstrained mechanisms can be handled. On the other hand, the actual load and motion
characteristics have not yet been included.
Deterministic methods. These try to calculate the actual error in the end-effector pose under the action of the load on the
manipulator. In order to do that, the rst step is a kinetostatic analysis to obtain actual reactions on the joints. Several
authors apply the principle of virtual work to assess the contribution of the clearances of each joint to the pose error, and
then sum these contributions. For example, in [9], the author uses a kinetostatic model but only considers mechanisms with
revolute pairs and the load on one link. This approach is completed, in [8], with the adding of the possibility of additional
loads and opening of the procedure to other pair types. In [10], the authors model the clearance of a revolute pair in the
plane with an equivalent link and use the principle of virtual work on a four-bar mechanism identifying sudden changes of
pose at dead centre positions. More recently, the same authors have extended their technique to multi-loop linkages [11].
In [12], kinetostatic analysis is used in such a way that an analytical model is obtained for the pose error of the end-effector
anywhere within the workspace. And nally, some authors add the effect of deformations of links to the analysis of the
effect of clearances (e.g., [13]).
Multibody methods. This kind of method is focused on describing contact forces inside imperfect joints. The response of a
mechanical system with clearance joints becomes chaotic for high speeds and low contact friction, whereas it remains periodic
for lower speeds and a low coefcient of restitution [14]. For revolute joints, three kinds of relative motion between the journal
and the bearing have been distinguished: free movement, impact and continuous contact with or without friction. Some authors
have modelled the impact with nite element models and others use analytical impact force models. The calculated joint
reactions have been introduced into the dynamic model as external generalised forces [1517]. Impact forces cause high peaks in
accelerations and staircase-shaped velocities. These phenomena are deadened by lubricated joints [17] or exible links [18].
Multibody methods are appropriate for detailed analysis of advanced design of mechanisms with high dynamic requirements.
However, in the early stages of design they have a reduced usefulness because of their high computational cost and demand for
detail.
In this paper, error pose due to clearances is calculated by a deterministic method, which can be applied in the early design
stage. In such an analysis, it is commonly assumed that there is always a frictionless contact inside the imperfect joints.
Nevertheless, contact loss is detected in the analysis and corresponding conditions where this happens are analysed. Analysing
such conditions can be considered more realistic, since the actual error is calculated based on clearance magnitude, which is
always uncertain. Bearing this fact in mind, this paper presents an analytical methodology for determining where the contact loss
occurs in applications in which inertial loads can be neglected.
Clearance magnitude is much smaller than other dimensions and consequently it is assumed to be innitesimal, which leads to
a linear analysis [59,12]. The use of virtual links proposed in [10,11] can overcome this simplication, but resultant hyper-degreeof-freedom linkages cannot be solved analytically so numerical methods are required. Another weakness of virtual links is the
complexity when applied to spatial mechanisms with any kind of joints.

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

579

The method developed uses loop closure equations expressed in screw coordinates as in [5,10,11]. Clearance take-up is
given from the results of a kinetostatic or dynamic analysis of the nominal mechanisms as in [8,9,12]. Then, loop closure
equations are solved with clearance freedoms in order to determine variations on passive freedoms, i.e., in order to know how
clearances affect them. We note that such an analysis could be improved by an iterative process such as the one in [13].
Further, the effect of contact modes described in [7,12] could be taken into account by establishing relations between different
clearances of a certain imperfect joint, as in [5]. The error calculation method proposed has characteristics taken from various
different published methods and exploits the strengths of these methods in an attempt to overcome the limitations that each
of them presents.
When the error calculation procedure is carried out on a path or on the workspace, it provides the locations where
discontinuities on the end-effector actual position occur due to sudden changes in contact modes. Merlet [19] mentions
ight simulators, pick-and-place, vibration simulation and even high-speed machining tools as applications where dynamics
play an important role. In such applications, locations where discontinuities occur have a strong dependence on inertial
loads. On the other hand, in applications where inertial effects can be neglected, such as machine tools, micro-positioning
and surgical robots, these locations only depend on the applied external loads and in the mass distribution of the
manipulator itself. This paper shows how this dependence on inertial loads vanishes when slowing down the velocity of
the manoeuvres.
In order to identify the locations where discontinuities occur in kinetostatic applications and to calculate them efciently, an
analytical methodology for the so-called discontinuity loci calculation is developed. Discontinuity loci and their neighbourhood
must be avoided because they correspond to congurations of the mechanism with a high level of uncertainty and calibration is
not efcient. The main contribution of this work is to provide a method to locate regions of the workspace where discontinuities
due to changes in the contact mode of the imperfect joints do not appear.
2. Clearance error analysis
The rst tool that we need to develop is a powerful procedure for the analysis of pose errors due to clearances. This procedure
includes the possibility of applying any external load to any link of the mechanism, evaluates reactions at joints to decide the way
clearances are actually taken up, and nds the end-effector pose error considering the joint clearances and allowing for pose
corrections in passive joints.
The procedure includes the following modules:
The inverse position problem. We will analyse the ideal mechanism to nd the nominal pose at one conguration, over a desired
path, or throughout the workspace.
An evaluation of joint reactions. By means of a dynamic analysis, the wrenches at the joints under the external loads of the
mechanism will be calculated.
An accuracy analysis. Our approach is to employ a deterministic method using the kinematic velocity relations. For every
posture, we will dene with the joint reactions the innitesimal displacements at joints due to clearances. With xed inputs and
the velocity equations at the posture analysed, we will obtain changes in the passive joint rates due to clearances, and then the
end-effector pose error.
Numerical analysis of discontinuities. The pose error will show if clearance take-up changes suddenly at some postures, which
clearance causes this, and how wrenches are transmitted through the mechanism before and after that. These sudden changes
appear at certain poses depending on the conguration of the mechanism and the applied external wrenches. Said poses are
found and determined so-called discontinuity loci.
2.1. Clearance denition
Before starting with the analysis of pose errors, it is necessary to dene how clearances are taken up. From now on, we will use
by way of example a planar parallel manipulator with two degrees of freedom (DOF) dened by ve links joined using revolute
pairs, commonly known as a ve-bar mechanism. Two revolute pairs are actuated and the other three are passive. It is supposed
that both actuated and passive pairs have clearances.
A passive revolute pair Ji with a clearance is shown in Fig. 1a; the pin can move freely inside the hub, but it is assumed that they
always remain in contact. As shown in Fig. 1b, in the model of an actuated joint Ji there are two kinds of clearances: radial
difference between the pin and the hub, and width difference between the cotter and its case. For simplicity, those different
clearances are separated and treated independently. We will call them radial clearance and angular clearance. The rst one
produces two clearance-freedoms with magnitudes xJi and yJi, and angular clearance due to width differences at the cotter has
one clearance-freedomJi. Note that angular clearance can be also produced by backlash inside the motor or reduction gear.
2.2. Inverse position problem
The rst module solves the inverse position problem in the nominal mechanism. It consists in nding the values of the actuated
joints corresponding to a known end-effector pose. A usual approach is the use of geometrical loop equations. The morphology of
the geometrical equations depends on the mechanism architecture.

580

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

Fig. 1. Clearance joints of the 5R planar mechanism. (a) Passive revolute pair with clearance. (b) Actuated pair with clearance.

In Fig. 2 we have depicted the kinematic model of the 5R planar mechanism with the parameters used in the analysis. In the
pick-and-place application, which is analysed in Section 3, the 5R planar manipulator works upside down, that is, its xed points
A1 and A2 are always over the reference point P. Actuated joints are at A1 and A2, respective angles are 1 and 2. Passive joints are at
B1 and B2, i.e., angles 1 and 2. The end-point is at P, and corresponding output variables are its coordinates x and y. Angle is a
passive parameter.
For given values of the output [xPyP], the solution of the inverse position problem consists in nding the intersection of two
pairs of circles, i.e., one centred at P with radius PBi and the other centred at Ai with radius AiBi. Each one produces two solutions to
the position of Bi, and hence, two different postures of each limb. There are 22 different combinations, and hence four working
modes for the mechanism (Fig. 3). It is common practice to keep the mechanism in one working mode, so we will always use the
working mode shown in Fig. 3a.
Using this approach on the mechanism shown in Fig. 2, where link lengths are all equal to L = 0.5 m and actuators A1 and A2 are
placed d = 0.5 m apart, the workspace of the mechanism is calculated (Fig. 4).
Further to this analysis, Jacobians can be evaluated to nd the singularity locus. The workspace is divided into two parts
depending on the sign of the determinant of the direct Jacobian, i.e., the assembly mode, as shown in Fig. 4. Limits between two
colours are direct singularities. The boundary of the workspace corresponds to inverse singularities found with the inverse
Jacobian.
The use of screw coordinates allows a compact matrix formulation of the velocity equations. Let P be the reference point of the
end-effector, and choose for convenience the link PB2 as the mobile platform. Then, the end-effector twist is dened by:

$P =


1

where is the angular velocity of the mobile platform, and p is the velocity of end-point P.

Fig. 2. Pose parameters of a 2DOF 5R planar parallel manipulator.

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

581

Fig. 3. Working Modes of the 5R planar mechanism.

The screws for the revolute pairs A1, B1 and P of the 5R planar mechanism are:


k

$ A1 = r k ; $ B1 = r k ; $ P =
0
A1
B1

where k is the direction of the revolute pairs in planar mechanisms, rA1 and rB1 are the position vectors of A1 and B1 from point P
respectively.
Applying Eq. (1), the end-effector twist at point P can be stated via links A1B1 and PB1 as:

$P = 1 $ A1 + 1 $ B1 + $ P

where 1 , 1 and are joint rates.

The same twist found through links A2B2 and PB2 can be expressed as:

$P = 2 $ A2 + 2 $ B2

where $ A2 and $ B2 are:

$ A2 = r k ;
A2

$ B2 = r k
B2


5

An independent loop can be drawn for this mechanism, and equating Eqs. (3) and (4) produces the following equation:
1 $ A1 + 1 $ B1 + $ P 2 $ B2 2 $ A2 = 0

Fig. 4. Workspace of a 2DOF 5R planar parallel manipulator.

582

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

The nominal velocity analysis is solved by obtaining 1 , and 2 for given inputs 1 and 2 in:
h

2 3
i 1
h
i

1
6 7
$ B1 $ P $ B2 4 5 = $ A1 $ A2
2

and back-substituting in Eq. (3) or (4) to give the twist of the end-effector.
Acceleration analysis is performed by differentiating Eq. (6) with respect to time. The accelerations calculated can be used
in the next sections to determine inertial loads.
2.3. Joint reaction evaluation
Once the inverse position problem is solved, dynamics of the mechanism can be analysed. Using the free-body diagram
approach and setting out NewtonEuler equations, it is possible to obtain a linear system of equations. Under a given load,
reactions in the joints and actuation forces or torques will be the unknowns of the system.
First, without loss of generality, a link of a spatial mechanism will be analysed. Let us consider the i-th link of a parallel
mechanism. At a point Q i of this link, a known external wrench, wei, is applied:


wei =

f ei
mei + rQ i f ei


8

where fei is the applied force and mei is the applied external couple; rQ i is the position vector of Q i with respect to a certain point,
and then rQ i fei is the moment of fei with respect to the same point.
Reactions in i-th and (i + 1)-th joints are represented by a set of forces and couples (fJi, mJi)T and (fJi + 1, mJi + 1)T. Let P be the
chosen reference point, from which the position vectors of the i-th and (i + 1)-th joints are rJi and rJi + 1 respectively.
Writing NewtonEuler equations for the i-th link in the general 3D case, Eq. (9) is obtained:

f Ji
mJi + rJi f Ji

+

mJi

+ 1

f Ji + 1
+ rJi + 1 f Ji

+

+ 1

f ei
mei + rQi f ei

=


::
mi c i
::
Ii i + i Ii i + rCi mi c i

::
where mi is the mass of link i, c i the acceleration of its centre of mass Ci, Ii its inertia tensor at Ci, and nally i and i are its angular
velocity and acceleration respectively. Note that depending on the joint type, some components of vectors fJi and mJi related to the
permitted relative movements may be null. For example, in a passive spherical joint, mJi is equal to zero.
Since the inverse kinematics have been solved previously, the right side of Eq. (9) is known and actuation loads and joint
reactions are the only unknowns. Hence, NewtonEuler equations can be expressed in a matrix form and unknowns can be simply
calculated:
AwJ = wi we

10

where wJ is the vector of joint reactions and loads withstood by actuators, vector wi contains the inertial loads and matrix A is built
from the NewtonEuler equations. we is a column vector which groups the external wrenches wei at any link of the mechanism, as
shown in Eq. (11).


T
T T
we = we1 ; ; wel

11

where l is the number of links of the mechanism.

It is remarkable that matrix A is a geometrical matrix, i.e., it depends only on the mechanism conguration. Note also that its
singularities are the same of those of the Direct Kinematic Problem. In fact, this matrix can be used for detecting the direct
singularities of a mechanism. By virtue of the duality of kinematics and statics, Hubert and Merlet [20] used the kinematic Jacobian
for detecting closeness to singularities, considering high values in joint reactions as the measure of closeness. Matrix A of Eq. (10)
could be used for the same purpose.
Regarding Eq. (10), for the 5R planar parallel manipulator, the twelve unknowns of vector wJ are two reactions on each passive
revolute joint and two reactions and one actuation torque on each actuated joint.

wJ = f TA1

f TB1

f TP

f TB2

f TA2

MA1

MA2

T

where f Ji is an xy component column vector and MJi is a scalar.

12

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

583

Assembly of A for the 5R planar mechanism leads to the next matrix:

2

1
0
1
0
0
0
0
0
0
0
6
0
1
0
1
0
0
0
0
0
0
6
6 yA1 yP xA1 xP yB1 yP xB1 xP
0
0
0
0
0
0
6
6
0
0
1
0
1
0
0
0
0
0
6
6
0
0
0
1
0
1
0
0
0
0
6
6
0
0
0
0
0
0
0
yB1 yP xB1 xP 0
6
A=6
0
0
0
0
1
0
1
0
0
0
6
6
0
0
0
0
0
1
0
1
0
0
6
6
0
0
0
0
0
0
0
0 yB2 yP xB2 xP
6
6
0
0
0
0
0
0
1
0
1
0
6
4
0
0
0
0
0
0
0
1
0
1
yB2 yP xB2 xP yA2 yP xA2 xP
0
0
0
0
0
0

0
0
1
0
0
0
0
0
0
0
0
0

3
0
07
7
07
7
07
7
07
7
07
7
07
7
07
7
07
7
07
7
05
1

13

Solving Eq. (10), joint reactions f Ji at every joint Ji are obtained, mJi being null for passive revolute joints. In the absence of
friction, as reactions in joints are produced by contact forces between the pin and hub, their directions will indicate how the
known radial clearance, i, is distributed in x and y, i.e., the actual values of xJi, yJi, namely:
2

6 fJ 7
xJi = i 4 i  i5
f Ji 

14

3
f
6 J 7
yJi = i 4 i  j5
f Ji 

15

where i and j are unitary vectors in Cartesian axes.

In Fig. 5 we have depicted how the direction of the joint reactions indicates the contact mode in the revolute joint. The joint
reaction denes the direction that the radial clearance has taken, and therefore the components in the xed frame XY. The angle
indicating the direction of clearance take-up Ji is a very important parameter because it will present sudden changes when the
take-up mode changes. Ji can be found from:
"
#
f Ji i

Ji = arctan

16

f Ji j

2.4. Kinematic accuracy analysis

After having solved the NewtonEuler equations of the preceding section, an accuracy analysis is carried out. In this, as in
Section 2.2, velocity equations are expressed in terms of screw coordinates in order to achieve a compact formulation. These
velocity equations are used to approximate displacements, which are assumed to be innitesimal.
First, let us consider the end-effector twist $P in Eq. (1). The $P of the nominal mechanism can be calculated by adding up
contributions from each kinematic chain, going from the xed frame to the end-effector:
Nj

$P = q ij $ Ji

17

i=1

where Nj is the number of joints of kinematic chain in limb j, and q i and $ Ji are, respectively, the rate and the screw at each joint i of
the said chain.
For a fully parallel manipulator (manipulators whose number of chains is strictly equal to the number of DOFs of the endeffector) with n kinematic chains, the number of independent loops will be equal to (n-1). $P can be calculated by means of the
vector loop equation. Thus, for every closed loop of a mechanism Eq. (18) can be written:
j

Nj

+ 1

Nj

j
j+1 j+1
$ Ji q ij $ Ji = 0
q i

i=1

where j = 1..(n 1).

i=1

18

584

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

Fig. 5. Model of clearance take up in a planar revolute joint.

Terms of Eq. (18) can be separated into those of the passive joints and those of actuated joints. Assembling equations of each
closed loop, denoting the passive joints with index P and the actuated or input joints with I , and reordering the terms, an equation
system can be built in order to solve the nominal velocity analysis:
P

JP q = JI q

19

where JI is a matrix with the screws of the input joints and JP groups the screws of the passive joints, being the latter a
square matrix for fully parallel manipulators. Note that this equation is equivalent to Eq. (7).
From Eq. (19) values at passive joints can be obtained. Matrix JP is conguration dependent and it becomes singular for direct
singularities of the mechanism, where Eq. (19) cannot be solved.
Additional freedoms due to clearances can be added to the mechanism and then to Eq. 19, yielding Eq. (20):

JP q = JI

J C 

I
q
C
q


20
C

where JC is a matrix with the screws from freedoms due to clearances and q are their velocities.
Assuming that displacements are innitesimal, they can be approximated by velocities. If the actuated joints are xed, only
freedoms due to clearance remain at the mechanism. As mentioned before, relative positions between the two parts of each joint
are determined by the reactions calculated in the dynamic analysis (Eqs. (14) and (15). Then the difference between nominal
position and actual position is considered as an innitesimal displacement of each freedom due to clearance. Eq. (21) is then used
instead of Eq. (19), where actuated joints have been eliminated and velocities of freedom due to clearance have been replaced by
innitesimal displacements qC :
P

JP q = JC q

21

Solving Eq. (21) gives the displacements of the passive joints qP due to clearance freedoms. Let us dene vector as the pose
error of the end-effector. By adding it to the nominal pose of a mechanism, the actual pose is obtained. With the calculated qP ,
can be computed with Eq. (17) as follows:

=
p


=

jJ

2
3
i qJj
4
5
$
qCj
j
C

22

where p is the error in position of point P due to clearance take up, is the change in the absolute orientation of the endeffector due to clearances, and qJj and qCj are innitesimal displacements of nominal joints (only passive joints if actuated
joints are xed) and clearance freedoms respectively. Superindex j means that can be calculated using any limb j of the
manipulator.

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

585

Let us apply the described error pose calculating procedure to the 5R planar mechanism studied, the nominal kinematics
having been analysed in Section 2.2. As innitesimal displacements can be approximated by velocities, the relation of Eq. (7) can
be applied to the analysis of the transmission of innitesimal changes of the inputs:
h

2
3

i 1
h
i

1
4
5 = $ $

$B1 $P $B2
A1 A2
2
2

23
a

Note that screws $ A1 and $ A2 which represent actuated pairs can be replaced by $ A1 and $ A2 in order to characterise the angular
clearance described in Section 2.1.
If radial clearances at revolute joints are modelled as innitesimal displacements of the pin in the hub, these can be
decomposed in a xed frame; for example, at joint A1 into xA1 and yA1. Corresponding screws of imperfect joints are shown in
the next equation:





x
x
x
x
x
y
y
y
y
0 y
0
$ A1 = $ A2 = $ B1 = $ B2 = $ P =
; $ A1 = $ A2 = $ B1 = $ B2 = $ P =

24

Adding all clearance freedoms to the loop closure equation, Eq. (6), yields:
y
a
x
1 $ A1 + xA1 $ A1 + yA1 $ A1 +

+ 1 $ B1 +

x
xB1 B1

y
yB1 B1

25
+

x
y
+ $ P + xP $ P + yP $ P
x
y
2 $ B2 xB2 $ B2 yB2 $ B2
a
x
y
2 $ A2 xA2 $ A2 yA2 $ A2 = 0

where 1 and 2 could be used to model an angular clearance of the rotary actuator, and 1, 2 and are innitesimal
changes in the passive angles.
Taking the results of the NewtonEuler equations to Eqs. (14) and (15), values of innitesimal displacements xAi, yAi, xBi,
yBi, xP and yP were obtained. Further, the sense of moments at links AiBi indicate how the angular clearances of actuators, 1
and 2, have to be dened. Then, using Eq. (25) we can solve the innitesimal changes in the passive angles, i.e., 1, 2 and ,
due to the taking up of clearances:
2

2
3
h
i 1
h
a
$ B1 $ P $ B2 4 5 = $ A1
2

$ A2 $ A1

3
1
6 2 7
7
i6
y 6 xA 7
1 7
::: $ B2 6
6 : 7
6
7
4 : 5
yB2

26

Once these are obtained, the end-effector pose error can be found using an expression analogous to Eq. (4) where clearances
are added:
2

=


=

$ A2

3
2
6 2 7
7
i6
6 xA 7
x
y
x
y
2 7

$B2 $ A2 $ A2 $ B2 $ B2 6
6 yA 7
2 7
6
4 xB 5
2
yB2

27

The procedure described calculates the pose error of the mechanism taking into account the inuence of clearance take-up on
passive freedoms. These passive freedoms have been determined from local errors at imperfect joints, which have been obtained
from a dynamic analysis of the nominal mechanism, not the actual one. Consequently, dynamics of the actual mechanism are not
analysed, but since clearance magnitudes are innitesimal, the procedure can be considered an appropriate approach.
Nevertheless, it could be made more complete by means of an iterative procedure. In such an iterative procedure, local errors at
imperfect joints and passive freedom errors could be used in order to update geometrical parameters of the nominal mechanism,
namely, lengths of the links, passive angles and angular variations of actuators. Then, the dynamics of the new nominal mechanism
could be calculated and, with them, local pose errors at imperfect joints and variations on passive freedoms, returning to the
starting point. However, implementation of this type of iterative procedure is not considered in this paper.

586

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

2.5. Discontinuity analysis

As well as the error calculation procedure, it is possible to think of congurations of the manipulatorwith applied external
loadsin which some joint reactions become zero. In such congurations the contact between the two parts of certain joint (or
joints) get lost, leading to a free relative movement between two links. The consequence is an uncertainty on the actual pose of the
end-effector or a discontinuity along a path. In this paper, a deep analysis of such discontinuities is carried out and a way to
calculate locations where they occur is presented in Section 4.

3. Error analysis on a pick-and-place trajectory

When the performance of the manipulator requires high accelerations, inertial loads play an important role. This is the case in
applications such as pick-and-place, ight manipulators or high-speed machining tools [19]. In the present section, a pick-andplace trajectory of the aforementioned 5R planar mechanism is analysed. Such a trajectory is considered for the moving of a heavy
metal object in a blast furnace. In the numerical example, the mass of the links is 1 kg.and of the metal object 10 kg, while link
lengths and distance d between points A1 and A2 are 0.5 m.
In this pick-and-place process, the manipulator picks the object from a certain position, raises it for 0.025 m, displaces it
horizontally for 0.5 m, lowers it for 0.025 m and returns to its starting position. Said path is made smooth by introducing rounded
corners between vertical and horizontal displacements. The object to be moved is picked up by a gripper located at point P. As the
end-point P belongs to links PB1 and PB2, it is assumed that the gripper is located on PB2, which has been previously chosen as the
mobile platform. The trajectory is a round trip with a rst part of the path, denoted as stage 1, in which the load is brought to the
end point; and a return path of the unloaded manipulator to the starting position identied as stage 2.
Such a trajectory is depicted on the XY plane in Fig. 6a and velocities and accelerations produced at point P through the whole
trajectory (complete trip) can be visualised in 6b.
Results of the error calculation procedure described in Section 2 applied to the pick-and-place trajectory are shown in Fig. 7,
where stage 1 is represented in blue and stage 2 in black. Clearance magnitude is exaggerated (10 mm for radial clearance and
0.005 rad for angular clearance) in order to show clearly the effect it produces. Apart from the calculated position error, sudden
changes appear in the actual position. These are due to sign changes in joint reactions, which lead to loss of contact at joints.
As mentioned in the literature review, one kind of relative motion between the two parts of a joint is the continuous contact.
The other two stages, free ight and impact, produce staircase-shaped variations of the mechanism velocity, according to the
multibody approach [16]. In a kinetostatic approach, it is assumed that there is always a contact between the two parts of a joint.
Under this assumption, some authors replace joints with clearance by ideal links, which can only carry a tension load. In [10],
authors showed that sudden changes in ideal link load lead to discontinuities in a four-bar linkage trajectory. Other authors
describe different contact modes between the two parts of a joint. In [21], the switch from tension to compression in the legs of the
Generalised Stewart Platform leads to discontinuities in position and orientation error. In a similar way, changes in the contact
modes have been shown to produce discontinuities in the position error in a 3-UPU robot in [12].
In our work, these staircase-shaped values or sudden changes in contact modes are represented as discontinuities in the actual
trajectory (Fig. 7). A discontinuity represents a loss of contact inside imperfect joints along the trajectory. This causes additional
undesirable effects such as noise and vibrations. In order to avoid these situations, discontinuities and their causes are analysed.

Fig. 6. Pick-and-place trajectory for a 5R mechanism. (a) XY trajectory. (b) Velocity and Acceleration plots.

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

587

Fig. 7. Nominal and actual trajectories.

Taking into account the importance of inertial loads on fast tasks like pick-and-place processes, the inuence of the cycle period
is analysed below. As a starting point, we choose the one in Fig. 7, which is the fastest with a period of 0.76 s. Subsequently, the
period was increased in order to analyse slower cycles and show the inuence of inertial loads on discontinuities.
Fig. 8a shows the positioning error as a function of distance covered in the pick-and-place trajectory for different cycle
periods. The distance covered in the cycle is represented on an abscissa axis and the cycle period on an ordinate axis, while
colour on the vertical axis represents positioning error with respect to the nominal path in the Y direction as shown in Fig. 7.
On Fig. 8b a planar view of the same error is depicted, and it can be seen that the coloured curves become closely packed at
discontinuities. For example, for a cycle with a period of one second, the rst discontinuity appears at a distance of 55 mm
from the starting point and the second one at 145 mm. The gures have two clearly different halves: the left corresponds to
stage 1 (loaded) and the right to stage 2 (unloaded). Notice that a discontinuity is produced between these two stages, when
the transported object is placed.
A deeper analysis of Fig. 8b shows how cycle period affects the location of the discontinuities. For fast cycles with cycle periods
lower than 1.2 s, locations of discontinuities remain almost constant. In stage 1 discontinuities disappear for cycle periods greater
than 1.5 s, while in stage 2 discontinuities remain until cycle periods of 2.2 s. Then, regardless of the unavoidable discontinuity
produced when placing the object, cycles which last more than 2.2 s avoid discontinuities. We conclude that discontinuities have
strong dependence on the inertial loads in high-speed tasks.
In the present section a pick-and-place trajectory has been analysed. Results show that inertial loads have a strong inuence on
the location of the discontinuities of the positioning error. For such an application, the location of the discontinuities in the
workspace depends on the specic trajectory1 performed by the mechanism. For cycle periods longer than a certain threshold,
inertial loads do not have inuence on the location of the discontinuities. Therefore, in applications where dynamics do not play an
important role, inertial loads can be neglected. In the next section, the so-called kinetostatic applications are analysed.
4. Workspace analysis in kinetostatic applications
Kinetostatic applications are those in which inertial loads can be neglected. Among them, the most common are machine tools,
micro-positioning or surgical robots. In this approach, the right side term of NewtonEuler equations (Eq. 9) is equal to zero.
Therefore, there is no dependence on velocities and accelerations, and the entire workspace of a certain working mode can be
analysed.
Applying the approach of Section 2 to the 5R planar mechanism loaded at P with a vertical and negative load of 100 N, the
position error is calculated within the workspace (Fig. 9) for given values of the clearances. Position errors depend on both the
direction of the applied external load and the magnitude of clearances, but the location of the discontinuities is not dependent on
the latter. This allows the discontinuity locus in the workspace to be assessed in spite of the usual uncertainty in the magnitude of
clearances.
As mentioned above, matrix A dened in Eq. (10) detects the direct singularities of the mechanism. Furthermore, the matrix is
poorly conditioned when it is close to a direct singularity. As was the case in the work of Hubert and Merlet [20], that means that in
order to balance external forces, reactions require high values, and then the mechanism could break down. This fact is clearly
shown in Fig. 9. Maximum positive and negative values are located in the same positions as the direct singularities, as can be seen
in Fig. 9a. Error in the x coordinate is depicted also in Fig. 9b, in which, apart from the direct singularities, discontinuities are also
noticeable. The latter form curves that divide the workspace into different areas where the position error is stable. As the external
force is applied in the y direction, reactions in this direction will always have the same sign, so discontinuities appear only in the x
coordinate error.
As far as discontinuities are concerned, they are due to sign changes in joint reactions. With only an external load applied at P,
these changes correspond to positions where links PB1 or PB2 are aligned with the external load. In these situations, reactions are
equal to zero on at least one joint, so the relative position between the pin and the hub is undened. In Fig. 10, one of these

The term trajectory takes into account not only the path (positions) but also the task velocities and accelerations.

588

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

Fig. 8. Cycle time inuence on positioning error. (a) Error in Y. (b) Error in Y.

congurations is shown. A vertical position of link PB2 implies that this link supports the external vertical load f e . Links whose
position is undened are plotted with dashed lines. They do not transmit any load and reactions in joints B1 and A1 are null.
Consequently, these links are not xed and accordingly the mechanism is unstable in the conguration shown in Fig. 10 and in
every conguration where PB1 or PB2 are vertical. The discontinuity is produced when the mechanism crosses these
congurations.
Discontinuities on the workspace divide the error plot into four different surfaces. These surfaces represent a region of the
workspace where there are no discontinuities. Designers may take these regions into account in order to create discontinuity-free
zones, where vibrations produced by clearance joint impacts are avoided. In the so-called discontinuity-free regions, errors due to
clearance can be calculated and corrected by means of a calibration process.
It is worth noting that the pick-and-place trajectory of Section 3 is located in a discontinuity-free region. That is the reason why
discontinuities disappear when inertial loads are negligible. Notice that the vertical load applied at P of the present section is
equivalent to the travelling load of Section 3.
The error plot of Fig. 9 changes if the direction of the external load is modied. In Fig. 11, the position error in P is shown when
the mechanism carries a horizontal load. It can be seen how the location of the discontinuities is different depending on the
applied load. Note also that, for this horizontal load, discontinuities now appear only in the y coordinate error.
Near direct singularities, the poor conditioning of matrix A leads to the methodology of error calculation performing
inadequately, since innitesimal displacements cannot be approximated by velocities. In order to overcome this limitation, an
analytical method for discontinuity detection is developed.

Fig. 9. Errors in the workspace. Load Vertical and Negative. (a) Error in X: plane representation. (b) Error in X: 3-D representation.

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

589

Fig. 10. 5R mechanism with clearance at a discontinuity.

So far, detecting discontinuities has been done by solving the NewtonEuler equations of the mechanism at discrete postures
on a grid of the workspace. Now, an analytical method to nd the discontinuity loci is presented. Since this method deals with
kinetostatic applications, inertial loads are neglected and these loci only depend on the applied external loads, the joints with
clearance and the conguration of the mechanism.
At this point, it should be emphasised that clearance magnitude is usually uncertain. Furthermore, wear phenomena produce
unmeasurable changes in this parameter. Hence, there is always an uncertainty when calculating the position error due to
clearance. Since discontinuity loci depend on the joints with clearance themselves and not on their magnitude, this uncertainty
disappears. Then, calculating the discontinuity loci of a mechanism can be more useful than calculating the positioning error.
From Eq. (10) reactions in the joints of a mechanism can be calculated. Using Cramer's rule to solve this equation system, it is
possible to calculate the component of the reaction in a joint Ji (Eq. 28):

k
wJi

 
 k
Ai 
jA j

28

where index i is referred to the i-th joint and k to the k-th component, with k = 1, , c; and c the number of restrictions imposed by
the joint Ji. Matrix Aki results from replacing the (6(i 1) + k)-th column of matrix A by the known column vector we , since vector
wi is neglected.
As mentioned before, A will be singular at the direct singularities, that is, jAj will be equal to zero. For other points, wkJi will be
equal to zero when the determinant jAki j is null. Solving this determinant symbolically, an analytical expression can be obtained.
This expression is used to describe the geometrical locus where reaction wkJi is equal to zero. When wkJi represents the reaction

Fig. 11. Errors in the workspace. Load Horizontal and Positive. (a) Error in Y: Plane representation. (b) Error in Y: 3-D representation.

590

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

Fig. 12. Locus of poses where Cartesian reactions in joint A1 are equal to zero. Load Vertical and Negative. (a) fAx1.(b) fA1.(c) fA1.

 
 
 
 
moment at A1, MA1, equating Aki  to zero gives the corresponding discontinuity loci of MA1. Specically, the determinant Aki  of
reaction MA1 is written in Eq. (29) supposing the mechanism is loaded at P with a vertical load.


1
0
1
0
0
0
0


0
1
0
1
0
0
0

 yA1 yP xA1 xP yB1 yP xB1 xP
0
0
0


0
0
1
0
1
0
0


0
0
0
1
0
1
0




0
0
0
yB1 yP xB1 xP 0
0


AMA1  = 
0
0
0
0
1
0
1


0
0
0
0
0
1
0


0
0
0
0
0
0 yB2 yP


0
0
0
0
0
0
1


0
0
0
0
0
0
0


0
0
0
0
0
0
yB2 yP

0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
xB2 xP
0
0
1
1
0
xB2 xP yA2 yP

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
fPy
0
0
0
0
1
0
xA2 xP 0


0

0 
0 
0 
0 
0 
0 
0 
0 
0 
0 
1

29
where coordinates xB1, yB1
 , xB2 and yB2 have to be written as functions of coordinates xP and yP. As discontinuity loci are curves
calculated equating AMA1  to zero, the obtained geometrical locus does not depend on the magnitude of the vertical load fPy.
On the other hand, supposing that fJxi and fJyi are reactions in a revolute joint Ji in Cartesian axes, Eq. (30) describes the
geometrical locus where the reaction in Ji becomes null, that is, where the sudden change in the relative position between the pin
and hub occurs.
 x 2
 
A  + Ay 2 = 0
i
i

30

Notice that a sudden change in a joint with clearance leads to a sudden change in the end-effector, i.e., the discontinuity locus of
joint Ji is also the discontinuity locus of the end-effector. In order to determine all end-effector discontinuity loci, it is necessary to
calculate Eq. (30) for each clearance joint in the mechanism. This way, the joint which causes the end-effector's discontinuity is
also identied.
For the mechanism studied with a vertical load applied at point P, results of the locus dened by Eq. (30) are shown in Fig. 12c
for sudden changes in the Cartesian position of the pin of joint A1. Fig. 12a and b depict loci where components of reaction fAx1 and
fAy1 respectively are equal to zero while Fig. 12c shows the condition for the reaction module. As discontinuity loci are calculated by
means of analytical expressions, the curves of Fig. 12 exceed the workspace limits (red colour area). In any case, points outside the
workspace must not be taken into consideration.
In Fig. 13 the locus of sudden changes of 1 and 2 due to the angular clearance at A1 and A2 can be seen. Discontinuities of
both 1 and 2 dene all the end-effector positions where the mechanism will undergo a sudden change. Discontinuity loci of
joints A1, B1 and P are the same as that of 1, while loci of A2 and B2 are the same as that of 2. This is due to the null reactions on the
whole kinematic chain and can be understood by looking at Fig. 10. When the link PB2 is in a vertical posture, the relative position
in joint P is undened because its reaction becomes null. As link PB1 does not transmit any load, the reaction at joint B1 is null and
in the same way the reaction at A1 is also null. Similarly, when link PB1 is vertical, reactions in joints B2 and A2 are equal to zero. As a
consequence, discontinuity loci for a vertical load are all the positions of the workspace where link PB2 or PB1 are vertical.

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

591

Fig. 13. Locus of poses where the reaction moment at joints A1 and A2 is equal to zero. Load Vertical and Negative. (a) MA1. (b) MA2. (c) MA1 and MA2.

Finally, the combination of discontinuity loci of Fig. 13a and b leads to Fig. 13c. It should be emphasised that this shows
discontinuities appearing in the same positions on the workspace as in Fig. 9a.
In the present section, geometrical loci of discontinuities are calculated, both numerically and analytically. These loci leave
discontinuity-free zones where stages of free ight and impact inside imperfect joints are avoided. In the literature, methods to
avoid sudden changes in actual position, such as additional counterweights, springs [22], redundant actuation [23], modication of
link shapes [24] and trajectory planning [25], have been proposed. Designing the mechanism in order to have a desired
discontinuity-free region could be another effective option.
5. Conclusions
In this paper, a numerical procedure for accuracy analysis of parallel manipulators has been presented. Errors in the pose of the
end-effector due to clearance are calculated by means of an approximation of the velocity analysis, while the relative position of
the two parts of the joints are determined by means of a dynamic analysis. It is possible to calculate the dynamics with the
application of external loads to every link. Adding the error pose to the nominal pose, the actual position has been obtained.
The error calculation analysis has been applied to a pick-and-place trajectory of the 5R planar parallel manipulator.
Discontinuities appear on the actual trajectory due to the loss of contact between the pin and the hub of a joint with clearance.
Analysing the trajectory at different cycle periods, it is possible to see how inertial loads affect the location of the discontinuities. It
is also shown that for slow cycles, inertial effects do not have an inuence and kinetostatic analysis is applicable.
A way to nd out the location of discontinuities has been presented, which is valid for applications where dynamics do not play
an important role. In these cases, the error analysis can be widened to the entire workspace for the given mechanism. Now, the
location of discontinuities depends on the mechanism conguration, the imperfect joints and applied external loads. With this
data, a computationally efcient analytical procedure has been developed in order to locate the discontinuities, which determines
the discontinuity loci depending on the applied external loads. Discontinuity loci divide the workspace into discontinuity-free
zones, within which errors can be corrected by calibration.
Acknowledgment
The authors wish to acknowledge the nancial support received from the Spanish Government via the Ministerio de Ciencia e
Innovacin (Project DPI2008-00159), the ERDF of the European Union and the University of the Basque Country (Project GIC07/78).
References
[1] P. Voglewede, I. Ebert-Uphoff, Application of workspace generation techniques to determine the unconstrained motion of parallel manipulators, Journal of
Mechanical Design 126 (2004) 283290.
[2] K.-L. Ting, J. Zhu, D. Watkins, The effects of joint clearances on position and orientation deviation of linkages and manipulators, Mechanism and Machine
Theory 35 (2000) 391401.
[3] J. Zhu, K.-L. Ting, Uncertainty analysis of planar and spatial robots with joint clearances, Mechanism and Machine Theory 35 (2000) 12391256.
[4] Z. Shi, X. Yang, W. Yang, Q. Cheng, Robust synthesis of path generating linkages, Mechanism and Machine Theory 40 (2005) 4554.
[5] C.R. Tischer, A.E. Samuel, Prediction of the slop in general spatial linkages, International Journal of Robotic Research 18 (1999) 845858.
[6] J. Meng, D. Zhang, Z. Li, Accuracy analysis of parallel manipulators with joint clearance, Journal of Mechanical Design 131 (2009)8 011013-19.
[7] S. Venanzi, V. Parenti-Castelli, A new technique for clearance inuence analysis in spatial mechanisms, Journal of Mechanical Design 127 (2005) 446455.
[8] V. Parenti-Castelli, S. Venanzi, Clearance inuence analysis on mechanisms, Mechanism and Machine Theory 40 (2005) 13161329.
[9] C. Innocenti, Kinematic clearance sensitivity analysis of spatial structures with revolute joints, Journal of Mechanical Design 124 (2002) 5257.
[10] M.-J. Tsai, T.-H. Lai, Kinematic sensitivity analysis of linkage with joint clearance based on transmission quality, Mechanism and Machine Theory 39 (2004)
11891206.

592

O. Altuzarra et al. / Mechanism and Machine Theory 46 (2011) 577592

[11] M.-J. Tsai, T.-H. Lai, Accuracy analysis of a multi-loop linkage with joint clearances, Mechanism and Machine Theory 43 (2008) 11411157.
[12] A.-H. Chebbi, Z. Af, L. Romdhane, Prediction of the pose errors produced by joints clearance for a 3-UPU parallel robot, Mechanism and Machine Theory 44
(2009) 17681783.
[13] S.R. Lim, K. Kang, S. Park, W.C. Choi, J.-B. Song, D. Hong, J.K. Shim, Error analysis of a parallel mechanism considering link stiffness and joint clearances, KSME
International Journal 16 (6) (2002) 799809.
[14] F. Farahanchi, S.W. Shaw, Chaotic and periodic dynamics of a slider-crank mechanism with slider clearance, Journal of Sound and Vibration 177 (3) (1994) 307324.
[15] P. Ravn, A continuous analysis method for planar multibody systems with joint clearance, Multibody System Dynamics 2 (1998) 124.
[16] P. Flores, J. Ambrsio, Revolute joints with clearance in multibody systems, Computers & Structures 82 (2004) 13591369.
[17] P. Flores, J. Ambrsio, J.C.P. Claro, H.M. Lankarani, C.S. Koshy, A study on dynamics of mechanical systems including joints with clearance and lubrication,
Mechanism and Machine Theory 41 (2006) 247261.
[18] I. Khemili, L. Romdhane, Dynamic Analysis of a exible slider-crank mechanism with clearance, European Journal of MechanicsA/Solids 27 (2008) 882898.
[19] J.-P. Merlet, Parallel Robots, 2nd edSpringer, 2006.
[20] J. Hubert, J.-P. Merlet, Static of parallel manipulators and closeness to singularity, ASME Journal of Mechanisms and Robotics 1 (2009).
[21] V. Parenti-Castelli, S. Venanzi, On the Joint Clearance Effects in Serial and Parallel Manipulators, Proceedings of the Workshop on Fundamental Issues and
Future Research Directions for Parallel Mechanisms and Manipulators, Quebec, Canada, 2002.
[22] O. Perera, W. Seering, Prevention of Impact in Bearings of Four-bar Linkages, ASME Paper No. 82-DET-134, 1982.
[23] A. Mller, Internal preload control of redundantly actuated parallel manipulatorsits application to backlash avoiding control, IEEE Transactions on Robotics
21 (4) (2005) 668677.
[24] B. Feng, N. Morita, T. Torii, A new optimization method for dynamic design of planar linkage with clearances at jointsoptimizing the mass distribution of
links to reduce the change of joint forces, Journal of Mechanical Design 124 (2002) 6873.
[25] W. Bu, Z. Liu, J. Tan, S. Gao, Detachment avoidance of joint elements of a robotic manipulator with clearances based on trajectory planning, Mechanism and
Machine Theory 45 (2010) 925940.

Oscar Altuzarra received his M. Sc. Mechanical Engineering degree from the Faculty of Engineering of Bilbao, Universidad del Pas
Vasco (UPV/EHU), Spain, in 1995 and the Ph.D. degree in Mechanical Engineering from the same University in 1999. He also attended a
course in Mechanical Engineering at Coventry University, Coventry, UK in 1992. Since 1996, he has been a lecturer and then an
Associate Professor in the Department of Mechanical Engineering at the Faculty of Engineering of Bilbao, Universidad del Pas Vasco
(UPV/EHU). His research interests are kinematics and singularities in parallel kinematic machines, robotics, and computational
applications to complex mechanical problems.

Jokin Aginaga received the M.E. degree in mechanical engineering, in 2004, from the Public University of Navarra, Navarra, Spain. He
is currently in the Ph.D. program of Mechanical Engineering in the Public University of Navarra. Since 2006 he has been an Assistant
Professor in the Public University of Navarra, in the Department of Mechanical Engineering. His research interests include kinematics,
dynamics and accuracy analysis of parallel manipulators.

Alfonso Hernndez received a degree in Mechanical Engineering from the Faculty of Engineering of Bilbao, University of the Basque
Country, Spain, in 1985, and a Ph.D. in Mechanical Engineering from the same University in 1988. He has developed lecture and
research activities at the University of the Basque Country, where he has been a Professor of Mechanical Engineering since 1995. He
has had educational materials published in the elds of Machine Theory, Machine Design and Theory of Vibrations. In addition, for
lecture (and research) purposes, he has designed various computer programmes for the analysis and design of mechanisms. He has
had numerous papers published in refereed journals in the elds of Reliability of the Finite Element Method, Non-linear Structural
Problems, and Mechanisms. Currently he is working on Advanced Mechanical Design and Analysis, and Parallel Robots for industrial
production. He is also a member of several technical and scientic associations and committees. Further details can be found at the
research group webpage http://www.ehu.es/compmech/.

Isidro Zabalza received the B.E. degree in Mechanical Engineering in 1968 and the M.E. degree in Electrical Engineering en 1985 in the
Polytechnic University of Catalunya and the PhD degree in Mechanical Engineering in 1999 in the Public University of Navarra. From
1968 to 1990 worked in mechanical enterprises and since 1990 has been a lecturer and then an Associate Professor in the Department
of Mechanical Engineering in the Public University of Navarra. His research activities include the kinematic and dynamic analysis of
parallel manipulators.