Advances in the Modelling and Control of

Elastic Parallel Manipulators

Krzysztof Stachera
Institute of Technology of Machines
and Automation of Production,
Czestochowa University of Technology, Poland

Frank Schreiber, Walter Schumacher

Institute of Control Engineering,
Technical University of Braunschweig, Germany

Introduction

Modern industrial processes require high performance manipulators, especially regarding the obtainable speed
and precision. Parallel manipulators have the potential to meet some of those goals. As opposed to their omnipresent serial counterparts, the drives of parallel mechanisms are mounted to the base or at least close to the
base allowing for a significant reduction of the moved structural masses and at the same time increasing the obtainable precision of the parallel manipulators. Since the heavy drives do not have to be moved, the links and
joints can consequently be designed lighter. The resulting light-weight manipulators are able to realize higher
accelerations and velocities, but tend to suffer from a reduced mechanical damping which may lead to unwanted
vibrations during and after the execution of motions. Since the following steps of a handling and assembly process require the decay of the end-effector oscillations below a certain level of desired accuracy, these oscillations
increase the necessary duration of each working cycle. To allow for the settling of the end-effector some of the
time gained by performing faster motions is unavoidably lost. Furthermore, the reduced stiffness of the structure
causes static deformations of the robot structure under the influence of gravity, which decrease the manipulators
precision and which have to be compensated by the control strategies.
Among other means, the vibrations in a manipulator structure can be suppressed, using integrated adaptronic components. These, along with the elasticities of the manipulator structure, introduce additional degrees
of freedom (Keimer & Sinapius, 2011). In comparison to rigid structures, these degrees of freedom increase the
mathematical and computational effort for the description and calculation of the manipulators kinematics and
dynamics. In order to use the advantages of the light-weight constructions, efficient algorithms for the derivation
of the manipulator models and their real time calculation for the motion control, as well as the structural control
have to be proposed. The current lack of appropriate methods is caused by the common assumption, that the
parallel manipulators are rigid structures, which is true for most, but not for all of them.
In order to solve the problems mentioned above and to ensure the high performance of the elastic parallel manipulators, computationally efficient algorithms for the calculation of the direct dynamics, based on the

Lagrange-DAlembert formulation, as well as for the derivation of the Jacobian of parallel elastic manipulators
are developed and presented in this chapter. To illustrate the advantages of the suggested approaches on a low
complexity example, the algorithms are demonstrated for the F IVE -BAR-elastic planar parallel manipulator and
evaluated by comparison to conventional techniques. Based upon the presented methods, a model-based robust
control strategy was developed and validated using experimental results obtained from the implementation for the
F IVE -BAR manipulator (Stachera, 2009).

Related works

Serial and parallel robots in general belong to the class of nonlinear systems (Murray et al., 1994). Their behavior
in the time-domain is described by second order nonlinear dynamics equations (Tsai, 1999). Several techniques
from analytical mechanics can be applied to derive the models of the manipulator dynamics (Schwertassek &
Wallrapp, 1999; Jarzebowska & Jarzebowski, 2000). One elegant and efficient method which is very widely
used in the field of parallel manipulators is the Lagrangian method (Beyer, 1928; Nakamura & Ghodoussi, 1989;
Spong & Vidyasagar, 1989). Here a scalar function, called the Lagrangian, is generated, which describes the
entire kinetic, potential and dissipative energy of the system in generalized coordinates. For parallel manipulators,
additional equations describing the closed kinematic loop constraints, must be provided.
For this purpose the Jacobian matrices of the kinematic constraints are used, which are parameterized by
the non-redundant coordinates (Yiu et al., 2001; Kock, 2001; Stachera, 2006b). The approach using the Jacobian
matrix has the significant advantage that for the modelling process, that the parallel manipulator can be divided
into many serial kinematic chains. Consequently, well known methods and techniques, which have already been
applied to serial elastic robots can be used for the description of the manipulators chain dynamics (Beres &
Sasiadek, 1995; Khalil & Gautier, 2000; Robinett et al., 2002).
The Lagrange-DAlembert formulation provides compact equations for the manipulators dynamics, which
are advantageous for system analysis and controller design. A problem occurs when calculating the direct dynamics e.g. for control purposes, since it requires the inversion of the inertia matrix. For matrices of higher order,
which is the case for elastic parallel manipulators, this step requires high computational effort and can thereby
constitute a limitation in the real-time calculation. The division of the dynamics equations into many groups and
their parallel calculation can reduce the computational effort. Approaches, which consider this problem can be
found (Saha, 1997; Wang & Gosselin, 2000; Wang et al., 2002).
As mentioned in Section 1, the reduced stiffness of the robot structure (Krefft & Hesselbach, 2005a;
Krefft & Hesselbach, 2005b) induces deformations and vibrations of the end-effector during and after the robots
movements. The Singular-Perturbation-method allows to separate the motion control of the manipulator from
the vibration suppression control for the structure, since both take place at different time scales (Levine, 1996;
Siciliano & Book, 1988; Hermle, 2001). The use of the robot drives for the purposes of vibration control can be
complicated for parallel elastic manipulators (De Luca et al., 1998; Yang et al., 2000; Moallem et al., 2001). In
contrast to their serial counterparts the drives of parallel manipulators are not always directly connected to the
elastic elements of the structure. Furthermore, since the forces generated by the drives are proportional to the
mass of the structure, they are consequently reduced for light weight parallel manipulators, therefore decreasing
their influence upon the elastic modes of the structure. Those problems can be overcome through the use of
so-called smart materials, which, in combination with the appropriate control algorithms, form the field of
adaptronics. Among those smart materials are piezo-electric actuators, which are manufactured in diverse
shapes (Wierach & Schonecker, 2005), allowing an easy integration into the robot structure. The actuators
placements can be optimized in order to maximize their authority over the elastic modes of the manipulator

(Gabbert et al., 1997; Gabbert et al., 2002; Rose, 2005). The separate control of the elastic structure using piezoelectric patches and of the robot motion using conventional drives has been studied for several serial kinematics
(Hermle, 2001). It was shown, that the control algorithms for the robot motion and for the structures vibrations
can be arbitrarily combined and laid out independently. First research on the separate control strategies has
been carried out by applying patches as well as stacks of piezoactive materials. These works were also able to
demonstrate the effectiveness of the adaptronics in this area (Wilson et al., 2000; Algermissen et al., 2004; Wang
& Mills, 2004b; Breitbach et al., 2005; Algermissen et al., 2008). The nonlinear model-based control concepts
for elastic manipulators have not yet been studied as thoroughly as in the field of serial robots. This provided the
motivation for the research in this area, described in this chapter.

Dynamics of elastic parallel manipulators

The elasticity of a manipulators links and of its joints can be considered as the main sources of structural vibrations. Significant progress was made to make the compliance of the manipulator joints accessible for control (Pavlovic et al., 2011). So, without loss of generality, in the presented approach the links of an elastic parallel
manipulator will be regarded as a main source of vibrations in the robot structure, since they are mainly built as
long and slender elements (Piedboeuf, 2001; ?). The analysis of the dynamics of elastic parallel manipulators will
be started with a brief discussion of the methods for the description of the elastic robot arms. The derivation of
the equations of motion of such a parallel manipulator will be based upon the Lagrange-DAlembert formalism
enhanced by elements to represent the links elasticities (Nakamura & Ghodoussi, 1989; Park et al., 1999). In
addition, a new method for the derivation of the Jacobians of elastic parallel manipulators (Stachera et al., 2007)
is introduced. Afterwards a new method for the simultaneous/distributed calculation of their direct dynamics
(SCDD) will be presented and verified (Stachera & Schumacher, 2008).
3.1

Description of the elasticities in the manipulators structure

The deformation of a continuum can be described by a partial differential equation:

i w i (x,t) + EIi wi (x,t)0000 + i i w i (x,t) + i EIi w 0000
i (x,t) = 0,

(1)

with i (x) = i Ai (x) the mass distribution, i the mass density, A the cross sectional area perpendicular to the
neutral fiber of the beam, EIi (x) the bending stiffness, Ii (x) the geometrical moment of inertia and wi (x,t) the
deformation vector of the ith arm (see Figure 1). The factors i and i have to be determined from experiment (Magnus & Popp, 1997). This type of description is only possible for geometrically simple shapes and only
in cases in which forces are exclusively applied to the end points of the element (Stachera, 2009). The same
limitations hold for the Ritz method with the global trial functions and its approximative solution
Mi~qie (t) + Di~qie (t) + Ki~qie (t) = ~0

(2)

in the form of ordinary differential equations (ODE), with Mi Rnn representing a symmetrical positive definite
mass or inertia matrix, Ki Rnn a symmetrical positive definite or semidefinite stiffness matrix and ~qie (t) Rn1
the vector of the elastic modes. Di = i Mi + i Ki denotes the attenuation matrix according to a specific Rayleigh
ansatz (Magnus & Popp, 1997).
The finite element method (FEM) allows an approximative description of more complicated continua (Czichos, 1989; Gross et al., 2007; Schwertassek & Wallrapp, 1999). The solution obtained with this method is also
formulated using ordinary differential equations (ODE), in a form corresponding to eq. (2). One disadvantage of
this approach is the high order of the obtained matrices, which might make a real time calculation of the equations

r, A
yi

Oi
xi
zi

ci
i

ri , T0

wi

O0

Os

z si+1 x si+1

z ei+1
z0

y si+1

y ei+1
Oe
x ei+1

y0

x0
Figure 1: Deformation of a elastic robot link

of motion unfeasible. One significant possibility to reduce the order of the model is the method of concentrated
parameters which also provides solutions of the form (2) (Benedict & Tesar, 1978; Isermann, 2002; Khalil &
Gautier, 2000). A reduction of the obtainable precision has to be taken into account, though. Depending upon
the precision demands and the goal of the modelling, one of the two at last mentioned methods is applied. The
resulting models obtained by these discretization methods share the same structure, but differ with regard to the
interpretation of the coordinates and the precision of the solution.
3.2

Derivation of the equations of motion

The Lagrange-DAlembert formulation can be used for the derivation of the dynamic equations of elastic parallel
manipulators (Nakamura & Ghodoussi, 1989; Nakamura, 1991; Park et al., 1999; Yiu et al., 2001). The model
forms a system in which the coordinates of the active, passive and elastic degrees of freedom are interdependent
due to the systems constraints. In order to derive the dynamic equations, this system is transformed into a representation in which the independent generalized coordinates are defined in such a way that the motion constraints
are implicitly fulfilled. The algorithm to derive the equations of motion in these coordinates is detailed below.
3.2.1

Inverse Dynamics

The procedure for the derivation of the inverse dynamics corresponds to methods, which are known from serial
manipulators and consists of the following three steps (Nakamura & Ghodoussi, 1989):
1. Transformation of the System: Each closed kinematic loop of the parallel manipulator is separated at a
passive joint, end-effector or link. The result is a reduced system which possesses a tree structure. Consequently, only serial kinematic chains are found in this system. Furthermore it is assumed that all remaining
passive joints possess virtual actuators.
2. Calculation of the Torques: The torques and forces of the real and virtual actuators are computed for
each kinematic chain. These torques and forces cause a movement in every chain, and these movements
correspond to the movement of the original closed-link structure.
3. Transformation of the Torques: The torques and forces of the original parallel manipulators actuators are
calculated from the forces and torques of the tree structure by considering the additional closed kinematic
loop constraints.

It is assumed that the manipulator consists of l closed kinematic loops and possesses a total of e+a+ p = none DOF joints, with e and p of them being discrete elastic DOF ~qe and passive DOF ~q p , respectively. The
coordinates of the active joints ~qa and elastic DOF ~qe form a set of non-redundant coordinates. The controllability
of the manipulator in absence of elasticity is assumed. According to the first step, this system is divided into a
tree structure. The dimensions of ~qa and ~qe remain identical to those of the original structure (na = (n p e) and
ne = e respectively). The number of passive joint coordinates ~q p amounts to n p = (p l). Thus the coordinates
of the tree structure are:
~qt = ~qt (~qa ,~q p ,~qe ),
(3)
where ~qa Rna 1 , ~q p Rn p 1 , ~qe Rne 1 and the dimension of ~qt Rnt 1 , with nt = na + n p + ne . The redundant
coordinates of the passive joints ~q p depend upon the coordinates of the active joints ~qa and the elastic DOF ~qe
~q p = ~q p (~qae ),

(4)

where ~qae Rnae 1 and nae = na + ne . Generally, the relation represented in (4) may not always exist analytically,
but a solution to determine the relationship between the velocities and accelerations of the active and passive
joints can always be found (Merlet, 2000; Stachera, 2005; Yiu et al., 2001). For this purpose the closed kinematic
loop constraints h of the parallel manipulator are introduced:
h(~qt ) = h(~qae ,~q p ) = 0.

(5)

After the differentiation of (5) a Pfaffian form of the constraints is obtained:

h
h
~qae + T ~q p = 0.
~qTae
~q p

(6)

In order to parameterize the configuration space of the manipulator, a Jacobian of the kinematic constraints
parameterized by the non-redundant coordinates has to be derived. For this purpose it will be assumed that the
robot is normally actuated and away from actuator singularity. Therefore, the matrix ~h
qTp from (6) is square and
invertible. The configuration space of the manipulator can be smoothly parameterized by the coordinates of the
active joints and the elastic DOF ~qae :

1 



~q p
h
h
~q p =
~
q
=
~qae .
(7)
ae
~qTp
~qTae
~qTae
Based on (7) the sought Jacobian of the parameterization matrix can be written as follows:
W=

~qt
.
~qTae

(8)

In accordance with the second step, the torques/forces of the tree-structure have to be calculated. They can be
derived from the equations of motion in matrix form:
Mt (~qt )~qt + Ct (~qt ,~qt )~qt +~t (~qt ) + Kt~qt + Dt~qt =~t ,

(9)

where Mt (~qt ), Ct (~qt ,~qt ) Rnt nt are the inertia matrix and the Coriolis matrix of the tree structure. These matrices
satisfy the following structural properties:
1. Mt (~qt ) is a symmetric and positive definite matrix,
t (~qt ) 2Ct (~qt ,~qt ) is a skew-symmetric matrix.
2. M

~t Rnt 1 represents all torques/forces of the real and virtual drives of the reduced system and ~t (~qt ) Rnt 1 is
the vector of the gravitational force projected into joint space. Kt Rnt nt and Dt Rnt nt represent the diagonal
matrices of the discrete elasticities and discrete damping elements in joint space. By using the matrix W from
(8) the third step of this method is performed to transform the equations of the dynamics of the tree structure (9)
into the equations of the closed-link mechanism. In the following, they are expressed only as a function of the
coordinates of the active joints ~qa and the elastic DOF ~qe :
Mc (~qt )~qae + Cc (~qt ,~qt )~qae +~c (~qt ) + Kc~qae + Dc~qae =~c ,

(10)

W T Mt W
Rnae nae ,
+ WT Ct W Rnae nae ,
Cc = WT Mt W
~c =
WT~t
Rnae 1 ,

(11)

with:
Mc =

Kc =

Dc =
~c =

WT K t W
WT D t W
WT~t

Rnae nae ,
R

nae nae
nae 1

(12)
(13)
(14)
(15)
(16)

Proofs of these transformations and their derivations have been published in (Nakamura & Ghodoussi, 1989;
Nakamura, 1991). The redundant coordinates of the passive joints, which are necessary for the computation of
the compact matrices, result from the closed kinematic loop constraints of the parallel manipulator (4) as well as
from their first (7) and second derivatives.
Using this method, the equations of the dynamics of the tree structure (9) are transformed into the compact
equations of the closed-link mechanism (10) and parameterized by the non-redundant coordinates ~qae (11)(16).
Consequently the drive torques can be calculated.
3.3

Derivation of the Jacobian matrix of the parallel manipulator

In conventional methods the Jacobian of the parallel manipulator is derived using the velocity vector-loop method
(Tsai, 1999) or by analyzing of the parallel manipulators statics (Chakarov, 1999; Merlet, 2000; Kock, 2001).
The L-DA formulation allows to systematically convert between the single models of serial kinematic chains and
the model of the compact parallel manipulator (Nakamura, 1991; Park et al., 1999; Yiu et al., 2001). However, it
was not yet shown how the Jacobians of the serial kinematic chains of the tree structure Ji can be transformed into
the Jacobian of the parallel manipulator G. An algorithm has been developed to perform this transformation. For
this purpose, the W matrix, representing the parameterization of the configuration space from (8) and the static
matrix S of the parallel manipulator are used. The static matrix S describes the relationship between the forces in
the arms of the parallel manipulators ~fb Rna 1 and the force ~fx on its end-effector e, shown in the Figure 2.
These forces in the branches ~fb can be calculated using the following relationship:


~fb = ~s1 . . . ~sna 1 ~fx = S1 ~fx .

(17)

If the matrix S is not square then S1 is the pseudo-inverse S+ . The elements of the S-matrix ~si (qai ,~q pi ,~qei )
comprise the vector that relates fbi , the branch force of the ith chain, and ~fxi , the Cartesian force. That relation
can be written:
~fxi =~si fbi .
(18)

fx
e

fb1

fbna
fbi

Figure 2: Distribution of the Cartesian force on the end effector ~fx into the branch forces fbi of the
parallel manipulators arms

and in the matrix form with matrix U this relation takes the form:

~s1 . . . ~0
.
.
~fbx =
.. . . . .. ~fb = U~fb .
~0 . . . ~sna
Now, the Jacobian Jt of the tree-structure is introduced:

J1 . . . 0

.. ,
Jt = ... . . .
.
0 . . . Jna

(19)

(20)

where Ji (~qai ,~q pi ,~qei ) are the na - Jacobians of the serial kinematic chains. In order to eliminate the dependencies
of the coordinates of the passive joint ~q p for the calculation of the Jacobian G of parallel manipulator, the matrix
Jt is parametrized with the matrix W. After this parametrization, the new matrix does not yet represent the
mapping between the joint and Cartesian velocities nor that between the joint torques/forces and end-effector
torques/forces of the parallel manipulator. In order to obtain this mapping, the matrices S and U have to be
introduced. Applying the transformations (17) and (19), the Jacobian of the parallel manipulator can be derived
from the following relation:
G+T = WT JTt US1 .
(21)
This pseudo-inverse Jacobian G+T represents the necessary mapping between the Cartesian force ~fx on the endeffector of parallel manipulator and the forces/torques ~ae Rnae 1 in the manipulators structure in the joint
space:
~ae = G+T ~fx
(22)
The presented method has the significant advantage that the derivation of the serial Jacobians is less complex in
comparison to the direct derivation of the compact Jacobian using standard methods.
3.4

Direct Dynamics

The equations of the direct dynamics are obtained from the compact equations (10) of the manipulators inverse
dynamics:
(23)
~qae = Mc (~qt )1 (~c Cc (~qt ,~qt )~qae ~c (~qt ) Kc~qae Dc~qae ),
This method therefore produces compact equations of the direct dynamics of the elastic parallel manipulator. The
disadvantage is that the computations cannot be executed in parallel. If more parameters need to be considered

for the modelling, e.g. Finite-Element-Method (Wang & Mills, 2004a), the matrices of the compact system may
no longer be feasibly calculated in real time due to their size and complexity.
3.4.1

Simultaneous calculation of the direct dynamics

The compact form of the direct dynamics equations, as mentioned above, can lead to complications in the real
time calculation, especially in control systems with short cycle times. In order to overcome this limitation, a
new method for the simultaneous/distributed calculation of the direct dynamics of the elastic parallel manipulator
- SCDD is suggested (Stachera, 2006a; Stachera et al., 2007; Stachera & Schumacher, 2008). In this method,
the manipulator is handled as a reduced system similar to the approach for the inverse dynamics of the L-DA
formulation. However, in this system the closed kinematic loop constraints of the elastic parallel structure are
represented by forces and torques, just like in the case of the Lagrangian equations of the first type (Miller &
Clavel, 1992; Tsai, 1999).
The equations of the tree-structure have the known form shown in (9). These equations will now be
factored into the equations of motion of the individual serial kinematic chains:
Mti (~qti )~qti + Cti (~qti ,~qti )~qti +~ti (~qti ) + JTi ~fxi + Kti~qti + Dti~qti =~ti ,

(24)



for i = 1 . . . na , where i denotes one kinematic chain with the associated variables ~qtiT = qai ~qTpi ~qTei and torques


~tiT = ai ~Tpi ~Tei of the active ai and passive ~ pi joints and additional structure torques ~ei . ~fxi represents an
external force acting on the end of the ith -branches of the tree-structure. The equations of the direct dynamics of
each such chain can then be formulated:

~qti = Mti (~qti )1 ~ti Cti (~qti ,~qti )~qti ~ti (~qti ) Kti~qti Dti~qti JTi ~fxi ,
(25)
for i = 1 . . . na . Thus, the direct dynamics of each serial kinematic chain can be calculated. The inputs of each
of these equations are the external forces acting upon the end of the particular serial kinematic chain ~fbxi and
the input torques ai and ~ pi . The torques of the elastic DOF ~ei result from the material properties like stiffness
and damping. The input of the tree-structure (9) and of the compact parallel manipulator (10) is the torque/force
vector c . According to the DAlembert principle the performed virtual work for both systems, the reduced and
the original one, has to be equal (Jarzebowska & Jarzebowski, 2000; Nakamura & Ghodoussi, 1989). Moreover,
the working conditions of both systems have to be equal. It implies following conditions (Stachera, 2009):
~a =~c ,
~ p = ~0.

(26)

In order to fulfill these conditions, the new torques/forces of the closed-loop constraints acting on the end of
each branch, must be calculated. Along with the drive torques they are applied to the partial equations of direct
dynamics (25).
The calculation of these torques/forces starts with the computation of the interdependent torques/forces
of the tree-structure resulting from the input torque vector ~c . The relation between them is established in (16).
All the torques/forces of the passive joints of the tree-structure have to be calculated. To perform this task, those
partial matrices and vectors of the equations of the reduced system (24) are used, which are associated with the
virtual torques of the passive joints:
~ pi = Mpij (~qti )~qti + Cpij (~qti ,~qti )~qti +~ pi j (~qti ) + Kpij~qti + Dpij~qti + JTij ~fbxi ,

(27)

for j = 1 . . . n p , i = 1 . . . na . In accordance with the conditions (26) and regarding relation (16) the difference
between acting torques/forces of the tree-structure and the compact parallel manipulator can be calculated:


~q p T
~ p .
~a =~c ~a =
(28)
~qTa
Since the influence of the torques/forces of the elastic DOF (~e ) upon the manipulators movement has already
been taken into account during the calculation of the virtual torques ~ p , only the virtual torques (27) of the
passive joints are used for the calculation of these differences ~a . Now, the new constraints torques/forces can
be calculated:
~fxi = JT [~ai ~ pi ] ,
(29)
i

for i = 1 . . . na . The distribution of this force (29) in the manipulators structure implies the compliance of the
condition (26). Now, the external forces acting on the end-effector of the compact parallel manipulator have to
be distributed between all the separate serial kinematic chains. The relations (17)(19) can be combined:
~fbx = US+ ~fx ,

(30)
i
T
T
T
. These forces have to be complemented with the resulting constraint forces (29)
= ~fx1
. . . ~fxiT . . . ~fbn
with ~fbx
a
and the force takes the end form:
~fxi = ~fxi + ~fxi ,
(31)
h

for i = 1 . . . na . The forces from (31) in combination with the torques/forces from (26) cause the same movement
of the tree-structure as the movement of the original parallel manipulator as well as the same force and torques
distribution in both structures.
3.5

Verification of the new calculation method

In the new method - the Simultaneous Calculation of the Direct Dynamics (SCDD), the system is divided into
a tree structure in order to accelerate the inversion of the inertia matrix. The most frequently used method, the
LU-Gaussian elimination, has the complexity O(n3 ); more sophisticated algorithms can reduce the complexity of
the inversion to O(ne ). For the computation of the direct dynamics in the joint space of large scale systems, it is
therefore better to process several small matrices rather than a single large one. Additionally, the computations
of the direct dynamics with this decomposition can be performed in parallel. In the Table 1 the complexity of the
matrix inversion indicated by the number of the necessary arithmetical operations is calculated for three different
manipulators of the Collaborative Research Center SFB562 (Hesselbach et al., 2005). These calculations were
carried out with the assumption that one elastic DOF nek exists in each kinematic chain.
na

n pk

nek

SCDD

L-DA

Reduction

F IVE -BAR

54

64

16 %

H EXA

384

1728

78 %

T RIGLIDE

162

729

78 %

Table 1: Complexity of the direct dynamics calculation

The results indicate a considerable reduction of the calculation complexity by using the proposed algorithm. Therefore each kinematic chain can be modeled with more parameters and accordingly higher accuracy
(Stachera & Schumacher, 2008).

The new SCDD method has been compared with the L-DA method in the simulation. In order to do this, a
model of the elastic planar parallel manipulator F IVE -BAR was derived (Stachera, 2009). The discrete elasticities
k14 = k24 = 5.464 106 [N/m] were considered in all upper arms of the manipulator, shown in Figure 3. Q11 and
Q21 represent the motors. The other parameters of this model are not relevant for this comparison.

0.8

y in m

0.6

pA

0.4

k 14

k 24

0.2
0

Q 11

0.4

Q 21

0
0.4
x in m

0.8

Figure 3: Trajectory and workspace of elastic planar parallel manipulator - F IVE -BAR

A straight line trajectory between two points pA and pE has been chosen. The models have then been
controlled by torques, which were created by a rigid body model without control. The black line represents the
reference trajectory. The dark gray line is a result of the L-DA model and the light gray line from the SCDD
model. These results show that the models both followed the trajectory with comparable accuracy. The existing
differences between the paths traveled by these two elastic models can be accounted for by the numerical precision
(Stachera & Schumacher, 2008). Figure 4 shows the distance between the endpoints of both kinematic chains.
The light gray line in Figure 4(a) shows the numerical error of the SCDD model and the black line of the L-DA
method. For better investigation, the error of the standard L-DA is additionally presented in Figure 4(b). Even for
this standard method, a small numerical error, which increases slowly with time, is identifiable. The error in the
SCCD method results from the parallel computation and rounding errors and depends upon the sample interval
of the simulation: the smaller the interval, the smaller the error. This error does not constitute any encumbrance
to implement this new method, because in the field of numerical methods algorithms are known that deal with the
stabilization of the numerical calculation and increasing of the computation accuracy (Baumgarte, 1972).

Control of the structures vibration of a parallel manipulator

One of the principal requirements for the structural, as well as the motion control is a constant control quality
all over the workspace. This demand can not be fulfilled using gain-scheduling-control (Wang & Mills, 2004b;
Algermissen et al., 2005; Algermissen & Sinapius, 2011). As the properties of the manipulator are pose dependent, this implies the necessity of the use of model-based control algorithms (Asada & Slotine, 1986; Moallem
et al., 2000). Furthermore the control algorithms have to be robust enough to be able to tolerate uncertainties of
the robot model, which can only be determined with a limited precision. According to the Singular-Perturbationmethod, the control law for vibration suppression of robot structure can be designed separately from the motion
control of the manipulator (Siciliano & Book, 1988; Levine, 1996; Hermle, 2001).
In this chapter a model-based control algorithm is described in Section 4.1 along with the update strategy

10

s in m

x 10

8
6
4
2
0
0

0.05

0.1

0.15

0.2

0.25

0.2

0.25

t in s
((a)) SCDD Method
14

s in m

x 10

8
6
4
2
0
0

0.05

0.1

0.15
t in s

((b)) L-DA Method

Figure 4: Numerical error of the SCDD and LDA method

in Section 4.2 to adjust the controller parameters to the varying nonlinear plant. The approach is validated in
Sects. 4.3 and 4.4 using the experimental results obtained with the F IVE -BAR manipulator.
4.1

Linear robust controller

The choice of the control structure was motivated by methods for passive vibration suppression, which are commonly applied to linear systems (Fischer & Stephan, 1993). These methods use additional spring-mass-dampersystems as dynamic vibration absorbers, which are installed with poorly damped systems and tuned in order to
suppress its natural frequency. In the selected approach the actuator is chosen to behave like a spring resulting in
a controller with a D2 T2 transfer function (Stachera & Schumacher, 2003):
Ga (s) =

L { fa } L {mmod xa }
mmod va s2
=
=
,
L { fs }
L { fs }
mmod s2 + da s + ka

(32)

with da denoting the virtual attenuation factor, ka the virtual stiffness, mmod the effective vibrating mass, xa the
actuator displacement and va the controller gain. The transfer function Ga (s) describes the relationship between
the force measured by the sensor fs (controller input) and the actuator forces caused by the control action fa
(controller output). Therefore, the effect of the controller upon the system mainly equals the product of the
accelerations produced by the actuator xa and the effective vibrating mass mmod . Figure 5 depicts the basic idea
for the controller design.
Each vibration mode possesses its own controller. The controller coefficients for each mode are calculated
with the assumption, that the structure exhibits a very low damping and its natural damping factors can be neglected (dmod 0). The parameters of the virtual spring-mass-damper-system formed by the actuator and each

sensor

fS
xa

actuator

n-D2T2

f12

kVS

f13

1-D2T2
dmod xr

kmod

mmod

fe y
0

f12

x0

f13
Controller

((a)) One-mode-model of the vibration

((b)) Superposition of the modal controllers

Figure 5: Basic ideas for the controller design

one-mode-system have to be chosen in a way that allows the closed-loop control system to return to its resting
point as quickly as possible without oscillating. Therefore the closed-loop system has to show a behavior of a
critically damped second order system. By comparing the coefficients of the closed-loop system with the desired
transfer function the parameters ka and da can be determined:
ka =

4
kmod = mod kmod ,
2 4

(33)

with mod the relation factor between both stiffnesses and the virtual attenuation factor, which has to ensure very
good damping of the system:
p
da = 2 ka mmod .
(34)
The parameters kmod and mmod represent the stiffness and the effective vibrating mass of a single structural mode
projected into the direction of the actuator for a complex systems. Any of those modal controllers is designed
for each of the controlled structural modes of the manipulator and for each of the used actuators. The actuator
response is computed from superposition of the modal controllers associated with him, as shown in Figure 5(b).
Further information regarding the used formula and the proof of robustness and stability are detailed in (Stachera
& Schumacher, 2003; Stachera, 2009). Since most manipulators belong to the family of nonlinear systems, the
frequencies of their respective modes change during motion. Consequently the parameters kmod and mmod have to
be updated according to the end-effector position.
The new D2 T2 -Controller (32) combined with the proposed parameters (33) and (34) was compared in
the simulation with LQ-Controller in order to investigate its performance. The simulations results for both
controllers are presented in Figure 6. The black line shows the behavior of the system under control, the light
gray line the undamped vibration without control, and the dark gray line the behavior of the actuator. The answer
of the controlled closed-loop system shows clearly, that the new controller possesses an efficiency comparable to
the reference control law derived for that specific point and at the same time requires a lower computational effort
than the LQ-Controller. This feature is very important for complex non-linear systems, where the controllers
parameters have to be calculated for a high number of controllers at every time step of the working digital control
system. Additionally, as can be seen by the symmetrical actuator action, the new control algorithm allows a better
usage of the actuators range than the reference control law.

40

30

30

20

f in N

f in N

40

10

10
0
10
0

20

10

10
0

10
15
t in ms

20

10
0

25

((a)) D2 T2 -Controller

10
15
t in ms

20

25

((b)) LQ-Controller

Figure 6: Comparison of the control strategies

4.2

Application of the manipulator characteristics for the adjustment of the controller parameter

A linear controller was proposed for the control of the structural vibrations of an elastic parallel manipulator. In
order to adjust the modal controller to the varying nonlinear plant, the features of the force and inertia ellipsoids
were utilized (Tsai, 1999; Nakamura & Ghodoussi, 1989; Gosselin, 1990; Chakarov, 1999; Ceccarelli & Carbone,
2005).
The force ellipsoids allow the determination of the robot properties with regard to the generation of forces
and the resistance against forces and, therefore, also the calculation of the structural stiffness in an arbitrary
direction in space. The stiffness can be derived as follows:
DIR
kmod = kxy
=

~fxT NTx Nx ~fx

(35)

with Nx = K1
x denoting the compliance matrix (the inverse of the stiffness matrix in the task coordinate frame),
the unit vector k~fx k = 1 describing the spatial direction. In some cases the generation of the fields for every mode
is advisable to be able to analyze them separately.
The inertial ellipsoid describes the ability of the manipulator to generate a unit force in an arbitrary spatial
direction, therefore it determines the inertia of the robot structure into a given direction. The effective vibrating
mass differs from the projected inertia by a factor and can be calculated from:
q
mmod = mDIR
=

~xxT MTx Mx ~xx

(36)
xy
with the unit vector k~xx k = 1 determining the spatial direction and MTx describing the inertia matrix of the manipulator in the task frame (Stachera, 2009). Here again, the analysis of separated modes can require the separate
calculation of each field.
~ x and the inertia matrix M
~ x are pose-dependent and have to be calculated in each
The compliance matrix N
control step using the non-linear model of the manipulators dynamics, as derived in Sec. 3. Thus the control
algorithm consists of the transfer function (32), the parameters of the controller (33) and (34) and the adaption
formulas (35) and (36).

4.3

Verification of the proposed control law

The proposed control algorithm was examined using the elastic planar parallel manipulator F IVE -BAR (see Figure
7(a)). In the robot model each upper arms elasticities are modeled by the lumped parameters k14 and k24 , as
depicted in Figure 7(b). The parameters have the following values: i = 1, 2, li1 = 0.3 [m], mi1 = 0.523 [kg] - stiff
me

O3
m14, k14
m13, k13
m12, k12

l14

l24

m23, k23
m22, k22
l23
l22

l12
m 11

O4

l21
y1

l11

Q11
O1
z1

((a)) Elastic planar parallel manipulator

F IVE -BAR

m24, k24

l13

O2

e
p
a

y 5 m 21
x1

O5
z5

Q21
x5

((b)) Model parameters of the F IVE -BAR

manipulator

Figure 7: Elastic planar parallel manipulator F IVE -BAR

crank , li2 = 0.016 [m], ki2 = 4.22 109 [N/m], mi2 = 0.063 [kg], di2 = 819.0 - piezo sensors, li3 = 0.08 [m], ki3 ,
mi3 = 0.125 [kg] - piezo actuators, li4 = 0.404 [m], ki4 = 5.464 106 [N/m], mi4 = 0.186 [kg], di4 = 10.0 - elastic
robot arms, me = 1.0 [kg] - additional mass at the end-effector, Idrive = 0.089 [kgm2 ] - moment of inertia at the
drives.

Adjustment of
the controllers

Parameters
States

Control of
the structure

Parameters
Parameters

f 12
f 22

Trajectory

x,y
Elastic
Parallel
Manipulator

t11
t21

Control of the
rigid body

f 13
f 23
Figure 8: Control Architecture implemented for the F IVE -BAR elastic planar parallel manipulator

The architecture of the proposed controller, as implemented to control the structural vibrations of the
elastic planar parallel manipulator F IVE -BAR is shown in the Figure 8. 11 and 21 are the torques of the left and
right drive respectively, f12 and f22 represent the sensors forces measured in the left and right upper link. f13 and
f23 are the actuators forces of the piezo-actuators integrated into the upper left and right link. The controllers
parameters are calculated in the block Adjustment of the controllers based on the measured or observed states
of the elastic manipulator at each time step. Consequently, the new parameters are transmitted to the controllers,

which compute the new output of the structural control.

250
f12 in N

125
0
125
250
0

0.05

0.1

0.05

0.1

0.15

0.2

0.25

0.15

0.2

0.25

250
f22 in N

125
0
125
250
0

t in s

Figure 9: Vibration suppression on the elastic parallel manipulator F IVE -BAR

The simulations were performed in the following way: The controllers were activated during a motion
along a straight line trajectory, as shown in Figure 3. Every actuator was driven by a one single-mode controller.
The force measured at the respective piezo sensor served as an input for the controller. Figure 9 depicts the
forces f12 and f22 at the sensors during the motion along the given trajectory with (black) and without (light gray)
structural control. The vibrations of the elastic modes were excited by the stepwise change of the accelerations
caused by the drive torques. Without the use of the controller, distinct vibrations are visible, whereas they are
efficiently attenuated after the activation of the control, during the motion as well as after the completion of the
trajectory. The magnifications in Figure 9 provide further details for two points on the trajectory. The boxes show
magnifications by 1.5 and 2.5, respectively. The effort of the actuators, as well as the control quality, remains
nearly constant in each point of the trajectory. This meets the demands placed upon the control algorithm.
4.4

Experimental results

The presented experiments were carried out for one pose of the manipulator (Q11 = 135 and Q12 = 45 ) and with
an additional payload at the end-effector me = 2 kg. These experiments have shown that the main sources of the
vibrations are located in the passive joints of the manipulator. The dominant oscillations were perpendicular to
the working plane and therefore a 3-axial acceleration sensor attached to the end-effector was used for the control
strategy. The results of the controller action are shown in Figure 10(a) for the piezo-sensors and in Figure 10(b)
for the 3-axial acceleration sensor. The attenuation of the z-components of the oscillations at the end-effector was
7.3 dB.
These experimental results confirm the applicability of the new control algorithm and the chosen design
strategy, although the experiments were performed only for static poses and the performance of the control was
not good enough. Responsible for this poor performance was a phase shift from the signal processing, which
was expected but not considered and modeled in this first implementation of the new control strategy 1 . These
implementation difficulties, e.g. the phase shift due to the digital signal processing, still need to be addressed in
1A

detailed discussion of the experimental and simulative results can be found in (Stachera, 2009)

2.5

2.5

3.5

3
3.5
t in s

((a)) Measurements using

piezosensors

a22 in m/s 2 a12 in m/s 2

80
40
0
40
80
120

az in m/s 2

f12 in N
f22 in N

80
40
0
40
80
120

50
0
50

2.5

3.5

2.5

3.5

2.5

3
3.5
t in s

50
0
50
5.0
0
5.0

((b)) Measurements using a 3-axial

accelerometer

Figure 10: Measurements

future work in order to apply the proposed controller scheme to a wider range of mechanisms.

Conclusions

In this chapter the modelling and control of structural vibrations of elastic parallel manipulators has been discussed. For the modelling of the manipulators dynamics, an enhancement of the Lagrange-DAlembert formulation has been proposed using a new algorithm for the derivation of the Jacobian of parallel manipulators. Based
upon these results, a method has been presented to simultaneously calculate the direct dynamics of parallel and
furthermore elastic parallel manipulators. It allows a significant reduction of the complexity of the calculations,
where this calculation time can be a critical point for the implementation of the systems observers or sophisticated
model-based control strategies. This procedure has been demonstrated for the F IVE -BAR manipulator.
The knowledge of manipulators dynamics has allowed the design of a new model-based control law for
vibration suppression in the manipulators structure. A robust linear controller combined with a pose-dependent
adjustment law has been conceptualized and presented. The performance and potential of the new algorithm has
been confirmed by simulative and experimental examinations. Future work will be directed towards applying the
presented methods to parallel kinematic machines offering more degrees of freedom.

References
Algermissen, S., Keimer, R., Rose, M., Breitbach, E., & Monner, H. P. (2005). Applied robust control for vibration suppression in
parallel robots. In Proc. of 22nd International Symposium on Automation and Robotics in Construction (ISARC) Ferrara, Italy.

Algermissen, S., Rose, M., & Keimer, R. (2004). Angewandte robuste Regelung zur Schwingungsunterdruckung am Parallelroboter
mit adaptronischen Komponenten. In Adaptronic Congress (pp. 17). Hildesheim, Germany.
Algermissen, S., Rose, M., Sinapius, M., & Stachera, K. (2008). Robust gain-scheduling control for parallel robots with smartstructure components. In D. Schuetz, A. Raatz, & F. M. Wahl (Eds.), Proc. of the 3rd International Colloquium of the Collaborative Research Center SFB 562 - Robotic Systems for Handling and Assembly, volume 14 of Fortschritte in der Robotik (pp.
191 206). Braunschweig: Shaker.
Algermissen, S. & Sinapius, M. (2011). Robust gain scheduling for smart-structures in parallel robots. In D. Schutz & F. Wahl
(Eds.), Robotic Systems for Handling and Assembly, volume 67 of Springer Tracts in Advanced Robotics (pp. 159174).
Springer Berlin / Heidelberg.
Asada, H. & Slotine, J. J. E. (1986). Robot analysis and control. John Wiley & Sons, Inc.
Baumgarte, J. (1972). Stabilization of constraints and integrals of motion in dynamical systems. Computer Methods in Applied
Mechanics, 1, 136.
Benedict, C. E. & Tesar, D. (1978). Model formulation of complex mechanisms with multiple inputs: Part II - The dynamic model.
Journal of Mechanical Design, 100, 755 761.
Beres, W. & Sasiadek, J. Z. (1995). Finite element dynamic model of multilink flexible manipulators. Applied Mathematics and
Computer Science, Technical University Press Zielona Gora, Poland, 5(2), 231 262.
Beyer, R. (1928). Dynamik der Mehrkurbelgetriebe. Zeitschrift fur angewandte Mathematik und Mechanik, 8(2), 122 133.
Breitbach, E., Algermissen, S., Keimer, R., Rose, M., & Stachera, K. (2005). Adaptive Tools in Parallel Robotics. In P. Last, C.
Budde, & F. M. Wahl (Eds.), Proc. of the 2nd International Colloquium of the Collaborative Research Center 562 - Robotic
Systems for Handling and Assembly (pp. 203 219). Aachen: Shaker.
Ceccarelli, M. & Carbone, G. (2005). Numerical and experimental analysis of the stiffness performances of parallel manipulators.
In P. Last, C. Budde, & F. M. Wahl (Eds.), Proc. of the 2nd International Colloquium of the Collaborative Research Centre
562 - Robotic Systems for Handling and Assembly (pp. 21 35). Aachen: Shaker Verlag.
Chakarov, D. (1999). Study of the passive compliance of parallel manipulators. Pergamon: Mechanism and Machine Theory, 34,
373389.

Czichos, H., Ed. (1989). HUTTE,

Die Grundlagen der Ingenieurwissenschaften. Springer, 29th edition.
De Luca, A., Panzieri, S., & Ulivi, G. (1998). Stable inversion control for flexible link manipulators. In Proc. of the IEEE
International Conference on Robotics and Automation (ICRA) (pp. 799 805). Leuven, Belgium: IEEE.
Fischer, U. & Stephan, W. (1993). Mechanische Schwingungen. Fachbuchverlag Leipzig - Koln.
Gabbert, U., Koeppe, H., & Nestorovic-Trajkov, T. (2002). Entwurf intelligenter Strukturen unter Einbeziehung der Regelung.
Automatisierungstechnik, Oldenbourg Verlag, 50, 432 438.
Gabbert, U., Schulz, I., & Weber, C.-T. (1997). Actuator placement in smart structures by discrete-continuous optimization. In
Symposium on Mechatronics and Smart materials, Conference on Mechanical Vibration and Noise (pp. 14 17). Sacramento,
CA, USA.
Gosselin, C. (1990). Stiffness mapping for parallel manipulators. 6(3), 377382.
Gross, D., Hauger, W., & Wriggers, P. (2007). Technische Mechanik 4. Springer, 6th edition.
Hermle, M. (2001). Hierarchische Regelung globaler Bewegungen elastischer Roboter, volume 8 of Fortschritt-Berichte. Stuttgart:
VDI.
Hesselbach, J., Bier, C., Pietsch, I., Plitea, N., Buttgenbach, S., Wogersien, A., & Guttler, J. (2005). Passive joint-sensor for parallel
robots. Mechatronics 15, S. 43 65, 2005, 15, 43 65.

Isermann, R. (2002). Mechatronische Systeme: Grundlagen. Springer, 1st edition.

Jarzebowska, E. & Jarzebowski, W. (2000). Mechanika Ogolna. Wydawnictwo Naukowe PWN, Warszawa.
Keimer, R. & Sinapius, M. (2011). Design and implementation of adaptronic robot components. In D. Schutz & F. Wahl (Eds.),
Robotic Systems for Handling and Assembly, volume 67 of Springer Tracts in Advanced Robotics (pp. 413427). Springer
Berlin / Heidelberg.
Khalil, W. & Gautier, M. (2000). Modelling of mechanical system with lumped elasticity. In Proc. of the IEEE International
Conference on Robotics and Automation (ICRA) (pp. 3965 3970). San Francisco, USA: IEEE.
Kock, S. (2001). Parallelroboter mit Antriebredundanz, volume 8 of Fortschritt - Berichte. Dusseldorf: VDI.
Krefft, M. & Hesselbach, J. (2005a). Elastodynamic optimization of parallel kinematics. In Proc. of IEEE International Conference
on Automation Science and Engineering (pp. 357 362). Edmonton, Kanada.
Krefft, M. & Hesselbach, J. (2005b). A way to the optimal design of parallel robots with high dynamic capabilities. In Proc. of
IASTED Int. Conf. on Robotics and Applications - RA (pp. 249 254). Cambrige, USA.
Levine, W. S., Ed. (1996). The Control Handbook. CRC Press and IEEE Press.
Magnus, K. & Popp, K. (1997). Schwingungen. Stuttgart: Teubner Studienbucher, Mechanik.
Merlet, J.-P. (2000). Parallel Robots. Kluwer Academics Publishers.
Miller, K. & Clavel, R. (1992). The Lagrange-based model of Delta-4 robot dynamics. Robotersysteme, Springer Verlag, Germany,
8, 4954.
Moallem, M., Patel, R. V., & Khorasani, K. (2000). Flexible-link robot manipulators: Control techniques and structural design.
Number 257 in Lecture Notes in Control and Information Sciences. London: Springer.
Moallem, M., Patel, R. V., & Khorasani, K. (2001). Nonlinear tip-position tracking control of a flexible-link manipulator: Theory
and experiments. Automatica, 37, 1825 1834.
Murray, R. M., Li, Z., & Sastry, S. S. (1994). A mathematical introduction to robotic manipulation. CRC Press LLC, 1st edition.
Nakamura, Y. (1991). Advanced robotics: redundancy and optimization. Addison-Wesley Publishing Company, USA.
Nakamura, Y. & Ghodoussi, M. (1989). Dynamics computation of closed-link robot mechanisms with nonredundant and redundant
actuators. PIEEEJ R A, 5(3), 294 302.
Park, F. C., Choi, J., & Ploen, S. R. (1999). Symbolic formulation of closed chain dynamics in independent coordinates. Pergamon:
Mechanism and Machine Theory, 34, 731751.
Pavlovic, N., Otremba, R., Inkermann, D., Franke, H.-J., & Vietor, T. (2011). Passive and adaptive joints for parallel robots. In D.
Schutz & F. Wahl (Eds.), Robotic Systems for Handling and Assembly, volume 67 of Springer Tracts in Advanced Robotics
(pp. 429444). Springer Berlin / Heidelberg.
Piedboeuf, J.-C. (2001). Six methods to model a flexible beam rotating in the vertical plane. In Proc. of the IEEE International
Conference on Robotics and Automation (ICRA) (pp. 2832 2839). Seul, Korea: IEEE.
Robinett, R. D., Dohrmann, C., Eisler, G. R., Feddema, J., Parker, G. G., Wilson, D. G., & Stokes, D. (2002). Flexible robot
dynamics and controls. New York: Kluwer Academic/Plenum Publishers: International Federation for System Research IFSR.
Rose, M. (2005). Modal based correction methods for the placement of piezoceramic modules. In Proc. of ASME International
Mechanical Engineering Congress and Exposition Orlando, Florida, USA.
Saha, S. K. (1997). A decomposition of the manipulator inertia matrix. IEEEJ R A, 13(2), 301 304.

Schwertassek, R. & Wallrapp, O. (1999). Dynamik flexibler Mehrkorpersysteme. Braunschweig/Wiesbaden: Vieweg.

Siciliano, B. & Book, W. J. (1988).
IEEEJ R A, 7(4), 79 90.

A singular perturbation approach to control of lightweight flexible manipulators.

Spong, M. W. & Vidyasagar, M. (1989). Robot dynamics and control. John Wiley and Sons, Inc.
Stachera, K. (2005). An approach to direct kinematics of a planar parallel elastic manipulator and analysis for the proper definition
of its workspace. In Proc. of the 11th IEEE International Conference on Methods and Models in Automation and Robotics
(MMAR) Miedzyzdroje, Poland: IEEE.
Stachera, K. (2006a). An approach for the simultaneous calculation of the direct dynamics of parallel manipulators. In Proc.
of the 12th IEEE International Conference on Methods and Models in Automation and Robotics (MMAR) (pp. 751 758).
Miedzyzdroje, Poland: IEEE.
Stachera, K. (2006b). A new method for the direct dynamics calculation of parallel manipulators. In Proc. of the 6th World Congress
on Intelligent Control Automation (WCICA) Dalian, China: IEEE.
Stachera, K. (2009). Konzepte fur elastische, parallele Manipulatoren zur Regelung der Strukturschwingungen. PhD thesis, Institute
of Control Engineering, TU Braunschweig, Germany.
Stachera, K. & Schumacher, W. (2003). Robust vibration control of flexible planar parallel robot. In Proc. of the 9th IEEE International Conference on Methods and Models in Automation and Robotics (MMAR) Miedzyzdroje, Poland: IEEE.
Stachera, K. & Schumacher, W. (2008). Automation and Robotics, chapter Derivation and calculation of the dynamics of elastic
parallel manipulators, (pp. 261 276). I-Tech Education and Publishing, Vienna, Austria, 1th edition. ISBN 978-3-90261341-7.
Stachera, K., Wobbe, F., & Schumacher, W. (2007). Jacobian-based derivation of dynamics equations of elastic parallel manipulators.
In Proc. of the IASTED Asian Conference on Modelling and Simulation (AsiaMS 2007) Beijing, China: IASTED.
Tsai, L.-W. (1999). Robot analysis: the mechanics of serial and parallel manipulators. John Wiley and Sons, Inc.
Wang, J. & Gosselin, C. (2000). Parallel computational algorithms for the simulation of closed-loop robotic systems. In Proc. of
the International Conference on Parallel Computing Applications in Electrical Engineering (PARELEC2000) (pp. 34 38).
Washington, DC, USA: IEEE Computer Society.
Wang, J., Gosselin, C. M., & Cheng, L. (2002). Modeling and simulation of robotic systems with closed kinematic chains using the
virtual spring approach. Kluwer Academic Publishers, Springer Netherlands, Multibody System Dynamics, 7(2), 145 170.
Wang, X. & Mills, J. K. (2004a). A FEM model for active vibration control of flexible linkages. In Proc. of the IEEE International
Conference on Robotics and Automation (ICRA) (pp. 43084313). New Orleans, USA: IEEE.
Wang, X. & Mills, J. K. (2004b). Substructuring dynamic modeling and active vibration control of a smart parallel platform. In
Proc. of 2004 ASME International Mechanical Engineering Congress (IMECE2004) (pp. 1 8). Anaheim: ASME.
Wierach, P. & Schonecker, A. (2005). Bauweisen und Anwendungen von Piezokompositen in der Adaptronik. In Adaptronic
Congress (pp. 1 11). Gottingen.
Wilson, D. G., Starr, G. P., Parker, G. G., & Robinett, R. D. (2000). Robust control design for flexible - link / flexible - joint robots.
In Proc. of the IEEE International Conference on Robotics and Automation (ICRA) (pp. 1496 1500). San Francisco, USA:
IEEE.
Yang, H., Krishnan, H., & Ang, M. H. (2000). Tip-trajectory tracking control of single-link flexible robots by output redefinition.
In Proc. of IEEE Control Theory and Applications, volume 147 (pp. 580 587).: IEEE.
Yiu, Y. K., Cheng, H., Xiong, Z. H., Liu, G. F., & Li, Z. X. (2001). On the dynamics of parallel manipulators. In Proc. of the IEEE
International Conference on Robotics and Automation (ICRA) (pp. 3766 3771). Seul, Korea: IEEE.