ROBOTICS RESEARCH AND TECHNOLOGY

PARALLEL MANIPULATORS
DESIGN, APPLICATIONS
AND DYNAMIC ANALYSIS

No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
 ROBOTICS RESEARCH
AND TECHNOLOGY

Additional books in this series can be found on Nova’s website

under the Series tab.

Additional e-books in this series can be found on Nova’s website

under the e-book tab.
 ROBOTICS RESEARCH AND TECHNOLOGY

PARALLEL MANIPULATORS
DESIGN, APPLICATIONS
AND DYNAMIC ANALYSIS

CECILIA NORTON
EDITOR

New York
Copyright © 2016 by Nova Science Publishers, Inc.

All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted
in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying,
recording or otherwise without the written permission of the Publisher.

We have partnered with Copyright Clearance Center to make it easy for you to obtain permissions to
reuse content from this publication. Simply navigate to this publication’s page on Nova’s website and
locate the “Get Permission” button below the title description. This button is linked directly to the
title’s permission page on copyright.com. Alternatively, you can visit copyright.com and search by
title, ISBN, or ISSN.

For further questions about using the service on copyright.com, please contact:
Copyright Clearance Center
Phone: +1-(978) 750-8400 Fax: +1-(978) 750-4470 E-mail: [email protected].

NOTICE TO THE READER

The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or
implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is
assumed for incidental or consequential damages in connection with or arising out of information
contained in this book. The Publisher shall not be liable for any special, consequential, or exemplary
damages resulting, in whole or in part, from the readers’ use of, or reliance upon, this material. Any
parts of this book based on government reports are so indicated and copyright is claimed for those parts
to the extent applicable to compilations of such works.

Independent verification should be sought for any data, advice or recommendations contained in this
book. In addition, no responsibility is assumed by the publisher for any injury and/or damage to
persons or property arising from any methods, products, instructions, ideas or otherwise contained in
this publication.

This publication is designed to provide accurate and authoritative information with regard to the subject
matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in
rendering legal or any other professional services. If legal or any other expert assistance is required, the
services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS
JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A
COMMITTEE OF PUBLISHERS.

Additional color graphics may be available in the e-book version of this book.

Library of Congress Cataloging-in-Publication Data

ISBN H%RRN

Published by Nova Science Publishers, Inc. † New York

 CONTENTS

Preface vii
Chapter 1 Adaptive Control of Parallel Manipulators: 1
Design and Real-Time Experiments
M. Bennehar, A. Chemori, S. Krut and F. Pierrot

Chapter 2 A Redundant Dynamic Modelling Procedure Based 41

on Extra Sensors for Parallel Robot Control
Asier Zubizarreta, Itziar Cabanes, Marga Marcos,
Charles Pinto, Javier Corral and Pablo Bengoa

Chapter 3 Structure Synthesis of Fully-Isotropic 81

Two-Rotational and Two-Translational Parallel
Robotic Manipulators
Yanbin Zhang

Chapter 4 The Parallel Two-Legged Walking Robot 101

CENTAUROB
Shucen Du, Josef Schlattmann, Stefan Schulz
and Arthur Seibel

Chapter 5 Sliding-Mode Tracking Control of the 6-DOF 143

3-Legged Wide-Open Parallel Robot
Mohammad H. Abedinnasab,
Jaime Gallardo-Alvarado,
Bahram Tarvirdizadeh
and Farzam Farahmand

Index 167
 PREFACE

Robots are a key element in current industrial processes, as they can be

applied to a number of tasks, increasing both quality and productivity.
Traditionally, serial robots have been installed in factories, as their wide
operating space allowed them to fulfill a number of tasks. However, due to
their high moving mass and single kinematic chain structure, these robots
present some disadvantages when high speed, accuracy or heavy load handling
tasks have to be executed. Parallel robots provide an interesting alternative to
these application fields, as their multiple kinematic chain structure offers
increased stiffness, allowing reduced positioning errors, lighter mechanisms
and increased load/weight ratios. In this book, Chapter One addresses a new
control strategy for parallel manipulators based on L1 adaptive control. This
latter is known for its decoupled control and estimation loops, enabling fast
adaptation and guaranteed robustness. Chapter Two focuses on the control of
parallel robots. Chapter Three reviews structure synthesis of fully-isotropic
two-rotational and two-translational parallel robotic manipulators. Chapter
Four reviews the new prototype of the two-legged, parallel kinematic walking
robot CENTAUROB, developed at Hamburg University of Technology.
Chapter Five analyzes and robustly controls the 6-DOF 3-legged Wide-Open
parallel manipulator, using a Lyapunov analysis approach.
In Chapter 1, the authors address a new control strategy for parallel
manipulators based on 1 adaptive control. This latter is known for its
decoupled control and estimation loops, enabling fast adaptation and
guaranteed robustness. To improve the tracking performance of parallel
manipulators, the authors propose in this work to include the dynamic model
of the robot in the control loop of 1 adaptive control. The additional
viii Cecilia Norton

dynamics-based term partially compensates for inherent nonlinear dynamics of

the robot in order to reduce the impact of uncertainties on the closed-loop
system and enhance the overall control performance. Real-time experiments,
conducted on a 4-DOF fast parallel manipulator developed in the laboratory,
demonstrate the effectiveness of the proposed controller and its superiority
compared to standard 1 adaptive control in terms of tracking performance.
As explained in Chapter 2, robots are a key element in current industrial
processes, as they can be applied to a number of tasks, increasing both quality
and productivity. Traditionally, serial robots have been installed in factories,
as their wide operating space allowed them to fulfill a number of tasks.
However, due to their high moving mass and single kinematic chain structure,
these robots present some disadvantages when high speed, accuracy or heavy
load handling tasks have to be executed.
Parallel robots provide an interesting alternative to these application
fields, as their multiple kinematic chain structure offers increased stiffness,
allowing reduced positioning errors, lighter mechanisms and increased
load/weight ratios. These capabilities make these robots suitable for those
tasks where high accelerations and speeds, micrometric precision or heavy
load handling are required.
In Chapter 2, a solution based on the use of extra sensors attached to the
unactuated joints of the parallel robot is detailed. The extra information
provided by these sensors can be used to implement a model-based control
approach that combines the performance of advanced control techniques and
the robustness of sensor redundancy. In order to implement this approach,
three major areas will be analyzed in the next sections: dynamic modelling,
control law definition, and sensitivity, considering the well-known Gough-
Stewart platform as study case.
Strong kinematic coupling is one of inherent properties for general parallel
robotic manipulators. Although it contributes high stiffness and large loading
capability, it also leads to complicated kinematic solutions, difficult
controlling design and path planning, which will take passive effects for the
actual applications of the parallel manipulators. A new systematic method for
structure synthesis of four-degree-of-freedom fully-isotropic parallel robotic
manipulators is proposed in Chapter 3. The moving platform of the mechanism
consists of two-rotational and two-translational (2R2T) motion components
with respect to the fixed base. Firstly, structure synthesis of uncoupled 2R2T
parallel robotic manipulator is set up based on the concept of hybrid
mechanism and the reciprocal screw theory for the first time. Then, type
 Preface ix

synthesis of fully-isotropic 2R2T parallel robotic manipulator is performed by

changing one kinematical chain’s structure of the uncoupled parallel robotic
manipulator obtained before. A lot of novel uncoupled and fully-isotropic
2R2T parallel robotic manipulators are synthesized. Kinematical Jacobian of a
fully-isotropic 2R2T parallel robot is a 4×4 identity matrix, so there is a one-
to-one linear mapping correspondence between the output velocity of the
platform and the input velocity of the actuated joints. Therefore, one
independent output of the moving platform can be controlled only by one
actuator.
In Chapter 4, the authors review the new prototype of the two-legged,
parallel kinematic walking robot CENTAUROB, developed at Hamburg
University of Technology. The main features of this robot are its modular,
symmetric structure, the construction of the legs as Stewart platforms
(hexapods), and the special C-shaped feet that allow a statically stable
movement at any time. The symmetric structure of the robot leads to a center
of gravity exactly in the middle of the hip and prevents from a preferred
moving direction or an asymmetric load of the legs. As a highly flexible
walking device, the CENTAUROB is able to walk in unpaved environment,
avoid and overcome obstacles, and even handle straight and circular stairs.
In Chapter 5, the authors analyze and robustly control the 6-DOF 3-legged
Wide-Open parallel manipulator, using a Lyapunov analysis approach. The
mechanism is evaluated in terms of kinematics, workspace, and dynamics. By
implementing the dynamic model, tracking control of the moving platform is
performed using the nonlinear sliding-mode control method. The numerical
simulations with desired trajectory inputs attest to the excellent performance
and robustness of the designed position tracking sliding-mode controller, in
presence of fast varying uncertain parameters and external disturbances, which
are common in parallel manipulators.
In: Parallel Manipulators ISBN: 978-1-63485-926-4
Editor: Cecilia Norton, pp. 1-39
c 2016 Nova Science Publishers, Inc.

Chapter 1

A DAPTIVE C ONTROL OF PARALLEL

M ANIPULATORS : D ESIGN AND R EAL -T IME
E XPERIMENTS
M. Bennehar∗, A. Chemori†, S. Krut‡ and F. Pierrot§
LIRMM, Montpellier, France

Abstract
In this chapter, we address a new control strategy for parallel manipu-
lators based on L1 adaptive control. This latter is known for its decoupled
control and estimation loops, enabling fast adaptation and guaranteed ro-
bustness. To improve the tracking performance of parallel manipulators,
we propose in this work to include the dynamic model of the robot in the
control loop of L1 adaptive control. The additional dynamics-based term
partially compensates for inherent nonlinear dynamics of the robot in or-
der to reduce the impact of uncertainties on the closed-loop system and
enhance the overall control performance. Real-time experiments, con-
ducted on a 4-DOF fast parallel manipulator developed in our laboratory,
demonstrate the effectiveness of the proposed controller and its superior-
ity compared to standard L1 adaptive control in terms of tracking perfor-
mance.

Keywords: Parallel manipulators, adaptive control, nonlinear system, dynamics

∗
E-mail address: [email protected]
†
E-mail address: [email protected]
‡
E-mail address: [email protected]
§
E-mail address: [email protected]
2 M. Bennehar, A. Chemori, S. Krut et al.

1. Introduction
Adaptive control of parallel manipulators has been a decades-long area of re-
search. It is a very relevant control strategy when it comes to control such
systems due to the abundance of uncertainties and variations of their dynamics
and the environment they interact with. Undoubtedly, using model-based con-
trollers with constant dynamic parameters (e.g. computed torque [1], augmented
PD [2], PD+ [3], etc.) significantly improves the overall performance of paral-
lel manipulators and enhance their tracking capabilities. However, such control
schemes are not endowed with automatic adjustment mechanisms (adaptation)
to cope with variations and uncertainties in the dynamics of the controlled sys-
tem and, hence, do not yield the expected performance. As a result, adaptive
control seems to be a promising solution to account for such limitation by es-
timating and compensating the uncertainties in real-time to provide improved
tracking and enhanced closed-loop performance.
The first attempts to apply adaptive control to mechanical manipulators were
based on restrictive assumptions and hypotheses about their dynamics not nec-
essarily reflecting the real properties of these systems. For instance, the pro-
posed controller in [4], which is based on Model Reference Adaptive Control
(MRAC), does not consider the coupling between the joints. Such hypothesis is
questionable since mechanical manipulators are known for their coupled non-
linear dynamics. To relax this requirement, another MRAC-based controller has
been proposed in [5]. The full dynamics of the manipulator as well as the cou-
pling between the joints were explicitly considered and new adaptive laws were
accordingly derived. Unfortunately for this controller, the number of parame-
ters to be estimated is high, even for simple manipulators (11 parameters for the
provided 3-DOF serial manipulator example).
One major reason of the failure of MRAC-based controllers lies in the fact
that the nonlinear structure of the dynamics of the system is not taken into ac-
count. Considering this weakness, many researchers have investigated alter-
native adaptive control architectures inspired from non-adaptive model-based
controllers. In [6], an adaptive version of the computed torque scheme has been
proposed. A stability analysis based on the Lyapunov theory was conducted to
derive the adaptation law that provided both the tracking of the desired trajec-
tories as well as the convergence of the estimated parameters. Nevertheless, in
this work, the control law showed two main limitations, namely the need of the
measured actual joint accelerations and the boundedness of the inverse of the
 Adaptive Control of Parallel Manipulators 3

estimated inertia matrix. These two limitations were bypassed in [7] by online
estimating the measured accelerations while the former issue has been solved
by following an approach inspired from robust control theory. Similar adaptive
approaches can be found in [8, 9].
Recently, a new control scheme called L1 adaptive control which seeks to
revisit some primary issues in MRAC has been proposed [10, 11]. This control
approach, being itself inspired from MRAC, includes a low-pass filtering tech-
nique that decouples the estimation and control loops. This is of a tremendous
importance since increasing the adaptation gain in conventional adaptive con-
trol may hurt the robustness of the closed-loop system. Indeed, control signals
in MRAC may have undesired high frequencies lying outside the control chan-
nel bandwidth. The low-pass filter in L1 adaptive control, which is designed
by using the L1 small gain theorem, is consequently introduced to remove these
undesired frequencies. The first validations of L1 adaptive control were through
numerical simulations and real-time experiments mainly in aerial vehicles [12].
It was afterwards extended to address more applications such as underwater
vehicles [13], mechanical manipulators [14], underactuated systems [15] and
parallel manipulators [16]. One major advantage of the L1 adaptive control is
its being model-free; i.e. not requiring any knowledge about the dynamics of the
system. However, it would be interesting to include, if available, some knowl-
edge about the dynamics of the system in the control loop in order enhance its
tracking performance and improve the overall closed-loop system.
This chapter focuses on the development of a new control scheme based
on the L1 adaptive control. The nominal nonlinear dynamics of the system are
included in the control loop in order to improve its performance. Specifically,
the proposed controller has a major advantage of improving the tracking perfor-
mance which is crucial in most robotic applications. In order to demonstrate our
claims, real-time experiments are conducted on Veloce; a 4-DOF parallel ma-
nipulator developed in our laboratory intended to be used for high-speed pick-
and-place applications. The remainder of this chapter is organized as follows.
In section 2, we provide a brief state of the art on adaptive controllers proposed
in the literature for parallel manipulators. Section 3 is devoted to the modeling
and description of Veloce robot. In section 4, the control problem formulation
and development of L1 adaptive control are provided. Section 6 is dedicated
to the development of the new proposed control scheme which is based on L1
adaptive control. Real-time experimental results are presented and discussed in
section 7 Finally, conclusions are drawn in section 8
4 M. Bennehar, A. Chemori, S. Krut et al.

2. Overview on Adaptive Control of Parallel

Manipulators
Adaptive controllers are those control schemes requiring a real-time adjustment
(i.e. adaptation) of their parameters in the aim of finding their steady-state val-
ues. The development of adaptive strategies is motivated by the abundance of
uncertainties in the controlled system and its environment that may deteriorate
the control performance. Adaptive controllers can be divided into two categories
depending on their requirement of a dynamic model of the robot; non-model-
based and model-based strategies.
In what follows, we give a breif overview on the most prominent adaptive
control schemes for parallel manipulators that can be found in the literature.

2.1. Non-Model-Based Adaptive Strategies

Strategies belonging to this class do not require the dynamic model of the robot
nor its structure to be known. Instead, they consider all the nonlinearities includ-
ing those of the dynamic model, uncertainties and possible disturbances, whose
structure is unknown, as a general disturbance term to be estimated in real-time.
Then, the control law is designed such that the effect of this disturbances is
minimized.

2.1.1. MRAC-Based Strategies

The main idea of MRAC is to obtain a closed-loop system with adjustable con-
troller parameters. These parameters have the ability of changing the behavior
of the closed-loop system. The time evolution of the adaptive parameters is
adjusted by comparing the output of the controlled system and the desired one
obtained from the reference model. The ultimate goal is to make the behavior
of the controlled system match a desired reference model despite the eventual
variations/uncertainties in the system or its environment [17].
While numerous MRAC-based control schemes for serial manipulators can
be found in the literature [19, 20], only few controllers were proposed for paral-
lel ones. In [18], an adaptive control scheme based on MRAC has been proposed
to control a 6-DOF parallel manipulator based on the Stewart platform prototype
used to emulate space operations. In this work, the control scheme consisted of
a joint space PD feedback controller with adjustable gains. The control scheme
 Adaptive Control of Parallel Manipulators 5

q̇(t)
q(t)

Xd (t) qd (t) − e(t) Γ(t) Parallel

Task-space Inverse + KP (t) +
trajectory + robot
Ẋd (t) kinematics q̇d (t)
generator +
−

ė(t)
KD (t)

MRAC
adaptation
law

Figure 1. Bloc diagram of the proposed MRAC-based control in [18].

is further endowed with an adaptation mechanism that controls the evolution of

the adaptive feedback gains. The joint tracking error of each axis was supposed
to follow a desired user-defined reference second order linear system character-
ized by its natural frequency and its damping ratio. Based on Lyapunov stability
analysis, the adaptation law stabilizing the error dynamics is derived. Exper-
iments were conducted on a Stewart-platform parallel manipulator to evaluate
the performance of the controller and its robustness towards sudden payload
changes. Two case studies were considered; mainly, a vertical motion of the
moving platform and a circular one. In both cases the proposed MRAC-based
strategy with adaptive gains outperforms the PD with constant feedback gains.
Figure 1 illustrates the block diagram summarizing the MRAC-based controller
proposed in [18].

2.1.2. Strategies Based on Artificial Neural Networks

Artificial Neural Networks (ANN) are known by their powerful universal ap-
proximation features [21]. This is why they attracted a big deal of interest in
various fields, and have been applied almost everywhere, from identification to
estimation and control. Artificial neural networks learn from experience rather
than programming, hence, they are typically used in repetitive tasks. Several
works can be found in the literature regarding the application of neural net-
works in control of parallel manipulators [22]. However they remain relatively
few compared to what can be found for serial manipulators [23].
6 M. Bennehar, A. Chemori, S. Krut et al.

Thanks to their learning ability, artificial neural networks are mostly used to
approximate the dynamics of the manipulator. Then, the learned dynamics can
be included in the control scheme to compensate for the uncertainties and dis-
turbances. [24] proposed to augment a decentralized Cartesian space PID con-
troller with an artificial neural network term. The main motivation of this work
is to improve the tracking capabilities of a 2-DOF redundantly actuated parallel
manipulator. The provided simulation results demonstrated the superiority of
the augmented controller with respect to the original PID. The maximum errors
were significantly reduced since the additional neural network term accounted
for the nonlinear dynamics of the manipulator.

2.2. Model-Based Adaptive Strategies

Model-based adaptive controllers explicitly include the dynamics of the robot
in the control loop. In contrast to their non-adaptive counterparts, the adap-
tive schemes adjust the parameters of the model-based loop in order to con-
verge them their best steady-space values. Consequently, if the dynamics of the
manipulator is uncertain or time-varying, adaptive controllers result in a better
closed-loop performance.

2.2.1. Strategies Based on Computed Torque

Computed-torque-based adaptive schemes are those controllers relying on the
classical non-adaptive computed torque controller. The control design starts by
considering the ideal non-adaptive controller that assumes perfect cancellation
of the system nonlinearities. Due to possible uncertainties and variations in the
system dynamics, the compensation of the nonlinearities of the system could
not be as perfect as the non-adaptive controller assumes. Then, an additional
adaptive estimation loop is considered to account for these uncertainties. The
role of the adaptive loop is to estimate, in real-time, the system’s parameters and
the obtained estimation are used in the controller.
The typical example of a computed-torque-based adaptive controller is the
one proposed in [25]. The first step of the control design in this work is to
consider a computed torque control law with uncertain parameters. Then, the
error equation resulting from the application of this control law is obtained.
Based on a thorough stability analysis using the Lyapunov theory, an adaptive
law for the parameters’ estimation is derived. The proposed adaptive controller
in conjunction with the adaptation law guarantee that the tracking errors vanish
 Adaptive Control of Parallel Manipulators 7

and that the estimated parameters converge to their best steady-state values.
The block diagram depicted in Figure 2 clarifies the principle of such control
strategy.
The computed-torque-based adaptive controller, proposed in [25], was
mainly developed for serial manipulators. However, it is known that parallel ma-
nipulators share many properties with their serial counterparts [26, 27]. Based
on this fact, the controller in [25] has been straightforwardly applied to PKMs in
[28]. In this work, an adaptive Cartesian space computed-torque-based scheme
is applied to a 2-DOF redundantly actuated parallel manipulator. Real-time
experiments were carried out in order to highlight the benefits of the adaptive
controller compared to its non-adaptive version. The adaptive controller shows
a net superiority and allows to estimate the system dynamic parameters.
A main drawback of computed torque based adaptive controllers is their de-
pendence on the real acceleration of the robot [25]. This shortcoming is crucial
from a practical point of view since measuring actual accelerations is tedious.

2.2.2. Based on Passivity of the System

Passivity-based controllers use the passivity property of the manipulator [7]. In
contrast with computed-torque-based ones, passivity-based controllers do not
assume perfect cancellation of the system nonlinearities, even in the ideal case
of perfect knowledge of the manipulator’s parameters. Hence, they do not lead
to a closed-loop linear error system. However, based on the passivity property
of the system, they result in a stable closed-loop system.
[8] proposed a passivity-based adaptive controller that holds many advanta-
geous features. This controller can be assimilated to an adaptive version of PD

d2
dt2

Task-space Xd Inverse qd e + Parallel

trajectory + KP + M̂ (q) +
generator kinematics − + + manipulator
q
d ė q, q̇
KD N̂ (q, q̇)
dt

Adaptation
algorithm

Figure 2. Bloc diagram of adaptive computed torque control.

8 M. Bennehar, A. Chemori, S. Krut et al.

control with computed feedforward. Indeed, the proposed control law consists
mainly in a PD feedback loop in addition to an adaptive feedforward term based
on the dynamics of the robot and the desired reference trajectories. The main
advantage of this scheme is that it does not require the joint acceleration mea-
surement. Moreover, the compensation of the system’s nonlinearities is based
on desired quantities that can be computed offline. This means that the con-
trol scheme does not require heavy computations. Furthermore, the use of the
desired values instead of the measured ones, which can be often noisy, may
enhance the robustness of the controller toward measurement noise.
In [29], the developed controller in [8] has been embraced for joint space
control of a 6-DOF parallel manipulator called the Hexaglide. The proposed
adaptive controller as well as a linear PD controller were implemented in real-
time for a comparison purpose. It was not surprising that the adaptive controller
provided better tracking results since it compensates for the nonlinear dynamics
of the manipulator.
In [30], a Cartesian space control strategy called Dual-Space Adaptive Con-
trol is proposed to control redundantly actuated parallel manipulators. The pro-
posed controller in this work share many similarities with the proposed control
strategy in [29]. The main difference is that the weighting gain of the position
error with respect to the velocity error is time-varying. The proposed controller
was experimentally validated on the R4 parallel manipulator [31]. Real-time
experiments showed that the proposed Dual-Space adaptive controller signifi-
cantly improves the tracking performance and allows the tracking of extremely
fast trajectories (up to 100G of maximum acceleration).
Sadegh [32] proposed to extend the joint-space passivity based adaptive
controller of [33], mainly developed for serial manipulators, to Cartesian-space
control of parallel manipulators with actuation redundancy. Moreover, to deal
with friction an additional control term has been added to the conventional feed-
back and adaptive loops. Real-time experiments, conducted on a 2-DOF redun-
dantly actuated parallel manipulator, showed the superiority of the proposed
adaptive controller compared to the augmented PD controller in terms of Carte-
sian tracking errors, at both low and high accelerations. Unfortunately, the evo-
lution of the estimated parameters was not shown in this work to see if they
really converge to their steady-state values.
 Adaptive Control of Parallel Manipulators 9

3. Modeling of Delta-Like Parallel Robots: Application

to Veloce
Veloce is a 4-DOF Delta-like [34] fully actuated parallel manipulator having
four identical kinematic chains [35]. Each kinematic chain can be seen as serial
arrangement of the actuator, a rear-arm and a forearm. Its articulated moving
platform illustrated in Figure 3, which is connected to the fixed-base through the
kinematic chains, can perform three spatial translations and one rotation around
the vertical axis. The rotation is obtained via the relative motion of the upper
and lower parts of the moving platform. The four actuators responsible for the
movement of the mechanical structure are all lying on the same plane. The CAD
of Veloce is illustrated in Figure 4.

Figure 3. Articulated moving platform of Veloce.

3.1. Inverse Kinematics Model

The inverse kinematics problem consists in finding the actuated joint vector
q = [q1 , q2 , q3 , q4 ]T ∈ R4 corresponding to a specific pose X = [x, y, z, s]T of
the moving platform of Veloce. Figures 5 and 6 in addition to Table 1 summarize
the various geometric parameters involved in the establishment of the inverse
kinematic model of Veloce. The actuated joints locations are represented by
points Ai , i = 1, . . . , 4, whose coordinates are given in the fixed Cartesian
frame O − xyz by:

Ai = rb [cos(αi ), sin(αi ), 0]T , (1)

10 M. Bennehar, A. Chemori, S. Krut et al.

Figure 4. CAD view of Veloce 4-DOF parallel manipulator.

where αi = (i − 1)π/2, i = 1, . . . , 4, is the orientation of the ith actuator axis

with respect to the fixed Cartesian frame. The locations of the centers of passive
ball joints are represented by points Bi and Ci which are expressed in the base
frame O − xyz by:

Bi =Ai + L [cos(αi ) cos(qi ), sin(αi ) cos(qi ), sin(qi )] , (2)

Ci = [rp cos(αi ) + x, rp sin(αi ) + y, z] . (3)

Table 1. Geometric parameters of Veloce robot

Parameter Description Value

Li Rear-arm length 200 mm
li Forearm length 530 mm
rb Base radius 135 mm
rp Moving Platform radius 48 mm
 Adaptive Control of Parallel Manipulators 11

To facilitate the subsequent mathematical development, a frame Pi − ui vi z

is attached to each actuator, where the vectors ui and vi are given by:

ui = [cos(αi ), sin(αi ), 0]T , (4)

vi = [− sin(αi ), cos(αi ), 0]T , (5)

where vi is always directed towards the ith actuator’s axis. The inverse kine-
matics solution of Veloce robot can be obtained by finding the intersection
of a circle and a sphere which represents the coordinates of the passive joints
Bi = [xBi , 0, zBi ]T in the corresponding Ai − ui vi z frame [36, 37]. Notice
that in this frame, the Bi points always satisfy yBi = 0.
From the motion of one rear-arm, we can write the equation of a circle in the
corresponding Ai − ui z plane of center Ai = [0, 0] and of radius L as follows:

x2Bi + zB
2
i
= L2 (6)

The movements of one forearm can be described by a sphere of center Ci

u2

A2
v2

y v1
rb

A3
u3
O x A1 u1

v3

v4

A4
u4

Figure 5. Geometric parameters of Veloce robot: top view.

12 M. Bennehar, A. Chemori, S. Krut et al.

z
x Ai ui
O ti qi
rb
Li

Bi
si

li
rp

X Ci

Figure 6. Geometric parameters of Veloce robot: side view.

and radius l. In the Ai − ui vi z frame, we can write the following equation:

(xBi − xCi )2 + yC
2
i
+ (zBi − zCi )2 = l2 (7)
Subtracting (6) from (7) yields:
2xCi xBi + 2zCi zBi = l2 − L2 + x2Ci + yC
2
i
2
+ zC i
(8)
Solving (8) for zBi yields:
Si − 2xCi
zB i = , (9)
2zCi
with Si , l2 − L2 + x2Ci + yC2 + z 2 . Substituting (9) in (6) yields:
i Ci
2
+ x2Ci x2Bi − 4Si xCi xBi + (Si2 − 4zC2
L2 ) = 0


4 zC i i
(10)
Solving for xBi results in:
r  
Si xCi ± (Si xCi )2 − zC2 + x2
i
2 2 2
Ci (Si − 4zCi L )
xBi =   (11)
2 zC2 + x2
i Ci
 Adaptive Control of Parallel Manipulators 13

Equations (11) and (9) result in the coordinates of the intersection of a circle
and a sphere. For geometrical reasons, only the largest xBi is kept. Once the
coordinates of the points Bi are found, the corresponding actuated joint value
can be obtained by:

qi = atan2 (zBi , xBi ) (12)

Applying (12) to each of the four kinematic chains results in the solution of
the inverse kinematic model of Veloce.

3.2. Jacobian Matrix

The Jacobian matrix J(q, X) ∈ R4×4 of Veloce relates its moving plat-
form’s velocity vector Ẋ = [ẋ, ẏ, ż, ṡ] to the actuated joints velocity vector
q̇ = [q̇1 , q̇2 , q̇3 , q̇4 ] such that:

Ẋ = J q̇ (13)

Note that q and X are related through the inverse kinematic model. The Ja-
cobian matrix J can be obtained based on the following kinematic relationship:

kBi Ci k2 − l2 = 0, (14)

which means that the length of each forearm remains constant independently of
the robot’s configuration. Differentiating (14) and rearranging the terms results
in:
Jx Ẋ = Jq q̇, (15)
where Jq and Jx are given as follows:

Jq = diag l1 w1T s1 , . . . , ln wnT sn ,

(16)
T
Jx = [s1 . . . sn ] , (17)

where

w i = ti × v i (18)
Ai Bi
ti = (19)
Li
Bi C i
si = (20)
li
14 M. Bennehar, A. Chemori, S. Krut et al.

Finally, the Jacobian matrix can be written as follows:

l1 w1T s1 li wiT si
 
J= ... (21)
s1 si

3.3. Inverse Dynamic Model

In this section, a brief description of the simplified dynamic model of Veloce
is presented in the sequel. But first, to simplify the motion equations of the
different parts of the mechanical structure of Veloce, the following assumptions,
commonly used in Delta-like robots, are considered [37]
Assumption 1. Both dry and viscous frictions in all passive and active joints
are neglected. This is mainly due to the fact that the joints are carefully designed
such that friction effects are minimized.
Assumption 2. The rotational inertia of the forearms is neglected and their
mass is split up into two equivalent parts, one is added to the mass of the arm
while the other one is considered with the moving platform. This hypothesis is
justified by the small mass of the forearms compared to other components.
The dynamics of the Veloce robot shows a lot of similarities with those
of the Delta robot. Nevertheless few differences arise due to the number of
kinematic chains and the additional rotational DOF of the moving platform.
Regarding the moving platform, we distinguish two kinds of forces acting
on it, the gravity forces Gp ∈ R4 and the inertial force Fp ∈ R4 , they are given
by

Gp = Mp [0 0 − g 0]T (22)
Fp = Mp Ẍ (23)

Table 2. Dynamic parameters of Veloce robot

Parameter Description Value

ma One Rear-arm’s weight 0.541 kg
mf One Forearm’s weight 0.08 kg
mp Platform’s weight 0.999 kg
 Adaptive Control of Parallel Manipulators 15

where Mp ∈ R4×4 is the mass matrix of the moving platform that also considers
the half-masses of the forearms, g = 9.81 m/s2 is the gravity constant and
Ẍ ∈ R4 is the Cartesian acceleration vector.
The contributions of Gp and Fp to each motor can be computed using the
Jacobian matrix J(q, X) ∈ R4×4 as follows

ΓGp = J T Mp [0 0 − g 0]T (24)

T
Γp = J Mp Ẍ (25)

From the joints side, the elements that contribute to the dynamics of the
actuators are the forces and torques resulting from the movement of the rear-
arms in addition to the half-messes of the forearms.
Applying the virtual work principle, which states that the sum of non-inertial
forces is equal to that of the inertial ones, and after rearranging the terms, we
get

J T Mp Ẍ + ΓGp + Ia q̈ + ΓGa = Γ (26)

being Ia ∈ R4×4 a diagonal inertia matrix of the arms accounting for the
rear-arms as well as the half-masses of the forearms, q̈ ∈ R4 the joint acceler-
ation vector and ΓGa ∈ R4 is the force vector resulting from gravity acting on
the arms being given by

ΓGa = ma rGa gcos(q) (27)

with ma the sum of the mass of one rear-arm and one half-mass of a forearm,
rGa the distance between the center of one axis and the center of mass of one
arm and q1 , q2 , q3 and q4 are the joints positions.
Given the kinematic relationship Ẍ = J q̈ + J˙q̇, (26) can be rewritten as
follows
 
Ia + J T Mp J q̈ + J T Mp J˙ q̇ − J T Gp + ΓGa + Γd = Γ,



(28)

where Γd is a disturbance term that accounts for non-modeled dynamics and

possible external disturbances. It can be written in a standard joint-space form
as follows
M (q)q̈ + C(q, q̇)q̇ + G(q) + Γd = Γ, (29)
with:
16 M. Bennehar, A. Chemori, S. Krut et al.

M (q) = Ia + J T Mp J the total mass matrix.

C(q, q̇) = J T Mp J˙ the Coriolis and centrifugal forces matrix.

G(q) = − J T Gp + ΓGa the gravitational forces vector.

The expression of the dynamics in (29) is suitable for joint-space control

since it is expressed in terms of the actuated joints.

4. Background on L1 Adaptive Control

In this section, the basic idea behind the development of the L1 adaptive control
theory is detailed. For the sake of simplicity, we consider a single-input single-
output system. Since L1 adaptive control is inspired from MRAC, we first in-
troduce MRAC for the considered system. Then the architecture of MRAC is
changed to achieve the decoupling of robustness and adaptation.

4.1. Control Problem Formulation

Consider the following single-input single-output linear time-invariant system

ẋ(t) = Ax(t) + bu(t), x(0) = x0

T
(30)
y(t) = c x(t)

where

x(t) ∈ Rn is the measurable state of the system,

u(t) ∈ R is the control input,

b, c ∈ Rn are known constant input and output vectors,

y(t) ∈ R is the output to be regulated by the controller.

The system state matrix A ∈ Rn×n is supposed unknown. The control

objective is to design an adaptive control signal u(t) such that the regulated
output of the system y(t) tracks a given reference signal r(t) ∈ R with desired
performance while guaranteeing that the system state and other signals remain
bounded under the following assumption
 Adaptive Control of Parallel Manipulators 17

Assumption 3. There exist a Hurwitz matrix Am ∈ Rn×n and a vector θ ∈ Rn

of perfect parameters such that the pair (Am , b) is controllable and Am − A =
bθT . Moreover, assume that this unknown perfect parameters vector θ belongs
to a given compact set Θ.
Under the above assumption, the system in (30) can be rewritten as follows

ẋ(t) = Am x(t) + b u(t) − θT x(t) , x(0) = x0

(31)
y(t) = cT x(t)

4.2. From Direct MRAC to Direct MRAC with a State Predictor

One of the key differences between MRAC and L1 adaptive control is the intro-
duction of a prediction-based adaptive architecture. In the following, this new
structure is highlighted and demonstrated that it leads to the same closed-loop
behavior as direct MRAC.

4.2.1. Direct MRAC

Consider the nominal nonadaptive controller given by

unom (t) = θT x(t) + kg r(t) (32)

where kg , 1/ cT A−1


m b is a feedforward gain that ensures zero steady-state
tracking error. Substituting (32) in the system dynamics in (31) leads to the
following desired reference system

ẋm (t) = Am xm (t) + bkg r(t), x(0) = x0

T
(33)
ym (t) = c xm (t)

where xm (t) is the state of the reference system and ym (t) its output. The
nominal control law is impossible to implement due to the uncertainties in the
system state matrix A (and, hence, θ) which should be estimated online.
The feasible direct MRAC control law is hence given by

u(t) = θ̂T x(t) + kg r(t) (34)

where θ̂ ∈ Rn is an estimate of θ whose evolution is governed by the following

adaptation rule
˙
θ̂(t) = γx(t)eT (t)P b, θ̂(0) = kx0 (35)
18 M. Bennehar, A. Chemori, S. Krut et al.

ẋm (t) = Am xm (t) + bkg r(t) xm (t)

ym (t) = cT xm (t)

ẋ(t) = Am x(t) + b u(t) − θ T x(t)

r(t) u(t) x(t) +
u(t) = θ̂T x(t) + kg r(t) T −
y(t) = c x(t)

θ̂(t) ˙ e(t)
θ̂(t) = γx(t)eT (t)P b

Figure 7. Block diagram of MRAC.

where γ ∈ R+ is the adaptation gain, e(t) , xm (t) − x(t) and P = P T > 0 is

the solution for the algebraic Lyapunov equation ATm P + P Am = −Q, for an
arbitrary choice of Q = QT > 0.
Substituting the adaptive control law (34) in the system dynamics in (31)
leads to the following error dynamics

ė(t) = Am e(t) − bθ̃T (t)x(t) (36)

where θ̃ , θ̂(t) − θ is the parameters estimation error. Figure 7 depicts the

overall block diagram of MRAC.

4.2.2. MRAC with a State Predictor

Now consider the following state predictor that mimics the behavior of the sys-
tem in (31) with the unknown parameter θ replaced by its estimate θ̂(t)

˙
x̂(t) = Am x̂(t) + b(u(t) − θ̂x(t)), x̂(0) = x0
T
(37)
ŷ(t) = c x̂(t)

where x̂(t) being the state of the predictor. The prediction error dynamics can
be obtained by subtracting the system dynamics from those of the predictor as
follows
˙
x̃(t) = Am x̃(t) − bθ̃(t)x(t) (38)
where x̃(t) , x̂(t)−x(t) is the prediction error. It can be seen that the prediction
error dynamics is identical to the obtained error dynamics with direct MRAC in
 Adaptive Control of Parallel Manipulators 19

(36). The adaptive law of the predictor-based controller is similar to that of

direct MRAC, the only difference is the use of the prediction error x̃(t) instead
of the tracking error e(t). It is thus given by
˙
θ̂(t) = γx(t)x̃T (t)P b, θ̂(0) = θ0 (39)

It can be noticed that the closed-loop state predictor mimics the behavior
of the reference model in (33). Indeed, substituting the control law (34) in the
predictor dynamics and upon the use of (39) we get the following closed-loop
dynamics
˙
x̂(t) = Am x̂(t) + bkg r(t), x̂(0) = x0,
(40)
ŷ(t) = cT x̂(t)
which is identical to the dynamics of the reference system of direct MRAC
in (33). It is demonstrated in [38] that the tracking (or prediction) errors are
upper bounded at any time by

kθ̃(0)k
ke(t)k(= kx̃(t)k) ≤ p (41)
λmin (P )γ

being θ̃(0) the initial parameters estimation error and λm in(P ) the minimum
eigenvalue of P . This means that the tracking error can be made arbitrarily
small by increasing the adaptation gain γ. However, it can be seen from the
control law in (34) and the adaptation laws in (35) and (39) that increasing
the adaptation gain lead to high gain feedback. The block diagram of MRAC
scheme with a state predictor is shown in Figure 8.

4.3. L1 Adaptive Control

In a real system, the unknown parameters vector θ may be a time-varying uncer-
tainty with frequencies that may lie outside the control channel bandwidth [38].
In this case, such frequencies are naturally attenuated due to hardware limita-
tions of the actuators. Consequently, the behavior of the closed-loop system will
be different than the dynamics of the reference model in (33).
The control problem in L1 adaptive control is formulated with the under-
standing that some uncertainties could never be perfectly compensated. Indeed,
while the control objective in MRAC is only stated asymptotically, the closed-
loop performance in L1 adaptive control is specified ∀t ≥ 0. This means that at
20 M. Bennehar, A. Chemori, S. Krut et al.

˙
x̂(t) = Am x̂(t) + b(u(t) − θ̂x(t)) x̂(t)
ŷ(t) = cT x̂(t)

ẋ(t) = Am x(t) + b u(t) − θ T x(t)

r(t) x(t) +
u(t) = θ̂T x(t) + kg r(t) −
u(t) T
y(t) = c x(t)

θ̂(t) x̃(t)
˙
θ̂(t) = γx(t)eT (t)P b

Figure 8. Block diagram of the control loop based on MRAC with state predic-
tor.

any time, the performance of the system can be predicted. Moreover, the control
signal never exceeds the available control channel bandwidth.
For the class of systems given by (31), consider the following state predictor
that mimics its behavior
˙
x̂(t) = Am x̂(t) + b(u(t) − θ̂x(t)), x̂(0) = x0
(42)
ŷ(t) = cT x̂(t)
In addition to the state predictor in (42), consider a projection-type adaption
law for the estimated parameter vector θ̂ expressed as
˙
 
θ̂(t) = γ Proj θ̂(t), x(t)x̃(t)P b , θ̂(0) = θ0 (43)

which is adjusted using the prediction error x̃(t) = x̂(t) − x(t). The projection
operator Proj avoids the parameters drift and ensures that they remain inside the
compact set Θ. In (43), γ is the positive adaptation gain and P is the symmetric
positive definite matrix, solution of the algebraic Lyapunov equation ATm P +
P Am = −Q for an arbitrary symmetric positive definite matrix Q.
The last stage which is one of the notable unique features of the L1 adaptive
control, is the control input characterized by the introduction of a low-pass filter.
It is given by its Laplace form as follows
u(s) = C(s) (η̂(s) + kg r(s)) (44)
where C(s) is a bounded-input bounded-output strictly proper transfer function
with C(0) = 1, r(s) is the Laplace transform of the reference trajectory r(t)
ˆ = θ̂T (t)x(t).
and η̂(s) is the Laplace transform of η(t)
 Adaptive Control of Parallel Manipulators 21

ẋm (t) = Am xm (t) + bkg r(t) xm (t)

T
ym (t) = c xm (t)

r(t) ẋ(t) = Am x(t) + b u(t) − θ T x(t)

x(t) +
u(t) = θ̂T x(t) + kg r(t) C(s) T −
u(t) y(t) = c x(t)

θ̂(t) ˙ e(t)
θ̂(t) = γ Proj(x(t)eT (t)P b)

Figure 9. Block diagram of L1 adaptive control.

It is demonstrated in [38] that the controlled system in (30) under the L1

adaptive control law in (44) is bounded-input bounded-state with respect to r(t)
and x0 if the following L1 -norm-based equality is satisfied

kG(s)kL1 L < 1 (45)

where

G(s) , H(s)(1 − C(s)), H(s) , (sI − Am )−1 b, L , max kθk1 (46)

θ∈Θ

The block diagram of the L1 adaptive control strategy for the class of sys-
tems given by (31) is displayed in Figure 9.

5. Application of L1 Adaptive Control to Parallel

Manipulators
Recall the inverse dynamics for parallel mechanical manipulators which are ex-
pressed as
M (q)q̈ + C(q, q̇)q̇ + G(q) + Γd (t) = Γ(t) (47)
In the aim of developing a L1 adaptive controller for the class of systems
given by (47), introduce the following combined position-velocity tracking error
given by
r = (q̇ − q̇d ) + Λ (q − qd ) (48)
where Λ ∈ R4×4 is a symmetric positive-definite weighting matrix.
22 M. Bennehar, A. Chemori, S. Krut et al.

5.1. Control Law

Consider the following control input vector Γ(t) consisting of a combination of
two distinct terms [39]

Γ(t) = Γm (t) + Γad (t), Γm (t) , Am r(t) (49)

where Γm ∈ Rn is a state-feedback term characterizing the desired transient

response of the tracking error r(t) and Γad is the adaptive control term which
will be designed subsequently. Taking the first time derivative of (48) with
respect to time we get

ṙ(t) = (q̈ − q̈d ) + Λ (q̇ − q̇d ) (50)

Substituting the control law (49) into the dynamics of the parallel manip-
ulator in (47) and solving for q̈ along with substituting into (50), we get the
following error dynamics

ṙ(t) = Am r(t) + Γad (t) − η(t, ζ(t)), r(0) = r0 (51)

where ζ = [r, q]T and η(t, ζ(t)) is a nonlinear function that gathers all the
nonlinearities of the system including possible external disturbances. Before
proceeding to the development of the adaptive control law, some important as-
sumptions and properties regarding the unknown function η(t, ζ(t)) have to be
made, they can be summarized as follows

Assumption 4. (Uniform boundedness of η(t, 0)) There exist B > 0 such that
∀t ≥ 0, kη(t, 0)k≤ B.

Assumption 5. (Semiglobal uniform boundedness of partial derivatives of

η(t, ζ(t))) The unknown nonlinear function η(t, ζ(t)) is continuous with re-
spect to its arguments, and for arbitrary δ > 0, there exist dηt (δ) and dηx (δ)
such that
∂η(t, ζ)
≤ dηt (δ), ∂η(t, ζ) ≤ dη (δ)


∂t ∂ζ ζ
(52)
∞ ∞

Assumption 6. for t ≥ 0, krt kL∞ ≤ ρ and kṙT kL∞ ≤ dr , for some positive
constants ρ and dr .
 Adaptive Control of Parallel Manipulators 23

It follows from Lemma A.9.2 in [38] that the unknown nonlinear function
η(t, ζ(t)) can be parametrized as follows

η(t, ζ(t)) = θ(t)krt kL∞ + σ(t) (53)

where θ(t), σ(t) ∈ Rn are continuous, piecewise-differentiable and uniformly

bounded unknown functions. Therefore, the error dynamics in (51) can be
rewritten as:

ṙ(t) = Am r(t) + Γad (t) − (θ(t)krt kL∞ + σ(t)) , r(0) = r0 . (54)

Notice that the substitution of Γad (t) = (θ(t)krt kL∞ + σ(t)) in the er-
ror dynamics in (54) results in the desired performance characterized by
ṙ(t) = Am r(t). However, since the nonlinear function η(t, ζ(t)) =
(θ(t)krt kL∞ + σ(t)) is unknown due to uncertainties and unmeasured distur-
bances, the control term Γad (t) should be designed in a way that it estimates the
unknown functions θ(t) and σ(t) in real-time.

5.2. State Predictor

To predict the behavior of the combined tracking error r(t) and according to
the L1 adaptive control theory, a state predictor that mimics the behavior of the
error dynamics in (54) is formulated by [39]:
 
˙ = Am r̂(t) + Γad (t) − θ̂(t)krt kL + σ̂(t) − Z r̃(t), r̂(0) = r0 (55)
r̂(t) ∞

where r̂(t) is the state of the predictor, r̃(t) , r̂(t) − r(t) is the prediction
error and θ̂(t), σ̂(t) are estimates of θ(t) and σ(t), respectively. Notice the
introduction of the additional term Z r̃(t) responsible of rejecting the estimation
error.

5.3. Adaptation Laws

The evolution of the adaptive estimates θ̂(t), σ̂(t), required for the L1 adaptive
control architecture, is governed by the following projection-type adaptation
laws
˙
 
θ̂(t) = Ξ Proj θ̂(t), P r̃(t)krt kL∞ , θ̂(0) = θ̂0 (56)
˙
σ̂(t) = Ξ Proj (σ̂(t), P r̃(t)) , σ̂(0) = σ̂0 (57)
24 M. Bennehar, A. Chemori, S. Krut et al.

˙ r̂(t)
Am r̂(t) = Am r̂(t) + Γad (t) − η̂(t) − K r̃(t)

rd = 0   + r(t) +
η̂(t) = θ̂(t)krt kL∞ + σ̂(t) C(s) + ṙ(t) = Am r(t) + Γad (t) − η(t, ζ(t)) −
Γad (t)

˙
 
θ̂(t), σ̂(t) θ̂(t) = Ξ Proj θ̂(t), P r̃(t)krt kL∞ r̃(t)

˙
σ̂(t) = Ξ Proj (σ̂(t), P r̃(t))

Figure 10. Block diagram of L1 adaptive control for parallel manipulators.

where Ξ > 0 is the adaptive gain, P = P T > 0 is the solution to the algebraic
Lyapunov equation ATm P + P Am = −Q for some arbitrary matrix Q = QT >
0.
Finally, the adaptive control term Γad (t) in (49) is given by its Laplace form
as follows
Γad (s) = C(s)η̂(s) (58)
 
where η̂(s) is the Laplace transform of η̂(t) = θ̂(t)krt kL∞ + σ̂(t) and C(s)
is a diagonal matrix whose elements are bounded-input bounded-output stable
strictly proper transfer functions satisfying C(0) = In and zero initialization for
their state-space realizations.
To sum up, the block diagram shown in Figure 10 illustrates the architecture
of L1 adaptive control for the class of systems given by (47).

6. Extended L1 Adaptive Control of Parallel

Manipulators
Let qd (t), q̇d (t), q̈d (t) ∈ R4 be the desired joint position, velocity and accel-
eration trajectories respectively. These reference trajectories to be tracked by
the active joints of Veloce, are generated in task-space according to a specific
task of the moving platform (e.g. pick-and-place). Then, adequate kinematic
relationships are used to compute the corresponding joint quantities. The con-
trol objective is to ensure that the traveling plate of the manipulator tracks, as
accurately as possible, the desired trajectories regardless of any inherent non-
linearities or external disturbances. To quantify the control objective, consider
 Adaptive Control of Parallel Manipulators 25

the combined joint tracking error r(t) ∈ R4 defined by

r(t) = (q̇ − q̇d ) + Λ (q − qd ) (59)

where Λ ∈ R4×4 is a symmetric positive-definite design matrix. Since −Λ is

Hurwitz, it follows that (59) is BIBO-stable; i.e. there exist L1 , L2 > 0 such
that [40]
kqτ kL∞ ≤ L1 krτ kL∞ + L2 (60)
where k(.)τ kL∞ denotes the truncated L∞ -norm of (.). The proposed control
law to achieve the desired tracking performance is given by:

Γ = M0 (qd )q̈d + C0 (qd , q̇d )q̇d + G0 (qd ) + Am r + Γad , (61)

The proposed control law in (61) can be thought of as the combination of

three distinct terms. Each term has a specific role in the closed-loop system:

• The first term (i.e. M0 (qd )q̈d + C0 (qd , q̇d )q̇d + G0 (qd )) is a nominal feed-
forward inverse dynamics term intended to minimize the effect of the
inherent nonlinearities of the system and hence, to improve the tracking
performance.

• The second term (i.e. Am r) consists of a stabilizing state-feedback term

specifying the desired closed-loop behavior of the system.

• More importantly, the last term Γad , which will be detailed in the subse-
quent development, is the adaptive signal responsible of rejecting all the
remaining disturbances such as friction effects, external disturbances, etc.

Substituting the control law (61) into (29) results in the following error dynam-
ics
ṙ(t) = Am r(t) + Γad (t) − η(t, ζ(t)), r(0) = r0 (62)
 T T T T
where η(t, ζ(t)), ζ = r , q , q̇ is a nonlinear function that gathers all the
remaining nonlinearities that result from applying the control law (61) to (29)
and is given by

η(t, ζ(t)) = M −1 (q)(M̃ (q)q̈d + Ñ (q, q̇))

+ (I − M −1 (q))(Am r + ΓAD ) (63)
− Λr + Λ2 (q − qd )
26 M. Bennehar, A. Chemori, S. Krut et al.

where M̃ (q) , M (q) − M̂ (qd ), Ñ (q, q̇) , N (q, q̇) − N̂ (qd , q̇d ) and I ∈ R4×4
denotes the identity matrix. For the subsequent parametrization of η(t, ζ(t)),
let’s consider the following non-restricting assumptions [41]

Assumption 7. There exist positive B such that kη(t, 0)k≤ B holds for all
t ≥ 0.

Assumption 8. η is continuous in its arguments, and for arbitrary δ > 0, there

exist dηt (δ) and dηx (δ) such that

∂η(t, ζ) ∂η(t, ζ)
∂t ≤ dηt (δ), ∂ζ ≤ dηζ (δ) (64)

∞ ∞

Assumption 9. for τ ≥ 0, krτ kL∞ ≤ ρ and kṙτ kL∞ ≤ dr , for some positive
constants ρ and dr . Now, for some arbitrary γ > 0, let
ρ̄
ρ̄ , max {ρ + γ, L1 (ρ + γ) + L2 } , Lρ , dη (ρ̄) (65)
ρ ζ

It follows from Lemma A.9.2 in [41] that η(t, ζ(t)) can be parametrized as
follows
η(t, ζ(t)) = θ(t)krτ kL∞ + σ(t) (66)
where θ(t), σ(t) ∈ R4 are differentiable functions. It follows that (62) can be
rewritten as

ṙ(t) = Am r(t) + ΓAD (t) − (θ(t)krτ kL∞ + σ(t)) , r(0) = r0 (67)

Since the nonlinear function η(t, ζ(t)) is unknown due to uncertainties and un-
measured disturbances, the control term Γad (t) should be adaptively designed
in order to obtain estimates of θ(t) and σ(t). Now, consider the following state
predictor of the combined tracking error
 
˙ =Am r̂(t) + Γad (t) − θ̂(t)krτ kL + σ̂(t)
r̂(t) ∞
(68)
− K r̃(t), r̂(0) = r0

where r̃(t) , r̂(t) − r(t) and K ∈ R4×4 is a design parameter introduced to

reject high-frequency noise [14]. θ̂(t) and σ̂(t) are real-time estimates of θ(t)
 Adaptive Control of Parallel Manipulators 27

and σ(t) respectively, their evolution is governed by the following projection-

based adaptive laws in order to ensure their boundedness [41]
˙
 
θ̂(t) = ΣProj θ̂(t), P r̃(t)krτ kL∞ , θ̂(0) = θ̂0 (69)
˙
σ̂(t) = ΣProj (σ̂(t), P r̃(t)) , σ̂(0) = σ̂0 (70)
where Σ ∈ R+ is the adaptive gain, P = P T > 0 is the solution to the algebraic
Lyapunov equation ATm P +P Am = −Q for some arbitrary Q = QT > 0. Since
the estimated parameters are bounded thanks to the projection operator, they
satisfy kθ̂(t)k∞ < θb , kσ̂(t)k∞ < σb , ∀t ∈ [0, T ]. The adaptive control
signal Γad (t) is the output of the following system given in Laplace domain
Γad (s) = C(s)η̂(s) (71)
 
where η̂(s) is the Laplace transform of η̂(t) = θ̂(t)krt kL∞ + σ̂(t) and C(s)
is a diagonal matrix of filters whose elements are BIBO-stable strictly proper
transfer functions satisfying a unity DC gain and zero initialization for their
state-space realizations.

7. Real-Time Experiments and Results

In order to demonstrate the relevance of the proposed extended L1 adaptive
controller, real-time experiments were conducted on an experimental testbed
consisting of the Veloce robot.

7.1. Experimental Testbed of the VELOCE Robot

The actuators of VELOCE are the TMB0140-100-3RBS ETEL direct-drive mo-
tors. They can provide a maximum peak torque of 127 Nm and they are able
to reach 550 rpm of speed. Each actuator is equipped with a non-contact incre-
mental optical encoder providing a total number of 5000 pulses per revolution.
The global structure of the manipulator is capable of reaching 10 m/s of maxi-
mum velocity of the traveling plate, 200 m/s2 of maximum acceleration and is
able to handle a maximum payload of 10 Kg. The control architecture is imple-
mented using Simulink from Mathworks and compiled using the XPC Target
real-time toolbox. The resulting low-level code is then uploaded to the target
PC; an industrial computer cadenced at 10 KHz (i.e. sample time of 0.1 ms).
The experimental testbed is displayed in Figure 11.
28 M. Bennehar, A. Chemori, S. Krut et al.

Actuator
Rear-arm

Forearm
Moving platform

Figure 11. View of the experimental setup of the VELOCE robot.

7.2. Real-Time Experimental Results

To demonstrate the relevance of the proposed contribution and to highlight its
benefits, both the L1 adaptive controller in [16] and the proposed extended one
were implemented in real-time on VELOCE. As a reminder, the implemented
control law in [16] consists only of the stabilizing term and the adaptive term, it

Table 3. Summary of the controllers’ parameters

Parameter Description Value

Λ position error weight 10 × diag(65, 65, 65, 65)
Am transient response matrix diag(−6, −6, −6, −6)
θmax upper bound on θ 50
σmax upper bound on σ 30
Σ adaptation gain 106
K noise rejection gain 103 × diag(6, 6, 6, 6)
C(s) L1 low-pass design filter 144/(s2 + 21.6s + 144)
 Adaptive Control of Parallel Manipulators 29

is expressed as
Γ(t) = Am r(t) + Γad (t) (72)
The traveling plate of the manipulator had to perform several spatial point-to-
point displacements as well as rotations inside the manipulator’s workspace.
We use 5th order polynomials to generate the desired trajectories in Cartesian
space, then the inverse kinematics problem is solved online to determine the
corresponding joint trajectories. The duration of each point-to-point trajectory
was fixed to 0.2 sec. The desired joint trajectories were obtained by solving
the Inverse Kinematics problem in real-time while the actual ones are available
from the encoders measurements. The manipulator is not equipped with ex-
ternal sensors, hence the actual Cartesian position is obtained by solving the
Forward Kinematics problem in real-time as well. We choose the same control
parameters for both controllers, they are summarized in Table 3. The estimated
functions were initialized to zero ( θ̂(0) = [0, 0, 0, 0]T , σ̂(0) = [0, 0, 0, 0]T ).
A comparison of the Cartesian tracking errors between both controllers is
depicted in Figure 12. The plots are zoomed in the interval [5, 7] seconds for
clarity. It can be clearly seen that the augmented controller performs much better
than the original L1 adaptive controller especially on translations on the x and
y-axes, while the enhancement is smaller on z-axis and the rotational DOF.
To quantify the improvement brought by the proposed extended L1 adaptive
controller, we formulate the following criteria based on the Root Mean Square
(RMS) of the tracking errors:

4
! 21
X
RMSJ = RMS2 (ǫqi ) (73)
i=1

1
RMSC = RMS2 (ǫx ) + RMS2 (ǫy ) + RMS2 (ǫz ) + + RMS2 (ǫs ) 2
(74)

where ǫ(.) is the error between the desired and actual position and RMS(.) is
the standard root mean square function. The obtained results are summarized in
Table 4. It can be noticed that the improvement is very significant regarding the
joint tracking errors (up to 52%) and the translational movements (up to 68.8%)
while only small improvement is observed on the rotation of the traveling plate
(only 4.2%). This result highlights the benefits of using the dynamics of the
manipulator in the control loop in terms of tracking performance.
The generated control inputs are shown in Figure 15. The control inputs
remain within the allowable range and do not exceed the limits of the actuators.
30 M. Bennehar, A. Chemori, S. Krut et al.
Extended L1 -AC Standard L1 -AC
1

ex [mm]
0
−1
5 5.5 6 6.5 7
1
ey [mm]

0
−1
5 5.5 6 6.5 7
ez [mm]

0
−0.2

5 5.5 6 6.5 7
es [mm]

0
−0.2

5 5.5 6 6.5 7
Time [s]

Figure 12. Cartesian tracking errors: Extended L1 adaptive controller (solid

line), L1 adaptive controller (dashed line).

Overall, it can be observed that the amplitudes of the control inputs in the case
of the augmented controller are slightly smaller than those of the standard L1
adaptive one. Hence, the following conclusion can be drawn: the proposed

Table 4. Tracking performance comparison

L1 -AC Extended L1 -AC Improvement

RMSJ [deg] 0.1012 0.0486 52 %
RMSC [mm] 0.4634 0.195 69.7 %
 Adaptive Control of Parallel Manipulators 31
Extended L1 -AC Standard L1 -AC

20

θ̂1 (t)
0
−20
5 5.5 6 6.5 7
40
θ̂2 (t)

20
0
−20
5 5.5 6 6.5 7
40
20
θ̂3 (t)

0
−20
5 5.5 6 6.5 7
40
θ̂4 (t)

20
0
−20
5 5.5 6 6.5 7
Time [s]

Figure 13. Evolution of θ̂ versus time.

controller significantly improves the tracking performance of the closed-loop

system while consuming less energy than the standard one.
The evolution of the estimated nonlinear functions θ̂(t) and σ̂(t) versus
time is depicted in Figure 13 and Figure 14 respectively. These figures clearly
demonstrate the relevance of the proposed contribution. Indeed, the addition
of the model-based feedforward considerably helps in compensating for the
modeling inherent nonlinearities of the manipulator. Therefore, the remain-
ing nonlinearities that have to be compensated for by the adaptive signal are of
smaller amplitudes. This result is further illustrated in Figure 16 where only the
evolution of the adaptive component of (61) and (72) are plotted (i.e Γad (t)).
We can see that the adaptive signal, needed to compensate for the remaining
nonlinearities, of the proposed controller is of lower amplitude than that of the
32 M. Bennehar, A. Chemori, S. Krut et al.

Extended L1 -AC Standard L1 -AC

20

σ̂1 (t)
10

0
5 5.5 6 6.5 7
20
σ̂2 (t)

10
0

5 5.5 6 6.5 7
10
σ̂3 (t)

−10
5 5.5 6 6.5 7

0
σ̂4 (t)

−10

5 5.5 6 6.5 7
Time [s]

Figure 14. Evolution of σ̂ versus time.

conventional one.

Conclusion
In this work, a new controller based on L1 adaptive control for parallel manipu-
lators is presented. Standard L1 adaptive control do not rely on any knowledge
about the dynamics of the controlled system which may result in a poor track-
ing perfromance. Since a simplified dynamic model for parallel manipulators is
relatively easy to derive, we propose to take advantage of the modelled dynam-
ics in the control loop to enhance the performance of L1 adaptive control. For
this reason, we augment the standard L1 adaptive controller with a computed
 Adaptive Control of Parallel Manipulators 33

Extended L1 -AC Standard L1 -AC

Γ1 (t) [Nm]
10
0
−10
5 5.5 6 6.5 7
Γ2 (t) [Nm]

10
0
−10
5 5.5 6 6.5 7
Γ3 (t) [Nm]

10
0
−10
5 5.5 6 6.5 7
Γ4 (t) [Nm]

10
0
−10
5 5.5 6 6.5 7
Time [s]

Figure 15. Evolution of the control input torques Γ versus time.

feedforward term based on the nominal dynamics of the parallel manipulator

and the desired trajectories. The main motivation behind such a proposition is
to improve the tracking performance of the manipulator. In fact, the included
dynamics partially compensates for the inherent nonlinear dynamics which re-
sults in an improved tracking performance. Real-time experiments on a 4-DOF
parallel manipulator show that the proposed augmented control scheme signifi-
cantly reduces the tracking errors as compared to standard L1 adaptive control.
Future work may consider evaluating the proposed controller in other working
scenarios and on other parallel manipulators prototypes.
34 M. Bennehar, A. Chemori, S. Krut et al.

Extended L1 -AC Standard L1 -AC

Γad1 (t) [Nm]

5
0
−5
5 5.5 6 6.5 7
Γad2 (t) [Nm]

5
0
−5
5 5.5 6 6.5 7
Γad3 (t) [Nm]

5
0
−5
5 5.5 6 6.5 7
Γad4 (t) [Nm]

5
0
−5
5 5.5 6 6.5 7
Time [s]

Figure 16. Evolution of the adaptive control input torques Γad versus time.

Acknowledgments
This research was supported by the French National Research Agency, within
the project ARROW (ANR-2011-BS3-006-01-ARROW).

References
[1] J. Luh, M. Walker, and R. Paul, “Resolved-acceleration control of me-
chanical manipulators,” IEEE Trans. Autom. Control, vol. 25, no. 3, pp.
468–474, Jun. 1980.
 Adaptive Control of Parallel Manipulators 35

[2] D. Koditschek, “Natural motion for robot arms,” in Proc. The 23rd IEEE
Conference on Decision and Control (CDC’84), Las Vegas, Nevada, USA,
Dec. 1984, pp. 733–735.

[3] B. Paden and R. Panja, “Globally asymptotically stable pd+ controller for
robot manipulators,” International Journal of Control, vol. 47, no. 6, pp.
1697–1712, Jun. 1988.

[4] S. Dubowsky and D. DesForges, “The application of model-referenced

adaptive control to robotic manipulators,” Journal of Dynamic Systems,
Measurement, and Control, vol. 101, no. 3, pp. 193–200, Jun. 1979.

[5] R. Horowitz and M. Tomizuka, “An adaptive control scheme for mechan-
ical manipulators - compensation of nonlinearity and decoupling control,”
Journal of Dynamic Systems, Measurement, and Control, vol. 108, no. 2,
pp. 127–135, Jun. 1986.

[6] J. Craig, P. Hsu, and S. Sastry, “Adaptive control of mechanical manipu-

lators,” The International Journal of Robotics Research, vol. 6, no. 2, pp.
16–28, Jun. 1987.

[7] R. Ortega and M. W. Spong, “Adaptive motion control of rigid robots: A

tutorial,” Automatica, vol. 25, no. 6, pp. 877–888, Oct. 1989.

[8] N. Sadegh and R. Horowitz, “Stability and robustness analysis of a class of

adaptive controllers for robotic manipulators,” The International Journal
of Robotics Research, vol. 9, no. 3, pp. 74–92, Jun. 1990.

[9] M. Bennehar, A. Chemori, and F. Pierrot, “A new extension of direct com-

pensation adaptive control and its real-time application to redundantly ac-
tuated pkms,” in IEEE/RSJ International Conference on Intelligent Robots
and Systems (IROS’14), Chicago, Illinois, USA, Sep. 2014, pp. 1670–
1675.

[10] C. Cao and N. Hovakimyan, “Design and analysis of a novel L1 adaptive

controller, part i: Control signal and asymptotic stability,” in American
Control Conference (ACC’06), Minneapolis, USA, Jun. 2006, pp. 3397–
3402.
36 M. Bennehar, A. Chemori, S. Krut et al.

[11] ——, “Design and analysis of a novel L1 adaptive controller, part

ii: Guaranteed transient performance,” in American Control Conference
(ACC’06), Minneapolis, USA, Jun. 2006, pp. 3403–3408.

[12] I. Gregory, C. Cao, E. Xargay, N. Hovakimyan, and X. Zou, “L1 adaptive

control design for nasa airstar flight test vehicle,” AIAA Guidance, Navi-
gation, and Control Conference, p. 5738, Aug. 2009.

[13] D. Maalouf, V. Creuze, and A. Chemori, “A novel application of multi-

variable L1 adaptive control: From design to real-time implementation on
an underwater vehicle,” in Proc. IEEE International Conference on Intel-
ligent Robots and Systems (IROS’12), Vilamoura, Algarve, Portugal, Jan.
2012, pp. 76–81.

[14] K. D. Nguyen, H. Dankowicz, and N. Hovakimyan, “Marginal stability in

L1 adaptive control of manipulators,” in Proceedings of the 9th ASME In-
ternational Conference on Multibody Systems, Nonlinear Dynamics, and
Control, Portland, OR, USA, 2013.

[15] L. Techy, C. K. Reddy, C. A. Woolsey, C. Cao, and N. Hovakimyan, “Non-

linear control of a novel two-link pendulum,” in American Control Con-
ference (ACC’07), New York, USA, 2007, pp. 19–24.

[16] M. Bennehar, A. Chemori, and F. Pierrot, “L1 adaptive control of paral-

lel kinematic manipulators: Design and real-time experiments,” in IEEE
International Conference on Robotics and Automation (ICRA’15), Seattle,
Washington, USA, May 2015, pp. 1587–1592.

[17] K. Astrom, Adaptive Control (Second Edition). Addison-Wesley, 1995.

[18] C. C. Nguyen, S. S. Antrazi, Z. L. Zhou, and C. E. Campbell, “Adaptive

control of a stewart platform-based manipulator,” Journal of Robotic Sys-
tems, vol. 10, no. 5, pp. 657–687, 1993.

[19] S. Dubowsky and D. DesForges, “The application of model-referenced

adaptive control to robotic manipulators,” Journal of Dynamic Systems,
Measurement, and Control, vol. 101, no. 3, pp. 193–200, Sep. 1979.
 Adaptive Control of Parallel Manipulators 37

[20] S. Nicosia and P. Tomei, “Model reference adaptive control algorithms for
industrial robots,” Automatica, vol. 20, no. 5, pp. 635 – 644, 1984.

[21] B. Yegnanarayana, Artificial neural networks. PHI Learning Pvt. Ltd.,

2009.

[22] X. Qingsong and L. Yangmin, “A 3-prs parallel manipulator control based

on neural network,” in Advances in Neural Networks ISNN 2007, D. Liu,
S. Fei, Z.-G. Hou, H. Zhang, and C. Sun, Eds. Springer Berlin Heidelberg,
2007, pp. 757–766.

[23] F. Lewis, S. Jagannathan, and A. Yesildirak, Neural network control of

robot manipulators and non-linear systems. CRC Press, 1998.

[24] Y. Li and Y. Wang, “Trajectory tracking control of a redundantly actuated

parallel robot using diagonal recurrent neural network,” in International
Conference on Natural Computation (ICNC’09), vol. 2, Aug 2009, pp.
292–296.

[25] J. J. Craig, P. Hsu, and S. S. Sastry, “Adaptive control of mechanical ma-

nipulators,” The International Journal of Robotics Research, vol. 6, no. 2,
pp. 16–28, 1987.

[26] H. Cheng, Y. Yiu, and Z. Li, “Dynamics and control of redundantly ac-
tuated parallel manipulators,” IEEE/ASME Transactions on Mechatronics,
vol. 8, no. 4, pp. 483–491, Dec. 2003.

[27] J. Merlet, Parallel Robots, Second Edition. Dordrecht, Netherlands:

Springer, 2006.

[28] W. Shang, S. Cong, and Y. Ge, “Adaptive computed torque control for a
parallel manipulator with redundant actuation,” Robotica, vol. 30, no. 3,
pp. 457–466, May 2012.

[29] M. Honegger, R. Brega, and G. Schweiter, “Application of a nonlinear

adaptive controller to a 6 dof parallel manipulator,” in IEEE International
Conference on Robotics and Automation (ICRA’00), vol. 2, San Francisco,
CA, USA, Apr. 2000, pp. 1930–1935.
38 M. Bennehar, A. Chemori, S. Krut et al.

[30] G. Sartori Natal, A. Chemori, and F. Pierrot, “Dual-space control of ex-

tremely fast parallel manipulators: Payload changes and the 100g experi-
ment,” IEEE Transactions on Control Systems Technology, vol. 23, no. 4,
pp. 1520–1535, Jul. 2015.

[31] D. Corbel, M. Gouttefarde, O. Company, and F. Pierrot, “Actuation redun-

dancy as a way to improve the acceleration capabilities of 3t and 3t1r pick-
and-place parallel manipulators,” Journal of Mechanisms and Robotics,
vol. 2, no. 4, pp. 1–13, 2010.

[32] W. Shang and S. Cong, “Nonlinear adaptive task space control for a 2-dof
redundantly actuated parallel manipulator,” Nonlinear Dynamics, vol. 59,
no. 1-2, pp. 61–72, 2010.

[33] J.-J. Slotine and L. Weiping, “On the adaptive control of robot manipula-
tors,” The International Journal of Robotics Research, vol. 6, no. 3, pp.
49–59, Sep. 1987.

[34] R. Clavel, “Device for the movement and positioning of an element

in space,” Dec. 1985, uS Patent 4,976,582. [Online]. Available:
https://patents.google.com/patent/US4976582A

[35] J. Merlet, Parallel Robots, Second Edition. Dordrecht, Netherlands:

Springer, 2006.

[36] A. Codourey, “Dynamic modeling of parallel robots for computed-torque

control implementation,” The International Journal of Robotics Research,
vol. 17, pp. 1325–1336, 1998.

[37] D. Corbel, M. Gouttefarde, and O. Company, “Towards 100g with pkm.

is actuation redundancy a good solution for pick and place?” in Proc.
IEEE International Conference on Robotics and Automation (ICRA’10),
Anchorage, Alaska, May 2010, pp. 4675–4682.

[38] N. Hovakimyan and C. Cao, L1 Adaptive Control Theory: Guaranteed Ro-

bustness with Fast Adaptation. Philadelphia, Pennsylvania, USA: SIAM,
2010.

[39] M. Bennehar, A. Chemori, and F. Pierrot, “L1 adaptive control of paral-

lel kinematic manipulators: Design and real-time experiments,” in IEEE
 Adaptive Control of Parallel Manipulators 39

International Conference on Robotics and Automation (ICRA’15), Seattle,

Washington, USA, May 2015, pp. 1587–1592.

[40] H. Khalil, Nonlinear Systems. Prentice-Hall, 2002.

[41] N. Hovakimyan and C. Cao, L1 Adaptive Control Theory: Guaranteed Ro-

bustness with Fast Adaptation. Philadelphia, Pennsylvania, USA: SIAM,
2010.
In: Parallel Manipulators ISBN: 978-1-63485-926-4
Editor: Cecilia Norton, pp. 41-80
c 2016 Nova Science Publishers, Inc.

Chapter 2

A R EDUNDANT DYNAMIC M ODELLING

P ROCEDURE B ASED ON E XTRA S ENSORS
FOR PARALLEL R OBOT C ONTROL

Asier Zubizarreta∗, Itziar Cabanes, Marga Marcos,

Charles Pinto, Javier Corral and Pablo Bengoa
Faculty of Engineering in Bilbao,
University of the Basque Country (UPV/EHU)
Bilbao, Spain

Keywords: parallel robots, Gough-Stewart platform, position control, com-

puted torque control, modelling

1. Introduction
Robots are a key element in current industrial processes, as they can be applied
to a number of tasks, increasing both quality and productivity. Traditionally, se-
rial robots have been installed in factories, as their wide operating space allowed
them to fulfill a number of tasks. However, due to their high moving mass and
single kinematic chain structure, these robots present some disadvantages when
high speed, accuracy or heavy load handling tasks have to be executed.
∗ Corresponding Author: [email protected]
42 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

Parallel robots [27] provide an interesting alternative to these applica-

tion fields, as their multiple kinematic chain structure offers increased stiff-
ness, allowing reduced positioning errors, lighter mechanisms and increased
load/weight ratios. These capabilities make these robots suitable for those tasks
where high accelerations and speeds, micrometric precision or heavy load han-
dling are required.
However, the complexity of their structure presents some drawbacks: com-
plex kinematic and dynamic modelling, lack of closed-form solution in the kine-
matic problem, presence of nonactuated joints, orientation and position cou-
pling, limbs interference, small workspace and presence of inner singularities.
These issues limit an extensive implementation of these robots in practical ap-
plications in industry. Moreover, being a recently rediscovered field, there are
still many areas in parallel robotics that have not been thoroughly studied. Thus,
in addition to the aforementioned problematic, other challenges in areas such as
control or calibration arise.
The present work focuses on the control of parallel robots. Although less
studied than areas such as kinematic analysis or synthesis, control plays an im-
portant role in parallel robots [26], as it allows to extract the dynamic capa-
bilities of the mechanical structure, ensuring that the design specifications are
met in real robots. Moreover, parallel robots are usually implemented in tasks
with high performance requirements where advanced control laws are required
to fulfill the required specifications. Hence, research in this area is a central
issue in parallel robotics.
However, most of control approaches in parallel robotics have been im-
ported from serial robots without considering the particularities of these robots.
One of the main issues is that while control laws are easily defined in the joint-
space in serial robots, in parallel robots the natural coordinates are the ones
defined in task space [31]. These, which represent the pose of the mobile plat-
form are not easily measured directly in the general case. Thus, it is common
to use estimators based on the kinematic model of the robot, whose resolution
is based on numerical procedures such as Newton-Raphson iterative approach.
Moreover, in parallel robots some joints are not actuated, and their motion also
has to be estimated.
In this work, a solution based on the use of extra sensors attached to the un-
actuated joints of the parallel robot is detailed. The extra information provided
by these sensors can be used to implement a model-based control approach that
combines the performance of advanced control techniques and the robustness of
 A Redundant Dynamic Modelling Procedure ... 43

sensor redundancy. In order to implement this approach, three major areas will
be analyzed in the next sections: dynamic modelling, control law definition, and
sensitivity, considering the well known Gough-Stewart platform as study case.

2. Modelling Methodology Based on Extra Sensors

2.1. Definition of the Dynamic Model in Parallel Robots
In serial robotics literature, Lagrangian and Newton-Euler formulations have
been widely used to calculate the dynamic model of n degrees-of-freedom serial
robots [40].
τ = D (q) q̈ + C (q, q̇) q̇ + G (q) (1)
where τ is the vector of the actuator torque/forces, q is the vector of the n articu-
lar coordinates, D is the symmetric and positive-defined inertia matrix, C is the
Coriolis matrix, where (D − 2 C) is skew-symmetric and G is the gravity terms
vector. Friction terms have been neglected in this model, although they can be
introduced in it as an additional term easily. This model has been used in the
literature to implement multiarticular model-based control techniques such as
the Computed Torque Control approach [9].
The dynamic model of parallel robots can also be defined in the joint space
as in Eq. (1), although some of the properties of the dynamic model do not
hold, as the dynamic model is not defined in Direct Kinematic Problem singu-
larities [23]. Being more complex than serial robots, the procedures applied to
serial robots cannot be used in parallel robots directly, as constraints due to the
multiple kinematic chains have to be introduced in the dynamic model.
In the literature, three main approaches have been proposed to obtain the
dynamic model of parallel robots: the traditional Newton-Euler formulation
[10, 14, 18], which usually leads to a large number of equations due to the inter-
nal forces of the mechanism; the Lagrangian formulation using Lagrange multi-
pliers [5,20], which is an energetic approach and allows to eliminate the need of
internal forces calculation; and the Principle of Virtual Work [12, 41, 43], which
is also an energetic approach. When modelling a particular parallel robot, any
approach can lead to a dynamic model as the one described in Eq. (1).
However, these approaches require the knowledge of all kinematic variables
of the robot in order to calculate the dynamic model of parallel robots. As the
measurement of the end effector pose, this is, the Tool Center Point location,
is not possible in the general case, the required kinematic variables need to be
44 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

estimated in terms of the actuated joints, which are usually sensorized. This ap-
proach usually requires iterative procedures to calculate the estimated variables,
and robustness of the estimation is based on the accuracy of the kinematic model
parameters.
In order to provide extra robustness to the approach and reduce compu-
tational time, extra sensors can be introduced in specific passive joints. The
advantages of the use of redundant data have been demonstrated in several
works [3,25] to improve estimation accuracy or decrease kinematic model com-
putational time. In this work, the use of explicit extra data in the dynamic model
is proposed, so that this extra information can be used to increase control per-
formance in model-based control approaches. This way, in the next section, an
extended procedure is detailed to define the dynamic model of a parallel robot
in terms of whatever generalized coordinates the designer selects, allowing to
introduce data from extra sensors directly in the dynamic model.

2.2. Dynamic Modelling Methodology for Full Parallel Robots

The proposed methodology is based in two main ideas: the use of the La-
grangian Formulation, which allows to define the dynamic model in terms of
any set of generalized variables; and the use of a subsystem approach, in which
the parallel robot is divided in the moving platform and the limbs, allowing to
calculate the model of each part easily. It should be noted that the proposed
approach is valid for fully parallel robots, i.e., robots in which the number of
limbs are equal to the number of the degrees of freedom of the robot. These
robots are very common in parallel robotic applications.
In addition, the following considerations apply, which are common in gen-
eral application cases: nondeformable elements are considered and friction
terms will be neglected, although their introduction is straightforward.
In order to illustrate the approach, the Gough-Stewart platform will be con-
sidered as study case (Fig. 1). The Gough-Stewart platform is one of the most
studied parallel robots, being composed by a moving platform and six UPS se-
rial chains. This structure provides 6 degrees-of-freedom, making it suitable
for tasks such as heavy load handling [37, 44], high precision [28], underwater
robotics [35], nuclear facilities inspection [34] and motion simulation [22].
 A Redundant Dynamic Modelling Procedure ... 45

Moving
platform
x P(u, v , w ) Bi subsystem

joint

actuated joint

! joint
"

#
Ai Serial chain
q O(x, y , z ) subsystem
base platform

Figure 1. Gough-Stewart platform.

2.2.1. Previous Considerations: Orientation Angles and Angular

Velocities
In the general case, for a 6 degrees-of-freedom motion, the output or task
coordinates x contain the cartesian position of the Tool Center Point (TCP)
 T  T
px = x y z , and the orientation angles ψ = α β γ of the moving
platform with respect to the fixed frame O(x, y, z),
 T
x= x y z α β γ (2)
However, the choice of the orientation angles is arbitrary, and depends on
the particular convention used. Hence, the time derivative of x only has physical
meaning for the translation terms [2] in the general case, as the derivative of
the orientation angles ψ̇ is not, in the general case, the angular velocity of the
moving platform, ψ̇ 6= ω.
Hence, as the dynamics of rotational motions are described in terms of the
angular velocity ω but traditionally orientation is defined in terms of Euler an-
gles ψ, both variables should be related if a general case is to be considered.
This relationship can be derived as [9],
ω̃ = Ṙ RT (3)
where R(ψ) is the rotation matrix of the moving platform with respect to the
fixed frame, Ṙ(ψ, ψ̇) is its time derivative and ω̃ is a skew-symmetric matrix
46 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

related to ω,
 
0 −ωz ωy  T
ω̃ =  ωz 0 −ωx  ω= ωx ωy ωz (4)
−ωy ωx 0

Eq. 3 can be rewritten in the following form, depending on the selected

Euler angle convention,  
α̇
ω = E ψ̇ = E  β̇  (5)
γ̇
where E depends on the selected Euler angle convention, and will be illustrated
for a specific case in the next subsection.
Hence, if a velocity vector v is defined in terms of the angular velocity terms,
 
ṗx
v= (6)
ω

then, ẋ and v can be related as,

 
I 0
v= ẋ = TR ẋ (7)
0 E
where TR (ψ) is a projection matrix defined in terms of the selected Euler an-
gles.
Based on this expression, and taking the time derivative, the angular accel-
eration and the second time derivatives of Euler angles can be related,
 
p̈x
a = v̇ = = TR ẍ + ṪR ẋ (8)
ω̇

where ṪR (ψ, ψ̇) depends on the Euler angles and their derivatives.
Eqs. 7 and 8 will be used in the next section to define the dynamic model of
the robot.

2.2.2. Subsystems and Variable Sets

A parallel robot is composed by two platforms, a fixed one and a moving one,
connected by several serial links (Fig. 1). If a fully parallel robot is considered,
 A Redundant Dynamic Modelling Procedure ... 47

such as the Gough-Stewart platform, the number of serial chains or limbs is

equal to the number of degrees-of-freedom (dof).
In order to calculate the dynamic model of the robot, the usual approach
is to consider the parallel robot as a group of independent subsystems that are
connected using kinematic constraints. Thus, two subsystems are considered:
the moving platform, where the TCP is located, and the serial chain subsystem,
i.e, the limbs. The dynamic model of each subsystem can be calculated easily
in terms of its natural coordinates. This way, the motion of the moving platform
can be easily defined in the task space in terms of the n = 6 output coordinates
 T
x = x y z α β γ , while the dynamics of the serial limbs can be cal-
culated using serial robot approaches, which are based on the use of nq = 18
joint variables q.
Joint variables can be divided in two sets: actuated and nonactuated joint
variables,  
qa
q= ∈ R18 (9)
qna
The n = 6 actuated joints qa , are associated to the prismatic actuators of
the Gough-Stewart platform, which are composed by a cylinder and a rod, and
whose relative motion varies the qai = li distance between points Ai and Bi for
each i = 1, . . .6 leg (Fig. 2).
 T
qa = l1 l2 l3 l4 l5 l6 (10)
Nonactuated joint variables qna , on the other hand, are not actuated although
their motion is required to be known in order to calculate the dynamic model of
the robot (Fig. 2),  
θ
qna = ∈ R12 (11)
ϕ
 T
where, θ = θ1 θ2 θ3 θ4 θ5 θ6 and ϕ =
 T
ϕ1 ϕ2 ϕ3 ϕ4 ϕ5 ϕ6 .
In this work, the introduction of sensors in a subset of the nonactuated joints
is considered. Hence, the joint variables in the nonactuated joint variable set
can be divided in ns sensorized qs and n p passive qp joint variables, qna =
 T T
qs qp T . This way, the set of sensorized and measurable nc = n + ns
48 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

dO
i Bi
P

li

!$
px
#$

! "$
# ai Ai
"
O
Figure 2. Closed loop kinematic equation.

joints will be denoted as control coordinates qc (Fig.3),

 
qa
qc = ∈ Rn c
qs

where nc ∈ [n, nq], depending on the number of extra sensors introduced to mea-
sure the variables in qna .

2.2.3. Kinematic Model

In order to define the dynamic model of the Gough-Stewart platform, its kine-
matic relations have to be calculated first.

Closure Loop Equation. The six vectorial closure loop equations are de-
duced from the mechanical structure of each limb and its connection to the fixed
and moving platform, as seen in Fig. 2. This way, if O(x, y, z) is the fixed refer-
ence frame, attached to the fixed platform, the equations defining the motion of
each limb can be written as follows,

Γi = ai + li − px − di O = 0 i = 1, . . .6 (12)
 A Redundant Dynamic Modelling Procedure ... 49

Passive

nonactuated
q p ∈ Rnp
qna ∈ Rnna Sensorized
Joint Variables
nq
q s ∈ R ns
q∈R
Dynamic Model
Actuated Actuated
Task Coordinates n
qa ∈ R q a ∈ Rn
x ∈ Rn
Control Coordinates

q c ∈ R nc

Figure 3. Variables in the dynamic modelling methodology.

 T
where px = x y z is the cartesian position of the TCP of the robot, which
is the origin of the moving frame; ai is the constant position vector of the U
joints located in Bi , which define the geometry of the fixed platform and li is the
vector that represents the relative position of Ai with respect to Bi , and it is de-
 T
fined in terms of the joint variables associated to the limbs qi = li θi ϕi ,
 
cosθi cos ϕi
li = li  sinθi cosϕi  (13)
˙ i
sinϕ
The orientation of the moving platform is given by di O , which is the projec-
tion of the constant vector di defined in the moving reference frame P(u, v, w)
with respect to the fixed one,

di O = R(α, β, γ) di (14)

where if roll-pitch-yaw (α, β, γ) angles are selected to define the orientation,

 
cγ cβ cγ sβ sα − sγ cα sγ sα + cγ sβ cα
R(α, β, γ) =  sγ cβ cγ cα + sγ sβ sα sγ sβ cα − cγsα  (15)
−sβ cβ sα cβ cα
where c and s stand for the trigonometric functions cosine and sine, respectively.
50 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

Note that each closure loop equation is defined using two sets of variables,
the output ones x, and the joint variables associated to each limb qi . The com-
bination of the closure loop equations 12, defines the constraint equation set of
the robot, that relates all variables,

[Γi(x, qi )]i=1,...6 = Γ(x, q) = 018×1 (16)

Inverse Kinematic Problem. The Inverse Kinematic Problem (IKP) allows

the calculation of the active joint variables qa in terms of the output ones x.
Thus, if x is known, the cartesian position of Bi = px + R di can be determined
easily. Hence, imposing the distance constraint li2 = ||li ||2 on the vectorial loop
closure equation Eq.12 and operating,

li2 = li T li = [px − ai + R(α, β, γ) di]T [px − ai + R(α, β, γ) di] i = 1 . . .6 (17)

Position Problem for Nonactuated Joints. The nonactuated joint variables

define the position of the elements of each serial chain. Hence, if vector li
associated to each limb is considered and Eqs. 12 and 13 are combined,
 
li cosθi cos ϕi
li = px − ai + R(α, β, γ) di =  li sinθi cos ϕi  (18)
li sinϕi

Considering qa and x known, each nonactuated joint variable can be calcu-

lated for i = 1, . . .6,
 q 
ϕi = atan2 liz , lix2 + liy2 θi = atan2 (liy , lix) (19)

Direct Kinematic Problem. The Direct Kinematic Problem (DKP) calculates

the pose of the moving platform x in terms of the actuated ones qa , whose actua-
tors are usually sensorized. This relationship is of great importance in robotics,
as x cannot be directly measured in the general case, and its estimation is re-
quired. However, this estimation is usually time consuming, as no analythical
solution exists in the general case for the DKP. Moreover, the estimation pro-
vided by the DKP is highly dependent on the accuracy of the kinematic param-
eters of the robot.
Redundant sensors can be used to enhance the calculation procedure of the
DKP [25, 32] and add more robustness. This way, if a subset of the nonactuated
 A Redundant Dynamic Modelling Procedure ... 51

joints are sensorized, qs , this data can be used to get a better estimation of the
output variables.
For this purpose, a selection vector Ic is defined. This vector contains the
indexes of the variables of q that are sensorized, considering that active ones qa
are always in this set. Hence, the set of control variables qc can be defined,

qc = [q]Ic (20)

Hence, two sets of equations can be used to estimate x. First, if Eq.17 is

considered, x and qa can be related,
 
||li (x)|| − li2 i=1...6 = 06×1 (21)

On the other hand, the sensorized nonactuated joint variables defined in Eq.
19 can be also used to estimate x,

hp [lix (x, qa ) tanθi − liy (x, qa )]ii=1...6 = 06×1

(22)
lix (x, qa )2 + liy (x, qa )2 tan ϕi − liz (x, qa ) = 06×1
i=1...6

Combining Eqs. 21 and 22, a 18 equation system can be defined, each

relating a variable from q with the output variables, F = 018×1 ,
   
||li|| − li2 i=1...6
F(x, q) =  hq [lix tanθi − liy ]i=1...6
 
i  = 018×1 (23)
lix2 + liy2 tanϕi − liz
i=1...6

Hence, selecting the sensorized variables using Ic , the equation system re-
lation all measurable variables qc and x can be obtained,

Fc (qc, x) = [F(q, x)]Ic = 0 (24)

which, as qc is known, defines the redundant DKP solving approach,

Velocity Problem. In order to calculate the velocity relations of the robot, the
vectorial Closure Loop equation for each limb Eq. 12 has to be differentiated
with respect to time,

l̇i = ṗx + Ṙ di = ṗx + ω × R di = ṗx − R di × ω (25)

52 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

where, R di = dO O T
i , defined in Eq. 14, satisfies d̃i = R d̃i R . Hence, the previous
equation can be written as,
 
l̇i = ṗx − R d̃i RT ω = I3×3 −R d̃i RT v (26)
As seen in Section 2.2.1,
 
l̇i = I3×3 −R d̃i RT TR ẋ (27)
where, if roll-pitch-yaw angles are used, TR can be calculated ( Eq. 7),
 
I 0
TR = (28)
0 E
and,  
1 0 − sin β
E = R  0 cos α cos β sinα  (29)
0 − sinα cos β cos α
Combining Eqs. 27 and 28,
 
  −c ϕi c θi li c ϕi s θi li c θi s ϕi
l̇i = I3×3 −R d̃i RT TR ẋ = −  −c ϕi s θi −li c ϕi c θi li s ϕi s θi  q̇i
| {z } −s ϕi 0 −li c ϕi
Jxi | {z }
J qi
(30)
Hence, the link velocity equation can be written in matrix form,
Jxi ẋ = −Jqi q̇i (31)
 T
where q̇i = l˙i θ̇i ϕ̇i .
Operating, the link Jacobian matrix associated with the serial chain i being
Ji can be calculated,
−Jqi q̇i = Jxi ẋ → q̇i = − Jqi −1 Jxi ẋ = Ji ẋ (32)
| {z } |{z} |{z}
3×3 3×6 3×6

All velocity relations can be deduced operating Ji . Note that each row of Ji
is associated to a joint variable contained in qi . Hence, if [A]i is defined as the
row operator that extracts the i-th row of a matrix A,
 T
q̇a = Jqa ẋ Jqa = [J1 ]T1 . . . [J6 ]T1
 T
q̇na = Jqna ẋ Jqna = [J1 ]T2 . . . [J6 ]T2 [J1 ]T3 . . . [J6 ]T3
(33)
 A Redundant Dynamic Modelling Procedure ... 53

and combining both active joint Jacobian Jqa and nonactuated joint Jacobian
Jqna , the joint variables Jacobian Jq can be calculated,
   
q̇a Jqa
q̇ = = ẋ = Jq ẋ (34)
q̇na Jqna |{z}
18×6

If only the control coordinates are considered, the control coordinates Ja-
cobian Jqc can be calculated by combining the rows of Jq that are related to
sensorized (vector Ic) joint variables,

q̇c = [Jq ]Ic ẋ = Jqc ẋ → ẋ = Jqc † q̇c (35)

where Jqc † is the Moore-Penrose generalized pseudoinverse of the Jacobian ma-

trix associated with the control coordinates.
The rows not included in Jqc , are associated to passive joint variables that
are not sensorized nor actuated, hence, the passive variable joint Jacobian can
be defined as,
q̇p = Jqp ẋ (36)
The relationship between the velocity of all joint coordinates q̇ and active
ones q̇a is required to calculate the dynamic model of the robot, and can be
calculated based on previously calculated Jacobians,
 T
I3
q̇ = q̇a = T q̇a (37)
Jqna Jqa −1

Finally, if control coordinates are to be considered, the relationship between

velocity of all joint coordinates q̇ and control coordinates q̇c ,
 T
 qc j ∈ qc
e , qi =
q̇ = Tq q̇c , [Tq ]i = j (38)
|{z} J qp p k , qi = q pk ∈ qp
J
18×nc

where 
1, i = j
[ej ]i =
0, i 6= j
54 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

Acceleration Problem. Acceleration relationships can be calculated by dif-

ferentiating the link velocity equation Eq. 32 with respect to time,


q̈i = Jqi −1 Jxi ẍ − Jqi −1 J̇xi + J̇qi Ji ẋ
(39)
q̈i = Ji ẍ + J̇i ẋ

If the procedure defined for velocity relations is followed, the acceleration

Jacobian associated to the joint variables J̇q can be calculated,
   
Jqa J̇qa
q̈ = ẍ + ẋ
Jqna J̇qna (40)
q̈ = Jq ẍ + J̇q ẋ

Similarly, if control coordinates are considered and the associated rows se-
lected from the previous equation,

q̈c = Jqc ẍ + J̇qc ẋ → ẍ = Jqc † q̈c − Jp J̇qc Jp q̇c = Jp q̈c + J̇p q̇c (41)

Finally, considering the time derivative of matrix Tq ,

q̈ = Tq q̈c + Ṫq q̇c (42)

|{z} |{z}
18×nc 18×nc

2.2.4. Dynamic Model

As stated previously, a subsystem approach will be used to calculate the dy-
namic model. Hence, the dynamic model of the moving platform and the limbs
will be calculated independently in terms of their natural coordinates.
The Lagrangian formulation with multipliers will be used, as it provides the
possibility to define the model in terms of any set of generalized coordinates.
For that purpose, the Lagrangian of the robot L has to be calculated in terms of
the kinetic KMP , KS and potential UMP , US energies of each subsystem,

L = (KMP −UMP ) + (KS −US ) (43)

The calculation of these terms will be detailed for the Gough-Stewart plat-
form next,
 A Redundant Dynamic Modelling Procedure ... 55

Kinetic and Potential Energies of the Moving Platform The moving plat-
form is considered a free body rotating and translating in space with mass m p
and a centroidal inertia Ip . Its center of mass (CM) is located in the origin of the
moving reference frame P(u, v, w) (Fig. 1). As the TCP is located in the moving
platform, the mass of the load mc and its inertia Ic have to be considered in its
dynamic model.
This way, if the energy of free motion in space is considered,
1 T
KMP = 2 ṗx ṗx (m p + mc ) + 12 ω T (Ip + Ic ) ω (44)

Taking into account Eq.7,

 
1 T (m p + mc ) I3 03
KMP = 2 ẋ TR T TR ẋ
03 Ic + Ip
(45)
1 T
= 2 ẋ Mx (x) ẋ
| {z }
6×6

being Mx (x) the inertia matrix of the moving platform, defined in terms of x.
The potential energy of the platform is related to its relative position with
respect to the fixed frame,

UMP = (m p + mc ) g z (46)

where g is gravity vector.

Kinetic and Potencial Energies of the Limbs Each of the limbs connecting
the moving platform and the fixed base are structurally identical, being their
parametric model identical. Each serial chain is a polar serial robot, presenting
 T
3 dof associated to the link joint variables qi = li θi ϕi (Fig. 2). As
seen in Fig. 4, the serial chain is composed by two elements: a cylinder of mass
mi1 ,attached using a universal joint to Ai , and a rod, of mass mi2 , which has a
prismatic motion with respect to the cylinder, and is attached to point Bi .
The center of mass (CM) of each cylinder is calculated in terms of the non-
actuated joint variables,
 
lci1 cosθi cos ϕi
CMi1 = ai +  lci1 sin θi cosϕi  (47)
lci1 sinϕi
56 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

Figure 4. Limb parameters.

Differentiating this expression with respect to time, the linear velocity of the
CM of each cylinder can be calculated,

∂CMi1
vi1 = q̇ = Evi1 q̇ (48)
∂q
where Evi1 is the Jacobian that relates the linear velocity of the CM to the deriva-
tives of the joint variables.
The angular velocity of the CM of the cylinder, ωi1 , can be calculated by
projecting the velocity vectors associated to the nonactuated joint variables θ̇i
and ϕ̇i in a moving frame attached to the CM of the cylinder,
 T
ωi1 = θ̇i sinϕi −ϕ̇i θ̇i cos ϕi (49)

If the terms are rearranged in terms of q,

ωi1 = Eωi1 q̇ (50)

where Eωi1 is the Jacobian matrix that relates the angular velocity to the deriva-
tives of the joint variables.
 A Redundant Dynamic Modelling Procedure ... 57
()*+,-./ 01*+,-./

! #

%
$ & $
# %
& ! $

’ %

" ’ "

Figure 5. Cylinder rotation.

Once defined the velocities for the cylinder, the kinetic energy expression
can be calculated,
1 T
1
Ksi1 = q̇ mi1 Evi1 T Evi1 + Eωi1 T Ii1 Eωi1 q̇ = q̇T Msi1 q̇ (51)
2 2
where Msi1 is the inertia matrix of the cylinder.
The kinetic energy of the rod can be calculated following the same proce-
dure. First, the position of the CM is calculated in the fixed frame,
 
(li − lci2 ) cos θi cos ϕi
CMi2 = ai +  (li − lci2 ) sin θi cos ϕi  (52)
(li − lci2 ) sin ϕi

Differentiating this expression with respect to time, the linear velocity of the
CM of each rod can be calculated,
∂CMi2
vi2 = q̇ = Evi2 q̇ (53)
∂q
where Evi2 is the Jacobian that relates the linear velocity of the CM to the deriva-
tives of the joint variables.
On the other hand, as the relative motion of the rod with respect to the
cylinder is a pure translation, the angular velocity of the rod will be the same as
the one of the cylinder,
ωi2 = ωi1 = Eωi2 q̇ (54)
where Eωi2 = Eωi1
58 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

Hence, the kinetic energy of the rod,

1
1
Ksi2 = q̇T mi2 Evi2 T Evi2 + ETωi2 Ii2 Eωi2 q̇ = q̇T Msi2 q̇ (55)
2 2
The potential energy of both the cylinder and the rod can be calculated con-
sidering the z coordinate of their CM position,
Usi1 = mi1 g lci1 sin ϕi
(56)
Usi2 = mi2 g (li − lci2 ) sinϕi
Thus, the total kinetic and potential energies of this subsystem is calculated
considering all limbs,
6 6
1 T
KS = ∑ Ks i1 + ∑ Ksi2 =
2
q̇ Ms q̇
i=1 i=1
6 6 6 (57)
US = ∑ Us i1 + ∑ Usi2 = ∑ [mi2 (li − lci2 ) + mi1 lci1 ] g sin ϕi
i=1 i=1 i=1
6
where Ms = ∑ (Ms i1
+ Msi2 ) is the total inertia matrix of the serial chain sub-
i=1
system.

2.3. Dynamic Model

Once defined all energies, the Lagrangian of the robot can be calculated,
L = (KMP −UMP ) + (KS −US ) = LS (q, q̇) + LMP (x, ẋ) (58)
Note that the Lagrangian associated to the serial chains is defined only in
terms of the joint variables, while the Lagrangian of the moving platform is
defined in terms of the output coordinates. These coordinates sum a total of
n + nq = 24. However, only 6 are independent, as the robot provides only 6
dof. Hence, a set of 18 constraints have to be applied to solve the model. By
using the Lagrangian Multiplier formulation, these constraints can be implicitly
introduced by using a set of 18 multipliers.
This way, if the formulation is applied to each subsystem, two vectorial
equations can be obtained,
  nq
d ∂LS ∂Ls ∂Γi (x, q)
− = ∑ λi + τq j , j = 1, . . .18
dt ∂q̇ j ∂q j i=1 ∂q j
  nq (59)
d ∂LMP ∂L p ∂Γi (x, qi )
− = ∑ λi , k = 1, . . .6
dt ∂ẋk ∂xk i=1 ∂xk
 A Redundant Dynamic Modelling Procedure ... 59

where τq j are the virtual torques associated to each joint variable in q, if all
were actuated, and Γi (x, q) is the vectorial loop equation for each kinematic
loop (Eq.12).
Operating on the left part of the equation system, and rearranging, the clas-
sical structured form of the dynamics of robots can be defined,
 
∂Γ(x, q) T
Dx ẍ + Cx ẋ + Gx = λ
 ∂x T (60)
∂Γ(x, q)
Ds q̈ + Cs q̇ + Gs = λ + τr
∂q
where Dx , Cx and Gx are the inertia, Coriolis and gravity terms associated with
the mobile platform, and defined in terms of x, Ds , Cs and Gs are the inertia,
Coriolis and gravity terms associated withthelegs, defined in terms of q, λ is
the vector of the Lagrange multipliers and τr j j=1,...18 = τr .
Combining both equations, Lagrange multipliers can be eliminated,
 T " T #†
∂Γ ∂Γ
τr = Ds q̈ + Cs q̇ + Gs − (Dx ẍ + Cx ẋ + Gx ) (61)
∂q ∂x
| {z } | {z }
18×18 18×6

and considering that Γ(x, q) = 018×1 is the set of all constraints of the robot (Eq.
16), the constraint Jacobian JC = ∂x/∂q can be defined,

τr = Ds q̈ + JCT Dx ẍ + Cs q̇ + JCT Cx ẋ + Gs + JCT Gx (62)

However, as previously stated, τr is the set of virtual torques associated with
all joint variables in q. In parallel robots, only the actuated set qa has an actuator
attached and can provide torque, hence, the model needs to be projected to the
actuated space using the Jacobian defined in Eq. 37 [29],
τqa = TT τr = TT Ds q̈ + TT JCT Dx ẍ + TT Cs q̇ + TT JCT Cx ẋ + TT Gs + TT JCT Gx (63)

where it can be easily demonstrated that TT JC T = [Jqa −1 ]T ,

 T  T  T
τqa = TT Ds q̈ + Jqa −1 Dx ẍ + TT Cs q̇ + Jqa −1 Cx ẋ + TT Gs + Jqa −1 Gx
(64)

The previous equation defines the dynamic model of the robot in terms of
all variables. If the redundant data is to be used to feed the dynamic model, the
60 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

model can be rewritten in terms of qc in compact form using the velocity and
acceleration expressions defined previously,

τqa = Dq̈c + Cq̇c + G (65)

where  T
D = TT Dq Tq + Jqa −1 Dx Jp

 T

C = TT Dq Ṫq + Cq Tq + Jqa −1 Dx J̇ p + Cx Jp
 T
G = TT Gq + Jqa −1 Gx
which defines the dynamic model of the robot in terms of any set of control
variables qc , which can include any set of sensorized variables. Note that the
matrices of the model are not regular, as the model is fed with more information
than strictly needed. This will be the basis of the Extended control approaches
analyzed in further sections.

2.4. Model Validation

In order to validate the approach, a particular case will be analyzed. The works
of Wang and Gosselin [43], Tsai [41] and Guo and Li [14] have stablished a
well-documented study case that provides all the required data for proper model
validation. Hence, the redundant model results will be compared with the ones
of these works in order to validate the approach.
The particular Gough-Stewart platform parameters used are shown in Table
1, and have been extracted from the aforementioned works.
The reference trajectories proposed by these authors are detailed next. The
first one is a constant orientation trajectory, where α = β = γ = 0 and
 
−1.5 + 0.2 sinωt
px =  0.2 sinωt 
1 + 0.2 sinωt

where ω = 3 rad/s and ωt = [0, 2π].

The second trajectory is a constant position trajectory px =
 T
−1.5 0 1 , where platform rotates with respect to the fixed z axis:
α = β = 0 and γ = 0.35 sinωt with ω = 3 rad/s and ωt = [0, 2π].
Using the dynamic model detailed in Eq. 65 the torques τqa required to ex-
ecute these trajectories have been calculated for the case in which all joint vari-
ables are sensorized, this is, qc = q. The obtained results can be seen in Figs.
 A Redundant Dynamic Modelling Procedure ... 61

Table 1. Dynamic and kinematic parameters (i = 1, . . .6 limbs)

Parameter value
 T
a1 −2.12 1.374 0 (m)
 T
a2 −2.38 1.224 0 (m)
 T
a3 −2.38 −1.224 0 (m)
 T
a4 −2.12 −1.374 0 (m)
 T
a5 0 −0.15 0 (m)
 T
a6 0 0.15 0 (m)
 T
d1 0.17 0.595 −0.4 (m)
 T
d2 −0.6 0.15 −0.4 (m)
 T
d3 −0.6 −0.15 −0.4 (m)
 T
d4 0.17 −0.595 −0.4 (m)
 T
d5 0.43 −0.445 −0.4 (m)
 T
d6 0.43 0.445 −0.4 (m)
li1 0.5 (m)
lcli 1 0.25 (m)
li2 0.5 (m)
lcli2 0.25 (m)
mp 1.5 (kg)
mc 0 (kg)
mi1 0.1 (kg)
mi2 0.1 (kg)
Ip 0.08 I3 ( kg m2 )
Ii1 0.00625 I3 (kg m2 )
Ii2 0.00625 I3 (kg m2 )
Ic 03 ( kg m2 )

6-7, and are identical to those obtained by Wang and Gosselin [43], Tsai [41]
and Guo and Li [14] . Hence, for the same reference trajectories and model
parameters, the redundant dynamic model yields the same results as those ob-
tained by the cited authors. Moreover, any set of qc that includes at least the
actuated variables qa yields the same result.
62 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.
18
τ
1
τ
2

16 τ
3
τ
4
τ
5
14 τ
6

12
Forces (N)

10

0 0.5 1 1.5 2 2.5

Time(s)

Figure 6. Forces for the constant orientation trajectory.

3. Control
3.1. Position Control in Parallel Robots
In the control area of parallel robots, the main trend has been to import directly
the control approaches from serial robots to parallel robots without considering
the specific features of the latter ones. This way, most of the position control
related works in the literature fall into two categories: local feedback control
approaches and model based control approaches.
Local feedback control approaches are based on the use of a local PID based
controller to control each joint independently from the rest of the mechanism.
In this group several approaches can be found: simple Proportional-Integral-
Derivative (PID) loop based schemes [7, 16, 36], joint dynamic model based
approaches [6,13], Proportional-Derivative (PD) plus gravitation compensations
schemes [12,39] or nonlinear PID based control laws [30,38] fall into this group.
In general, these approaches are easy to implement due to their simple control
law. However, as they do not consider the dynamic coupling between the joints,
they do not provide good control performance when high speed and accuracy
 A Redundant Dynamic Modelling Procedure ... 63

τ
11 1
τ
2
τ3
τ4
10
τ5
τ6

✁
Forces (N)

0 0.5 1 1.5 2 2.5

Time(s)

Figure 7. Forces for the constant position trajectory.

are required.
Model based approaches, on the other hand, require the dynamic model of
the parallel robot to define the control law and provide better performance when
high speed, accelerations and accuracy are needed. One of the most extended
approach is the Computed Torque Control (CTC) scheme [1, 8, 11]. However,
obtaining an accurate dynamic model is a challenging task, as simplifications
are usually made and some phenomena, such as friction or backlash, are not
considered. Thus, in order to ensure the dynamic performance of the controller
in presence of model parameter uncertainties, some authors have also proposed
robust [13, 17, 19, 21] or adaptive control [15, 33] approaches.
From the results offered by the works in the literature, model-based schemes
seem to be the best approach when high-speed and accuracy requirements must
be met in parallel robots. However, the correct calibration of the dynamic model
is difficult, as parameter identification errors arise, and their computational cost
is higher than the simpler, local feedback controllers.
64 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

3.2. The Extended CTC Approach

An interesting approach to increase control performance in parallel robots is the
introduction of extra sensors in the nonactuated joints of the mechanism, so that
the redundant data can be used to minimize the effect of parameter uncertain-
ties and to implement a more efficient redundant control law in parallel robots.
Although this approach increases the cost of the robot, the advantages of using
redundant sensor data have been demonstrated by authors such as Merlet [25],
Baron and Angeles [3], Marquet et al [24], and Bauma et al [4].
The Extended CTC approach is based on this idea [45]. This position con-
trol scheme is based on a modified Computed Torque Control approach that
makes use of the redundant dynamic model defined in Section 2.2. This way,
the Extended CTC approach combines the performance of the classical CTC
with the robustness of sensor redundancy, providing better trajectory tracking
than the classical nonredundant CTC approach.
The control law is defined as (Fig. 8),
τqa = D̂(qc , qp , x) (q̈cd + Kv ėqc + Kp eqc ) + Ĉ(qc , qp, x, q̇c, q̇p, ẋ) q̇c + Ĝ(qc , qp, x)
(66)
where Kp and Kv are the position and velocity matrix gains, eqc = qcd − qc is
the tracking error of the control coordinates and ėqc its time derivative, qcd is
the reference of the control coordinates, and D̂, Ĉ, Ĝ are the identified matrices
of the dynamic model defined in Eq. 65
This way, all the measurable variables contained in qc are used to fed the
model, and in case of errors in model parameters, the controller tries to minimize
not only the errors in active joints, but also the sensorized ones, achieving better
dynamic performance.

3.3. Extended CTC Control Validation

The Gough-Stewart platform defined in Section 2.2 is used to validate the per-
formance of the Extended CTC approach over the classical CTC scheme. For
that purpose, the redundant dynamic model defined in Eq. 65 is calculated for
the four different sensor configurations detailed in Table 2. As it can be seen,
Configuration 1 corresponds to the classical CTC approach in which only the
active joints are sensorized, while Configuration 4 corresponds to the case in
which all joint variables are measured.
A complex, high speed helicoidal trajectory defined in the non-singular
workspace of the Gough Platform (Fig. 9) is defined to compare the dynamic
 A Redundant Dynamic Modelling Procedure ... 65

!678398:;!68- <=!>:- ?’8@

2
;!68- <=!>:
- ?’8@

!"#$%&$’(")* +,- ./$0

1 3.4.5

1 2

Figure 8. Extended Computed Torque Control.

Table 2. Extended CTC Configurations

Config. qc
1 qc = qa ∈ R6
 T
2 qc = qa T θ1 θ3 ϕ1 ϕ3 ∈ R10
 T  T
3 qc = qa ϕT ∈ R12
4 qc = q ∈ R 18

performance of the approach. This trajectory, defined in Eq. 67, allows the
controller to be tested in a wide workspace area.
 
0.4 cos(10t)

 0.4 sin(10t) 

 1.2 + 0.1592t 
xref = 
  (67)
 10 π/180 sin(12t) 

 10 π/180 cos(12t) 
10 π/180 sin(12t)
The model parameter set for the comparative analysis is the one detailed in
66 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

1.5

1
z(m)

0.5

0.5
0

0 -0.5

-1
-0.5
-1.5
y(m)
-1
-2 x(m)

Figure 9. Helicoidal Reference Trajectory.

Table 1 where, in addition, a load of mass mc = 1 kg and inertia Ic = 0.06 I3

kg m2 has been attached to the platform. Moreover, a friction term Fr is in-
troduced into the direct dynamic model to simulate the dynamics of the robot,
which is not considered in the control law,
Fr = TT Fv q̇ + TT Fc sign(q̇) (68)
where Fv is the viscous friction term and Fc is the Coulomb term. The Fv
terms associated with the active joints qa are assigned a value of 0.83 kg m/s,
and those associated with the non-actuated joints are assigned a value of
0.09 kg m/s. Further, the terms of Fc associated with the active joints qa are
0.42 N , while those associated with the non-actuated joints are 0.05 N.
The proportional Kp and velocity Kv gains have been tuned to obtain a max-
imum overshoot of 5% and a peak time of 0.05 s in the classical CTC controller.
The same gains are used for all controllers, and the cycle time is set to 5 ms.
In order to carry out the comparative analysis, the Integral of the Absolute
Error (IAE) performance index of the end-effector positioning and orientation
 A Redundant Dynamic Modelling Procedure ... 67

error, defined as 0∞ |e(t)|dt, will be analyzed. Specifically, for the proposed

R

reference trajectories the real end-effector coordinates x are measured and com-
pared with the desired ones in order to calculate the tracking accuracy. In order
to show the effect of parameter uncertainty in the performance of the controllers,
the real robot parameters have been randomly modified from their nominal val-
ues by a maximum value of 0.1 mm in distance parameters, 0.01 kg in mass
parameters and up to the 10% in inertia terms. Five parameter cases have been
considered, from 0% to 100% of theses maximum values. In order to obtain
statistically relevant data, each simulation iteration and case has been repeated
350 times using different parameter variations within the desired bounds and
the average dynamic performance is considered as representative for a given
controller.
Results are summarized in Figs. 10 and 11. As it can be seen, the IAE
error index increases with the parameter variation percentage, this is, errors
increase as uncertainties arise. However, the relative increase is much lower
when redundant data is used (Configurations 2, 3 and 4).
!"(
+,-./01#1
+,-./01$1
+,-./01%1
!"’
+,-./01&1

!"&
&’
!" #$%

!"%

!"$

!"#

!"!
!) $’) ’!) *’) #!!)
( ) *+) , -. / ) $01$2 . 34$( . *. &) -) *$#**0*

Figure 10. Average positioning IAE.

The relative performance also varies if different sensor configurations are

considered. For instance, it seems logical to think that the more sensors used,
the better the performance will be. However, this is not always straightforward,
as Configuration 2 and 3 have very similar performance with different sensor re-
quirements. Thus, it is clear that it is necessary to identify an appropriate sensor
distribution in order to achieve the best performance and exploit the potential of
the Extended CTC approach.
68 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.
!"$$
*+,-./0!0
*+,-./0#0
!"$& *+,-./0%0
*+,-./0’0

!"’$
&’ ( )
!" #$%

!"’&

!"%$

!"%&

!"#$
&( #$( $&( )$( !&&(
* +&, +- .’ / +$01$2 ’ 34$* ’ &’ 5 +.+&$#&&0&

Figure 11. Average orientation IAE.

As a more detailed example, is shown in Figs. 12 and 13 , where the error

over time associated with the two TCP locations is shown for a case in which
a helicoidal trajectory is followed and maximum errors are introduced in the
nominal parameters. Configuration 1 and 4 are compared, being Configuration
1 the equivalent of the classical CTC approach and Configuration 4 the one with
all joint variables sensorized. If errors in the joint variables are considered, it
can be seen that the classical approach provides better tracking in the active
joints, as the controller tries to minimize these errors, while the Extended CTC
seems to present higher errors in the sensorized joints, as errors are distributed
among the measured variables. However, this error distribution provides better
results in the task space, where Configuration 4 presents a smaller trajectory
tracking.
Thus, in summary, it has been demonstrated that the use of the redundant
Extended CTC configurations can provide better performance than the classical
CTC approaches even in the presence of model parameter errors and nonlinear
phenomena such as friction. However, it is clear that in order to achieve the best
results the sensor configuration has to be selected with care.
 A Redundant Dynamic Modelling Procedure ... 69
’ ($!
ï& (0 123*4($(56 7 8! "! ! ! 9#: $; ï(0 123*4(<(56 7 8! "! ! ! <! =; # ’ ($!
ï& (0 123*4($(56 7 8! "! ! ! : $&9ï(0 123*4(<(56 7 8! "! ! ! ; =; %&
%"# ( >123*4($ < (
>123*4(<
% %

, _(-ABC/
, ’ (-+ /

$"# !

$ ï%

! "# ï<

! ( ï= (
! ! "# $ $"# % %"# & &"# ! ! "# $ $"# % %"# & &"#
) *+ , (-. / ) *+ , (-. /
’ ($! (0 123*4($(56 7 8; "$=; : , ï! #ï(0 123*4(<(56 7 8<"#%&%, ï! # (0 123*4($(56 7 8! "! ! &%9: =ï(0 123*4(<(56 7 8! "! ! ! =: #$;
ï& ï&
’ ($!
% ( : (

=
$

, `(-ABC/
, (-+ /

<
!
?

%
ï$
!

ï%( ï%(
! ! "# $ $"# % %"# & &"# ! ! "# $ $"# % %"# & &"#
) *+ , (-. / ) *+ , (-. /
’ ($!
ï& (0 123*4($(56 7 8! "! ! ! $$&! $ï(0 123*4(<(56 7 8%"$=#; , ï! # ’ ($!
ï& (0 123*4($(56 7 8! "! ! $: %; ; ï(0 123*4(<(56 7 8! "! ! $&: ! 9
$"# ( = (

$ <

, a(-ABC/
, @(-+ /

! "# %

! !

ï! "# ï%

ï$ ( ï< (
! ! "# $ $"# % %"# & &"# ! ! "# $ $"# % %"# & &"#
) *+ , (-. / ) *+ , (-. /

Figure 12. Time evolution of the TCP position and orientation error.

( )$!
ï& )2 345+6)$)78 9 :! "! ! ! $; ’ ! ’ ï)2 345+6)’ )78 9 :! "! ! ! #! <=; ( )$!
ï& )2 345+6)$)78 9 :! "! ! ! %! ##%ï)2 345+6)’ )78 9 :! "! ! ! #! ; &>
’ ) ?345+6)$ % )

& ?345+6)’ $
%
!
%
$
1

-1

$
-

ï$
!

ï$ ï%

ï%) ï& )
! ! "# $ $"# % %"# & &"# ! ! "# $ $"# % %"# & &"#
* +, - )./ 0 * +, - )./ 0

( )$!
ï& )2 345+6)$)78 9 :! "! ! ! %#=#<ï)2 345+6)’ )78 9 :! "! ! ! %&>#;
% ) ( )$!
ï& )2 345+6)$)78 9 :! "! ! ! &>; >>ï)2 345+6)’ )78 9 :! "! ! ! ’ >=>#
% )

$ $

!
-1
’
&
1

!
-

ï$

ï$ ï%

ï& )
! ! "# $ $"# % %"# & &"#
ï%)
! ! "# $ $"# % %"# & &"# * +, - )./ 0
* +, - )./ 0

( )$!
ï& )2 345+6)$)78 9 :! "! ! ! ’ #<$$ï)2 345+6)’ )78 9 :! "! ! ! ’ <>; = ( )$!
ï& )2 345+6)$)78 9 :! "! ! ! ’ ’ $’ ; ï)2 345+6)’ )78 9 :! "! ! ! =’ #$>
& ) ’ )

%
%
$
>
#
1

-1

! !
-

ï$
ï%
ï%

ï& ) ï’ )
! ! "# $ $"# % %"# & &"# ! ! "# $ $"# % %"# & &"#
* +, - )./ 0 * +, - )./ 0

Figure 13. Time evolution of joint variables error.

70 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

4. Sensitivity
4.1. Sensitivity-Based Approach to Evaluate Sensor Configurations
One of the central issues when implementing redundant sensor based ap-
proaches such as the Extended CTC is determining the number and location
of the extra sensors. This is an important step in the design of the mechanism,
as the introduction of extra sensors is not a trivial task. Hence, determining
how many sensors are needed and where are they located is very useful when
designing passive joints.
Note that the Extended CTC is defined in a general way in terms of qc .
Thus, multiple sensor configurations can be used to implement it. Depending
on the complexity of the mechanism, the number of possible combinations can
increase significantly.
In order to evaluate in an easy way the different sensor configurations for a
given platform and task, a sensitivity based procedure is proposed. This analysis
allows to determine the influence of the variation of each model parameter in
the position of the TCP. Thus, if the sensitivity of all parameters for a given
Extended CTC configuration is calculated, a measure of the relative robustness
to model parameter uncertainties can be obtained. Hence, if different Extended
CTC configurations are analyzed, an insight into the relative robustness of each
configuration can be calculated, so that the most suitable sensor distribution can
be selected for a given robot and trajectory.
The measure of the sensitivity of a sensor configuration is based on the
procedure proposed in [42] and modified to suit the needs of parallel robots.
This procedure is based on the calculation of the sensitivity function ξxθk that
calculates the deviation of the TCP δx for the Extended CTC approach along a
nominal trajectory Pd when the parameter θk is uncertain. This is calculated as
the difference between the deviation of the task coordinates in the nominal case,
δx, and the deviation of the task coordinates when the parameter θk varies from
its nominal (or ideal) value, δxθk ,

ξxθk = δx − δxθk (69)

where the calculation of both δx and δxθk for a given trajectory and Extended
CTC configuration requires the linearization of the closed loop dynamics along
the nominal trajectory Pd .
 A Redundant Dynamic Modelling Procedure ... 71

Hence, if the dynamics of the robot (Eq. 65) and the Extended CTC control
law (Eq. 66) are considered,

D q̈c + h = D̂ (q̈cd + Kv ėq + Kp eq ) + ĥ (70)

where D̂(x̂, q̂p , qc ) is the estimated inertia matrix and ĥ =

ˆ q̇ˆ p , q̇c ) q̇c + Ĝ(x̂, q̂p , qc ) groups the estimated Coriolis and
Ĉ(x̂, q̂p , qc, ẋ,
gravity terms of the dynamic model implemented in the Extended CTC
controller. As these matrices are estimated, their calculation will be made in
terms of uncertain parameters, whose value will vary from the nominal value.
On the other hand, the inertia matrix D and the nonlinear vector h characterize
the real dynamics of the robot, whose calculation is done using the nominal
parameters.
In order to calculate the linearized dynamics of the closed-loop system, first
the left term of Eq. 70 will be linearized considering a nominal trajectory Pd ,
  
∂D q̈c  ∂D q̈c  ∂h 
δτ = D(x, qp , qcd ) δq̈c +  δx + ∂q  δq + ∂x  δx
  ∂x Pd  Pd Pd
(71)
∂h  ∂h  ∂h 
+  δq + δẋ + δq̇
∂q Pd ∂ẋ Pd ∂q̇ Pd

If the linearized model is to be defined in terms of the control coordinates

qc , the velocity Jacobians detailed in Section 2.2.3 can be used to project the
model,
δτ = Am δq̈c + Bm δq̇c + Cm δqc (72)
where,
Am = D|Pd 
∂h  ∂h 
Bm = Jp + Tq
∂ẋ Pd ∂q̇ Pd
 
Cm = ∂D∂xq̈c  Jp + ∂D∂qq̈c  Tq
 
Pd  Pd
∂h 
∂h 
+ ∂x  Jp + Tq
Pd ∂q Pd
∂h  ∂h 
+  J̇p + Ṫq
∂ẋ Pd ∂q̇ Pd
where Eq. 72 is defined in terms of the nominal parameters of the model, i.e,
the real ones.
72 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

If the same procedure is applied to the right term of Eq. 70,

δτ = Âc δu + B̂c δq̇c + Ĉc δqc (73)

where δu = −Kv δq̇c − Kp δqc [42]. Note that Eq. 73 is defined in terms of the
uncertain parameters of the model.
Combining the linearized Eqs. 71 and 73, the linearized closed loop dynam-
ics can be calculated as,

0 = A δq̈c + B δq̇c + C δqc (74)

where A = Am, B = Bm − B̂c + Âc Kv and C = Cm − Ĉc + Âc Kp .

In order to calculate the deviation of the nominal TCP δx pose required to
analyze sensitivity, the kinematic relations of Section 2.2.3 can be used,

δqc = Jp † δx = Jqc δx (75)

Hence, Eq. 74 can be rewritten as,



δẍ = (−A Jqc )−1 2 A J̇qc + B
Jqc δẋ
(76)
+ A J̈qc + B J̇qc + C Jqc δx

that defines an ODE system, that can be solved for an initial δx0 and an arbitrary
nominal reference trajectory Pd (t). If the nominal parameters are used, Eq. 76
will be used to calculate δx, while if a parameter θk is modified, i.e. is uncer-
tain, the modified parameter will be used to calculate Âc , B̂c and Ĉc, and this
expression will be used to calculate δxθk . Then, applying Eq. 69, the sensitivity
function over the reference trajectory will be obtained.
Note that the absolute value of the sensitivity function ξxθk will depend on
the selection of the initial δx0 used to integrate it. Thus, the analysis of the
normalized sensitivity function is proposed ξxθk /δx0 . This way, the characteristic
sensitivity value will be considered as the maximum of the absolute value of the
normalized sensitivity function.

4.2. Sensitivity Based Extended CTC Sensor Configuration

Selection
In order to validate the sensitivity based Extended CTC sensor configuration
selection methodology, the Gough-Stewart platform analyzed in Section 2.2 will
be used as a study case.
 A Redundant Dynamic Modelling Procedure ... 73

In Table 3, the ten different sensor configurations to be evaluated are shown,

being Config. 1 the traditional CTC approach with no extra sensors, and Config.
10 the one in which all joints are sensorized. In order to compare the sensitivity
results of these configurations, the controller matrices Kp and Kv have been
tuned to obtain an overshoot of 5% and a peak time of 0.1s for all configurations.
The nominal values of the robot are those detailed in Table 1, considering a
maximum identification error of 1mm for lengths, 10g for masses and a 10%
variation of the inertia nominal value.

Table 3. Extended CTC configurations to be evaluated

Config 1 q =q
 c T a T T
Config 2 qc = qa θ
 T T
Config 3 qc = qa ϕT
 T
Config 4 qc = qa T [θ]T1,3,6 [ϕ]T1,3,6
 T
Config 5 qc = qa T θ1 θ3 ϕ1 ϕ3
 T
Config 6 qc = qa T θ1 θ3 ϕ1
 T T
Config 7 qc = qa θ1 θ3 ϕ3
 T
Config 8 qc = qa T θ1 θ3 θ6
 T
Config 9 qc = qa T ϕ1 ϕ3 ϕ6
Config 10 qc = q

As the proposed methodology requires a nominal trajectory to be evaluated,

the helicoidal trajectory detailed in Eq. 67 (Fig. 9) will be used.
The sensitivity related to orientation and position coordinates will be evalu-
ated separately. For that purpose, the following expressions will be used,
q
ξPOS
θk = (ξx )2 + (ξyθk )2 + (ξzθk )2
q θk (77)
ξORI = (ξα )2 + (ξβ )2 + (ξγ )2
θk θk θk θk

This way, the characteristic sensitivity value of each Extended CTC config-
uration will be calculated as the sum of the absolute maximum sensitivity values
of all model parameters when an initial TCP pose deviation of δx0 = 0.001 m is
considered.
74 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

Results are shown in Figs. 14 and 15. As it can be seen, the sensitivity
of the Extended CTC approach varies depending on the sensor distribution, as
seen previously. However, it can be seen that the dynamic performance and ro-
bustness of the use of redundant data is higher than the classical CTC approach
(Config. 1).
The appropriate selection of the Extended CTC sensor configuration can
be carried out applying different criteria. This way, if economic criteria are
considered, a less number of sensors will reduce the cost of the robot. In these
cases, Config. 3 and Config 4. seem to be the best alternatives considering
the number of sensors vs performance criteria. However, if the robustness is
to be maximized, the full sensorization of the robot (Config. 10) is the best
alternative.
As it can be seen, it is shown that the characteristic sensitivity value calcu-
lated for the Extended CTC approach varies depending on the selected sensor
configuration, illustrating the need of a preliminary analysis to select the most
suitable sensor configuration for a given robot and task.
!"!!*

!"!!)

!"!!(
3,4/5/,-167-4/5/8/59

!"!!’

!"!!&

!"!!%

!"!!$

!"!!#

!"!!!
+,-./01# +,-./01$ +,-./01% +,-./01& +,-./01’ +,-./01( +,-./01) +,-./01* +,-./012 +,-./01#!

 
 
Figure 14. Normalized position Sensitivity ∑ max ξPOS
θk /δx 0 .

Conclusion
Control is a central issue in parallel robos, as these mechanisms are usually
used in high performance tasks where heavy load handling, precision and/or
speed are required. However, being a recently rediscovered field, there are still
many areas in parallel robotics that need further research.
 A Redundant Dynamic Modelling Procedure ... 75
!"!)

!"!(

!"!’
34.5,676.+,0859 :.6.;.6<

!"!&

!"!%

!"!$

!"!#

!"!!
*+,-./0# *+,-./0$ *+,-./0% *+,-./0& *+,-./0’ *+,-./0( *+,-./0) *+,-./01 *+,-./02 *+,-./0#!

 
 
Figure 15. Normalized orientation Sensitivity ∑ max ξORI
θk /δx 0 .

In this paper, the use of extra sensors in passive joints of parallel robots
is proposed for position control purposes. The extra data provided by these
sensors can reduce the computational cost of the kinematic and dynamic model,
increase the robustness of the controller and provide better dynamic behaviour.
These advantages are demonstrated in the Extended CTC approach, a con-
trol scheme that combines the advantages of redundant sensors and advanced,
model-based controllers. In this paper, the three steps for implementing this
approach are analyzed briefly: the calculation of a generic, redundant, recon-
figurable dynamic model, the definition of the control law, and its implementa-
tion. Moreover, a procedure to select the most suitable configurations is detailed
based on a sensitivity approach. A study case based on the Gough-Stewart plat-
form is proposed.
The procedures and control laws defined in this paper can be used in those
applications where high performance criteria need to be met, such as medi-
cal applications or medium load pick-and-place tasks. Based on these ideas,
new control approaches can be derived, modifying the formulation to consider
sensors that measure other magnitudes such as acceleration or force. Further
research is being carried on focusing in industrial applications such as test rigs
for vehicle suspension based on parallel robots.
76 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

References
[1] H. Abdellatif and B. Heimann. Computational efficient inverse dynamics
of 6-dof fully parallel manipulators by using the lagrangian formalism.
Mechanism and Machine Theory, 44:192–207, 2009.

[2] J. Angeles. Fundamentals of Robotic Mechanical Systems: Theory, Meth-

ods, and Algorithms. Springer, 2007.

[3] L. Baron and J. Angeles. The kinematic decoupling of parallel manipula-

tors using joint-sensor data. IEEE Transactions on Robotics and Automa-
tion, 16(6):644–651, 2000.

[4] V. Bauma, M. Valášek, and Z. Sika. Increase of pkm positioning accuracy

by redundant measurement. Proceedings of the 5th Chemnitzer Paral-
lelkinematik Seminar, pages 547–564, 2006.

[5] S. Bhattacharya, H. Hatwal, and A. Ghosh. An on-line parameter estima-

tion scheme for generalized stewart platform type parallel manipulators.
Mechanism and Machine Theory, 32(1):79–89, 1997.

[6] C. Brecher, T. Ostermann, and D.A. Friedrich. Control concept for pkm
considering the mechanical coupling between actuator. Proceedings of the
5th Chemnitz Parallel Kinematics Seminar, pages 413–427, 2006.

[7] P. Chiacchio, F. Pierrot, L. Sciavicco, and B. Siciliano. Robust design of

independent joint controllers with experimentation on a high-speed par-
allel robot. IEEE Transactions on Industrial Electronics, 40(4):393–403,
1993.

[8] A. Codourey. Dynamic modeling of parallel robots for computed-torque

control implementation. The International Journal of Robotics Research,
17(12):1325–1336, 1998.

[9] J.J. Craig. Robotica. Prentice Hall, 2006.

[10] B. Dasgupta and T.S. Mruthyunjaya. A Newton-Euler formulation for

the inverse dynamics of the Stewart Platform. Mechanism and Machine
Theory, 33(8):1135–1152, 1998.
 A Redundant Dynamic Modelling Procedure ... 77

[11] I. Davliakos and E. Papadopoulos. Model-based control of a 6-dof elec-

trohydraulic Stewart Gough platform. Mechanism and Machine Theory,
43(11):1385–1400, 2008.
[12] F. Ghorbel, O. Chtelat, R. Gunawardana, and R. Longchamp. Modeling
and set point control of closed-chain mechanisms: Theory and experiment.
IEEE Transactions on Control System Technology, 8(5):801–815, 2000.
[13] H. Guo, Y. Liu, and H. Liu, G. ; Li. Cascade control of a hydraulically
driven 6-dof parallel robot manipulator based on a sliding mode. Control
Engineering Practice, 16(9):1055–1068, 2008.
[14] H.B. Guo and H.R. Li. Dynamic analysis and simulation of a six degree
of freedom Stewart Platform manipulator. Proceedings of IMechE Part C:
J. Mechanical Engineering Science, 220:61–72, 2006.
[15] M. Honegger, A. Codourey, and E. Burdet. Adaptive control of the
Hexaglide, a 6 dof parallel manipulator. IEEE International Conference
in Robotics and Automation, 1997.
[16] B. R. Hopkins and R. L. Williams II. Kinematics, design and control of the
6-PSU platform. Industrial Robot: An international Journal, 29(5):443–
451, 2002.
[17] C-I. Huang, C-F. Chang, M-Y. Yu, and L-C. Fu. Sliding-mode tracking
control of the stewart platform. 5th Asian Control Conference, 1:562–569,
2004.
[18] W. Khalil and O. Ibrahim. General solution for the dynamic modeling of
parallel robots. Journal of Intelligent Robot Systems, pages 19–37, 2007.
[19] D. H. Kim, Yang J-Y., and K-I. Lee. Robust tracking control design for a
6 dof parallel manipulator. Journal of Robotic Systems, 17(10):527–547,
2000.
[20] K.-M. Lee and D. K. Shah. Dynamic analysis of a three-degree-of-freedom
in-parallel actuated manipulator. IEEE Journal of Robotics and Automa-
tion, 4(40):361–367, 1988.
[21] S-H. Lee, J-B. Song, W-C. Choi, and D. Hong. Position control of a Stew-
art Platform using inverse dynamics control with approximate dynamics.
Mechatronics, 13:605–619, 2003.
78 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

[22] D. Li and S.E. Salcudean. Modeling, simulation and control of a hydraulic

stewart platform. IEEE International Conference in Robotics and Automa-
tion, 4:3360–3366, 1997.

[23] E. Macho, O. Altuzarra, C. Pinto, and A. Hernández. Workspaces associ-

ated to assembly modes of the 5R planar parallel manipulator. Robotica,
36(3):395–403, 2008.

[24] F. Marquet, O. Company, S. Krut, and F. Pierrot. Enhancing parallel robots

accuracy with redundant sensors. Proceedings of the 2002 IEEE Interna-
tional Conference on Robotics and Automation, pages 4114–4119, 2002.

[25] J-P. Merlet. Closed form resolution of the direct kinematics of parallel
manipulators using extra sensors data. Proceedings IEEE International
Conference in Robotics and Automation, pages 200–304, 1993.

[26] J-P. Merlet. Still a long way to go on the road for parallel mechanisms.
Proceedings of the ASME 2002 DETC Conference, 2002.

[27] J. P. Merlet. Parallel Robots (Second Edition). Kluwer, 2006.

[28] R. Nakadate, H. Uda, H. Hirano, J. Solis, A. Takanishi, E. Minagawa,

M. Sugawara, and K. Niki. Development of a robotic carotid blood mea-
surement WTA-1RII: Mechanical improvement of gravity compensation
mechanism and optimal link position of the parallel manipulator based
on GA. IEEE/ASME International Conference on Advanced Intelligent
Mechatronics, pages 717–722, 2009.

[29] Y. Nakamura and M. Godoussi. Dynamics computation of closed-link

robot mechanisms with nonredundant and redundant actuators. IEEE
Transactions on Robotics and Automation, 5(3):294–302, 1989.

[30] P.R Ouyang, W.J Zhang, and F.X Wu. Nonlinear PD control for trajec-
tory tracking with consideration of the design for control methodology.
Proceedings of the 2002 IEEE International Conference in Robotics and
Automation, 4:4126–4131, 2002.

[31] F. Paccot, N. Andreff, and P. Martinet. A review on the dynamic control of

parallel kinematic machines: Theory and experiments. The International
Journal of Robotics Research, 28(3):395, 2009.
 A Redundant Dynamic Modelling Procedure ... 79

[32] V Parenti-Castelli and R Di Gregorio. A new algorithm based on two

extra-sensors for real-time computation of the actual configuration of the
generalized stewart-gough manipulator. Journal of Mechanical Design,
122(1):294–298, 2000.
[33] I.T. Pietsch, M. Krefft, O.T. Becker, C. C. Bier, and J. Hesselbach. How
to reach the dynamic limits of parallel robots? an autonomous control
approach. IEEE Transactions on Automation Science and Engineering,
2:369–380, 1995.
[34] J. Sabater, R. Saltarn, R. Aracil, E. Yime, J. Azorin, and M. Hernández.
Teleoperated parallel climbing robots in nuclear installations. Industrial
Robot: An International Journal, 33(5):381–386, 2006.
[35] R Saltaren, R Aracil, P Cardenas, C Pena, E Yime, and C Alvarez. Un-
derwater parallel robot for oceanic measuring and observations-remo i:
development and navigation control advances. Proceedings of OCEANS
2009-EUROPE, pages 1–9, 2009.
[36] S. Stan, M. Manic, M. Maties, and R. Balan. Kinematics analysis, design,
and control of an isoglide3 parallel robot (IG3PR). Proceedings of the
34th Annual Conference of the IEEE Industrial Electronics Society, page
12651275, 2008.
[37] Y Su and B Duan. The application of the stewart platform in large spheri-
cal radio telescopes. Journal of Robotic Systems, 17(7):375–383, 2000.
[38] Y. X. Su, B. Y. Duan, C. H. Zheng, Y. F. Zhang, G. D. Chen, and J. W. Mi.
Disturbance-rejection high-precision motion control of a Stewart Platform.
IEEE Transactions on Control System Technology, 12(3):364–374, 2004.
[39] Y. X. Su, D. Sun, L. Ren, X. Wang, and J. K. Mills. Nonlinear pd syn-
chronized control for parallel manipulators. Proceedings of the 2005 IEEE
International Conference on Robotics and Automation, pages 1374–1379,
2005.
[40] L-W. Tsai. Robot Analysis. Wiley-Interscience, 1999.
[41] L-W. Tsai. Solving the inverse dynamics of a stewart-gough manipulator
by the principle of virtual work. ASME Journal of Mechanical Design,
122:3–9, 2000.
80 Asier Zubizarreta, Itziar Cabanes, Marga Marcos et al.

[42] M. Vukobratovic and M. Filipovic. Dynamic accuracy of robotic mech-

anisms. part 1: Parametric sensitivity analysis. Mechanism and Machine
Theory, 35:221–237, 2000.

[43] J. Wang and C. Gosselin. A new approach for the dynamic analysis of
parallel manipulator. Multibody System Dynamics, 2:317–334, 1998.

[44] C Yang, Q Huang, H Jiang, O Ogbobe Peter, and J Han. PD control with
gravity compensation for hydraulic 6-dof parallel manipulator. Mechanism
and Machine Theory, 45:666–677, 2010.

[45] A. Zubizarreta, I. Cabanes, M. Marcos-Muoz, and Ch. Pinto. Control of

parallel robots using passive sensor data. Proceedings of the IEEE/RSJ
International Conference on Intelligent Robots and Systems, 2008. IROS
2008, pages 2398–2403, 2008.
In: Parallel Manipulators ISBN: 978-1-63485-926-4
Editor: Cecilia Norton, pp. 81-100 © 2016 Nova Science Publishers, Inc.

Chapter 3

STRUCTURE SYNTHESIS OF
FULLY-ISOTROPIC TWO-ROTATIONAL
AND TWO-TRANSLATIONAL PARALLEL
ROBOTIC MANIPULATORS

Yanbin Zhang
School of Mechatronics Engineering,
Henan University of Science and Technology, Luoyang, China

Abstract
Strong kinematic coupling is one of inherent properties for general parallel
robotic manipulators. Although it contributes high stiffness and large loading
capability, it also leads to complicated kinematic solutions, difficult controlling
design and path planning, which will take passive effects for the actual
applications of the parallel manipulators. A new systematic method for structure
synthesis of four-degree-of-freedom fully-isotropic parallel robotic manipulators
is proposed in this paper. The moving platform of the mechanism consists of
two-rotational and two-translational (2R2T) motion components with respect to
the fixed base. Firstly, structure synthesis of uncoupled 2R2T parallel robotic
manipulator is set up based on the concept of hybrid mechanism and the
reciprocal screw theory for the first time. Then, type synthesis of fully-isotropic
2R2T parallel robotic manipulator is performed by changing one kinematical


E-mail address: [email protected]; Address: School of Mechatronics Engineering, Henan
University of Science and Technology, Luoyang, China.
82 Yanbin Zhang

chain’s structure of the uncoupled parallel robotic manipulator obtained before.

A lot of novel uncoupled and fully-isotropic 2R2T parallel robotic manipulators
are synthesized. Kinematical Jacobian of a fully-isotropic 2R2T parallel robot is
a 4×4 identity matrix, so there is a one-to-one linear mapping correspondence
between the output velocity of the platform and the input velocity of the actuated
joints. Therefore, one independent output of the moving platform can be
controlled only by one actuator.

1. Introduction
A parallel robotic manipulator is a typically closed-loop mechanism which is
composed of a moving platform connected to a fixed base by several serial open-
single kinematical chains (or called limbs) in parallel [1-3]. In order to achieve
certain special properties, the serial open-single kinematic chain was replaced by a
hybrid chain [4, 5], in which there is at least one leg composed of a closed loop
structure. Compared with traditional serial mechanisms, parallel manipulators
have many merits, such as high stiffness, large load carrying capacity, high
accuracy and high velocity. Therefore, parallel robotic manipulators have
potential applications in the fields of industrial robots, flight simulators, parallel
machine tools, medical devices, radar antennas and telescopes. Delta Robot [6] is
one of the most famous parallel manipulators and applied successfully in the
industrial fields. However, parallel manipulators also have some drawbacks, such
as, small workspace, high coupling of kinematics and dynamics, difficult control
design and path planning.
Four-degrees of freedom parallel robotic manipulators can be divided into
three types based on the different motion output characteristics of the moving
platform, i.e., one-rotational and three-translational type (1R3T), three-rotational
and one-translational type (3R1T), two-rotational and two-translational type
(2R2T). The former two types have been well studied and many novel
manipulators were designed [7-9]. As for 2R2T parallel manipulators, only a few
of them were reported in literatures. Lu [10] presented a 2R2T parallel
manipulator with three legs, in which one of the legs provides two actuators to
drive the platform. Abbasnejad [11] designed a 2R2T parallel manipulator with
five limbs, in which one limb only restricts the output motion properties of the
platform, but it does not control them.
Strong kinematic coupling is one of inherent properties for general parallel
robotic manipulators. Although it contributes high stiffness and large loading
capability, it also leads to complicated kinematic solutions, difficult controlling
design and path planning, which will take passive effects for the actual
 Structure Synthesis of Fully-Isotropic Two-Rotational … 83

applications of the parallel manipulators. Recently, many scholars have paid more
and more attentions on decoupled parallel robotic manipulators [12-17], which
have simple linear kinematic solutions. The most remarkable characteristic for the
decoupled parallel robotic manipulators is that their Jacobian is a triangular
matrix. The mechanism is called uncoupled parallel robotic manipulator when its
Jacobian is a diagonal matrix [18]. In this case, there exists a one-to-one
relationship between the outputs of the moving platform and the inputs of
actuators. If the Jacobian of a parallel manipulator is an identical matrix, it is
defined as a fully-isotropic parallel robotic manipulator, because the condition
number is constantly equal to 1 throughout the entire workspace. So this kind of
manipulator has well performance in kinematic and dynamic transmission
capabilities. Kong and Gosselin [19] addressed type synthesis of linear
translational parallel manipulators based on the geometrical approach, among
which some are uncoupled parallel manipulators and others are fully-isotropic
parallel manipulators. Kuo and Dai [20] designed a fully-isotropic parallel
orientation manipulator composed of three active legs and one passive spherical
joint. Carricato [21] put forward type synthesis of fully-isotropic 3T1R parallel
manipulators in terms of the screw theory. Planar 2T1R fully-isotropic parallel
manipulators were investigated based on the screw theory and many novel
manipulators were obtained in [22]. Gogu [23, 24] has completed type synthesis
of fully-isotropic parallel manipulators based on linear transformation theory.
Notably, structure synthesis of fully-isotropic 2R2T parallel manipulators has
been performed as well [25].However, each manipulator consists of one complex
leg. This complicated structure probably leads to some problems in order to obtain
the expected motion property and precision.
A new systematic method for type synthesis of the fully-isotropic 2R2T
parallel manipulators is presented based on the concept of hybrid manipulator and
the reciprocal screw theory in this paper. Other contents are arranged as
following: structure design of a fully-isotropic 2R2T hybrid manipulator is
discussed in section 2, there exists a one-to-one corresponding relationship
between the output velocity of the moving platform and the input velocity of the
actuated joints for these mechanisms. So each output motion of the platform is
only controlled by one actuator.
84 Yanbin Zhang

2. Structure Design and Kinematic Analysis of a

Fully-Isotropic 2R2T Hybrid Mechanism
A hybrid mechanism is defined as a mechanism which is composed of a parallel
manipulator connected with a single open kinematic chain or two parallel
manipulators assembled serially. In this paper, suppose each fully-isotropic 2R2T
hybrid mechanism includes one parallel component and one serial component. The
parallel part is a fully-isotropic 2T1R planar parallel manipulator selected from [22]
and the serial part is a single open kinematic chain which only consists of one
revolute joint. A new fully-isotropic 2R2T hybrid mechanism is shown in Figure 1.
The parallel component of the hybrid mechanism is made of three legs. The first leg
is composed of two prismatic pairs and one revolute joint. Their axes are
perpendicular to each other. Thus, the structure of the first leg can be defined as
P1  P2  R 3 from the base to the platform in sequence. The second consists of two
prismatic pairs and three revolute joints. The axes of the first prismatic pair and the
first two revolute joints are parallel to each other and perpendicular to the second
prismatic pair, and the axis of the last revolute joint is perpendicular to the first
prismatic pair, but neither parallel to nor perpendicular to the second pair, i.e.,
P4 / /R 5 ( P6 ) / /R 7  R8 . The third consists of two revolute and two universal
joints. The axes of two revolute joints are orthogonal. Adjacent rotary axes of two
universal joints are parallel and the other axes of them are parallel to the axis of
second revolute joint. The sequence of these joints from the base to the output link
OL is R 9  R10 - U11 - U12 . Here, P, R, C and U denote the prismatic, revolute,
cylindrical and universal (or Hooke) joints, respectively. The assembly conditions
of three legs are as follows: the axes of joints P1 , P4 and R 9 mounted on the base

Meanwhile, their axes are parallel to the x, y and z axes of the fixed frame O  xyz ,
are normal to each other and these joints are selected as the actuated joints.

respectively. Axes of joints R 3 and R 8 are collinear and perpendicular to the

adjacent axis within U12 joint. Joints R 3 , R 8 and U12 are all connected with the
output link OL. When three actuated joints are driven, the output link will translate
along x-, y-axes and rotate around z-axis, i.e., the parallel part is a 3-DOF planar
manipulator. Jacobian of this planar parallel manipulator is a 3×3 identity matrix
[20]. In other words, this mechanism shows fully-isotropic throughout the entire
workspace because the condition number of its Jacobian is identically equal to 1.
 Structure Synthesis of Fully-Isotropic Two-Rotational … 85

M
r
w

P
M
u
v
R7 R13
P Actuator
OL R8
P6 R3 q4
2
U12
P2
R5 3
U11
a
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 1. Fully-isotropic 2R2T hybrid manipulator.

In order to develop a hybrid 2R2T mechanism, a revolute joint, R13 , is

connected to the output link OL of the planar parallel manipulator in serial. The
assembly condition is that the axis of joint R13 is perpendicular to the axis of joint
R 8 (or R 3 ). In the initial installing position, the axis of R13 is parallel to y-axis.
In addition, joint R13 is also an actuated joint and driven by an actuator. Then a
2R2T hybrid mechanism is obtained, shown in Figure 1. Its moving platform
(MP) with four degrees of freedom is connected to joint R13 .
Another coordinate frame (or called moving frame), P  uvw , is attached on
the platform. Origin point P is located on the intersection between the axes of
joints R13 and R 8 , v -axis is aligned with the axis of joint R13 , w -axis is
perpendicular to the plane of the moving platform, u -axis obeys the right-hand
rule.

velocity equation of point P with respect to the fixed frame O  xyz can be
Based on the architecture and assembly conditions of the mechanism, the

derived as follows
86 Yanbin Zhang

 vx   1 0   q1 
 v  0 0   q2 
0 0
 y
 z  0
1 0
0   q3 
(1)
    
v  0
0 1
0 0 1   q4 

Where vx and v y are the linear velocities along x- and y-axes, respectively.
v and  z are the angular velocities around v- and z-axes, respectively. q1 , q2 , q3
and q4 are the generalized input velocities of four actuated joints. Obviously, the
Jacobian is a 4×4 identity matrix, so this hybrid mechanism shows fully-isotropic
too. There exists a linear mapping relationship between the input velocity space
and the output velocity space.

3. Position Analysis
Referring to Figure 1, let point M is arbitrary on the moving platform and
M P  (r,0,0)T is its coordinate with respect to the frame P  uvw . The coordinate
of the origin point P within the fixed frame is PO  ( x, y, z)T  (q1 , q2 , a)T , where q1
and q2 are the input displacements of the actuated joints P1 and P4 , respectively,
a is the structure size denoting the distance from origin P to the fixed plane
determined by x- and y-axes. The coordinate of point M with respect to the fixed
frame can be calculated as follows

MO  RM P  PO (2)

where MO  ( xM , yM , zM )T , R is the rotational matrix from the moving frame to

Assume  and  are the posture angles of the moving platform. Since the
the fixed one.

hybrid mechanism is uncoupled, both posture angle  and  are only relate to
the actuated inputs q4 of joint R13 and q3 of joint R 9 , respectively. Thus, the
rotational matrix R can be derived

cos   sin  0  cos  0 sin   cos  cos   sin  sin  cos  

R   sin  cos  0   0 0    cos  sin  cos  sin  sin  

1 (3)
 0 0 1    sin  0 cos     sin  0 cos  
 Structure Synthesis of Fully-Isotropic Two-Rotational … 87

Substituting Equation (3) into (2), yields

 xM   r cos  cos   q11 

 y    r cos  sin   q 
 M  12  (4)
 zM   r sin   a 

Based on Equation (4), one can see that xM relates to the actuated input q1 of
the first leg and two pose angles  and  of the platform. Similarly, yM is the
function of the variables of q2 ,  and  . However, zM only depends on the
variable of  , which means that if the position coordinate zM is given, pose angle
 will be measured inversely. In other words, if there is another leg which is used
to control coordinate zM of point M independently, one uncoupled parallel
manipulator will be obtained.

4. Structure Synthesis of the Fourth Leg

In order to design the uncoupled 2R2T parallel robotic manipulators, the fourth
leg should be designed to provide single driving force for the platform along
z-axis. Without loss of generality, suppose there are only two kinds of single-DOF
joint in the fourth leg, i.e., prismatic and revolute joints. Structure synthesis of
fully-isotropic translational parallel manipulators was addressed by use of the
reciprocal screw theory in [26]. Research results show that the actuation screw,
which provides the direct actuated force along one moving direction for the
platform, of each leg for the fully-isotropic translational parallel manipulator,
must be a pure force screw. At the same time, the actuated screw of the same leg
should be an infinite-pitch screw and its axis is parallel to the moving direction.
Since the fourth leg is used to control the moving displacement of the
platform along z-axis, the actuation screw $a 4 should be a zero-pitch screw
parallel to z-axis and the actuated screw $4 is an infinite-pitch screw parallel to z-
axis [24], respectively. Therefore,

$a 4 = [0,0,1; Pa 4 , Qa 4 ,0]T (5)

$4 = [0,0,0;0,0,1]T (6)
88 Yanbin Zhang

where Pa 4 and Qa 4 are the position parameters of $a 4 .

Equation (6) suggests that the actuated joint of the fourth leg is a prismatic
pair and its moving direction is along z-axis.
In terms of the reciprocal law, the actuation screw is reciprocal to all
kinematic screws of other joints with the exception of the actuated screw within
the identical leg. Therefore, all the possible active non-actuated screws of the
fourth leg can be listed below:

1. Zero-pitch screws parallel to z-axis. They cannot be directly connected to

the moving platform or through the infinite-pitch screw. There must be at
least one zero-pitch screw and at most three in the fourth leg.
2. Infinite-pitch screws perpendicular to z-axis. They can be placed at any
position in the leg. There at most two infinite-pitch screws and their axes
should not be parallel to each other.
3. Zero-pitch screw. It intersects the axis of the actuation screw $a 4 and is
parallel to the axis of joint R13 . There exists one and only one zero-pitch
screw in the fourth leg.

In addition, the zero-pitch screw, which intersects the axis of the actuation
screw $a 4 and is perpendicular to the axes of the screws defined in steps (1) and
(3), is called idle screw. This kind of screw can be inserted at any place and its
number does not exceed one.
For the uncoupled 2R2T parallel manipulator, it is obvious that the platform
has two rotational degrees of freedom. At the same time, three coordinates of
point M are variable during the moving process, as shown in Equation (4).
Therefore, the terminal link of the fourth leg should have at least three-
translational and two-rotational degrees of freedom with respect to the initial link
(or fixed base). If the connectivity of the fourth leg is equal to six, there must be
one idle revolute joint. Otherwise, a redundant DOF exists.
According to the types and assembly conditions of the actuated joint and non-
actuated screws, all potential kinematic chains of the fourth leg can be enumerated
based on different connectivity (see Table 1). The first joint in each leg is selected
as the actuator. As far as C joint connected to the base is concerned, its linear
DOF will be chosen as the actuated input. Let subscripts, i, j and k, denote the
translational or rotational directions of the corresponding joints and the unit vector
space formed by three directions is linear independent. As for the P-pair following
the subscript n, their moving directions must be perpendicular to u-axis. Let X
represent a joint with an idle rotational DOF in the leg, where its connectivity is
 Structure Synthesis of Fully-Isotropic Two-Rotational … 89

equal to 6. In order to simplify the structure of the kinematic chains, the axes of
the adjacent joints are regarded as parallel or perpendicular to each other. Such as
the kinematic chain Ci R i Uik R j , the first joint is a cylindrical pair parallel to i-axis
and its linear DOF is chosen as the actuated input, the second is a revolute joint
parallel to i-axis too, the third is a Hooke joint with one axis parallel to the i-axis
and another axis parallel to the k-axis, the rotational DOF about the k-axis is idle,
and the last is a revolute joint parallel to the j-axis, which implies the last joint
axis is normal to all others axes in the chain.

Table 1. Kinematic chains of the fourth leg

Chains with complex

Fc Type Order Basic chains
joints
5 1P4R 1-4 PiRiRiRiRj PiRiRiUij; CiRiRiRj
CiRiUij
2P3R 5-16 PiPnRiRiRj; PiRiPnRiRj PiPnRiUij; PiRiPnUij
PiRiRiPnRj; PiRiRiRjPj PiRiRiCj; CiPnRiRj
CiRiPnRj; CiRiRjPj
CiPnUij; CiRiCj
3P2R 17-22 PiPjPkRiRj; PiRiPkPjRj PiPjPkUij; CiPkPjRj
PiRiPkCj; CiPkCj
． ． ． ．
6 1P5R 23-47 PiRkRiRiRiRj; PiRiRkRiRiRj PiUkiRiRiRj; PiRiUkiRiRj
． ． ． ． ．
PiRiRiRkRiRj;PiRiRiRiRkRjPiRiRiRiRjRk PiRiRkRiUij; PiRkRiRiUij
． ．
PiRiRiUkiRj; PiRiRiRkUij
． ．
PiRiRiRiUkj; PiRiUkiUij
． ．
PiUkiRiUij; CiRkRiRiRj
． ．
CiRiRkRiRj; CiRkRiUij
． ．
CiRiRiRkRj; CiUkiRiRj
． ．
CiRiRkUij; CiRiUkiRj
． ．
CiRiRiUkj; CiUkiUij
． ．
PiRiRiS; CiRiS
． ． ． ．
2P4R 48-77 PiPnRkRiRiRj; PiPnRiRkRiRj PiPnRkRiUij; PiCkRiRiRj
． ． ． ．
PiPnRiRiRkRj; PiPnRiRiRjRk PiPnRiRkUij; PiPnRiUkiRj
． ． ． ．
PiRkRiPnRiRj; PiRiRkPnRiRj PiUkiPnRiRj; PiRkRiPnUij
． ． ． ．
PiRiPnRkRiRj; PiRiPnRiRkRj PiRiCkRiRj; PiRiRwPnUij
． ． ． ．
PiRiPnRiRjRk; PiRkRiRiPnRj PiRwRiRiCj; PiUkiRiPnRj
． ． ． ．
PiRiRkRiPnRj; PiRiRiRkPnRj PiRiUkiPnRj; PiRiRkRiCj
． ． ． ．
PiRiRiPnRkRj; PiRiRiPnRjRk PiRiRiPnUjk; PiRiRiCkRj
． ．
CiPnRkRiRj; CiRkRiPnRj
90 Yanbin Zhang

If i-axis is parallel to z-axis of the fixed frame and j-axis is parallel to O  xy

plane, each kinematic chain in the Table 1 can be regarded as the fourth leg of a
2R2T parallel manipulator. It only provides the direct driving force for the
platform along z-axis.

5. Structure Synthesis of Uncoupled 2R2T Parallel

Robotic Manipulators
Uncoupled parallel robotic manipulator is a mechanism, in which a one-to-one
corresponding control relationship exists between the output motion components
of the platform and the inputs of the actuated joints. Each actuator of the
mechanism controls one independent output motion component of the platform.
An uncoupled parallel manipulator has the distinct characteristics that its velocity
Jacobian is a diagonal matrix. Structure synthesis of uncoupled 2R2T parallel
manipulators can be realized by use of one uncoupled 2R2T hybrid mechanism
designed in Section 2 and one kinematic chain listed in Table 1. For example, one
may choose the mechanism shown in Figure 1 and a Pi R i R i R i R j chain in Table 1
to generate a new 2R2T parallel robotic manipulator named UPM2R2T-I (see
Figure 2).
Some assembly conditions should be satisfied in order to achieve the
uncoupled parallel robotic manipulator. Firstly, the axis of the actuated joint P14
located at the base is parallel to z-axis. Secondly, the axes of joints R13 and R18
attached on the platform are parallel to each other. In addition, the actuator used to
drive the joint R13 in Figure 1 is removed, which means that the revolute R13
becomes a passive joint. The fourth leg controls the rotational DOF about the v -
axis of the platform in Figure 2.
Referring to Figure 2, q4 is the linear input displacement of the actuated joint
P14 .r is the length of the platform’s structural size. The velocity equation of point
P on the platform within the fixed frame is

 vx   1 0   q1 
 v  0 0   q2 
0 0
 y b

1
 z  0
1 0
  
(7)
    
0 q3 r cos
v  0
0 1
0 0 b   q4 
 Structure Synthesis of Fully-Isotropic Two-Rotational … 91

R 17
R 18
R 16
w P

Mu
P14

v
r R 15
R7 R13
P
R8 4
OL
P6 R3
2
q4
U12
P2
R5 3
U11
a
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 2. Uncoupled 2R2T parallel robotic manipulator UPM2R2T-I.

R 17
R 18
R 16

P14
P
M

R 15
R7 R13
4
P6 R 19
2
q4
U12 P2
R5 3
U11
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 3. Uncoupled 2R2T parallel robotic manipulator UPM2R2T-II.

92 Yanbin Zhang

The Jacobian in Equation (7) is a diagonal matrix, so this manipulator is an

uncoupled 2R2T parallel robotic manipulator. An one-to-one mapping

the inputs velocity vector space of the actuated joints, i.e., vx  q1 , vy  q2 , z  q3

relationship exists between the output velocity vector space of the platform and

and  y  bq4 .
Similarly, other kinematic chains in Table 1 may be selected to replace the
fourth leg of the manipulators in Figures 2 and 3.Thus, many uncoupled 2R2T
parallel manipulators can be synthesized and two novel manipulators are listed in
Figure 4.
Even though the axes of joints R 3 and R 8 are collinear, there is no any
relative motion between them. Therefore, these joints can be replaced by one
revolute joint. Then another new uncoupled 2R2T parallel robotic manipulator
named UPM2R2T-II is obtained, shown in Figure 3. Its output characteristics and
kinematic performances are similar to the original.

R17
R18 P15 P14
P

R16
M

4
R7 R13 q4
R8
P6 R3
2
U12
P2
R5 3
U11
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 4. (Continued)
 Structure Synthesis of Fully-Isotropic Two-Rotational … 93

R15
S 16
C14

P
M
4
R7 R13
q4
R8
P6 R3
2
U12
P2
R5 3
U11
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

b
Figure 4. Other two novel 2R2T uncoupled parallel robotic manipulators.

R 17
R 18
R 16
w R 15
P


M

u

v R 14
r
4 r
R7 R13
P q4
OL R8 R 19
P6 R3
2
U12
P2
R5 3
U11
a
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 5. Fully-isotropic 2R2T parallel robotic manipulator.

94 Yanbin Zhang

6. Structure Synthesis of Fully-Isotropic T2R2 Parallel

Robotic Manipulators
Replacing the prismatic pair of the fourth leg in UPM2R2T-I with a simple
kinematic chain containing two revolute joints (R14 and R19) with parallel axes
will yield a new 2R2T parallel robotic manipulator, in which the axis of joint R14
(or R19 ) is perpendicular to the axis of joint R15 and not parallel to joint R18 . In
order to realize the expected target, there are some structural conditions to be
satisfied. On the one hand, the plane defined by the axes of R13 and R19 is
parallel to the O  xy plane. On the other hand, link size between joints R19 and
R14 is equal to the length of platform (see Figure 5). If joint R19 is selected as the
actuated joint of the fourth leg and q4 is the actuated input angle, then velocity
equation of the verified manipulator in Figure 5 is as follows:

 vx   1 0   q1 
 v  0 0   q2 
0 0
 y
 z  0
1 0
0   q3 
(8)
    
v  0
0 1
0 0 1   q4 

It is very interesting that the Jacobian in Equation (8) is a 4×4 identical matrix
and its condition number is constantly equal to 1 throughout the whole workspace
of the manipulator. This property reveals that the manipulator shows fully-
isotropic in kinematics. So the mechanism is a fully-isotropic parallel robotic
manipulator. Meanwhile, there is no singular configuration for the manipulator in
the workspace since the special Jacobin.
According to above method, if the prismatic pair of the fourth leg in
UPM2R2T-I is replaced by a parallelogram with four revolute joints and the side
link length is equal to the platform size r, a new fully-isotropic 2R2T parallel
manipulator is obtained (see Figure 6).
Similarly, for all uncoupled 2R2T parallel manipulators synthesized in
Section 5, if the linear actuated joint in the fourth leg is replaced by a kinematic
chain with two revolute joints or a parallelogram structure with four revolute
joints, they can be evolved into fully-isotropic 2R2T parallel manipulators as well.
It is true that the special structure size is necessary in order to obtain the expected
 Structure Synthesis of Fully-Isotropic Two-Rotational … 95

R 17
R 18
R 16 R 15
R20

P
M
r
r 4 R 19
R7 R13 R21
P
R8 q14
P6 R3
2 R14
U12
P2
R5 3
U11
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 6. Another fully-isotropic 2R2T parallel robotic manipulator.

R16 R15
S 17 R20

r
R 19
P

4
M

R21
q14
R7 R13
R8 R14
P6 R3
2
U12
P2
R5 3
U11
R10 1
R9 q3
P4
y z
q2 x P1
q1
O

Figure 7. Novel fully-isotropic 2R2T parallel robotic manipulator with a parallelogram

structure.
96 Yanbin Zhang

mechanisms. Furthermore, if the connectivity of the fourth leg is 6, its linear input
joint can only be replaced by a parallelogram structure. Otherwise, the leg will
include one redundant joint. For example, as far as the mechanism in Figure 5(b)
is concerned, if the linear input DOF of the fourth leg is replaced by a
parallelogram, a new fully-isotropic 2R2T parallel robotic manipulator can be
developed (shown in Figure 7).

Conclusion
This paper presents a novel and systematic method for type synthesis of fully-
isotropic 2R2T parallel robotic manipulators. The uncoupled 2R2T parallel
robotic manipulators are synthesized based on the concept of hybrid mechanisms
and the reciprocal screw theory. The fully-isotropic 2R2T parallel robotic
manipulators are developed by using a kinematic chain with two revolute joints or
a parallelogram structure to replace the linear input actuated joint in the fourth leg
of each uncoupled 2R2T parallel mechanisms. For all of the proposed parallel
mechanisms, their Jacobian that maps the input velocity vector space to the output
velocity vector space of the platform is a 4×4 identity matrix. This leads to a one-
to-one correspondence between the output motion components of the platform and
the inputs of the actuators. Both the condition number and the determinant of the
Jacobian matrix are equal to one. Therefore, there is no any singular configuration
throughout the entire workspace and these mechanisms possess good performance
characteristics in force and motion transmission. Control design and path planning
of these mechanisms become very simple. All the uncoupled and the fully-
isotropic 2R2T parallel manipulators proposed in this paper are original.
Moreover, each parallel manipulator only consists of the prismatic, revolute,
cylindrical, universal, or spherical joints, without the homokinetic joints or
redundancy actuation. All these features will reduce the cost of manufacture and
assembly.

Acknowledgments

The first author wishes to acknowledge the supports by the National Natural
Science Foundation of China under grant 50905055, Program for Innovative
Research Team (in Science and Technology) in University of Henan Province
under grant15IRTSTHN008, and Program for Innovative Research Team of
Henan University of Science and Technology under grant 2015XTD012.
 Structure Synthesis of Fully-Isotropic Two-Rotational … 97

References
[1] Tahmasebi F. (2007). Kinematics of a new high-precision three-degree-of-
freedom parallel manipulator. ASME Journal of Mechanical Design, vol.
129, pp. 320-325.
[2] Zhu S. J., Huang Z. and Zhao M. Y., (2009). Singularity analysis for six
practicable 5-dof fully-symmetrical parallel manipulators. Mechanism and
Machine Theory, vol. 44, pp. 710-725.
[3] Gallardo J., Orozco H. and Rico J. M., (2008). Kinematics of 3-RPS parallel
manipulators by means of screw theory. International Journal of Advanced
Manufacturing Technology, vol. 36, pp. 598-605.
[4] Pierrot F., Marquet F., Company O. and Gil T., (2001). H4 parallel robot:
modeling, design and preliminary experiments. Proceedings, IEEE
International Conference on Robotics and Automation, Seoul, Korea, pp.
3256-3261.
[5] Jin Q. and Yang T., (2004). Synthesis and analysis of a group of 3-degree-
of-freedom partially decoupled parallel manipulators. ASME Journal of
Mechanical Design, vol. 126, pp. 301-306.
[6] Calvel R., (1988). DELTA, a fast robot with parallel geometry. Proceedings
of the International Symposium on Industrial Robotics, Switzerland, pp. 91-
100.
[7] Kong X. and Gosselin C. M., (2004). Type synthesis of 3T1R 4-dof parallel
manipulators based on screw theory. IEEE Transactions on Robotics and
Automation, vol. 20, no.2, pp. 181-191.
[8] Li Q.and Huang Z., (2003). A family of symmetrical lower-mobility parallel
mechanisms with spherical and parallel subchains. Journal of Robotic
Systems, vol.20, no.6, pp. 297-305.
[9] Gallardo-Alvardo J., Rico-Martinez J. M. and Alici G., (2006). Kinematics
and singularity analyses of a 4-dof parallel manipulator using screw theory.
Mechanism and Machine Theory, vol. 41, pp. 1048-1061.
[10] Lu Y., Hu B. and Shi Y., (2007). Kinematics analysis and statics of a
2SPS+UPR parallel manipulator. Multibody System Dynamics, vol. 18, pp.
619-636.
[11] Abbasnejad G., Daniali H. M. and Fathi A., (2011). Architecture
optimization of 4PUS+1PS parallel manipulator. Robotica, vol. 29, no. 5,
pp. 683-690.
[12] Zhang Y., Liu H. and Wu X., (2009). Kinematics analysis of a novel parallel
manipulator. Mechanism and Machine Theory, vol. 44, pp. 1648-1657.
98 Yanbin Zhang

[13] Briot S. and Bonev I. A., (2010).Pantopteron: a new 3T1R decoupled

parallel manipulator for pick-and-place applications. Mechanism and
Machine Theory, vol. 45, pp. 707-721.
[14] Zhang D. and Zhang F., (2011). Design and analysis of a totally decoupled
3-dof spherical parallel manipulator. Robotica, vol. 29, no. 7, pp. 1093-
1100.
[15] Zeng D. and Huang Z., (2011) Type synthesis of the rotational decoupled
parallel mechanism based on screw theory. Science China technological
Sciences, vol. 54, no. 4, pp. 998-1004.
[16] Zeng D.-X., Hou Y. and Lu W.-J., (2014). Type synthesis method for the
translational decoupled parallel mechanism based on screw theory. Journal
of Harbin Institute of Technology, vol. 21, no.1, pp. 84-91.
[17] Yue C., Su H.-J. and Kong X., (2013). Type synthesis of 3-dof translational
compliant parallel mechanisms. Proceedings of the ASME 2013
International Design Engineering Technical Conference and Computers
and Information in Engineering Conference, August, 2013, Portland,
Oregon, US, DETC2013-12718.
[18] Bouzgarrou B. C., Dauroux J. C. and Gogu G. and Heerah Y., (2004).
Rigidity analysis of T3R1 parallel robot with uncoupled kinematics.
Proceedings of the 35th international Symposium on Robotics, Paris, France,
pp. 1-6.
[19] Kong X. and Gosselin C. M., (2002). Type synthesis of linear translational
parallel manipulators. Advances in Robot Kinematics, In: J. Lenarčič, P.
Thomas (Eds.), Advances in Robot Kinematics: Theory and Application,
Kluwer Academic Publishers, pp. 453-462.
[20] Kuo C. H. and Dai J. S., (2011). A fully-isotropic parallel orientation
mechanism. Proceedings 13th World Congress in Mechanism and Machine
Science, Guanajuato, Mexico, pp. 1-7.
[21] Carricato M., (2005). Fully isotropic four-degrees-of-freedom parallel
mechanisms for Schoenflies motion. International Journal of Robotics
Research, vol. 24, pp. 397-414.
[22] Zhang Y., Wang H. and Wu X., (2012). Structure synthesis of fully-
isotropic 3-dof planar parallel manipulators. Optics and Precision
Engineering, vol. 20, no.3, pp. 579-586.
[23] Gogu G., (2004). Structural synthesis of fully-isotropic translational parallel
robots via theory of linear transformations. European Journal of Mechanics
A/Solids, vol. 23, pp. 1021-1039.
[24] Gogu G., (2007). Structural synthesis of fully-isotropic parallel robots with
Schonflies motions via theory of linear transformations and evolutionary
 Structure Synthesis of Fully-Isotropic Two-Rotational … 99

morphology. European Journal of Mechanics, A/Solids, vol. 26, pp. 242-

269.
[25] Gogu G., (2005). Fully-isotropic parallel robots with four degrees of
freedom T2R2-type. Proceedings of IEEE/RSJ International Conference on
Intelligent Robots and Systems, Alberta, Canada, pp. 960-965.
[26] Zhang Y., Liu H. and Wu X., (2008). Type synthesis of non-singular
full-isotropic translational parallel mechanisms based on theory of
reciprocal screw. Chinese Journal of Mechanical Engineering, vol. 44, no.
10, pp. 83-88.

Biographical Sketch
Yanbin Zhang is a Professor at the School of Mechatronics Engineering, Henan
University of Science and Technology, China. He received his PhD in Mechanical
Engineering from Xi’an University of Technology in 2008. His research interests
include mechanisms, parallel robotic manipulator, mechanical design theory. He
has published 30 academic papers in journals such as Mechanism and Machine
Theory, Int. J. Advanced Robotic Systems, Journal of Mechanical Engineering (in
Chinese), Optics and Precision Engineering (in Chinese), China Mechanical
Engineering (in Chinese), Transaction of the Chinese Society for Agricultural
Machinery (in Chinese), etc.
In: Parallel Manipulators ISBN: 978-1-63485-926-4
Editor: Cecilia Norton, pp. 101-141
c 2016 Nova Science Publishers, Inc.

Chapter 4

T HE PARALLEL T WO -L EGGED WALKING

R OBOT CENTAUROB
Shucen Du, Josef Schlattmann, Stefan Schulz∗ and Arthur Seibel
Workgroup on System Technologies and Engineering
Design Methodology, Hamburg University of Technology,
Hamburg, Germany

Abstract
In this chapter, we review the new prototype of the two-legged, paral-
lel kinematic walking robot CENTAUROB, developed at Hamburg Uni-
versity of Technology. The main features of this robot are its modular,
symmetric structure, the construction of the legs as Stewart platforms
(hexapods), and the special C-shaped feet that allow a statically stable
movement at any time. The symmetric structure of the robot leads to a
center of gravity exactly in the middle of the hip and prevents from a pre-
ferred moving direction or an asymmetric load of the legs. As a highly
flexible walking device, the CENTAUROB is able to walk in unpaved en-
vironment, avoid and overcome obstacles, and even handle straight and
circular stairs.

Keywords: parallel robots, service robots, biped walking, statically stable walk-
ing, sensor architecture, control concepts, walking scenarios
∗
E-mail address: [email protected] (Corresponding author).
102 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

1. Introduction
Industrial robots are well established in production processes. Since recent
years, there has also been a growing demand for service robots both for private
use, for example, as a ride-on device for the mobility assurance of immobile
people or as an interactive care giving robot in the household, and for use in
industrial environments, for example, as a simple industrial carrying device or
as an autonomous robot in rough or dangerous areas like mine fields or burn-
ing houses. Market studies show that the demand in robotic systems will rise
steadily in the coming years. Due to higher performance-requirements with
respect to dynamics, precision, velocity, weight, control intelligence, suitabil-
ity for automation, price, etc., most of the conventional robot solutions have
reached their limits.
Biped robots can be classified by their kinematic structure into serial and
parallel ones. Serial kinematic robots are most often used due to the lower con-
trol effort. Well known examples for humanoid robots with a serial kinematic
are Honda’s humanoid robot Asimo [1], the humanoid robot LOLA [2], and the
WABIAN series [3]. Parallel robots offer special advantages including higher
stiffnesses, smaller overall machine sizes, lower inertial forces, faster processing
velocities, and compensatory dimensional errors. The operating load occurring
at the end effector is divided among several links which results in a high stiffness
of the structure. The two-legged, parallel kinematic walking robot CENTAU-
ROB, developed in 1997 [4], was by then the first biped robot using two Stewart
platforms as leg structures. It was the first parallel walking robot that was able to
walk statically stable even in unpaved environments, avoid and overcome obsta-
cles, and even handle straight and circular stairs. A second realization of a biped
robot with a parallel kinematic, which exists up to now, is the Waseda-Leg [3].
In this chapter, we review the new prototype of the CENTAUROB [5], de-
veloped at Hamburg University of Technology. Just like the first prototype [6],
the redeveloped construction is based on the Stewart platform as well. Several
modifications were made due to the moving technique standard and to solve
disadvantages and problems [7].
The new prototype represents a universal moving device which convinces
by its high stability and payload and offers plenty of application possibilities.
In future, the robot will be able to operate autonomously and automatically
recognize obstacles and overcome them. Possible fields of usage that come into
mind are, for example, as a ride-on device for immobile people, as a service or
 The Parallel Two-Legged Walking Robot CENTAUROB 103

interactive care giving robot in the household, as a simple industrial carrying

device in unpaved terrain, or as an autonomous robot in rough or dangerous
areas like mine fields or burning houses.
The remainder of this chapter is organized as follows. In Section 2, we dis-
cuss parallel robots and their applicability as biped walking robots. In Sections
3 and 4, we review the first prototype, the redesign process, and the new proto-
type of the CENTAUROB. In Section 5, we briefly present the inverse kinematic
model of the robot followed by the motion control strategy in Section 6. Section
7 presents several walking scenarios and other motions to show the agility of the
robot. In Section 8, possible applications and fields of usage for the robot are
discussed. Finally, in Section 9, this chapter is summarized and an outlook to
future work is given.

2. Parallel Walking Robots

2.1. Parallel Robots
Currently, the field of industrial robots is dominated by articulated robots where
the driven axes are arranged in series, so-called serial robots. A well known
example is the SCARA robot, short for ’Selective Compliance Assembly Robot
Arm’. Serial robots owe their success mainly to the high degree of flexibility,
the high workspace-to-footprint ratio, as well as the low calculation and control
effort due to the easily solvable direct kinematics. Although the serial structure
leads to high flexibility, this is also a main disadvantage since positioning errors
in the driven joints as well as the elasticity of each component of the structure
accumulate and lead to lower positioning accuracies and vibration problems.
In order to compensate these problems with higher driving forces and inertia
torques, drives that are further away from the end-effectors are scaled larger.
Parallel robots can overcome these drawbacks. The name results from their
parallel kinematic structure in which the drives operate side by side. The load
is distributed over multiple paths, and the drives do not have to hold each other.
This results in high rigidity, similar to a framework. The positioning accuracy
is also very high because the positioning errors of the individual drives do not
accumulate. The moving masses are smaller, so that the machine dynamics can
be controlled by the use of lower power ratings. As a result of these advantages,
parallel robots have become more relevant in fields where high process forces
and special accuracies are required. The main applications for parallel robots
104 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

are widely ranged and spread from motion simulators and test benches over
machine tools for handling and manufacturing to brain surgery. However, the
mobility and flexibility of serial robots cannot be obtained.
There are many possible configurations of parallel robots which differ, in
particular, by the type of drive (rotary or linear) and the degrees of freedom.
One of the most universal parallel robot is the hexapod. The hexapod is also
called Gough-Stewart platform according to the developers who have used these
parallel kinematics for the first time [8], [9].
The function of a hexapod is to move the end-effector in all translational and
rotational axes in space. Six parallel drives connect the stationary base platform
with the mobile tool platform. On top of the tool platform, the end-effector is
mounted for specific tasks. If, for any particular purpose, it is not necessary to
reach the full dimension of the workspace, the number of drives can be reduced
as long as additional joints or supports are introduced. For example, when using
a rotary tool, such as a milling cutter, no rotatory degree of freedom is necessary.
A well known example of a parallel manufacturing robot is the three degree of
freedom delta robot [10], which uses only three drives and an additional passive
guidance.

2.2. Service Robots

Generally, biped robots will take their place in the near future in the area of mo-
bile service robots and will act as a substitute for humans or support them. The
tasks will range from the operation in hazardous environments up to a general
support in the household [11]. Especially in Japan, the realization of this vision
is already well advanced. In the field of service robotics, a dichotomy between
specialized and universal robots can be observed. Specialized service robots
already have a high practical relevance and were tested for various distribution
functions, for example, in hospitals [12], [13], [14]. Here, the drive kinematics
has minor importance since navigation is only necessary in known barrier-free
environments. In addition to these robots, for specific tasks, there is the aim to
construct a universal robot. Generally, a human-like form is used as a model.
According to the definition of the International Federation of Robotics, a
service robot is described as follows. ’A service robot is a robot which operates
semi- or fully autonomously to perform services useful to the well-being of
humans and equipment, excluding manufacturing operations.’
Currently, autonomy is the biggest problem. Further challenges are the
 The Parallel Two-Legged Walking Robot CENTAUROB 105

power supply and the autonomous navigation. By methods of image recognition

and processing, robots are independently able to move. The energy consump-
tion of the drives, however, still leads to short periods of use. Optimally, battery
charging stations are installed that are automatically visited by robots with a
low energy level. The wider the field of applications get, the larger and heavier
become the batteries the robot has to carry. Especially for humanoid robots,
this is a huge problem because they usually carry the batteries as a backpack.
With increasing weight of the batteries, the center of mass is shifted upwards,
resulting in a lower tilt stability.

2.3. Humanoid Robots

In order to construct a robot, the human locomotive system is often used as a
model. These robots are called humanoid robots. Such a transfer of biologically
successful principles into a technical application is called bionics.
The motion of a biped is one of the major challenges of research because it
allows optimal locomotion in unstructured environments. Following the guiding
principle of bionics according to the model of joint and muscle arrangements of
humans, a robot with a comparable consolidation achieves an approximately
equal mobility. An essential feature of the construction of the legs are the joints
at the hips, the feet, and especially the knees. Compared to human joints, they
are limited in the degrees of freedom, number, size, and resilience. As a result,
only a simplified transfer to the robot is possible. However, it should be noted
that bionics is only implicitly mentioned as a motivation for the design of a
robot with knee joints. Equally important is certainly the vision of maintaining
or increasing the workforce of a human by a robot, which is optimally solved
when the robot is built like a human. Moreover, the acceptance of a humanoid
robot would probably be higher since it would act more as a person and less as
a machine.
The biggest problems of bipeds are balance control and coordination of mo-
tions, which have already been treated in many works. Even for well-developed
models, fast walking is still challenging. This fails mostly due to the high
computation burden. The best-known representatives of humanoid robots are
ASIMO [1], the model of the company Honda, and LOLA [2], a very advanced
model with automatic obstacle detection from the Technical University of Mu-
nich.
The overall conclusion is that many research groups and institutions are
106 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

concerned with the research and development of humanoid robots. Aside from
a few models of large corporations, only a few bipeds have left the prototype
stage yet. The reason is, probably, the relatively low interest of the industry
since the way to a fully operational, robust, and secure biped is still far away.
ASIMO, for example, which has already achieved a worldwide recognition, is
more used to serve the image of the company than to execute household’s duties.
However, the theoretical fields of usage and applications for humanoid robots
are very large, for example, in dangerous environments. Here, bipeds could be
invaluable, so that their development is of great benefit.

2.4. Alternative Leg Structures

Usually, the joints of the robot are actuated directly. The stabilizing effect of
muscles, which are connected to the bone, is not given. The problem of the
lower stiffness of serially robots thus transfers to the robot’s knee. This prob-
lem can probably be solved constructively or by reducing the walking speed.
As a reasonable alternative, parallel kinematics can be used which offer high
stiffness and stability. However, the flexibility and the large workspace of serial
kinematics cannot be reached because the stability allows only small walking
steps.

3. First Prototype of the CENTAUROB

The first prototype of the CENTAUROB, shown in Fig. 1, consists of two
hexapods as leg structures. This feature is a fundamental difference to other
humanoid robots. It establishes new ways of movement strategies and tasks.
The aim is to use the robot in specific fields where the advantages of its parallel
kinematic structure can be fully exploited. These advantages mainly include the
high payload capacity, the high rigidity, as well as the possible large step size
and step height. High steps and staircases can be climbed by alternately extend-
ing and retracting the telescopic legs. Moreover, passing low openings is also
not a problem because both legs can be simultaneously retracted to reduce the
body height.
The robot has two specially designed C-shaped feet to ensure a statically
stable walking at any time. This means that the ground projection of the overall
center of mass of the robot must always remain in the area spanned by the feet,
 The Parallel Two-Legged Walking Robot CENTAUROB 107

Figure 1. The world’s first spiral stair climbing robot, own patent.

the so-called stability region. Figure 2 shows the robot’s stability region with
one foot and two feet on the ground.

Figure 2. Stability region of the CENTAUROB with one foot (left) and two feet
on the ground (right).

The joints of each leg are arranged in circles with different radii. A reduction
108 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

of the upper joints’ radius would move the static and kinematic properties of the
legs to an unfavorable area. The requirements for small space and high stability
are divergent and cannot be optimally achieved simultaneously.
The robot is operated by a pure feed-forward or open loop control. Be-
cause for statically stable motion, dynamic influences such as accelerations are
disregarded, the robot’s movement can only be performed very slowly.
The CENTAUROB’s advantages and disadvantages compared to other hu-
manoid robots are summarized in Table 1.

Table 1. Advantages and disadvantages of the CENTAUROB compared to

other robots

Advantages Disadvantages
high payload to weight ratio limited workspace due to leg colli-
sions
high feet flexibility operating close to singular configu-
rations
ability to overcome complex obsta- large leg cross sections
cles
statically stable motion high calculation effort in control
accurate and stable alignment friction afflicted operation of all ac-
tuators
of the hip while standing for the movement in only one de-
gree of freedom
retracting and extending of legs
leg structure modularization
failure save system

3.1. Structural Information

3.1.1. Workspace
Investigating the kinematics of a robot, the workspace is of great interest. Here,
a distinction is made between three different workspaces:

• The maximum workspace indicates the points that can be achieved with
 The Parallel Two-Legged Walking Robot CENTAUROB 109

at least one orientation of the platform.

• The total orientation workspace is a subspace of the maximum workspace
and indicates the points that can be achieved in any orientation of the
platform.
• The dexterous workspace indicates the points that can be achieved at a
given fixed orientation of the platform.

For hexapods used as machine tools, the calculation of the workspace plays
an important role. This leads to great interest in capturing tool shifts or inclina-
tions. The large orientation angles mark out the area of operation of a machine
and prove its competitiveness. In case of the CENTAUROB, however, these
extreme positions are never approached because the robot might lose its stable
state.
For each leg of the CENTAUROB, the calculation of the workspace de-
scribes the positions and orientations the corresponding foot can reach in a sin-
gle step. Investigating only straight planes, the calculation of the dexterous
workspace is sufficient. Here, it can be assumed that the orientation of the foot
platforms are always horizontal, respectively, parallel to the floor. Climbing
stairs is also covered in this case. In addition, studies can be carried out with
inclined feet. However, this scenario hardly shows significant changes in most
of the parameters. It is related to the elongated shape of the hexapod, where
the orientation of the foot platform has low impact on the length and inclina-
tion of the individual drives. The dexterous workspace of the CENTAUROB is
illustrated in Fig. 3.
With the help of the calculated workspace, two parameters can be derived.
These are the step height and the step size of the leg. Particularly, for deter-
mining the maximum surmountable obstacle height, the step height is impor-
tant. This applies even more to the CENTAUROB as its telescopic leg kine-
matics leads to a special mobility. The step height is derived from the vertical
workspace limits for x = 0 and y = 0. This corresponds with the maximum
retracted and extended leg in a horizontal state. Is the vertical position known,
the possible step size can be determined from the boundary of the workspace.

3.1.2. Singularities
The workspace specifies only the constructive constraints. Whether the drives
can reach a specific point or not, is not described by the workspace. Singularities
110 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

Figure 3. Dexterous workspace of one leg of the CENTAUROB.

can occur at points inside the workspace and cause severe problems. Mathemat-
ically, a singularity is a point at which a model is not defined or actually does not
work. In a singular configuration, the system’s behavior is not predictable and
the corresponding motions and forces become uncontrollable. Singularities and
even configurations close to the singularities should be avoided due to the se-
vere effect on the input-output transmission and possible failures or permanent
damages.
The case of losing a degree of freedom occurs, for example, when the drives
shall act in a perpendicular direction to the drives. In this direction, no forces
can be applied, with the result that a virtual displacement is possible. This is
a critical situation for a hexapod. In particular, the extended structure of the
CENTAUROB nearly leads to parallelly arranged actuators that can take forces
only slightly in horizontal direction.
Degrees of freedom can be obtained when a movement of the structure is
possible for fixed positions of the end-effector. This typically occurs in serial
 The Parallel Two-Legged Walking Robot CENTAUROB 111

robots when joint axes of drives coincide in the structure. Then, for example, the
total rotation angle is known while the single rotation angles cannot be uniquely
determined. The occurrence of singularities can be observed in the kinematic
model when the JACOBI matrix loses rank.

3.1.3. Stability
Stability, especially tilt stability, has high priority for legged robots. It is an
important part in mainly all phases of the robot’s movement and can be di-
rectly influenced by the structural design, especially of the feet, and the walking
pattern. Interferences, such as uncertainties in the environment or calculation
errors, shall not unbalance the robot. Both are grouped under the term stability.
If a foot is placed on the ground, a contact situation occurs. In this con-
tact situation, tangential and normal forces as well as moments act on the foot.
They result from internal and external forces on the robot’s body. For the foot,
there exists a point in which the moments in the plane of contact accumulate
to zero. This zero-moment point is consistent with the point where the result-
ing compressive force acts on the foot. The stability of the robot is ensured if
the zero-moment point lies within the stability region, which is formed by the
convex hull of all points of the contact surface.
For the CENTAUROB, a statically stable walking pattern is embedded. This
means that dynamic influences are neglected and only slow, quasi-static move-
ments are possible. Therewith, only the weight of the robot influences the
robot’s stability. Static stability is thus given if the ground projection of the
robot’s center of mass lies within the area formed by the feet in contact with the
ground, see Fig. 2.

3.1.4. Agility
The locomotion of the CENTAUROB mainly depends on the relative movement
between the hip and the foot platform. The transmission of movements from the
drives to the platforms is given by the kinematics, see Section 5. Therewith, the
maximum platform velocities can be determined from the maximum velocity of
a single drive.
For serial robots, the maximum velocities can usually be determined directly
from the performance of the drives, as these are each associated with one degree
of freedom alone. The velocities of an industrial robot cannot be achieved with
112 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

the CENTAUROB. However, these velocities are not necessary, as long as a

statically stable movement is required.

3.1.5. Stiffness
In real environmental situations, static and dynamic excitations can act on a
robot. A static excitation can be, for example, a load that is carried by the
robot. Impulses that occur when placing the foot on the ground lead to dynamic
excitations. Further impulses can occur in case of interactions with objects. The
stiffness of the leg structure determines how it reacts on these excitations. By
using a hexapod, the robot’s stiffness is inherently higher than the stiffness of
humanoid robots with knees.
The stiffness is determined by a global stiffness matrix, which is derived
from the resilience of the drives. Since the drives are connected to universal
joints at their ends, no bending moments appear. The purely axial loads allow
to model the drives as tie bars or struts. Here, the elasticities of further compo-
nents are not considered explicitly. Instead, they are considered by a high safety
factor. From the diagonal elements of the stiffness matrix, the rigidities can be
determined in every direction of the world coordinate system, and the influence
of off-diagonal elements is disregarded.

4. New Prototype of the CENTAUROB

4.1. Redesign Process
The first prototype of a parallel two-legged walking robot was a breakthrough
in technology and base for further engineering. The vision of creating a mo-
bile robot which can operate autonomously and automatically recognize obsta-
cles as well as overcome them implies using an independent power supply, a
stand-alone control system, and compliance to essential safety requirements in
human surroundings. For this reason, the CENTAUROB underwent a complete
redesign in which every mechanical component was thoroughly reviewed to re-
duce weight as well as increase dynamics and efficiency [7].
As a first step in the redesign process, an optimization of the Stewart plat-
form architecture was suggested. The circular shape of the Stewart platforms
attached to the hip was changed to an elliptical shape with a smaller diameter
only in the lateral direction, see Fig. 4. Therewith, the lateral dimension of the
 The Parallel Two-Legged Walking Robot CENTAUROB 113

(a) (b)

Figure 4. Redesign of the robot’s hip: (a) original, circular hip, and (b) opti-
mized, elliptical hip with reduced lateral dimension.

hip can be reduced to a more human-like one. Furthermore, the width of the hip
determines the flexibility of the robot as it limits the opportunity to pass through
narrow passages. By this means, the good performance in the dominant forward
direction can be preserved in combination with a narrow hip size. Other changes
in the mechanical components include a replacement of the formerly used step-
per motors by high performance electronically commuted servomotors and the
sliding screws by low-friction ball screws, which require brakes in case of a
power or control failure since the new mechanism is not self-locking. Due to the
low coefficient of performance, instead of worm gears, we now use belt drives
and planetary gears between the spindles and motors. In order to compensate
the passive rotation of the upper cylinder around the spindle axis [15], an upper
joint with an extra degree of freedom of rotation is possible. A more sophisti-
cated possibility is a compensation of the passive rotation by using quaternions
as proposed in [16]. Therewith, universal joints can be used for the upper and
lower joints as well. The new prototype of the CENTAUROB is illustrated in
Fig. 5.

4.2. Structural Design of the New Prototype

The main modules of the CENTAUROB are the hip, the upper joints, the legs,
the lower joints, and the feet. The main components of the robot are briefly in-
114 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

Figure 5. New prototype of the CENTAUROB.

troduced in the following. A more detailed description of the robot’s mechanical

structure is given in [5].

4.2.1. Hip
The hip, which is the connecting structure between the legs of the robot, con-
tains the motor cards, the power supply, and the on-board control system. The
size of the hip is 720 mm in the lateral direction and 500 mm in the longitudinal
direction and is adapted to the positions of the upper joints. The upper joints
of each hexapod are arranged side by side in an ellipse instead of a circle to
provide a narrow hip size as described in Section 4.1.

4.2.2. Joints
As upper and lower joints, we use state-of-the-art universal joints made of stain-
less steel with two degrees of freedom and a pivot angle of 45◦ . In this case,
the passive rotation that results from the asymmetric transformation of rotations
in the universal joints, must be mathematically formulated and computationally
compensated [15], [16].
 The Parallel Two-Legged Walking Robot CENTAUROB 115

4.2.3. Legs
The legs of the robot are composed of six linear units that are combined to a
hexapod. Each linear unit itself consists of a ball screw linear actuator and a
parallelly arranged motor unit. Both are connected by a cam belt.
The motor unit consists of an electrically commutated DC motor, a motor
card, which is placed on the hip, a break, and a planetary gear. The motor card
allows a position, velocity, and current control. Furthermore, the motor contains
Hall sensors and a three-channel encoder to detect the angular position of the
motor shaft and therewith the length of the linear actuator. The brake is used to
hold the position in case of an electricity failure, and the interposed planetary
gear leads to a higher torque in the linear actuator.
The cam belt transmits the motor’s torque one-to-one to the linear actuator.
Then, the ball screw converts the torque into a linear movement. The linear
actuator consists of an upper cylinder with an integrated spindle and a lower
cylinder, which can move in linear direction. Strain gauge force sensors are
used at the end of each linear unit for measuring the forces along the spindle.

4.2.4. Feet
The special C-shape of the feet leads to a convex standing surface containing the
ground projection of the robot’s center of mass, see Fig. 2. The hip of the robot
can be moved as long as the ground projection of the center of mass remains in
the standing surface. Due to the special shape of the feet, it is also possible to
place the smaller foot inside the bigger foot. When the feet are placed inside
each other, the height of the CENTAUROB from the ground to the bottom edge
of the hip is 872 mm in the retracted and 1269 mm in the extended condition.

4.3. Sensor Architecture

The global sensor concept for the CENTAUROB contains the so-called pose
detection, the force-measurement, and the detection of the ground projection of
the robot’s center of mass.

4.3.1. Pose Detection

Six minimal coordinates are necessary to determine the pose of a six degree
of freedom parallel mechanism. In order to determine the pose and orienta-
tion of such a mechanism, generally, the lengths of the six linear actuators are
116 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

measured. This so-called direct kinematics problem is quiet difficult to solve

because there is usually not a unique solution. It is necessary to search iter-
atively for the platform’s positions and orientations with the need of an initial
pose estimate. The accuracy and convergence time depends on the choice of this
initial pose estimate. Many studies have been made in this field. In order to find
a unique solution for a given set of actuator lengths, it is possible to add further
sensors on the passive joints of the hexapod [17], [18], [19], and others. Two
possibilities of the placement of sensors have been investigated. One possibility
is to add rotary sensors on existing passive joints, and the other possibility is to
add passive links whose lengths are measured with linear sensors.
For an electric driven linear actuator, it is indeed possible to detect its length
with a rotary encoder. Rotary encoders provide precise values of the angular po-
sition of the motor shaft, which, in turn, is related to the linear actuator’s length.
This length, on the other hand, needs to be linked with the encoder position by
performing an initial length measurement. Due to a possible backlash in the
gear or the spindle, the flexibility of the belt, or the so-called passive rotation in
unguided ball screw linear actuators, the obtained data do not have to comply
with the actual linear actuator’s lengths necessarily [15], [16], [20]. An addi-
tional compensation or a numerical real-time estimator, as suggested in [20],
needs to be implemented.
Although the lengths of the linear actuators are needed, for example, as
target values for the motors, the position and orientation of the platforms are
more valuable. One reason for this is, for example, the use of these values as a
reference signal in control. Furthermore, using inverse kinematics (see Section
5), the linear actuator’s lengths can be obtained from the platform’s positions
and orientations.
As mentioned before, for incremental length sensors, an initial length mea-
surement or a so-called homing, an initialization run to predefined reference
signals, is necessary. For absolute length sensors, the accuracy decreases with
increasing measuring range. Furthermore, the installation range and the cost
also increase.
For the two-legged walking robot CENTAUROB, several sensor concepts
have been studied. Four sensor concepts using only the minimum number of
sensors were proposed and presented in detail in [21]. These concepts still need
further numerical methods to find the unique pose of the robot. They are de-
scribed in the following.
 The Parallel Two-Legged Walking Robot CENTAUROB 117

Six angle sensors are mounted on the six linear actuators of the hexapod to
detect the orientation of each linear actuator. Therewith, manufacturing costs
can be reduced because the linear actuators can be constructed identically. A
problem here is that even small angle changes can result in major changes in the
linear actuators’ length.
Six displacement sensors are mounted on the six linear actuators of the
hexapod. These sensors detect length changes in the ball screws of the linear
actuators. In this context, different types of displacement sensors can be imple-
mented. For example, it is possible to implement displacement sensors where
the measuring armature dips into a core, a magnet is moved along a track, or the
length is detected with a rope. The advantage of using displacement sensors is
the high and uniform accuracy of the measurement. Even for simple executions,
these measuring units achieve an accuracy of up to 0.25 mm. This accuracy is
approximately constant for the entire stroke of the linear actuator. Moreover,
displacement sensors can directly detect length changes of the linear actuators.
The main disadvantage of displacement sensors, however, is that most of them
are quiet large and must be mounted directly on the ball screw linear actuators.
Three angle sensors and three displacement sensors are mounted on three
linear actuators to combine the advantages of both sensor concepts. Thus, the
lengths of the linear actuators can be measured with displacement sensors and
the orientation of the foot can be determined by angle sensors. Thereby, the
space required for the sensors is minimized so that the workspace is only as lim-
ited as possible. In particular, due to the serial arrangement of the two hexapods
of the CENTAUROB, the workspace of the legs is already limited. In order to
achieve an optimal sensor placement, the sensors can be mounted on the outer
linear actuators of each leg so that collisions or interference with other leg’s
sensors can be excluded.
Six inclination sensors are mounted on the six linear actuators of the hexa-
pod. With these sensors, the position of the feet in relation to the hip can be
determined. In addition, this sensor concept even provides information about
the position and orientation of the robot in space.
In order to elude the direct kinematics problem, further concepts were stud-
ied in [21]. In these concepts, the unique solution of the direct kinematics prob-
lem can be obtained by using only the sensor data. This is generally achieved
by means of additional sensors. Five different sensor concepts were proposed
and presented in detail, which are described in the following.
118 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

Five angle sensors and one displacement sensor: By attaching one angle
sensor on the linear actuator, two angle sensors on the lower joint, three angle
sensors on the upper joint, and one displacement sensor on the linear actuator,
the pose of this linear actuator is clearly determined just like the pose of the
end-effector. The disadvantage of this concept is that all the sensors have to be
placed on just one linear actuator. This might lead to difficulties in mounting
the angle sensors on the joints or to limitations of the workspace.
Five angle sensors and one displacement sensor on an additional leg:
An additional leg can be integrated in the middle of the hexapod. The leg can
be made of a non-rotatable telescopic cylinder mounted between the upper and
the lower platform. The length of the additional leg can be measured by the
displacement sensor. The advantage of this concept is the possibility to avoid
sensors on the linear actuators. In this context, angle sensors can be integrated
much easier in the additional leg than in the joints of the linear actuators.
Four angle sensors and three displacement sensors: In this concept, two
angle sensors and one displacement sensor are mounted on two linear actuators
of the hexapod. Therewith, two vectors of known lengths and orientations are
defined. The endpoints lie both on a plane indicating the orientation of the
foot. Here, a straight line between these two points can be drawn. However,
the rotation around the longitudinal axes of these two linear actuators is still
undetectable. With an additional displacement sensor on another linear actuator,
the pose of the foot can be obtained.
Four angle sensors, two displacement sensors, and three one-
dimensional inclination sensors: Similar to the previous concept, two linear
actuators are each equipped with a displacement sensor and two angle sensors
at the universal joints. Therewith, again, a line on the foot platform is known.
In this context, the plane through this line can only rotate around this line. The
only remaining degree of freedom can be detected by means of an inclination
sensor on the hip. However, this is only possible if the orientation of the foot
platform with respect to the global coordinate system is known. Since this is
not always the case, another inclination sensor on the foot is necessary to obtain
meaningful data for calculating the hexapod’s pose. Comparing the inclination
of the hip and the foot platform, an angle between the two planes can be de-
termined. The inclination sensors must be positioned so that they are mounted
on the hip and the foot each in a perpendicular direction to the connecting line
between the two linear actuators equipped with the other sensors.
 The Parallel Two-Legged Walking Robot CENTAUROB 119

Six displacement sensors and two angle sensors: In a concept with six
displacement sensors, where each sensor is mounted on one linear actuator, di-
rect kinematics offers up to 40 different poses for the robot [22]. With two addi-
tional angle sensors on two of the linear actuators, wrong poses can be excluded
until only the correct solution remains.

4.3.2. Load Detection

Static stability can only be guaranteed if the ground projection of the robot’s
center of mass constantly remains in the stability region spanned by the C-
shaped feet of the CENTAUROB, see Fig. 2. This point can be calculated from
the forces along the linear actuators. Therefore, strain gauge force sensors are
installed at the end of every linear unit. This makes it possible to measure the
load of every leg. Furthermore, by knowing the actual pose of the CENTAU-
ROB, the ground projection of the center of mass can be calculated using static
equilibrium equations. In a static situation, the robot’s center of mass can be
calculated by the weights and positions of every center of mass of the individual
component.

5. Kinematic Model
In this section, we review the kinematics of the linear actuators of the CEN-
TAUROB. The complete description of the kinematics can be found in [5].

5.1. Kinematics of the Free Rigid Body

The kinematics of a body describes its possible movement without considering
the cause of this movement, that is, the forces acting on the body. The move-
ment of a single, free rigid body can be described by two different coordinate
systems, a body-fixed one and an inertial one. The coordinate systems of the
CENTAUROB are illustrated in Fig. 6. Here, {0} denotes the inertial frame and
{i}, i = 1, . . . , 4, denote the body-fixed frames.
The position and orientation of a platform can be described by the following
vector of minimal coordinates:
 ⊤  ⊤
yi = px,i py,i pz,i αi βi γi = yt,i y r,i , i = 1, . . ., 4 , (1)
where yt,i contains the translatory and yr,i the rotatory degrees of freedom of
the platform. Furthermore, px,i shows in the forward, py,i in the sideward, and
120 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

{2} {3}

{0}

{1} {4}

Figure 6. Coordinate systems of the CENTAUROB.

pz,i in the upper direction of the robot. The rotation angles αi , βi , and γi can be
defined, for example, by CARDAN or EULER angles.
On the other hand, we can also use the lengths of the linear actuators qk,i ,
k = 1, . . ., 6, for defining the minimal coordinates:
 ⊤
q i = q1,i q2,i q3,i q4,i q5,i q6,i , i = 1, . . ., 4 . (2)
In order to transform an arbitrary vector i p described in the body-fixed
frame {i}, i = 1, . . ., 4, into the inertial frame {0}, we use the following rela-
tion:
p = p i + Ri · i p , (3)
where pi denotes the vector connecting the origins of frame {0} and frame {i}
and Ri the rotation matrix from frame {i} into frame {0}. Note that, here,
pi = yt,i . (4)
Because the rotation angles in Eq. (1) do not span an orthogonal coordinate
system, their temporal derivatives do not represent a geometrically meaningful
quantity. For this reason, we have to introduce a new vector of minimal coordi-
nates which only contains velocities:
 ⊤  ⊤
θ̇ i = ṗx,i ṗy,i ṗz,i ωx,i ωy,i ωz,i = ẏ t,i ω i , i = 1, . . ., 4 .
(5)
 The Parallel Two-Legged Walking Robot CENTAUROB 121

Here, ωx,i , ωy,i , and ωz,i denote the angular velocities around the the x, y, and
z direction, respectively. Note that

ω i 6= ẏ r,i . (6)

However, between ω i and ẏr,i , the following relationship holds:

     
ωx,i 1 0 sin βi α̇i
    
ω i = ωy,i = 0 cos αi − sin αi cos βi · β̇i  = ẏ r,i , (7)
ωz,i 0 sin αi cos αi cos βi γ̇i

assuming that αi , βi, and γi are defined by CARDAN angles.

5.2. Inverse Kinematics of the CENTAUROB

The inverse kinematics describes the relationship between the actuator coordi-
nates of the active bodies q i , i = 1, . . ., 4, and the world coordinates y i , as
illustrated in Fig. 7. For a hexapod, the world coordinates y i cannot be com-
puted directly. For this reason, we use the actuator coordinates q i as a reference
signal in our control system.

actuator forward kinematics world

coordinates coordinates
qi inverse kinematics yi

Figure 7. Relationship between the world and the actuator coordinates.

According to Fig. 8, the length of the kth linear actuator

qk,1 = qk,2 = |pJk,2 Jk,1 | = |pJk,1 Jk,2 | , k = 1, . . ., 6 , (8)

can be determined from

2
pJk,2 Jk,1 = 2 p21 + 2 R1 · 1 p1Jk,1 − 2 p2Jk,2 (9)

with respect to the temporarily fixed frame {2}. Here, 2 R1 denotes the rotation
matrix from frame {1} into frame {2}.
122 Shucen Du, Josef Schlattmann, Stefan Schulz et al.
y2
z2 x2
p {2}
Jk,2 2Jk,2

p21
pJk,2 Jk,1

y1
Jk,1 x1
p1Jk,1
z1{1}

Figure 8. Schematic diagram for determining the length of a linear actuator.

For a given velocity vector θ̇ i , i = 1, . . ., 4, the velocities of the linear

actuators q̇ i can be always determined by means of the inverse (geometrical)
JACOBI matrix J −1 i :
dq i dθ i
= J −1
i · . (10)
dt dt
Because of the modular design of the CENTAUROB, only four structurally
different inverse JACOBI matrices are needed:

• inverse JACOBI matrix of the linear actuators,

• inverse JACOBI matrix of the joints,

• inverse JACOBI matrix of the upper cylinder units,

• inverse JACOBI matrix of the lower cylinders.

In the following, we will review the inverse JACOBI matrix of the linear
actuators. For the derivaion of the other inverse JACOBI matrices, the reader is
referred to [5].
 The Parallel Two-Legged Walking Robot CENTAUROB 123

5.3. Inverse JACOBI Matrix of the Linear Actuators

The velocities of the linear actuators q̇k,1 , k = 1, . . ., 6, can be determined from

d 1
q̇k,1 = |pJk,2 Jk,1 | = · p⊤ · ṗ . (11)
dt |pJk,2 Jk,1 | Jk,2 Jk,1 Jk,2 Jk,1

By using the relation

 
0 −ωz,i ωy,i
Ṙi · R⊤ e i =  ωz,i
i =ω 0 −ωx,i  , (12)
−ωy,i ωx,i 0

the temporal derivative ṗJk,2 Jk,1 can be calculated from

e 1 · p1Jk,1 .
ṗJk,2 Jk,1 = ṗ21 + ω (13)

Substituting in Eq. (11) and simplifying results in

 ⊤
1
q̇k,1 = · p21 + p1Jk,1 − p2Jk,2 · ṗ21
|pJk,2 Jk,1 |
  ⊤  (14)
+ p e1Jk,1 · p21 − p2Jk,2 · ω1 .

Finally, the inverse JACOBI matrix of the linear actuators J −1

A1 can be com-
puted from the velocities of all six linear actuators:
   
q̇1,1 j 1,1 j 1,2  
 ..   .. ..  · ṗ21 ,
 . = . .  ω (15)
1
q̇6,1 j 6,1 j 6,2 | {z }
| {z } | {z } θ̇1
q̇1 J −1
A1

where  ⊤
p21 + p1Jk,1 − p2Jk,2
j k,1 = , (16)
|pJk,2 Jk,1 |
  ⊤
pe 1Jk,1 · p21 − p2Jk,2
j k,2 = . (17)
|pJk,2 Jk,1 |
124 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

6. Control Concepts
The motion control strategy for the CENTAUROB consists of three parts: the
execution of user commands or wishes, the stability control, and the position
control. Although the core of the strategy is formed by the execution of user
commands or wishes, the parts are structured as a three-step cascade. The sta-
bility control is designed to ensure static stability at all times while the position
control should make sure that the target position is always reached. All parts
work independently of each other by using different control parameters as well
as query and intervention times.

6.1. Execution of User Commands

Generally, the movement of the CENTAUROB can be classified into statically
stable and dynamic movement. Here, we will concentrate on the statically stable
movement. Furthermore, it has to be distinguished between a single-step and a
multi-step movement. In the single-step movement, the robot makes only one
step, whereas, in the multi-step movement, the CENTAUROB can reach a target
position out of his working area by making several steps. This requires a more
complex program than in the single-step movement. However, both movements
are nearly the same except that in the multi-step movement, the CENTAUROB
still has a command in mind after one step was made.
The overall user command execution concept is structured as follows. After
a command is entered, it is checked for its sense. Furthermore, it is necessary
to check if the parameters (these are, for example, the maximum and minimum
velocity, the acceleration, and the step size) remain in their boundaries. For this
reason, the robot’s pose must be known exactly. The step size, for example,
depends on the height of the hip and the step height, as illustrated in Fig. 9.
We can see that the maximum step size of 680 mm can be reached at a hip
height of 937 mm. In the retracted or extended hip condition, no step size can
be approached.
The verification of the user command is followed by the calculation of the
motion trajectories for every part of the motion by using the inverse kinematic
model from Section 5. The calculated trajectories are used to simulate the mo-
tion and to check the parameters one more time. After the trajectory is suc-
cessfully checked, the trajectories are enabled and sent to the motors. After
the user command is executed, the CENTAUROB returns to the initial position,
 The Parallel Two-Legged Walking Robot CENTAUROB 125

Figure 9. Step size of the CENTAUROB in longitudinal direction as a function

of the hip height and the step height.

phase zero, and the program is waiting for the next command. The overall user
command execution concept can be viewed as a loop starting with an entered
command and returning to the initial position after executing this command, see
Fig. 10.

6.2. Stability Control Concept

Static walking can be realized if the acceleration of the hip and the resulting
forces of the moving bodies can be neglected. This is possible if the weight of
the hip is significantly higher than the weight of each leg and the movement of
the hip is relatively slow. Static stability can then be reached if the ground pro-
jection of the robot’s center of mass constantly remains in the stability region,
see Fig. 2. It can be calculated from the data of the force sensors as described
in Section 4.3.2.

6.3. Position Control Concept

The position control concept is used to make sure that the CENTAUROB
reaches his target position. This can be realized, for example, by a very long
interference time. This means that the execution of user commands and the sta-
126 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

Figure 10. Overall user command execution concept for the motion control.

bility control act faster than the position control. Therewith, it is ensured that
stability is given and the user commands are executed properly. This concept
leads to long follow-up times until the target position is finally reached. Further-
more, large differences between the actual and the target position are possible,
which could cause collisions.

6.4. Power Control Concept

On top of the hip, two rechargeable batteries with 24 V DC are installed to
supply the electrical components. These are the servo motors supplied by their
motor cards and the breaks that are needed to hold the motor’s position in case of
an electricity failure. The batteries use lithium anodes and nickel-manganese-
cobalt cathodes which leads to a very high specific energy. In order to have
two independent legs with their own power supply, it is useful to separate the
power supply of each leg. Because the non-constant load on the feet during one
step results in a non-constant power consumption, it is necessary to analyze the
 The Parallel Two-Legged Walking Robot CENTAUROB 127

consumed current during one step.

In this context, three kinds of foot loads must be distinguished that appear
in the following phases: the double support phase, the single support phase,
and the swing phase. In the double support phase, both feet stand still on the
ground, and the load of the hip is distributed more or less symmetrically on the
feet, depending on the position of the hip and the ground conditions. In this
configuration, each leg requires nearly the same current, which can be easily
calculated for static conditions. Here, the whole load is concentrated on only
one leg because the other leg is in the so-called swing phase. In the swing phase,
one leg is moved to its target position, so the motors of this leg only have to
carry the weight of the corresponding foot. In this phase, the leg is stressed with
a tensile load whereas in the other phases, the leg is stressed with a compressive
load.

Figure 11. Overall current consumption I over time t during a walking step.

The three phases alternate during the motion process. The overall current
consumption during a walking step is shown in Fig. 11. In order to reduce
the average current that lays between the currents of the double support phase
and the single support phase, the holding breaks can be used during the single
support phase. Another strategy to reduce the average current is the optimization
of the walking pattern and trajectory by the parameters step size, step height,
step velocity, and acceleration.
128 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

7. Walking Scenario Simulations

In order to prove the agility of a robot with two serially arranged parallel kine-
matics, several walking scenarios were simulated. Although the robot’s focus
is walking forwards, it is still possible to walk sidewards, climbing straight and
spiral stairs, overcome obstacles, turning around, and some more special fea-
tures. The workspace boundaries are illustrated in Fig. 12.

Figure 12. Visualization of the robot’s workspace boundaries: ±680 mm in lon-

gitudinal and lateral direction with respect to the center points of the hexapods.

7.1. Walking Pattern

The statically stable movement of the CENTAUROB can be divided into six
phases, where phase zero builds the so-called initial situation. In this initial
situation, the CENTAUROB has finished the last movement or is freshly started,
and the program is waiting for a user command. The phases one to five are
part of every movement of the CENTAUROB. The only parameters that change
are the hip height, the step length and height, the speed, and the orientation of
the feet. Starting from the initial position where the CENTAUROB stands still
with both feet central under his hip (see Fig. 5), in phase one, the hip is moved
over the bigger foot to increase the stability and to relieve the other foot, see
Fig. 13(a).
In phase two, the released foot moves along the calculated trajectory towards
the target position, where the hip and the other foot remain still, see Figs. 13(a)–
(b). This trajectory is quite complex because in the initial position of the feet, the
 The Parallel Two-Legged Walking Robot CENTAUROB 129

(a) (b) (c) (d) (e)

Figure 13. Motion states of the CENTAUROB.

bigger foot nearly surrounds the smaller foot. There are three different possible
trajectories for reaching the target position without any foot collisions. One
possibility is to lift the smaller foot to the height of the bigger foot and then to
start the trajectory. The second possibility is to move the smaller foot sideways
out of the bigger foot and then to start the trajectory. Finally, it is also possible
to move one of the feet with a complex trajectory that prevents all collisions by
precalculation. Therefore, the actual position and the size of the feet have to be
known exactly.
In phase three, the hip moves from the position above the bigger foot to the
one above the smaller foot, see Figs. 13(b)–(d). Depending on the weight of the
hip, it might not be necessary to place the hip directly above the smaller foot.
This would make it possible to move the hip straight in the direction of the target
without any lateral movement.
In phase four, the bigger foot is moved under the hip, see Figs. 13(d)–(e). As
mentioned in phase two, a complex trajectory is useful to prevent from any foot
collision. Therefore, a fifth phase is necessary to move the hip into the initial
position in the middle of the feet if a hip movement above the stable standing
foot was necessary in the former phases, see Fig. 13(e).
In order to achieve an ’economic’ walking, several modifications on the
walking pattern described above were made. In this context, ’economic’ means
to use an optimized force distribution on the linear actuators to reduce the cur-
rent consumption of the batteries and increase the walking range. Furthermore,
’economic’ stands for a walking pattern that is definite, easily adjustable, with
low complexity, and identical for both feet.
In phase one and five, a hip movement between zero and 160 mm is required
130 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

(a) (b) (c)

Figure 14. Ground projection of the robot’s center of mass (GPCoM) at hip
displacements of (a) 0 mm, (b) 100 mm, and (c) 160 mm.

to stabilize the motion process. This hip movement over the bigger foot reduces
the forces along the linear actuators in the supporting leg. The ideal amount
of hip movement depends on the weight of the hip because it affects the forces
along the linear actuators as well as the lever arm of the tipping torque by the
height of the center of mass. For example, when moving the hip for 100 mm, the
forces along the linear actuators are reduced to 55% of the forces with no hip
movement. For a hip movement of about 160 mm, the forces along the linear
actuators are even reduced to a third. Figure 14 shows the ground projection of
the robot’s center of mass for different hip displacements.
There are a few parameters influencing the foot trajectory. The step size, for
example, has two limits, the minimum step size, which is 238 mm and results
from the shape of the feet, and the maximum step size that depends on the step
height, the hip movement during a walking step, and the height of the hip as
shown in Fig. 9.
Phase three, where the hip is moved from the position above the bigger foot
to the one above the smaller foot, is necessary to move the center of mass from
the position above one foot to the position above the other foot. The trajectory
is statically stable because in this so-called double support phase, the ground
projection of the center of mass always remains in the stability region formed
by both feet standing on the ground as illustrated in Fig. 15(a). Figure 15(b)
shows the ground projection of the robot’s center of mass for a whole walking
step.
For huge step sizes, for example, 680 mm in longitudinal or 640 mm in lat-
eral direction, in phases one and four, it gets necessary to move the hip contrary
to the moving direction of the CENTAUROB. This ensures that the ground pro-
 The Parallel Two-Legged Walking Robot CENTAUROB 131

(a) (b)

(c)

Figure 15. Ground projection of the robot’s center of mass (GPCoM): (a) in
phase three, (b) for a whole walking step, and (c) for huge walking steps with
and without additional hip displacement.

jection of the center of mass remains in the stability region. Figure 15(c) shows
the differences of the ground projection of the center of mass with and without
an additional hip movement in the phases one and five for forward walking.

7.2. Walking Forwards and Sidewards

The forward walking of the CENTAUROB is already described above. Addi-
tionally to this movement, where the second foot only follows the first one, it
is also possible to perform human-like steps where one foot overtakes the other
one. Both forward movements have their advantages. For example, the first sce-
nario leads to lower forces in the linear actuators and therewith to a lower current
consumption. Furthermore, it preserves a higher amount of stability since the
displacement of the ground projection of the center of mass in the walking di-
rection in phase two depends on the step size. The second scenario is a more
familiar motion. Due to the omission of an intermediate step, a higher moving
velocity for the same step sizes and step velocities can be achieved. This second
132 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

moving scenario, however, has a disadvantage in terms of an economic walking

pattern. After one step, the robot is not in its initial pose. In order to return to
the initial, stable pose, another step is necessary. Both walking scenarios with
its differences are illustrated in Figs. 16 and 17.

(a) (b) (c) (d)

(e) (f) (g) (h)

(i)

Figure 16. Moving forwards, sequence one: Moving the hip above the bigger
foot (a)–(b), lifting the smaller foot (b)–(c), moving the smaller foot forwards to
the target position (c)–(d), moving the hip above the smaller foot (d)–(e), lifting
the bigger foot (e)–(f), moving the bigger foot forwards to the position next to
the smaller foot (f)–(g), moving the hip to the initial position above the middle
of the feet (g)–(h), starting the next step again with the hip movement (h)–(i).

In order to avoid obstacles, it might be useful to walk sidewards instead

of turning around and walking forwards. The sideward movement, shown in
Fig. 18, mainly equals the forward movement. The main difference between
these movements in the walking pattern is the larger hip movement that is nec-
 The Parallel Two-Legged Walking Robot CENTAUROB 133

(a) (b) (c) (d) (e)

(f) (g) (h) (i)

Figure 17. Moving forwards, sequence two: Moving the hip above the bigger
foot (a)–(b), lifting the smaller foot (b)–(c), moving the smaller foot forwards to
the target position (c)–(d), moving the hip above the smaller foot (d)–(e), lifting
the bigger foot (e)–(f), moving the bigger foot forwards to the target position in
front of the smaller foot (f)–(h), moving the hip above the bigger foot (h)–(i).

essary to achieve stability during walking. This hip movement is contrary to the
moving direction.

7.3. Special Features

Due to the six degrees of freedom of the hip and each foot, the robot provides
plenty of special moving features. In case an obstacle shows up or it becomes
necessary to climb steps or stairs, the robot can change the height of the hip as
well as the step height. Therewith, it is not only possible to overcome small
obstacles with heights up to 350 mm and lengths up to 300 mm, as shown in
Fig. 19, the robot can also climb steps and stairs with rises up to 400 mm, as
shown in Fig. 20.
For statically stable walking, two cases for overcoming an obstacle are pos-
sible. In the first case, the obstacle is small enough to overcome it within one
step. Here, no additional or only little hip movement is necessary so that the
ground projection of the robot’s center of mass remains in the stability region.
In the second case, the obstacle has an even surface to place a foot on top of
the obstacle. Here, the obstacle is passed within two or more steps. It is also
possible to overcome bigger obstacles but not for statically stable walking.
134 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

(a) (b) (c) (d)

(e) (f) (g) (h)

(i)

Figure 18. Moving sidewards: Moving the hip above the bigger foot (a)–(b),
lifting the smaller foot (b)–(c), moving the smaller foot sidewards to the target
position (c)–(d), moving the hip above the smaller foot (d)–(e), lifting the bigger
foot (e)–(f), moving the bigger foot sidewards to the position next to the smaller
foot (f)–(g), moving the hip to the initial position above the middle of the feet
(g)–(h), starting the next step again with the hip movement (h)–(i).

In order to operate in any environment, it is essential not only being able to

walk a curve but also being able to turn around. For example, in tight hallways,
there is not enough space to turn in a curve. Due to the specially C-shaped feet
of the CENTAUROB, it is possible to turn around nearly at a point and with only
a few steps. Figure 21 shows the scenario for the robot rotating by 90 degrees.
By combining the two motions, climbing stairs and turning around, the robot
is able to walk spiral staircases as shown in Fig. 22. This is only possible be-
cause of the special C-shaped feet. For the shown movement, the center of the
hip never leaves the stability region formed by the feet. This motion makes
the CENTAUROB a unique moving device with several possible applications as
 The Parallel Two-Legged Walking Robot CENTAUROB 135

(a) (b) (c) (d) (e) (f)

(g) (h) (i)

Figure 19. Overcoming obstacles: Lifting the hip to the designated height (a)–
(b), moving the hip above the bigger foot (b)–(c), lifting the smaller foot above
the height of the obstacle (c)–(d), moving the smaller foot above the obstacle to
the target position (d)–(e), moving the hip above the smaller foot (e)–(f), lifting
the bigger foot above the height of the step (f)–(g), moving the bigger foot to the
position next to the smaller foot (g)–(h), moving the hip to the initial position
above the middle of the feet (h)–(i).

will be discussed in the next section.

8. Fields of Usage and Applications

As a universal moving platform, the CENTAUROB can be used in several ap-
plications and environments. One possible field of usage is, for example, the
industry. An assisting or fully autonomous employment of robots in unpaved
terrain will play an important role in the future. Despite of the working area, a
lot of jobs will be transferred to robots due to their precision, velocity, safety,
and endurance. Here, parallel two-legged robots can be used, for example, as
industrial carrying devices in undefined environments ore even as autonomous
robots in rough or dangerous areas.
Another aspect, that must be thoughtfully treated, is the increasing average
136 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

(a) (b) (c) (d) (e)

(f) (g) (h) (i)

Figure 20. Climbing straight stairs: Lifting the hip to the designated height (a)–
(b), moving the hip above the bigger foot (b)–(c), lifting the smaller foot above
the height of the step (c)–(d), moving the smaller foot to the target position (d)–
(e), moving the hip above the smaller foot (e)–(f), lifting the bigger foot above
the height of the step (f)–(g), moving the bigger foot to the position next to the
smaller foot (g)–(h), moving the hip to the initial position above the middle of
the feet (h)–(i).

age of our society. Despite of this fact, the aim to preserve a self-determined life
with coping daily routines becomes stronger. However, for older people, after a
certain point in time, assistance and/or care become inevitably necessary. In this
context, a survey in Germany found out that the majority of the German citizens
does not want to move into a retirement home in old age [23]. Two-thirds of
the respondents want to live in an apartment or a house when they are old and
only 15% prefer a nursing or retirement home. Nowadays, mobility and certain
independence have a high individual and social value. Furthermore, they are an
important aspect of freedom and self-determination, that is, dignity.
One possibility to retain the mobility of elderly people is by means of techni-
cal support. Wheelchairs and rollators usually require flat environments without
fixed or exposed obstacles. Fixed obstacles are, for example, door thresholds in
the apartment or terrace entries. Making things worse, wheelchairs and rolla-
tors have often to be lifted or carried to get, for example, in the apartment or
 The Parallel Two-Legged Walking Robot CENTAUROB 137

(a) (b) (c) (d) (e) (f)

(g) (h) (i) (j) (k) (l)

Figure 21. Rotating around 90 degrees: Moving the hip above the bigger foot
(a)–(b), lifting the smaller foot (b)–(c), turning the smaller foot (c)–(d), rotating
the hip and moving it above the smaller foot (d)–(e), lifting the bigger foot (e)–
(f), rotating and moving the bigger foot to the target position (f)–(h), rotating the
hip and moving it above the smaller foot (h)–(i), lifting the smaller foot (i)–(j),
turning the smaller foot (j)–(k), moving the hip to the initial position above the
middle of the feet (k)–(l).

in a public transportation vehicle. In case of using a stair lift, the person even
has to change seats which might be a huge effort or even not be possible at
all. However, so far, existing system improvements have not shown any sig-
nificant progress in overcoming these issues. For this reason, using wheels as
locomotion concept for elderly people might be reconsidered on behalf of using
a two-legged walking robot.

Conclusion
In this chapter, we reviewed the parallel two-legged walking robot named CEN-
TAUROB, a novel robot system for statically stable walking. Due to the uni-
versal moving platform, the robot offers plenty of possible applications, for
example, as a sustainable mobility assurance in the assistance and care or as
an assisting or fully autonomous industrial robot in unstructured or dangerous
environments. By using two serially arranged parallel kinematics (hexapods) as
138 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

(a) (b) (c) (d) (e) (f)

(g) (h) (i)

Figure 22. Climbing spiral stairs: Lifting the hip to the designated height (a)–
(b), moving the hip above the bigger foot (b)–(c), lifting the smaller foot above
the height of the step (c)–(d), turning the smaller foot and moving it to the target
position (d)–(e), turning the hip and moving it above the smaller foot (e)–(f),
lifting the bigger foot above the height of the step (f)–(g), turning the bigger
foot and moving it to the position next to the smaller foot (g)–(h), turning the
hip and moving it to the initial position above the middle of the feet (h)–(i).

leg structures, this system totally renounces the use of wheels as the locomotion
concept which provides high mobility in every direction and enables walking in
unpaved terrain as well as the handling of steps, stairs, and low clearances. In
order to prove the agility of the universal robot, several walking scenarios such
as moving forwards, backwards, and sidewards as well as other motions were
presented.
We also described the generation of an economic walking pattern for stati-
cally stable walking of the CENTAUROB. In order to minimize the calculation
effort, the movement of the robot is separated into as few segments as possi-
ble. This is why for a whole step of the robot, only five different trajectories
are needed, whereby the trajectories for both feet are identical. Furthermore,
we described the necessary hip movement to reduce the forces among the linear
actuators and to improve the stability.
In order to detect the actual pose of the CENTAUROB and its orientation
relative to the ground, several sensor concepts were presented and discussed.
 The Parallel Two-Legged Walking Robot CENTAUROB 139

Knowing the pose of the robot, it is possible to focus on the localization and path
planning in an unknown environment. An algorithm for robot path planning and
obstacle avoidance was already developed at our workgroup [24].
At the current project level, we consider a walking pattern based on feed
forward control using precalculated steps. In the next level, the development of
a much more sophisticated closed-loop control system is in the focus accompa-
nied by a significantly extended sensor structure that shall provide the position,
acceleration, as well as visual information. The aim is to establish a real-time
control that allows an online computation of the entire robot motion with respect
to its environment.

References
[1] Hirose, M., and Ogawa, K., 2007. “Honda humanoid robots development”.
Philosophical Transactions of the Royal Society A, 365, pp. 11–19.
[2] Lohmeier, S., Buschmann, T., and Ulbrich, H., 2009. “Humanoid robot
LOLA”. In Proceedings of the IEEE International Conference on Robotics
and Automation (ICRA), pp. 775–780.
[3] Lim, H., and Takanishi, A., 2007. “Biped walking robots created at
Waseda University: WL and WABIAN family”. Philosophical Transac-
tions of the Royal Society A, 365, pp. 49–64.
[4] Schlattmann, J., 1997. “Development of a highly flexible two-legged walk-
ing device”. In Proceedings of the 4th International Conference on Com-
puter Integrated Manufacturing, pp. 1287–1294.
[5] Du, S., Schlattmann, J., Schulz, S., and Seibel, A., 2015. “New prototype
of the two legged robot CENTAUROB”. In Proceedings of the ASME
2015 International Mechanical Engineering Congress & Exposition, Paper
ID 50354.
[6] Schlattmann, J., and Hampel, J., 2000. “Performance of the CENTAUROB
at production plants”. In Proceedings of the 5th International Conference
on Computer Integrated Manufacturing, pp. 425–433.
[7] Schlattmann, J., Tesfu, M. T., and Ziemen, L., 2007. “New steps in the
development of the two legged robot CENTAUROB”. The International
Journal of Modern Engineering, 7(2).
140 Shucen Du, Josef Schlattmann, Stefan Schulz et al.

[8] Gough, V. E., and Whitehall, S. G., 1962. “Universal tire test machine”. In
Proceedings of the 9th International Technical Congress FISITA, Vol. 117,
pp. 117–137.
[9] Stewart, D. A., 1966. “Platform with six degrees of freedom”. Proceedings
of the Institution of Mechanical Engineers, 180(15), pp. 371–386.

[10] Clavel, R., 1988. “Delta, a fast robot with parallel geometry”. In Proceed-
ings of the 18th International Symposium on Industrial Robots, pp. 91–
100.
[11] Garcia, E., Jimenez, M. A., Santos, P. G. D., and Armada, M., 2007. “The
evolution of robotics research”. IEEE Robotics & Automation Magazine,
14(1), pp. 90–103.
[12] Evans, J. M., 1994. “Helpmate: an autonomous mobile robot courier for
hospitals”. In Proceedings of the IEEE/RSJ International Conference on
Intelligent Robots and Systems (IROS), pp. 1695–1700.
[13] Baalbaki, H., Lamiri, M., and Xie, X., 2008. “Joint location and con-
figuration of mobile service robots for hospitals”. In Proceedings of the
IEEE International Conference on Automation Science and Engineering
(CASE), pp. 615–620.
[14] Ozkil, A., Fan, Z., Dawids, S., Aanes, H., Kristensen, J., and Christensen,
K., 2009. “Service robots for hospitals: A case study of transportation
tasks in a hospital”. In Proceedings of the IEEE International Conference
on Automation and Logistics (ICAL), pp. 289–294.
[15] Riebe, S., and Ulbrich, H., 2003. “Modelling and online computation
of the dynamics of a parallel kinematic with six degrees-of-freedom”.
Archive of Applied Mechanics, 72(11–12), pp. 817–829.

[16] Du, S., Schlattmann, J., Schulz, S., and Seibel, A., 2016. “Passive rotation
compensation in parallel kinematics using quaternions”. Proceedings in
Applied Mathematics and Mechanics, 16. To appear.
[17] Arai, T., Cleary, K., Nakamura, T., Adachi, H., and Homma, K., 1990.
“Design, analysis and construction of a prototype parallel link manipula-
tor”. In Proceedings of the IEEE International Workshop on Intelligent
Robots and Systems (IROS), pp. 205–212.
 The Parallel Two-Legged Walking Robot CENTAUROB 141

[18] Stoughton, R., and Arai, T., 1991. “Optimal sensor placement for forward
kinematics evaluation of a 6-dof parallel link manipulator”. In Proceed-
ings of the IEEE/RSJ International Workshop on Intelligent Robots and
Systems (IROS), pp. 785–790.

[19] Merlet, J.-P., 1993. “Closed-form resolution of the direct kinematics

of parallel manipulators using extra sensors data”. In Proceedings of
the IEEE International Conference on Robotics and Automation (ICRA),
pp. 200–204.

[20] Riebe, S., 2005. Aktive Schwingungsisolierung und Bahnregelung von

Hexapodsystemen. Fortschritt-Bereichte VDI, Düsseldorf, Germany.

[21] Tesfu, M. T., 2013. Systematic Development of a Real-time Motion Con-

trol System for a Coupled Parallel Kinematics Robot. Der Andere Verlag,
Uelvesbuell, Germany.

[22] Raghavan, M., 1993. “The stewart platform of general geometry has 40
configurations”. Journal of Mechanical Design, 115(2), pp. 277–282.

[23] Welt, D., 2011. Deutsche wollen im Alter selbstständig wohnen, January.
Online; posted 17-January-2011.

[24] Shehata, H., and Schlattmann, J., 2012. “Reactive algorithm for mobile
robot path planning among moving target, obstacles by means of dynamic
virtual obstacle concept”. In Proceedings of the International Conference
on Flexible Automation and Intelligent Manufacturing (FAIM).
In: Parallel Manipulators ISBN: 978-1-63485-926-4
Editor: Cecilia Norton, pp. 143-166
c 2016 Nova Science Publishers, Inc.

Chapter 5

S LIDING -M ODE T RACKING C ONTROL OF

THE 6-DOF 3-L EGGED W IDE -O PEN
PARALLEL R OBOT
Mohammad H. Abedinnasab1,∗, Jaime Gallardo-Alvarado2
Bahram Tarvirdizadeh3 and Farzam Farahmand4,5
1 Department of Biomedical Engineering, Rowan University,

Glassboro, New Jersey, US

2 Department of Mechanical Engineering,

Instituto Tecnológico de Celaya, TNM,

Celaya, GTO, México
3 Faculty of New Sciences and Technologies,

University of Tehran, Tehran, Iran

4 School of Mechanical Engineering,

Sharif University of Technology, Tehran, Iran

5 Research Center of Biomedical Technology and Robotics,

Tehran University of Medical Sciences, Tehran, Iran

Abstract
In this chapter, we analyze and robustly control the 6-DOF 3-legged
Wide-Open parallel manipulator, using a Lyapunov analysis approach.
The mechanism is evaluated in terms of kinematics, workspace, and dy-
namics. By implementing the dynamic model, tracking control of the
∗ E-mail address: [email protected]
144 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

moving platform is performed using the nonlinear sliding-mode control

method. The numerical simulations with desired trajectory inputs attest to
the excellent performance and robustness of the designed position track-
ing sliding-mode controller, in presence of fast varying uncertain param-
eters and external disturbances, which are common in parallel manipula-
tors.

1. Introduction
Parallel manipulators have many distinct advantages over their serial counter-
parts; such as greater stiffness, accuracy, speed, and payload capabilities. Those
are mainly due to the parallel linkages which distribute the payload and aver-
age out the positioning errors, since without parallel chains, the payload and
positioning errors accumulate [1, 2, 3, 4].
Many researchers have been studying on those manipulators after Gough
and Whitehall [5] first introduced a 6-DOF parallel manipulator with an ap-
plication in tire-testing equipment, followed by Stewart [6], who designed a
similar robot to be used as a flight simulator, known as the Gough-Stewart plat-
form. The Gough-Stewart platform consists of two platforms, one fixed and the
other moving, connected with six extensible legs. Many variants of the Gough-
Stewart platform have been introduced in the literature for different applications
[1, 7, 8, 9, 10, 11, 12, 13, 14, 15].
However, the lower number of legs induce less interferences between the
kinematic chains, enlarging the workspace of the robot. Monsarrat and Gosselin
[16] designed a 3-legged 6-DOF mobile platform connected to the base by three
identical kinematic chains using five-bar linkages. Badescu and Mavroidis [17]
performed the workspace optimization of 3-legged translational UPU and orien-
tational UPS parallel platforms under joint constraints. Lu et al. [18] proposed a
4-DOF over-constrained RRPU + 2UPU parallel manipulator (PM) with 3 legs.
Zhang and Zhang [19] proposed a novel 2-DOF parallel manipulator with 3 legs,
which is applied in vehicle driving simulators. Coppola et al. [20] designed a 3-
legged self-reconfigurable architecture. It addresses the realm of reconfigurable
6-DOF parallel mechanisms, for sustainable manufacturing. Azulay et al. [21]
proposed a 3-legged 6-DOF architecture based on a 3-PPRS topology that can
achieve high (end-effector) tilt angles with enhanced stiffness.
The purpose of the present study is to analyze and control the 6-DOF 3-
legged Wide-Open parallel robot for being used in various applications. The
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 145

proposed parallel manipulator is a 6-DOF spatial platform mechanism. It con-

sists of a fixed base platform and a moving platform, which are connected by
three extensible legs. The mechanism enjoys a unique non-symmetric architec-
ture with a frontally open half-plane, enabling it to easily embrace and manipu-
late column-shape objects.
On the other hand, accurate trajectory tracking control of parallel robots is a
key system requirement, as these devices must often follow prescribed motions,
and high performance control strategies can significantly improve the tracking
performance [22, 4]. Tracking control of parallel robots has been approached
using both linear and nonlinear control laws [23, 4].
There are two design frameworks for the control scheme of the parallel
robots. One is to design a controller based on the joint space coordinate and the
other is based on the workspace coordinate. The former is a conventional con-
trol idea, which is used to design a controller to set the actuators’ displacements
based on the desired displacements computed from the position command of
the manipulator by inverse kinematics. If the actuators’ displacements are well
controlled, the moving platform of the parallel manipulator moves based on the
reference trajectory. Many controllers in the literature are based on the joint
space coordinate [22, 24, 2]. However, the controllers based on the workspace
coordinate are directly controlled based on the position of the moving platform,
and not the actuators.
The rest of the chapter is organized as follows. In Section 2, the Wide-Open
parallel mechanism is introduced and described. The architecture of the mech-
anism is analyzed and an experimental evaluation is performed. The kinematic
analysis of the mechanism is performed in Section 3. Workspace analysis with
considering different constraints on the spherical joints is presented in Section
4. In Section 5, inverse and forward dynamic analyses of the mechanism are
discussed, and a numerical simulation is performed. Position control of the
robot using the sliding-mode algorithm, as well as illustrative numerical simu-
lations are reported in Section 6. Finally, a general conclusion on the chapter is
provided in Section 7.
146 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.
☛ ☞ ✌ ✍ ✎ ✏ ✑

T ✄

✂ ✁

\ ✄
✁

T ✄
✁

J ✄
✞

✂
✁
✟

✠
\ ✄
✝

✒ ✓ ✔ ✒ ✕ ✔

Figure 1. (a) Illustration of the 6-DOF 3-legged Wide-Open parallel manipula-

tor. There are two actuators per leg. Rotary actuators are resting on the fixed
platform, and linear actuators are moving. (b) Kinematic variables of i-th leg
are shown. Active rotation is θi around xi axis, along with the passive rotation
of ψi around ni axis. A linear actuator sets the length di , which is the distance
between Ai and Bi .

2. The Wide-Open Robot Description

2.1. The Manipulator’s Architecture
The schematics of the 6-DOF 3-legged Wide-Open mechanism is shown in Fig-
ure 1 (a). The mechanism is another version of the basic 3-leggd parallel manip-
ulator with symmetrical legs [25, 26, 27]. Each leg is composed of three joints;
universal, prismatic, and spherical (Figure 1 (b)). A rotary and a linear actuator
are used to actuate each leg. The rotary actuators, whose shafts are attached to
the lower parts of the linear actuators through universal joints, are placed on a
semicircle on the fixed platform. The spherical joints connect the upper parts of
the linear actuators to the moving platform.
By replacing the passive universal joints in the Stewart mechanism with ac-
tive joints in the proposed mechanism, the number of legs could be reduced
from 6 to 3. This change makes the mechanism lighter, since the rotary actua-
tors are resting on the fixed platform, which allows for higher accelerations to
be available due to smaller inertial effects. More importantly, in the proposed
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 147

parallel manipulator, the legs are configured non-symmetrically on a semicir-

cle on the base and moving platforms. This frontally wide open architecture,
enables the mechanism to embrace and manipulate column-shape objects. The
applications of such mechanism are versatile, from fracture reduction of long
bones in surgical robotics to column climber robots in industrial robotics.
As shown in the Figure 1, coordinate Ci (Ai ,xi ,yi ,zi ) is attached to the base
platform with its xi axis aligned with the rotary actuator in the xi direction, and
its zi axis perpendicular to the fixed platform. xi is rotated by γi from the X
direction of fixed platform coordinate A(O, X,Y, Z). The rotary actuators are
located at the positions Ai (for i = 1, 2, 3) on the base platform and each shaft
is connected to the lower part of the linear actuators through a universal joint
(Figure 1). The upper parts of linear actuators are connected to the moving
platform, Bi points, through spherical joints (Figure 1).
Cartesian coordinates A(O, x, y, z) and B(P, u, v, w) represented by {A} and
{B} are attached to the base and moving platforms, respectively. In Figure 1,
si represents the unit vector along the axes of i-th rotary actuator and di is the
vector along Ai Bi with the length of di . Assuming that each limb is connected
to the fixed base by a universal joint, the orientation of i-th limb with respect to
the fixed base can be described by two successive rotations, rotation θi around
the axis si , followed by rotation ψi around ni , which is itself perpendicular to
both di and si . It is to be noted that θi and di are active joints, actuated by the
rotary and linear actuators, respectively, while, ψi is an inactive joint.

2.2. Experimental Evaluation

The performance of the Wide-Open mechanism was evaluated in an experimen-
tal study on a fully functional prototype of the robot. As shown in Figures 2 and
3, this prototype has full-circular fixed and moving platforms. Each of the three
legs is equipped with a rotary actuator located on the base platform. This rotary
actuator consist of a stepper motor followed by a low-ratio gearbox to enlarge
the shaft torque. The gearbox shaft is then connected to the lower part of the
leg using a revolute joint which is made of a miniature needle bearing. Linear
actuation at each leg is provided using a ball screw system powered by another
smaller stepper motor. Finally, the spherical joints connect the upper parts of
the legs to the moving platform.
In the experimental tests, an optical stereoscopic vision system (Micron-
Trackers, Claron Technology Inc., Ontario, Canada) was used as an external ob-
148 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

Figure 2. The fabricated prototype of the 3-legged Wide-Open robot. Full rings
have been used for both fixed and moving platforms. At each leg, the ball screw
actuator, driven by a stepper motor, is providing the precise linear motion. The
rotary motion of each leg is generated by another stepper motor, attached to a
low-aspect-ratio gearbox.

server to track the robot’s end-effector, i.e., the center of the moving platform.
In the setup, the 3D camera detects a set of markers attached to the moving
platform to find its position and orientation with respect to a fixed Cartesian
coordinate system, defined using another set of markers attached to the frame.
In the experimental tests, at first, the workspace of the robot was measured
by running its actuators in different directions to the end of the moving range,
and recording the end-effector position by the optical tracker. Trajectory track-
ing experiments were performed to verify the robot’s kinematical model and to
assess the effects of the friction and backlash. In each test, a desired trajec-
tory, e.g., an offset circle, was defined for the end-effector to be followed and
the corresponding joint trajectory for each actuator was computed, using the in-
verse kinematic equations, to generate the position data of the actuators. The
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 149

Figure 3. Top view of the fabricated Wide-Open robot, Figure 2. Since the
legs are positioned in one half plane, the other half plane remains open. This
non-symmetric placement of the legs, increases the workspace and gives a wide
usable space between the two plates.

trajectory tracking experiments of the robot revealed reasonably acceptable re-

sults. The robot was capable of moving the end-effector on a 40 (mm) radius
circle with a maximum deviation from the reference trajectory of about 8%.

3. Kinematic Analysis
One of the most important issues in the study of parallel manipulators is the
kinematic analysis, where the generated results determine the applicability of
the designed mechanism. In what follows, the inverse kinematic analysis of the
parallel manipulator is formulated. It is to be noted that, the forward kinematic
problem of this robot has 16 solutions at most, and its complete analysis is out
of scope of this chapter.
Referring to Figure 1, ai which represents OAi can be written as

ai = g[ cos γi sinγi 0 ]T ,
150 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

where g and h are the radius of the fixed and moving platforms, respectively.
B b represents the position of the i-th joint on the platform in the moving frame
i
{B}, B bi = PBi )B . B bi is constant and is equal to B bi = h[ cos γi sinγi 0 ]T .
We can represent AB R = [ri j ], the rotation matrix from A to B , using Euler angles
as  
cα2 cα3 −cα2 sα3 sα2
A (1)
BR = cα3 sα2 sα1 + sα3 cα1 −sα3 sα2 sα1 + cα3 cα1 −cα2 sα1  ,

−cα3 sα2 cα1 + sα3 sα1 sα3 sα2 cα1 + cα3 sα1 cα2 cα1

where sα1 = sinα1 and cα1 = cos α1 , and so on. α1 , α2 , and α3 are three Euler
angles defined according to the x − y − z convention. Thus, the vector B bi would
be expressed in the fixed frame {A} as
bi = AB R PBi)B . (2)
Let p and ri denote the position vectors for P and Bi in the reference frame
{A}, respectively. From the geometry, it is obvious that
ri = p + bi . (3)
Subtracting vector ai from both sides of (3) one obtains
ri − a i = p + bi − ai . (4)
Left hand side of (4) is the definition of di , therefore
di2 = (p + bi − ai ) · (p + bi − ai ). (5)
Using Euclidean norm di can be expressed as
r 
di = (x − xi )2 + (y − yi )2 + (z − zi )2 , (6)

in which 
 xi = −h (cos γi r11 + sin γi r21 ) + g cos γi
yi = −h (cos γi r12 + sin γi r22 ) + g sin γi . (7)
zi = −h (cosγi r13 + sin γi r23 )


Coordinates Ci (Ai , xi , yi , zi ) are connected to the base platform with their xi

axes aligned with the rotary actuators in the si directions, with their zi axes
perpendicular to the fixed platform. Thus, one can express vector di in {Ci } as
 
sinψi
Ci
di = di  − sin θi cosψi  , (8)
cosθi cos ψi
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 151

and from the geometry is clear that

ri = ai + CAi RCi di , (9)

where CAi R is the rotation matrix from {Ci } to {A},

 
cos γi − sin γi 0
A
Ci R =
 sinγi cosγi 0  . (10)
0 0 1

By equating the right sides of (3) and (4) ), and solving the resultant equa-
tion, ψi and θi can be calculated as follows:
 
−1 cos γi (x − xi ) + sin γi (y − yi )
ψi = sin , (11)
di

and  
−1 sin γi (x − xi ) − cos γi (y − yi )
θi = sin . (12)
di cosψi

4. Workspace Analysis
Consider the Wide-Open mechanism with g = 1 (m) and h = 0.5 (m), where g
and h are the radii of the fixed and moving platforms, respectively. By assum-
ing a cubic with 1 (m) length, 1 (m) width and 1 (m) height located 0.25 (m)
above the base platform, we are interested in determining the volume percentage
in which the mechanism can successfully reach the locations within this cubic
space.
The experimental workspace of the Wide-Open robot is illustrated in Figure
4. There were two main mechanical constraints in the fabricated prototype that
limited its workspace. The first constraint was the length change of the linear
actuators, di , being between 0.22 (m) to 0.40 (m). The other restriction was
imposed by the spherical joints, θSJ
i . The spherical joints, used in our prototype,
could rotate only up to ±25 (deg). The workspace of the mechanism can be
enlarged to the theoretical one if these constrains are removed. Figure 4 also
illustrates the two potential enlarged workspaces of the mechanism; one with
the increased spherical rotation limits of ±50 (deg), and the other with an ideal
spherical joint without a rotational constraint.
152 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

Figure 4. Workspaces of the Wide-Open parallel mechanism with different con-

straints on the spherical joint angles. Spherical joints are considered to have no
limit, restricted to rotate up to ±50 (deg), and to ±25 (deg); respectively.

5. Dynamic Analysis
5.1. Inverse Dynamics
Due to the numerous constraints imposed by closed loops of a parallel manip-
ulator, deriving explicit equations of motion in terms of a set of independent
generalized coordinates become a prohibitive task. In this regard, the principle
of virtual work appears to be the most efficient method of analysis. In this sec-
tion we apply the principle of virtual work to derive a transformation between
joint torques/forces and end-effecter forces/moments.
For convenience, we introduce a six-dimensional wrench F̂i as the sum of
applied and inertia wrenches about the center of mass of link i. Similarly, we
assume Fˆp to denote the sum of applied and inertia wrenches about the center of
mass of the moving platform. Then the principle of virtual work for a parallel
manipulator can be stated as
δ qT τ + δ xT Fˆp + ∑ δ xTi F̂i = 0, (13)
i

where, τ is the wrench applied to the actuators. δ x is the virtual displacement

and rotation of the center of the moving platform. δxi is the virtual displace-
ment and rotation of the center of link i, and δq is the virtual displacement and
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 153

rotation of q. One can relate δ q and δ x as follows

δq = Jp δ x, Jp = Jq−1 Jx . (14)

By introducing Ji as link’s jacobian matrix, we have

δ xi = Ji δx. (15)

Substituting (14) and (15) into (13), results in

JpT τ + Fˆp + ∑ JiT F̂i = 0. (16)

i

If Jp is a square matrix, solving for τ in (16) is straight forward, which

results in !
τ = −Jp−T Fˆp + ∑ JiT F̂i . (17)
i

It is to be noted that, errors in the system parameters may result in deviation

of the platform position and orientation from those desired. Therefore, in Sec-
tion 6, a robust sliding-mode controller is designed to cancel out the effects of
parameters uncertainties and external load disturbances.

5.2. Forward Dynamics

In this section, the forward dynamic analysis of the Wide-Open parallel robot is
summarized. The robot is a 6-DOF closed kinematic chain mechanism consist-
ing of a fixed base, a moving platform, and 3 legs connecting the two platforms.
The basic procedure in deriving the forward dynamic model is using Newton-
Euler method, which incorporates the constraint forces in a compact form. The
general equations of motion for the mechanism, in the workspace coordinates,
could be written as [2]

M(x)ẍ + H(x, ẋ) = JpT τ , (18)

where M(x) is the 6 × 6 mass matrix and H(x, ẋ) is a 6 × 1 nonlinear vector
icluding coriolis, centrifugal and gravitational effects. Forward dynamic prob-
lem would be to solve (18) to obtain ẍ as

ẍ = −M(x)−1 H(x, ẋ) + M(x)−1 JpT τ . (19)

154 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.
20 -10 50

15 -15 45
x (cm)

y (cm)

z (cm)
10 -20 40

5 -25 35

0 -30 30
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

-5 25 35

-10 20 30
α (deg)

β (deg)

γ (deg)
-15 15 25

-20 10 20
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 5. Reference trajectory of the center of the moving platform. Frequency

of the sinusoidal movements is 2π/10. x, y, and z are translational motions and
α, β, and γ are the roll, pitch, and yaw angles; respectively.
30 35 0

20 -10
30
θ1 (deg)

θ2 (deg)

θ3 (deg)

10 -20
25
0 -30
20
-10 -40

-20 15 -50
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

20 10 20

10 5 10
θ̇1 (deg/s )

θ̇2 (deg/s )

θ̇3 (deg/s )

0 0 0

-10 -5 -10

-20 -10 -20

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 6. Time history of the angular rotations, θi , and angular velocities, θ̇i , of
the rotary actuators, based on the reference trajectory in Figure 5.

5.3. Dynamic Simulations

In this section, we numerically investigate the inverse dynamic solution of the
Wide-Open mechanism. The mass of the moving platform is 0.3 (kg). The
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 155
55 48 60

46
50
d1 (cm) 55

d2 (cm)

d3 (cm)
44
45
42
50
40
40

35 38 45
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

10 4 5

5 2
ḋ1 (cm/s)

ḋ2 (cm/s)

ḋ3 (cm/s)
0 0 0

-5 -2

-10 -4 -5
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 7. Time history of the linear displacements, di , and linear velocities, d˙i ,
of the linear actuators, based on the reference trajectory in Figure 5.
3 -2 3

2 2.5
T 1 (N.m )

T 2 (N.m )

T 3 (N.m )

-2.5
1 2
-3
0 1.5

-1 -3.5 1
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

9 3 6

8 5
2
F 1 (N )

F 2 (N )

F 3 (N )

7 4
1
6 3

5 0 2
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 8. Inverse dynamic solution, i.e., actuators’ torques and forces, based on
the reference trajectory in Figure 5.

masses of the lower part and upper part of each leg are 0.9 (kg) and 0.4 (kg);
respectively. radii of the moving platform and the fixed platform are 0.14 (m)
and 0.18 (m); respectively. The input reference trajectory is shown in Figure
156 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

5. As shown in the figure, the oscillations have the same frequency of 2π/10,
in all translational and rotational degrees of freedom. All three translational
movements have the same amplitudes of 0.1 (m), while their initial conditions
are different. Similarly, the rotational movements have the same amplitudes of
0.1 (rad) or 5.73 (deg), with different initial conditions.
Based on the same reference trajectory, time history of the rotations of the
three active rotary actuators are shown in Figure 6. Both angular rotations, θi ,
and angular velocities, θ̇i , are shown in the figure. Time history of the displace-
ments of the three linear actuators are shown in Figure 7. Linear displacements,
di , and linear velocities, d˙i , are shown in the figure.
Inverse dynamic solution of the reference trajectory of Figure 5, is shown in
Figure 8. The actuators’ torques and forces, τ , are calculated from Eq. (17). It
is to be noted that, the results for the two time periods of 10 (s) are identical.

6. Position Control Analysis

6.1. Sliding-Mode Control Design
In this section, the sliding-mode control technique, which is an effective way
for controlling nonlinear systems with un-modeled uncertainties [4], is used to
develop a controller allowing tracking of the reference inputs for the 6-DOF
Wide-Open mechanism. Desired Cartesian trajectories are used as input com-
mands to the controller. The controller then sets the actuators torques and forces
in the real-time.
In absence of disturbances and uncertain terms, the control problem would
be easily accomplished by replacing τ from inverse dynamic solution, (17), into
(19). However, in real situations, one should consider the effects of disturbances
and uncertainties by introducing the 6 × 1 vector ∆ into (19) as

ẍ = −M(x)−1 H(x, ẋ) + M(x)−1 JpT τ + ∆, (20)

in which ∆ = [∆1 , ∆2 , ∆3 , ∆4 , ∆5 , ∆6 ]T , with the following boundaries,

|∆i | ≤ δi , (21)

for i = 1 to 6. It is to be noted that, it would be difficult to simultaneously

guarantee the accuracy and robustness of a controller, when the system load is
uncertain and varying. A lot of research studies design robust controllers to deal
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 157

with the external disturbances. However, it depends on the assumptions that the
external disturbances are bounded [4].
To derive input actuators forces and torques, τ , for the general system (20),
one can use the following definition,

τ = τ inv − τ ∆ , (22)

where τ inv stands for the input forces and torques required for the nominal sys-
tem, (19), and τ ∆ is present to cancel the effects of the disturbances and uncer-
tainties, ∆. Replacing (22) into (20), one obtains

ẍ = ẍdes + ∆ − M(x)−1 JpT τ ∆ , (23)

in which, xdes is the desired 6 × 1 vector of position and orientation of the center
of the moving platform. By defining W as

W = M(x)−1 JpT τ ∆ , (24)

τ ∆ will be obtained as
τ ∆ = Jp−T M(x)W. (25)
Replacing (24) into (23) yields to

ẍ = ẍdes + ∆ − W. (26)

The displacement error variable zi , and its derivative zdi , are respectively
defined as,
zi = xi − xdes
i , (27)
and
zdi = ẋi − ẋdes
i . (28)
Using Eqs. (26), (27), and (28), we can reach to the following second-order
state-space system,
żi = zdi ,

(29)
żdi = ∆i −Wi .
One can define the sliding surface Si as follows,

Si = (d/dt + λi )zi = zdi + λi zi , (30)

158 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

where the λi parameter is chosen based on the desired convergence speed of the
controller. The derivative of the sliding surface is obtained as,

Ṡi = ∆i −Wi + λi zdi . (31)

In what follows, we use saturation function, sat(.), for omitting the noises
generated from the sign(.) function,
eq
Wi = Wi + βi sat(Si /φi), (32)

in which, smaller values for φi result in lower errors by increasing the oscilla-
tions, while larger values for φi increase the error and decrease the oscillations.
One may use the following formulation for φi selection based on the maximum
resultant error of displacement, zmax
i ,

λi zmax
i ≤ φi . (33)
eq
According to the definition, Wi is calculated from (31) by setting Ṡi and ∆i
to zero,
eq
Wi = λi zdi . (34)
We define the Lyapunov function as,

S2i
Vi = , (35)
2
where its derivative is bounded by

V̇i = Si Ṡi ≤ −ηi |Si |. (36)

By replacing Eqs. (30) and (31) into (36), and after simplifications, we reach
to the following condition for βi ,

δi + ηi ≤ βi . (37)

As a result, all the design parameters have defined boundaries. Finally, by

replacing (32) into (25), the vector of input forces and torques due to existence
of disturbances and uncertainties is determined. The general algorithm of the
proposed closed-loop sliding-mode controller is illustrated in Figure 9.
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 159
x des inv Forward Dynamics + x
Cartesian Inverse Dynamics +  Disturb. & Uncertain.
Desired Trajectory x des Eq. (17) _ Eq. (22) Eq. (20) x

_ z
Eq. (27)
Sliding-mode 
+ Controller 
_ z d Eq. (25)
+ Eq. (28)

Figure 9. Proposed closed-loop tracking control schematics. The sliding-mode

controller generates the required forces and torques to cancel out the negative
effects of the disturbances and uncertainities on the system.

6.2. Position Control Simulations

To verify the effectiveness and robustness of the proposed control algorithm,
the tracking performance of the controller is evaluated in this section. The pro-
posed control method is numerically evaluated in three case studies. The control
parameters λi , φi , and βi are respectively set to 5, 2, and 50 through all the nu-
merical simulations.

6.2.1. Initial Condition Alteration

Initial condition alteration, is a classic way of analyzing the robustness and per-
formance of a controller. In this section, we have altered the initial position
of each of x, y, and z translational degrees of freedom by 0.2 (m) from that of
the reference trajectory in Figure 5. As seen from the Figure 10, the controller
is well capable of following the reference trajectory with zero error, after the
first few seconds. The corresponding actuators’ forces and torques are shown
and compared to those of the reference trajectory in Figure 11. As expected, an
overshoot is seen right after the start point, then the forces and torques rapidly
converge to those of the reference trajectory.

6.2.2. Parameters Uncertainties

As another simulation, the robustness of the controller can be demonstrated by
applying the controller to the system with different parameter values. Parame-
ters uncertainties could be well modeled using this approach. To this end, every
component’s mass in the system is considered to be 20% larger than that re-
ported and initiated in the controller. This way, the controller will be faced to
160 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

the presence of a relatively large parametric uncertainty. The trajectory track-

ing results are shown in Figure 12. As seen in the figure, even though there
is a strong parametric uncertainty in the system, the controller has successfully
tracked the reference trajectory in a satisfactory fashion. It is to be noted that,
unlike the initial condition alteration, the parametric difference of 20% is always
present in the controller, and therefore the output trajectory can not converge to
the reference trajectory after awhile. The corresponding actuators’ forces and
torques are shown and compared to those of the reference trajectory in Figure
13. As seen in the figure, an offset is seen between most pairs of the gray and
black lines. This simulation, illustrates the robustness of the proposed sliding-
mode controller in the existence of parameters uncertainties.
30 -10 60

20 -20 50
x (cm)

y (cm)

z (cm)
10 -30 40

0 -40 30
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

0 35
30

-5 30
25
α (deg)

β (deg)

γ (deg)

-10 20 25

-15 15
20
10
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 10. Initial condition of the system is altered with that of the reference
trajectory. The initial position of each of x, y, and z differs 0.2 (m) with that
of the reference trajectory. The initial orientation of each of α, β, and γ differs
0.2 (rad) or 11.46 (deg), with that of the reference trajectory. Black line: output
trajectory of the system. Gray line: reference trajectory, i.e., Figure 5.

6.2.3. External Disturbances

Load disturbances are the main external disturbances of the parallel manipula-
tors, and greatly impact the system performance [4]. To study the robustness of
the proposed controller when an external disturbance occurs, the moving plat-
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 161
6 3

4 -5
T 1 (N.m )

T 2 (N.m )

T 3 (N.m )
2

2
-10 1
0
-15 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

10 0 5

-10
F 1 (N )

F 2 (N )

F 3 (N )
0

5 -20 -5

-30 -10
0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 11. Black line: actuators’ forces and torques corresponding to the ini-
tial condition change, i.e., Figure 10. Gray line: actuators’ forces and torques
corresponding to the reference trajectory, i.e., Figure 8.

20 -10 50

15 -15 45
x (cm)

y (cm)

z (cm)

10 -20 40

5 -25 35

0 -30 30

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

24
-5 35
22
20
α (deg)

β (deg)

γ (deg)

-10 18 30

16

-15 14 25
12
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 12. Mass of the components of the system is increased by 20% compared
to the data initiated in the controller model. Black line: output trajectory of the
system. Gray line: reference trajectory, i.e., Figure 5.

form is hit with an external load from three directions of x, y, and z. The load
162 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.
3
-2.5 3
2
T 1 (N.m )

T 2 (N.m )

T 3 (N.m )
-3 2.5
1

0 2
-3.5
-1 1.5
-4
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

11
6
10 2
9 5
F 1 (N )

F 2 (N )

F 3 (N )
8 1
4
7
0 3
6
2
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 13. Black line: actuators’ forces and torques corresponding to the mass
change simulation, i.e., Figure 12. Gray line: actuators’ forces and torques
corresponding to the reference trajectory, i.e., Figure 8.

hits the platform at t = 5 (s) and is detached after 0.2 (s). The magnitude of
the load is 9.81 (N), in each direction. The results of the simulation is shown
in Figure 14. The corresponding actuators’ forces and torques are shown and
compared to those of the reference trajectory in Figure 15. It is obvious from
the figures that the controller is able to cancel out the sudden impact of the ex-
ternal disturbance by properly overshooting the actuators’ forces and torques,
right after the incident. Shortly after, the output trajectory is identical with the
reference trajectory.

Conclusion
The 6-DOF 3-legged Wide-Open parallel mechanism was analyzed. In compar-
ison with the well-known Gough-Stewart platform, the passive universal joints
are replaced with active joints, reducing the number of legs from 6 to 3. Since
the active rotary actuators are resting on the fixed plate, the moving platform of
the robot can achieve higher accelerations, due to the smaller inertial effects.
The development of a sliding-mode tracking controller for the 6-DOF Wide-
Open mechanism was investigated, using the dynamic model of the parallel
robot. Lyapunov analysis approach was used in designing the sliding-mode
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 163
20 -10 50

15 -15 45
x (cm)

y (cm)

z (cm)
10 -20 40

5 -25 35

0 -30 30
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

-10
20 35
α (deg)

β (deg)

γ (deg)
-20
15 30
-30
10 25
-40

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 14. An external load hits the moving platform in all three x, y, and z
directions. The load hits the platform at t = 5 (s) and is detached after 0.2 (s).
The magnitude of the load is 9.81 (N), in each direction. Black line: output
trajectory of the system. Gray line: reference trajectory, i.e., Figure 5.
3
4 -1.5
2
T 1 (N.m )

T 2 (N.m )

T 3 (N.m )

-2
1
2
-2.5 0

0 -1
-3
-2
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

10 4

8 2 6
F 1 (N )

F 2 (N )

F 3 (N )

6 0
4
4 -2
2
-4 2

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

Figure 15. Black line: actuators’ forces and torques corresponding to the ex-
ternal load impact, i.e., Figure 14. Gray line: actuators’ forces and torques
corresponding to the reference trajectory, i.e., Figure 8.
164 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

controller. The trajectory tracking control of the 6-DOF parallel robot with un-
certain load disturbances and parameter uncertainties was demonstrated through
three numerical simulations. The simulation results show that the proposed non-
linear control strategy is robust in presence of disturbances and uncertainties.

References
[1] B. Dasgupta, T. Mruthyunjaya, The stewart platform manipulator: a re-
view, Mechanism and machine theory 35 (1) (2000) 15–40.

[2] H. Guo, Y. Liu, G. Liu, H. Li, Cascade control of a hydraulically driven

6-dof parallel robot manipulator based on a sliding mode, Control Engi-
neering Practice 16 (9) (2008) 1055–1068.

[3] I. Davliakos, E. Papadopoulos, Model-based control of a 6-dof electrohy-

draulic stewart–gough platform, Mechanism and machine theory 43 (11)
(2008) 1385–1400.

[4] Y. Pi, X. Wang, Trajectory tracking control of a 6-dof hydraulic parallel

robot manipulator with uncertain load disturbances, Control Engineering
Practice 19 (2) (2011) 185–193.

[5] V. E. Gough, S. G. Whitehall, Universal tyre test machine, in Proc 9th Int.
Tech. Congress FISITA (1962) 117–137.

[6] D. Stewart, A platform with six degrees of freedom, Proc. Inst. Mech. Eng.
180 (1965) 371–386.

[7] B. M. St-Onge, C. M. Gosselin, Singularity analysis and representation of

the general gough-stewart platform, Int. J. Rob. Res. 19 (3) (2000) 271–
288.

[8] H. Li, C. M. Gosselin, M. J. Richard, Determination of the maximal

singularity-free zones in the six-dimensional workspace of the general,
Mech. Mach. Theory 42 (4) (2007) 497–511.

[9] Q. Jiang, C. M. Gosselin, Determination of the maximal singularity-free

orientation workspace for the gough-stewart platform, Mech. Mach. The-
ory 44 (6) (2009) 1281–1293.
 Sliding-Mode Tracking Control of the 6-DOF 3-Legged ... 165

[10] Q. Jiang, C. M. Gosselin, Maximal Singularity-Free Total Orientation

Workspace of the GoughStewart Platform, J. Mech. Rob., Trans. ASME
1 (3) (2009) 1–4.

[11] Q. Jiang, C. M. Gosselin, Evaluation and Representation of the Theoreti-

cal Orientation Workspace of the GoughStewart Platform, J. Mech. Rob.,
Trans. ASME 1 (2) (2009) 1–9.

[12] B. Inner, S. Kucuk, A novel kinematic design, analysis and simulation tool
for general stewart platforms, Simulation (2013) 0037549713482733.

[13] G. Liu, Z. Qu, X. Liu, J. Han, Singularity analysis and detection of 6-

ucu parallel manipulator, Robotics and Computer-Integrated Manufactur-
ing 30 (2) (2014) 172–179.

[14] A. Karimi, M. T. Masouleh, P. Cardou, Singularity-free workspace analy-

sis of general 6-ups parallel mechanisms via convex optimization, Mecha-
nism and Machine Theory 80 (2014) 17–34.

[15] W. Zhou, W. Chen, H. Liu, X. Li, A new forward kinematic algorithm

for a general stewart platform, Mechanism and Machine Theory 87 (2015)
177–190.

[16] B. Monsarrat, C. M. Gosselin, Workspace analysis and optimal design of a

3-leg 6-dof parallel platform mechanism, Robotics and Automation, IEEE
Transactions on 19 (6) (2003) 954–966.

[17] M. Badescu, C. Mavroidis, Workspace optimization of 3-legged upu and

ups parallel platforms with joint constraints, Journal of Mechanical De-
sign 126 (2) (2004) 291–300.

[18] Y. Lu, Y. Shi, Z. Huang, J. Yu, S. Li, X. Tian, Kinematics/statics of a

4-dof over-constrained parallel manipulator with 3 legs, Mechanism and
Machine Theory 44 (8) (2009) 1497–1506.

[19] C. Zhang, L. Zhang, Kinematics analysis and workspace investigation of

a novel 2-dof parallel manipulator applied in vehicle driving simulator,
Robotics and Computer-Integrated Manufacturing 29 (4) (2013) 113–120.

[20] G. Coppola, D. Zhang, K. Liu, A new class of adaptive parallel robots,

Journal of Mechanisms and Robotics 6 (4) (2014) 041013.
166 M. H. Abedinnasab, J. Gallardo-Alvarado, B. Tarvirdizadeh et al.

[21] H. Azulay, M. Mahmoodi, R. Zhao, J. K. Mills, B. Benhabib, Comparative

analysis of a new 3× pprs parallel kinematic mechanism, Robotics and
Computer-Integrated Manufacturing 30 (4) (2014) 369–378.

[22] A. Ghobakhloo, M. Eghtesad, M. Azadi, Position control of a stewart-

gough platform using inverse dynamics method with full dynamics, in:
Advanced Motion Control, 2006. 9th IEEE International Workshop on,
IEEE, 2006, pp. 50–55.

[23] M. R. Sirouspour, S. E. Salcudean, Nonlinear control of hydraulic robots,

Robotics and Automation, IEEE Transactions on 17 (2) (2001) 173–182.

[24] J. Rosario, E. Oliveira, D. Dumu, Conception of stewart-gough plat-

form with reconfigurable control using integrate prototyping, in: Power
Electronics, Electrical Drives, Automation and Motion, 2006. SPEEDAM
2006. International Symposium on, IEEE, 2006, pp. 1039–1044.

[25] M. H. Abedinnasab, Y.-J. Yoon, H. Zohoor, Exploiting Higher Kine-

matic Performance–Using a 4-Legged Redundant PM Rather than Gough-
Stewart Platforms, InTech, 2012.

[26] M. H. Abedinnasab, G. R. Vossoughi, Analysis of a 6-dof redundantly

actuated 4-legged parallel mechanism, Nonlinear Dyn. 58 (4) (2009) 611–
622.

[27] O. Aghababai, Design, Kinematic and Dynamic Analysis and Optimiza-

tion of a 6 DOF Redundantly Actuated Parallel Mechanism for Use in
Haptic Systems, MSc Thesis, Sharif University of Technology, Tehran,
Iran (2005).
 INDEX

batteries, 105, 126, 127, 130

# biped walking, 101, 103
2R2T, viii, ix, 81, 82, 83, 84, 85, 87, 88, 89,
91, 92, 93, 94, 95, 96, 97 C
6-DOF 3-legged Wide-Open parallel
manipulator, 146 CAD, 9, 10
calibration, 42, 63
case studies, 5, 158
A case study, 141
CDC, 35
actuation, 8, 37, 38, 87, 88, 97, 147 CENTAUROB, vii, ix, 101, 102, 103, 105,
actuators, 9, 15, 19, 27, 29, 47, 50, 78, 82, 106, 107, 108, 109, 110, 111, 112, 113,
83, 97, 108, 110, 115, 116, 117, 118, 119, 114, 115, 116, 117, 119, 120, 121, 122,
120, 122, 123, 130, 132, 138, 145, 146, 123, 124, 125, 127, 128, 129, 131, 132,
147, 148, 150, 151, 152, 154, 155, 156, 133, 134, 135, 137, 138, 139, 140, 141
157, 159, 160, 161, 162, 163 challenges, 42, 104, 105
adaptation, vii, 1, 2, 3, 4, 5, 6, 16, 17, 18, closed-loop system, 1, 3, 4, 7, 19, 25, 71
19, 20, 23, 28 closure, 48, 50
adaptive control, vii, viii, 1, 2, 3, 4, 6, 7, 8, collisions, 117, 126, 129
16, 18, 21, 22, 24, 27, 32, 34, 35, 36, 37, comparative analysis, 65, 66
38 compensation, 6, 8, 35, 78, 80, 113, 116,
agility, 103, 128, 138 141
algorithm, 78, 139, 142, 145, 158, 165 competitiveness, 109
amplitude, 31 complexity, 42, 70, 130
Artificial Neural Networks, 5 compliance, 112
automation, 102 computation, 78, 105, 139, 141
autonomous navigation, 105 computed torque control, 41, 43, 63, 64, 65
computer, 27
configuration, 13, 68, 70, 72, 73, 74, 78, 95,
B 97, 110, 127, 141
connectivity, 88, 89, 96
backlash, 63, 116, 148
constant load, 127
bandwidth, 3, 19, 20
construction, ix, 101, 102, 105, 141
base, viii, 9, 10, 45, 55, 81, 82, 84, 88, 90,
consumption, 127, 130, 132
104, 112, 144, 145, 147, 150, 151, 153
control concepts, 101
168 Index

convention, 45, 46, 150 freedom, viii, 43, 44, 45, 47, 77, 81, 82, 85,
convergence, 2, 116, 157 88, 98, 99, 104, 105, 108, 110, 111, 113,
cost, 63, 64, 73, 75, 97, 116 114, 115, 118, 119, 132, 136, 140, 141,
156, 159, 164
friction, 8, 14, 25, 44, 63, 66, 68, 108, 113,
D 148
fully-isotropic, 83, 85, 87, 89, 91, 93, 94,
damping, 5 95, 97, 99
decoupling, 16, 35, 76
Delta, 9, 14, 82, 140
derivatives, 22, 46, 56, 57, 120 G
detection, 105, 115, 164
deviation, 70, 72, 73, 149, 153 geometry, 49, 98, 140, 142, 150, 151
dichotomy, 104 google, 38
displacement, 87, 91, 110, 117, 118, 131, Gough-Stewart platform, 41, 43, 44, 45, 47,
132, 152, 157, 158 48, 54, 60, 64, 72, 75, 104, 144, 161
distribution, 67, 68, 70, 73, 104, 130 gravitation, 62
dynamic control, 78 gravitational effect, 153
dynamics, viii, ix, 1, 2, 3, 5, 6, 8, 14, 15, 16, gravitational force, 16
17, 18, 19, 21, 22, 23, 25, 29, 32, 33, 45, gravity, ix, 14, 15, 43, 55, 59, 70, 78, 80,
47, 59, 66, 70, 71, 75, 76, 77, 79, 82, 102, 101
103, 112, 141, 144, 165 guidance, 104

E H
electricity, 115, 127 height, 106, 109, 115, 124, 125, 128, 129,
energy, 31, 55, 57, 58, 105, 127 130, 131, 132, 136, 137, 139, 151
energy consumption, 105 hybrid, viii, 81, 82, 83, 84, 85, 86, 89, 96
engineering, 112
environment, ix, 2, 4, 101, 111, 133, 139
environments, 102, 104, 105, 106, 134, 135, I
136, 137
equilibrium, 119 ICRA, 36, 37, 38, 39, 140, 141
equipment, 104, 144 industrial environments, 102
evolution, 4, 5, 8, 17, 23, 27, 31, 69, 140 industry, 42, 106, 134
excitation, 112 inertia, 3, 15, 43, 55, 57, 58, 59, 66, 67, 70,
execution, 124, 125, 126 71, 72, 103, 152
extracts, 52 inertial effects, 146, 163
intelligence, 102
interference, 42, 117, 126
F iteration, 67

factories, vii, viii, 41

filters, 27 J
flexibility, 103, 104, 106, 108, 112, 116
flight, 36, 82, 144 joints, viii, ix, 2, 9, 10, 11, 13, 14, 15, 16,
force, 14, 15, 75, 87, 89, 97, 111, 115, 119, 24, 42, 44, 47, 48, 49, 51, 62, 64, 66, 68,
125, 130 70, 72, 75, 82, 83, 84, 85, 86, 87, 88, 89,
90, 91, 92, 94, 95, 96, 97, 103, 104, 105,
 Index 169

106, 107, 108, 112, 113, 114, 116, 118, morphology, 99

122, 145, 146, 147, 151, 152, 161, 163 motion control, 35, 79, 103, 124, 126
motivation, 6, 33, 105
muscles, 106
K
kinematic chain structure, vii, viii, 41, 42 N
kinematic walking robot, ix, 101, 102
neural networks, 5, 6, 37
nickel, 127
L nonlinear dynamics, viii, 1, 2, 3, 6, 8, 33
nonlinear system, 1, 156
L1 adaptive control, vii, 3, 16, 17, 19, 21,
23, 24, 28, 29, 30, 32, 33, 36, 38
Lagrange multipliers, 43, 59 O
Lagrangian formulation, 43, 54
laws, 2, 19, 23, 27, 42, 62, 75, 145 obstacles, ix, 101, 102, 112, 128, 132, 133,
legs, ix, 59, 82, 83, 84, 101, 105, 106, 108, 136, 142
113, 114, 115, 117, 127, 144, 145, 146, omission, 132
147, 149, 153, 163, 165 operations, 4, 104
linear systems, 37 optimization, 98, 112, 127, 144, 165
lithium, 127
localization, 138
Lyapunov function, 158 P
parallel, vii, viii, ix, 1, 2, 3, 4, 5, 6, 7, 8, 9,
M 10, 21, 22, 24, 32, 33, 36, 37, 38, 42, 43,
44, 46, 47, 59, 62, 63, 64, 70, 75, 76, 77,
magnet, 117 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 89,
magnitude, 160, 162 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100,
manufacturing, 104, 117, 144 101, 102, 103, 104, 106, 109, 112, 115,
mapping, ix, 82, 86, 91 128, 134, 137, 140, 141, 143, 144, 145,
mass, vii, viii, 14, 15, 16, 41, 55, 66, 67, 146, 149, 152, 153, 160, 161, 163, 164,
105, 106, 111, 115, 119, 125, 130, 131, 165, 166
132, 133, 152, 153, 154, 159, 162 parallel manipulators, 1, 144
matrix, ix, 3, 13, 14, 15, 16, 17, 20, 21, 24, parallel platform, 144, 165
25, 26, 27, 28, 43, 45, 46, 52, 53, 54, 55, parallel robots, vii, viii, 41, 42, 101, 102,
57, 58, 64, 70, 71, 82, 83, 84, 86, 89, 91, 103
95, 97, 111, 112, 120, 121, 122, 123, 150, parameter estimation, 76
151, 153 patents, 38
measurement, 8, 29, 43, 76, 78, 115, 116, path planning, viii, 81, 82, 97, 139, 142
117 pitch, 49, 52, 87, 88, 154
medical, 75, 82 platform, viii, ix, 4, 5, 9, 13, 14, 15, 24, 36,
methodology, 44, 49, 72, 78 41, 42, 43, 44, 45, 47, 48, 49, 50, 54, 55,
modelling, viii, 41, 42, 43, 44, 45, 47, 49, 58, 59, 60, 64, 66, 70, 72, 75, 76, 77, 79,
51, 53, 55, 57, 59, 61, 63, 65, 67, 69, 71, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91,
73, 75, 77, 79, 141 94, 95, 97, 102, 104, 109, 111, 112, 116,
models, 105, 106 118, 119, 134, 137, 142, 144, 145, 146,
modifications, 102, 129 147, 148, 149, 150, 151, 152, 153, 154,
modules, 113 155, 157, 160, 161, 162, 163, 164, 165
170 Index

polar, 55 signals, 3, 16, 116

Portugal, 36 simulations, ix, 3, 6, 44, 67, 77, 144, 145,
position control, 41, 62, 125, 156, 158 158, 160, 162, 163, 164
positioning errors, vii, viii, 42, 103, 144 society, 135
programming, 5 solution, viii, 2, 11, 13, 18, 20, 24, 27, 38,
project, 34, 71, 139 42, 50, 77, 116, 117, 119, 154, 155, 156
proposition, 33 specifications, 42
prototype, vii, ix, 4, 33, 101, 102, 103, 106, spindle, 113, 115, 116
112, 113, 114, 140, 141, 147, 148, 151 stability, 2, 5, 6, 35, 36, 102, 105, 106, 107,
108, 111, 119, 124, 125, 126, 129, 131,
132, 133, 134, 138
R state(s), 3, 4, 7, 8, 15, 16, 17, 18, 19, 20, 21,
22, 23, 24, 25, 26, 27, 109, 114, 129, 157
radar, 82 statically stable walking, 101, 111, 132,
radius, 10, 11, 12, 108, 149 133, 137, 138
reciprocal screw theory, viii, 81, 83, 87, 96 structure, vii, viii, ix, 2, 4, 9, 14, 17, 27, 41,
redundancy, viii, 8, 38, 43, 64, 97 42, 44, 48, 81, 82, 83, 84, 86, 89, 96, 97,
reference frame, 48, 49, 55, 150 101, 102, 103, 106, 108, 110, 112, 113,
reference system, 17, 19 114, 139
rejection, 28, 79 structure synthesis, 83, 85, 87, 89, 91, 93,
requirements, 2, 4, 42, 63, 67, 102, 108, 94, 95, 97, 99
112, 145 synthesis, vii, viii, ix, 42, 81, 83, 87, 89, 96,
resilience, 105, 112 98, 99
resolution, 42, 78, 141
response, 22, 28
retirement, 135, 136 T
rings, 148
robotic manipulators, vii, viii, ix, 35, 36, 81, target, 27, 94, 116, 124, 125, 126, 127, 129,
82, 83, 87, 93, 96 133, 134, 135, 136, 137, 138, 139, 142
robotics, 42, 43, 44, 50, 75, 104, 140, 147 technical support, 136
rotational inertia, 14 techniques, viii, 42, 43
rotational matrix, 86 technology, 112
rotations, 29, 114, 147, 154, 156 testing, 144
time periods, 156
topology, 144
S tracking performance, vii, viii, 1, 3, 8, 25,
29, 31, 33, 158
safety, 112, 134 tracks, 16, 24
saturation, 158 trajectory, ix, 5, 20, 29, 60, 62, 63, 64, 65,
schema, 146, 159 68, 70, 71, 72, 78, 124, 128, 129, 131,
semicircle, 146 144, 145, 148, 149, 154, 155, 156, 159,
sensitivity, viii, 43, 70, 72, 73, 74, 75, 79 160, 161, 162, 163
sensor architecture, 101 transformation, 83, 99, 114, 152
sensors, viii, 29, 42, 43, 44, 47, 48, 50, 64, translation, 45, 57
67, 68, 70, 72, 73, 75, 78, 115, 116, 117, transmission, 83, 97, 110, 111
118, 119, 125, 141 transportation, 136, 141
serial robots, vii, viii, 41, 42, 43, 62, 103, trigonometric functions, 49
104, 111
service robots, 101, 102, 104, 141
shape, 109, 112, 115, 131, 145, 147
 Index 171

vibration, 103
V vision, 104, 105, 112, 147
validation, 60
variables, 43, 44, 45, 47, 48, 49, 50, 51, 53, W
54, 55, 56, 57, 58, 59, 60, 61, 64, 68, 69,
87, 146 walking, vii, ix, 101, 102, 103, 105, 106,
variations, 2, 4, 6, 67 111, 112, 116, 125, 127, 128, 129, 130,
vector, 9, 13, 15, 16, 17, 19, 20, 22, 43, 46, 131, 132, 133, 137, 138, 139, 140
49, 50, 51, 53, 55, 59, 71, 89, 91, 97, 119, walking scenarios, 101, 103, 128, 132, 138
120, 122, 147, 150, 153, 156, 157, 158 weakness, 2
velocity, ix, 8, 13, 21, 24, 27, 45, 46, 51, 52, weight ratio, vii, viii, 42, 108
53, 54, 56, 57, 60, 64, 66, 71, 82, 83, 86, workforce, 105
89, 91, 94, 97, 102, 111, 115, 122, 124,
128, 132, 134