P

Introduction
Parallel Robots
A parallel robot refers to a kinematic chain in
Frank C. Park
which a fixed platform and moving platform
Robotics Laboratory, Seoul National University,
are connected to each other by several serial
Seoul, Korea
chains, or legs. The legs, which typically have
the same kinematic structure, are connected to
the fixed and moving platforms at points that are
Abstract distributed in a geometrically symmetric fashion.
The Stewart-Gough platform (Fig. 1) is a well-
Parallel robots are closed chains consisting of a known example of a parallel robot: each of the
fixed and moving platform that are connected by six legs is a UPS structure (i.e., consisting of
a set of serial chain legs. Parallel robots typically rigid links serially connected by a universal, pris-
possess both actuated and passive joints and may matic, and spherical joint), with the prismatic
even be redundantly actuated. Although more joint actuated. Other examples of parallel robots
structurally complex and possessing a smaller include the 6  RUS platform of Fig. 2, the
workspace, parallel robots are usually designed haptic interface device of Fig. 3, and the eclipse
to exploit one or more of the natural advantages mechanism of Fig. 4.
they possess over their serial counterparts, e.g., Parallel robots can be regarded as a special
higher stiffness, increased positioning accuracy, class of closed chain mechanisms (i.e., chains
and higher speeds and accelerations. In this chap- that contain one or more closed loops) and are
ter we provide an overview of the kinematic purposely designed to exploit the specific ad-
and dynamic modeling of parallel robots, a de- vantages afforded by the closed chain structure,
scription of their singularity behavior, and basic e.g., for improved stiffness, greater positioning
methods developed for their control. accuracy, or higher speed. Parallel robots should
be distinguished from two or more cooperating
serial robots that may form closed loops during
Keywords execution of a task (e.g., a robotic hand grasping
an object). Some of the fastest velocities and
Closed kinematic chain; Closed loop mechanism; accelerations recorded by industrial robots have
Parallel manipulator been achieved by parallel robots, primarily by

J. Baillieul, T. Samad (eds.), Encyclopedia of Systems and Control, DOI 10.1007/978-1-4471-5058-9,

© Springer-Verlag London 2015
1032 Parallel Robots

of freedom may exceed the kinematic degrees of

freedom, in which case we say that the robot is
redundantly actuated.
A parallel robot with a designated end-effector
frame also has a notion of forward and inverse
kinematics. While for serial chains the forward
kinematics is a well-defined mapping and the
inverse kinematics can typically have multiple
solutions, for parallel robots the situation is less
straightforward. For the Stewart-Gough platform
of Fig. 1, in which the leg lengths can be adjusted
by actuating the prismatic joints, the inverse kine-
matics is unique and straightforward to obtain,
whereas the forward kinematics will have multi-
ple solutions. For other types of parallel robots
Parallel Robots, Fig. 1 Stewart-Gough platform
in which the legs themselves contain one or
more closed loops, both the forward and inverse
placing the actuators on the fixed platform and kinematics can have multiple solutions.
thereby minimizing the mass of the moving parts. The notion of kinematic singularities for par-
Many of the model-based techniques devel- allel robots is also much more involved than
oped for the control of traditional serial chain the case for serial robots. Whereas kinematic
robots are also applicable to a large class of singularities for serial chain robots are charac-
parallel robots. On the other hand, kinematic and terized by configurations at which the forward
dynamic models for parallel robots are inher- kinematics Jacobian (i.e., the linear mapping re-
ently more complex. Parallel robots also pos- lating joint velocities to end-effector frame ve-
sess features not found in serial robots, e.g., locities) becomes singular, for parallel robots and
passive joints, the possibility of redundant actu- closed chains in general, there exist other notions
ation, and a diverse range of singularity behav- of singularities not found in serial chains. For
ior, that need to be considered when designing example, given a parallel robot with kinematic
a control law. We therefore begin with a brief mobility m – if the parallel robot consists only
overview of the kinematic and dynamic mod- of one degree-of-freedom joints, this implies that
eling of parallel robots before discussing their exactly m joints can be actuated – there may exist
control. configurations in which these m joints cannot be
independently actuated. Conversely, even if the m
actuated joints are each fixed to some value, the
Modeling parallel robot may fail to be a structure, i.e., some
of the links may be able to move.
Kinematics In the above scenario, choosing a different set
Whereas the kinematic degrees of freedom, or of m actuated joints may remedy this situation,
mobility, of a serial chain robot can be obtained as in which case such singularities are referred to
the sum of the degrees of freedom of each of the as actuator singularities. Configurations at which
joints, the situation is somewhat more complex singularity behavior occurs regardless of which
for parallel robots and closed chains in general, joints are actuated are denoted configuration sin-
since only a subset of the joints can be indepen- gularities. The final class of singularities are end-
dently actuated. The mobility of a parallel robot effector singularities, which correspond to the
corresponds to the total degrees of freedom of usual serial chain notion of kinematic singularity,
the joints that can be independently actuated. In in which the end-effector loses one or more
some cases the number of actuated joint degrees degrees of freedom of available motion.
Parallel Robots 1033

Revolute joint

Universal joint
Ball joint

Parallel Robots, Fig. 2 6  RUS platform

Prismatic Joint Universal Joint

Universal Joint

P
Parallel Robots, Fig. 3 A 3  P U U haptic interface

Dynamics
In the case of a parallel robot whose actuated
degrees of freedom coincides with its kinematic
Parallel Robots, Fig. 4 The 3  PPRS eclipse parallel
mobility m, it is possible to choose an indepen-
mechanism
dent set of generalized coordinates of dimension
m, denoted q 2 Rm and typically identified with
the actuated joints, and to express the dynamics denotes the vector of gravitational forces. The
in the standard form structure of the dynamic equations is identical
to that for serial chain robots. Also like the case
M.q/qR C C.q; q/
P qP C G.q/ D ; (1) for serial chain robots, the Coriolis matrix term
P 2 Rmm is not unique, so that one
C.q; q/
where  2 Rm denotes the vector of input should ensure that the correct C.q; q/ P is used
joint torques, M.q/ denotes the n  n mass ma- in, e.g., any control law whose stability depends
P qP denotes
trix, the matrix-vector product C.q; q/ on the matrix MP .q/  2C.q; q/ P being skew-
the vector of Coriolis terms, and G.q/ 2 Rm symmetric.
1034 Parallel Robots

It is also important to keep in mind that the laws for serial robots are also covered in this
q must satisfy the kinematic constraint equations handbook, and we refer the reader to  Linear
imposed by the loop closure constraints. That is, Matrix Inequality Techniques in Optimal Control
if  2 Rn denotes the vector of all joints (both for the essential details. Here we summarize
actuated and passive), then q 2 Rm , m  n, the most basic control laws and point out any
will be a subset of  whose values can only additional computational or other requirements
be obtained by solution of the kinematic con- that are needed when applying these laws to
straint equations; depending on the nature of the parallel robots. Note that other control laws and
kinematic constraints, one may have to resort to techniques developed for serial chain robots, e.g.,
iterative numerical methods. robust, sliding mode, can also be applied with the
If the parallel robot is redundantly actuated, same additional considerations and requirements
then the dynamics are subject to a further set of outlined below:
constraints on the input torques. Letting qe denote 1. Computed torque control: Computed torque
the set of independent generalized coordinates control for parallel robots has the same control
and qa be the vector of all actuated joints, the law structure as for serial robots, i.e.,
vector of actuated joint torques a must then
further satisfy S T a D W T , where  denotes  
 D M.q/ Kp e  Kv eP C ff ; (3)
the vector of joint torques for an equivalent tree
structure system that moves identically to the
redundantly actuated parallel robot and W and S where e denotes the tracking error, Kp and Kv
are defined, respectively, by are the proportional and derivative feedback
gain matrices, and ff denotes the feedfor-
@ @qa ward term required to cancel the nonlinear dy-
SD ; W D : (2)
@qe @qe namics. Robust versions of computed torque
control are also applicable to parallel robots
Compared to the dynamics for serial chain robots, under the same set of conditions, e.g., estab-
the dynamics for parallel robots is, in general, lishing appropriate bounds on the mass matrix
considerably more complex and computation- eigenvalues and on the norm of the Coriolis
ally involved. The recursive algorithms that are matrix.
available for computing the inverse and forward 2. Augmented PD control: The augmented PD
dynamics of serial chain robots can also be used control law for serial robots is also applicable
to develop similar recursive algorithms for par- to parallel robots, i.e.,
allel robot dynamics; however, the computations
will be considerably more involved and require
multiple iterations.  D Kp e  Kv eP C M.q/qRd C C.q; q/
P qP C G.q/;
(4)
where qd is the reference trajectory to be
Motion Control tracked and Kp and Kv are the proportional
and derivative feedback gains. Asymptotic sta-
Exactly Actuated Parallel Robots bility is also established under the same con-
For parallel robots whose actuated degrees of ditions.
freedom match the kinematic mobility (this ex- 3. Adaptive control: Because the dynamic equa-
cludes the set of all redundantly actuated parallel tions for parallel robots are also linear in the
robots), most control laws developed for serial link mass and inertial parameters, i.e.,
chain robots are also applicable. This is not al-
together surprising in light of the similarity in
the structure of the kinematic and dynamic equa- M.q/qR C C.q; q/
P qP C G.q/ D ˆ.q; q;
P q/p;
R
tions between serial and parallel robots. Control (5)
Parallel Robots 1035

where p denotes the vector of link mass and are necessary to account for the different structure
inertial parameters, adaptive control laws de- of the dynamic equations.
veloped for serial robots can also be used. Like all model-based control algorithms, the
4. Other control methods: There exist above control laws are subject to model un-
numerous control methods developed for certainties. Whereas in serial chains the effects
serial robots, e.g., task space or operational of model uncertainty simply lead to errors in
space control, sliding mode control, and tracking, for redundantly actuated parallel robots,
various nonlinear control techniques; with the consequences can lead to internal forces in
few exceptions most of these algorithms can addition to end-effector tracking errors. Perhaps
also be applied to exactly actuated parallel the most significant effect of any modeling errors
robots with minimal modification. is that, unlike the serial chain case, the kine-
matic errors can potentially alter the shape of
the configuration space (recall that the configu-
ration space will in general be a curved space
Redundantly Actuated Parallel Robots for closed chains) and also interfere with any
As described earlier, parallel robots exhibit a PD feedback introduced into the control. The
much more diverse range of singularity behavior development of control laws that are robust to
than their serial counterparts, many of which such modeling errors and disturbances remains
depend on the choice of actuated joints (actua- an open and ongoing area of research in parallel
tor singularities). One way to eliminate actuator robot control.
singularities is via redundant actuation, i.e., the
total degrees of freedom of the actuated joints ex-
ceeds the kinematic mobility of the mechanism. Force Control
Redundant actuation offers some protection in
the event of failed actuators and, when combined Both hybrid force-position control and impedance
with an appropriate control law, offers an effec- control are well-known and widely applied
tive means of reducing joint backlash, increas- concepts in serial robots and can be extended
ing speed and payload and stiffness, controlling in a straightforward manner to exactly actuated
compliance through the generation of internal parallel robots. Recall that the basic feature of
forces, and even improving power efficiency (as hybrid force-position control is that the task
P
an analogy, the human musculoskeletal system space is decomposed into force- and position-
is redundantly actuated by antagonistic muscles). controlled directions, whereas in impedance
Of course, redundant actuation introduces a new control, the goal is have the robot maintain
set of control challenges, since the control in- a certain desired spatial stiffness in the task
puts must be designed so as not to conflict with space. Controllers that combine aspects of force-
the kinematic constraints inherent in the parallel position and impedance control have also been
robot; loosely speaking, the actuated joints can proposed and developed for both serial robots
no longer be independently controlled, since the and exactly actuated parallel robots. Modeling
consequences of unintended antagonistic actua- errors will cause deviations in both the force-
tion may be catastrophic. and position-controlled directions – leading
The control of cooperating manipulators (see to motions in force-controlled directions and
 Optimal Control and Mechanics) has a long forces in position-controlled directions – which
history in robotics, and many of the control tech- can be addressed by, e.g., a switching control
niques developed for such multi-arm systems can strategy.
also be applied to redundantly actuated parallel The problem of force control for redundantly
robots. One can also apply the control strategies actuated parallel robots, which encompasses
developed for exactly actuated parallel robots to both force-position and impedance control, has
the redundantly actuated case, but modifications also received some attention in the literature.
1036 Parallel Robots

The main difference with the exactly actuated Bibliography

case is that internal forces can now be generated,
which requires a more detailed and coordinate- Anderson RJ, Spong MW (1988) Hybrid impedance con-
trol of robotic manipulators. IEEE J Robot Autom
invariant examination of stiffness. Control
4:549–556
methods that combine elements of force-position Caccavale F, Siciliano B, Villani L (2003) The Tricept
and impedance control for redundantly actuated robot: dynamics and impedance control. IEEE/ASME
parallel robots have received only limited Trans Mechatron 8:263–268
attention in the literature. Chakarov D (2004) Study of the antagonistic stiffness
of parallel manipulators with actuation redundancy.
Mech Mach Theory 39:583–601
Cheng H, Yiu Y-K, Li Z (2003) Dynamics and control
of redundantly actuated parallel manipulators. IEEE
Cross-References Trans Mechatron 8(4):483–491
Fasse ED, Gosselin CM (1998) On the spatial impedance
control of Gough-Stewart platforms. In: Proceedings
 Linear Matrix Inequality Techniques in Opti- of the IEEE international conference on robotics and
mal Control automation, Leuven, pp 1749–1754
 Optimal Control and Mechanics Ghorbel F (1995) Modeling and PD control of closed-
 Optimal Control via Factorization and Model chain mechanical systems. In: Proceedings of the
IEEE conference on decision and control, New Or-
Matching leans
 Optimal Sampled-Data Control Honegger M, Codourey A, Burdet E (1997) Adaptive
control of the Hexaglide, a 6-DOF parallel manipu-
lator. In: Proceedings of the IEEE international con-
ference on robotics and automation, Washington, DC,
pp 543–548
Recommended Reading Kim J, Park FC, Ryu SJ, Kim J, Hwang JC, Park C,
Iurascu CC (2001) Design and analysis of a redun-
The monograph (Merlet 2006) offers a detailed dantly actuated parallel mechanism for rapid machin-
and comprehensive treatment of all aspects of ing. IEEE Trans Robot Autom 17(4):423–434
Merlet JP (1988) Force feedback control of parallel
parallel robots, with a particularly thorough treat- manipulators. In: Proceedings of the IEEE interna-
ment of the kinematics and singularity analysis. tional conference on robotics automation, Philadel-
Mueller (2008) provides an excellent survey of phia, pp 1484–1489
Merlet JP (2006) Parallel robots. Springer, Heidelberg
the dynamics and control of redundantly actuated
Mueller A (2005) Internal prestress control of redun-
parallel robots and is based on the preceding dantly actuated parallel manipulators and its applica-
work (Mueller 2005). Cheng et al. (2003) exam- tion to backlash avoiding control. IEEE Trans Robot
ines in detail the dynamic model for redundantly 21(4):668–677
Mueller A (2008) Redundant actuation of parallel manip-
actuated parallel robots and the basic control
ulators. In: Wu H (ed) Parallel manipulators: towards
strategies; Nakamura and Ghodoussi (1989) also new applications. I-Tech Publishing, Vienna
examines dynamic models for redundantly actu- Murray R, Li ZX, Sastry S (1994) A mathematical in-
ated parallel robots. Stiffness analysis and con- troduction to robotic manipulation. CRC Press, Boca
Raton
trol of redundantly actuated parallel robots are
Nakamura Y, Ghodoussi M (1989) Dynamics computation
addressed in Yi and Freeman (1993), Chakarov of closed-link robot mechanisms with nonredundant
(2004), and Fasse and Gosselin (1998). Analysis and redundant actuators. IEEE Trans Robot Autom
of specific parallel robots engaged in various 5(3):294–302
Satya SM, Ferreira PM, Spong MW (1995) Hybrid control
control tasks includes Caccavale et al. (2003),
of a planar 3-DOF parallel manipulator for machin-
Honegger et al. (1997), Kim et al. (2001), and ing operations. Trans N Am Manuf Res Inst/SME
Satya et al. (1995). The basic references on robot 23:273–280
control are Murray et al. (1994), Spong et al. Spong MW, Hutchinson S, Vidyasagar M (2006) Robot
modeling and control. Wiley, New York
(2006), Anderson and Spong (1988), and Ghor- Yi B-J, Freeman RA (1993) Geometric analysis of an-
bel (1995) focuses on PD control for closed tagonistic stiffness in redundantly actuated parallel
chains. mechanisms. J Robot Syst 10:581–603