engineering from the University of Systems with nonholonomic con-

California, Berkeley, in 1989 and Nonholonomic Mechanics and straints are not only well motivated
1992, respectively. After receiving Control by A.M. Bloch with the col- by applications such as mobile
his Ph.D., he was a postdoctoral fel- laboration of J. Baillieul, P. Crouch, robots but are intellectually interest-
low at the Ecole des Mines de Paris and J. Marsden. Springer, 2003. ISBN ing as well. From an analytical and
in Fontainebleau, France. In Septem- 0-387-95535-6, US$69.95. Reviewed by geometric point of view, nonholo-
ber 1992, he joined the faculty of the A.J. van der Schaft. nomic constraints present a class of
Electrical Engineering Department at systems that cannot be described in
the University of Minnesota, where Holonomic and the standard Lagrangian or Hamiltonian
he was an assistant professor until Nonholonomic Constraints framework; the powerful machinery
September 1997. In 1997, he joined As aptly formulated in its preface, developed for the standard case must
the faculty of the Electrical and Com- Nonholonomic Mechanics and Control be modified and extended. At the
puter Engineering Department at the links control theory with a geometric same time, nonholonomic constraints
University of California, Santa Bar- view of classical mechanics in both its may actually help in controlling a sys-
bara, where he is currently a profes- Lagrangian and Hamiltonian formula- tem. In particular, for underactuated
sor and director of the Center for tions and, in particular, with the theo- mechanical systems, the presence of
Control Engineering and Computa- ry of mechanical systems subject to nonholonomic constraints in direc-
tion. He has received the National kinematic motion constraints. The tions that are not directly actuated is
Science Foundation Research Initia- analysis and control of mechanical helpful for controllability. Interesting-
tion and CAREER Awards, the 1998 systems has been an active research ly enough, the geometric/analytical
IEEE Leon K. Kirchmayer Prize Paper area over the last several decades. and nonlinear control points of view
Award, and the 1998 George S. Axel- The book aims to present some of this appear to be intimately linked, giving
by Outstanding Paper Award. He was material, often scattered throughout rise to a rich theory that can be illus-
the recipient of the first SIAM Con- the literature, in a cohesive manner. trated on simple mechanical exam-
trol and Systems Theory Prize in Kinematic constraints are classically ples. It is reasonably “easy” to write a
1998. He was also the recipient of the divided into two classes: constraints difficult and obscure book about this
1999 Donald P. Eckman Award and that can be integrated to yield con- theory, but it is much more difficult
the 2001 O. Hugo Schuck Best Paper straints on the position variables, to write a book that provides clear
Award, both given by the American called holonomic constraints, and con- access to the topic, which intrinsical-
Automatic Control Council. He is a straints for which this integration is ly has many facets. The author has
Fellow of the IEEE. not possible, called nonholonomic con- invested considerable effort in
straints. A typical example of a nonho- achieving this goal.
Luca Zaccarian graduated with a lonomic constraint is a wheel rolling
degree in electronic engineering from vertically without slipping on a surface.
the University of Roma Tor Vergata in The constraint on the allowable veloci-
1995, where he also received his ty (the point of contact of the wheel
Ph.D. degree in computer science and with the surface cannot slip in all
control engineering in 2000. In directions) cannot be integrated to
1998–2000, he spent two years as a yield a constraint on the position of
visiting researcher at the Center for the wheel. This nonintegrability is
Control Engineering and Computation intuitively clear, as illustrated by the
of the University of California, Santa fact that an automobile can go any-
Barbara. During the summers of 2003 where it pleases by suitable maneuver-
and 2004, he was a visiting professor ing. (This example hints at the
at the Electrical and Electronic Engi- intimate relationship between nonho-
neering Department of the University lonomic constraints and controllabili-
of Melbourne (Australia). Since 2000, ty.) Loosely speaking, mechanical
he has been an assistant professor systems with holonomic constraints
(“ricercatore’’) in control engineering can be reduced to lower dimensional
at the University of Roma Tor Vergata. mechanical systems without con-
He was the recipient of the 2001 straints; for systems with nonholonom-
O. Hugo Schuck Best Paper Award ic constraints, this reduction is not
given by the American Automatic possible, and as a result some distin-
Control Council. guishing features arise.

0272-1708/05/$20.00©2005IEEE
June 2005 IEEE Control Systems Magazine 99
Contents ture, some of which is quite recent. tual displacements) satisfying the
The text begins with four introductory Chapter 6 deals with the control of non- kinematic constraints. This subtlety
chapters. After introducing Hamilton’s holonomic systems, and the control of is not easy to grasp and may seem to
principle and the Lagrange- d’Alembert mechanical systems subject to nonho- be a purely academic discussion (in
principle, Chapter 1 presents a collec- lonomic constraints. The terminology the bad sense of the word). However,
tion of mechanical systems subject to nonholonomic (control) systems, which it turns out that the equations of
nonholonomic kinematic constraints. may be somewhat confusing, originates motion obtained from Hamilton’s
The treatment of these examples illus- from the kinematic model of a mechani- principle are generally different from
trates some of the ideas and phenome- cal system subject to nonholonomic the “correct” equations of motion
na discussed in detail in the later kinematic constraints, for which the fea- obtained by applying the Lagrange-
chapters. As such, this chapter gives sible space of velocities on the configu- d’Alembert principle; in fact, these
the basic flavor of the material and ration space of the mechanical system equations are the same if and only if
provides an ideal introduction to the is represented as the span of input vec- the constraints are holonomic!
book. Chapters 2, 3, and 4, respectively tor fields. Pure nonholonomicity of the On the other hand, the framework
titled “Mathematical Preliminaries,” kinematic constraints (in the sense that of Hamilton’s principle is linked to
“Basic Concepts in Geometric Mechan- no part of them can be integrated to nonlinear optimal control theory,
ics,” and “Introduction to Aspects of yield constraints on the configuration while the special case of optimal con-
Geometric Control Theory,” give a variables) then corresponds to trol of nonholonomic control systems
brief overview of the mathematical controllability of the ensuing control sys- leads to the interesting mathematical
background needed for the rest of the tem. Controllability can be character- topic of sub-Riemannian geometry.
book and introduce key definitions ized by a full-rank condition on the Chapter 8 deals with the key problem
and notation. These chapters can be space of vector fields generated by tak- of stability of mechanical systems with
skipped by readers who are acquainted ing (possibly repeated) Lie brackets of nonholonomic constraints and shows
with this material and can be returned the input vector fields. Thus, nonholo- how the delicate stability theory of ordi-
to as needed. nomicity is intimately linked to control- nary Hamiltonian systems can be
Although, unavoidably, the open- lability. On the other hand, the linear extended to this case. Finally, Chapter
ing chapters provide only a “crash control paradigm of equivalence 9 provides a control counterpart by
course” at some points, the material between controllability and pole place- studying the stabilization or setpoint
has been written with much care. In ment breaks down in this case in the regulation of underactuated mechani-
fact, in many cases, the clarity of the sense that there does not exist an cal systems. The main method is the
presentation is unmatched elsewhere asymptotically stabilizing continuous theory of controlled Lagrangians. (The
in the literature. An example is the state feedback law. The second part of inclusion of gyroscopic forces in the
treatment of Ehresmann connections Chapter 6 deals with the control of Lagrangian description is equivalent to
in Section 2.9, where the geometric dynamical models of mechanical sys- interconnection-damping assignment
definition and interpretation is linked tems subject to kinematic constraints control of port-Hamiltonian systems.)
to the algebraic definition in a very and with external input forces. This method is based on the use of
insightful way. Also, the relation with Chapter 7 describes another inter- state feedback to shape another virtual
linear (Koszul) connections on tan- esting connection between nonlinear mechanical system with energy that
gent bundles is clearly explained. (I control theory and nonholonomic has a critical point at the desired set-
would have also linked the definition constraints, namely, the variational point. Another method treated is the
of a linear connection on a tangent formulation of mechanical systems theory of averaging for underactuated
bundle to the geometric interpreta- subject to nonholonomic constraints, mechanical systems.
tion of an Ehresmann connection.) As as in Hamilton’s principle as opposed
a result, these introductory chapters to the Lagrange-d’Alembert principle. Summary
provide stimulating reading for a The difference is that, in the first This book is a welcome addition to the
reader well versed in this material as case, the time-integral of the existing literature and will certainly
well as for the novice. Lagrangian is extremized over all become a standard reference. Related
A core chapter of the text is Chapter configuration trajectories whose tan- books with a different emphasis and
5, which deals with the geometric for- gent vectors satisfy the kinematic partly complementary choice of topics
mulation of mechanical systems sub- constraints, while in the latter case include [1] and the recent [2]. It is to be
ject to nonholonomic kinematic the conditions for extremality of an expected that Bloch’s book will be a
constraints, both in the Lagrangian and arbitrary curve in the configuration continuing source of inspiration for fur-
Hamiltonian frameworks. This chapter space is checked with respect to all ther research in this area. Indeed, many
provides a nice synthesis of the litera- tangent vectors (classically called vir- aspects remain to be investigated,

100 IEEE Control Systems Magazine June 2005

especially from the systems and con- Letters and editor-at-large for the Euro- h, one may try to find the roots of the
trol point of view. While important pean Journal of Control. He is coauthor gradient of h. However, the gradient of
basic understanding has been reached of System Theoretic Descriptions of h is not available when only noisy
for trajectory planning and classical Physical Systems (1984), Variational measurements of h can be used. In
control problems like setpoint stabiliza- and Hamiltonian Control Systems (1987, 1952, Kiefer and Wolfowitz [3] pro-
tion, the general problem of controlling with P.E. Crouch), Nonlinear Dynami- posed another recursive algorithm for
(by state feedback laws or by intercon- cal Control Systems (1990, with H. solving this problem, in which the gra-
nection with other dynamical systems) Nijmeijer), L2 -Gain and Passivity Tech- dient of h is replaced by its noisy gra-
the dynamical behavior of the physical niques in Nonlinear Control (2000), and dient estimate of finite difference type.
system is still largely open. Important An Introduction to Hybrid Dynamical Since stochastic approximation
issues here are understanding the Systems (2000, with J.M. Schumacher). methods were first introduced, there
robust dynamical behavior of systems He is a Fellow of the IEEE. has been significant progress in the
with nonholonomic constraints and the development of more sophisticated
exploration of new control paradigms. algorithms. Much of this development
In summary, this is a delightful Introduction to Stochastic Search has originated from, and has been inter-
book that will be valuable for both and Optimization by James C. Spall, twined with, applications in optimiza-
the control community and researchers Wiley, Hoboken, NJ, 2003, ISBN 0-471- tion, control theory, economic systems,
working on the geometric theory of 33052-3, US$105.00. Reviewed by G. Yin. signal processing, communication theo-
mechanical systems. With its exten- ry, learning, pattern classification, neur-
sive illustrations and exercises, this Search and al networks, and related fields.
book is eminently suited for a gradu- Optimization Methods Emerging applications have been found
ate course. The author should be con- James Spall’s book provides a survey in wireless communications, repeated
gratulated for such an admirable job. of random search (called stochastic stochastic games, and financial engi-
search by the author) and optimization neering. For instance, consider a pro-
References methods, including stochastic approxi- duction planning problem with
[1] J. Cortés Monforte, Geometric, Control and mation algorithms, evolutionary com- unreliable machines. Under certain con-
Numerical Aspects of Nonholonomic Systems
putation, Monte Carlo simulation, and ditions, it can be shown that the opti-
(Lecture Notes in Mathematics 1793). Berlin:
Springer-Verlag, 2002. related statistical methods. Stochastic mal control is of the threshold type. For
approximation algorithms include multiple machine problems, it is virtual-
[2] F. Bullo and A.D. Lewis, Geometric Control
of Mechanical Systems. Modeling, Analysis, recursive least squares, adaptive algo- ly impossible to find closed-form solu-
and Design for Simple Mechanical Control Sys- rithms, and reinforcement learning. tions. However, using stochastic
tems (Texts in Applied Mathematics, 49). New One of the original aims of stochas- approximation methods, the problem
York: Springer-Verlag, 2005.
tic approximation is to find roots of a can be recast as a parametric optimiza-
continuous function f(·), where either tion problem that can be solved by
A.J. (Arjan) van der Schaft received the precise form of f is not known or f
his undergraduate and Ph.D. degrees is too complicated to compute and
in mathematics from the University of only noisy measurements are avail-
Groningen, The Netherlands, in 1979 able. That is, at iteration n (often
and 1983, respectively. In 1982, he referred to as time n) and design point
joined the Department of Applied xn (often referred to as state xn ), it is
Mathematics, University of Twente, possible to obtain only a noisy mea-
Enschede, The Netherlands, where he surement yn = f(xn ) + ξn , where ξn
is presently a full professor in mathe- denotes random noise. In 1951, Rob-
matical systems and control theory. bins and Monro [2] introduced the
His research interests include the recursive algorithm xn+1 = xn + an yn ,
mathematical modeling of physical where an is an appropriately chosen
and engineering systems and the con- step size. Robbins and Monro coined
trol of nonlinear and hybrid systems. the name stochastic approximation for
He has served as associate editor for this procedure. Another one of the
Systems & Control Letters, Journal of original aims of stochastic approxima-
Nonlinear Science, SIAM Journal on tion is to determine the minimizer of a
Control, and IEEE Transactions on Auto- real-valued cost function h using only
matic Control. Currently, he is associ- noisy measurements of h(xn ). To find
ate editor for Systems and Control local minimizers of a smooth function

0272-1708/05/$20.00©2005IEEE
June 2005 IEEE Control Systems Magazine 101