Browse Books

Reviewer: George A. Bekey

A wide sweep of the design and implementation issues (as well as some theory) surrounding wheeled mobile robots is presented in this book. The major strength of the book lies in its discussion of 13 case studies, involving the design and implementation of robots by the author and his students, mostly in his own laboratory at the University of Manchester. One of the book's weaknesses is that it largely ignores legged, crawling, or climbing mobile robots, except for an example of robot learning, using the MIT hexapod named GENGHIS. I was not able to find any mention of the remarkable work on humanoid robots taking place primarily in Japan, but also in Germany and the United States. The book also completely ignores dynamics. Other weaknesses are discussed below. Nine chapters make up the book. Chapter 1 introduces the book, and presents the author's view of the importance of building one's own mobile robot. Chapter 2, "Foundations," discusses the applications of robots, and includes a brief history of mobile robots. Chapter 3, "Robot Hardware," presents a nice discussion of many common sensors for robots, but spends only one page on actuators. The discussion is somewhat superficial, since sensor and actuator dynamics are not mentioned, odometry errors are not discussed, and novel sensors (such as sniffers) are not mentioned. Chapters 4 through 8 are the meat of the book. Chapter 4, "Robot Learning, " discusses reinforcement learning and connectionist learning in some detail, with several case studies of actual learning robots. Chapter 5 concerns navigation, and presents several navigation strategies for robots, and compares them to biological models. This is also the longest chapter in the book; it presents four additional case studies. Chapter 6, "Novelty Detection," is one of the strongest and most unique parts of the book. Yet, while the chapter is concerned with identifying novel elements in the robot's environment, it does not mention similar work in artificial intelligence (AI). Chapter 7 concerns simulation as a tool for rapid design and experimentation. The author properly discusses the differences between model performance and real world performance, but the issues in the construction of models for simulation are not mentioned. Chapter 8, "Analysis of Robot Behavior," is another one of the book's contributions. It presents a principled way of performing analyses of robot behavior. These analytical methods are illustrated with three case studies. Chapter 9, the final chapter of the book, presents the author's view of some of the successes and challenges of work in mobile robotics. This chapter is somewhat superficial, and does not discuss the major research challenges in such areas as distributed control, human-robot interaction, reconfigurable mobile robots, and so on. Clearly, the book lives up to its title, being a "practical introduction." It also, however, presents some fairly difficult concepts in connectionist learning, partially observable Markov processes, and others. My major complaint is that the dynamics of moving robots are completely ignored. Granted, when they move slowly, it may be reasonable to omit inertial forces, but then sticking friction may be an important phenomenon. Dynamics will become increasingly important as robots move faster. The word "control" is used repeatedly throughout the book, but nowhere is there any mention of low level control, proportional, integral, derivative (PID) controllers, and so on, even though these features are essential in a practical design. Finally, the discussion of simulation appears to put the emphasis in the wrong place. The chapter discusses the wonderful things that are possible with simulation, but not in the real world. For example, at the beginning of chapter 7, the author mentions one of his laboratory robots that had the "annoying property" of veering left when the battery charge was high, and veering right when it was low. The problem was solved in simulation, where the robot moved straight as intended. There is no mention of the fact that the problem was not with the "annoying" robot, but rather based in the fact that the model representing its behavior was incomplete, and did not reflect the effect of battery charge on steering. Who knows what other aspects of the robot behavior were omitted in the simulation__?__ The control design was inadequate, since it did not reflect this dynamic effect, and the simulation was not a valid representation of the actual robot. In summary, this is an interesting and useful book for people interested in a broad, introductory view of many issues in the design of wheeled robots. While it suffers from some serious deficiencies, such as the lack of coverage of dynamics and legged locomotion, it presents numerous detailed case studies that can be useful to readers interested in building their own robots, or in implementing the author's strategies on commercial mobile robots. Online Computing Reviews Service