Behavior-Based Robotics

by Ronald C. Arkin, Michael Arbib (Foreword)

This introduction to the principles, design, and practice of intelligent behavior-based autonomous robotic systems is the first true survey of this robotics field. The author presents the tools and techniques central to the development of this class of systems in a clear and thorough manner. Following a discussion of the relevant biological and psychological models of behavior, he covers the use of knowledge and learning in autonomous robots, behavior-based and hybrid robot architectures, modular perception, robot colonies, and future trends in robot intelligence.

The text throughout refers to actual implemented robots and includes many pictures and descriptions of hardware, making it clear that these are not abstract simulations, but real machines capable of perception, cognition, and action.

Genres: Technology, Robots

491 pages, Hardcover

First published May 22, 1998

Reviews

[Aadesh · Rating 4 out of 5]

It was good. But I think all the robots and the techniques are outdated

[Mangoo · Rating 3 out of 5]

Though dated, Arkin's book provides a good baseline to understand what takes to make reactive robots acting only in response to stimuli through (the combination of) behavioral rules. This approach embodies the behaviorist point of view which was a reaction to the failure of hard artificial intelligence in that it rejected the necessity of intelligent agents having to build and act upon an internal representation of outer environment - that is, planning based on pregressed knowledge was considered unnecessary when not even deteriorating the performance of the robots. This extreme position was particularly supported by R. Brooks' proposal of subsumption architectures as compared to hierarchical (nested) control architectures. Subsumption, together with motor schemas, constitute the main behavior-based approaches discussed in the book (as Arkin is a main player in the development of the latter).
The book is rather pedantic, and with the ambition of full-fledged definition and classification of any aspect of the field, a tendency which makes the book overtly long as compared to what could be, ultimately, the strict minimum: at times it even makes the subject matter (which is conceptually straightforward) looking hermetic - though this does not mean it is not articulated, on the contrary. The discussions are on the contrary always very good and clear, as well as the short references to collateral study fields as biology (for neural correlates and social agent behaviors), psychology (for the ontology of information, action-oriented perception, support for control approaches, inspiration), control theory, artificial intelligence (learning and adaptation). The book also features many pictures of actual robots and diagrams implementing the ideas discussed.
What comes out of the read is the feeling that behavior-based robotics sounds reasonable as much as it hides a lot of fundamental issues (e.g. how to partition the sensory inputs, what is information and what to make of it, how to best combine behavioral responses, the relation between information, perception and action), well worth being digged. For that reason, the book needs an update, too.

[Arnoud Visser · Rating 5 out of 5]

It is impressive how a book that is 17 years old still rings truth. The pictures of old robots are mainly historical, but most of the research questions discussed in the book are still relevant. This can be explained by the fact that the main field of robotics stopped with the debate about architectures and just build the components that worked, but reading about this grand designs gives a lot of inspiration.