Robots for the real world

In the past decade, Philip Agre's arguments against the plausibility of unembodied planning and Rodney Brooks's delightful mobile robots have helped behaviour-based robots to colonise artificial intelligence laboratories. Ironically, while the concept of building robots to resemble the behaviour, physiology and evolution of real creatures was fundamental to nouveau AI in the 1990s, the inspiration for this approach dates back to the 1950s and the writings of Warren McCulloch. So the publication of a textbook that explains how to build robots that can "survive" in the real world with reference to such historical influences is timely for both historians of technology and educators of would-be roboticists.

Ronald Arkin's Behaviour-Based Robotics begins by explaining that the design of mobile robots should take a Gibsonian ecological approach such that mechanisms and algorithms are matched to real environments. Biological inspirations for robot design are therefore surveyed to provide working definitions of features, such as intelligence and perception, that are relevant to "survival" in the world. Arkin then describes robot behaviours in analogous terms and explains how these can be reproduced by a combination of algorithms and mechanisms.

Important architectures are surveyed, with a particular focus on Brooks's influential subsumption architecture, showing how these can implement a behaviour such as "foraging". Knowledge representation is also discussed with reference to symbolic and non-symbolic approaches. The pros and cons of pure planning, pure reaction and hybrid strategies are illustrated in a non-partisan way that typifies Arkin's objective approach.

This leads to discussions of the need for adaptability to cope with uncertain environments. Learning is discussed at two levels: the simulated evolution of a robot "species", and online learning within an individual robot in an environment.

The final chapters address more esoteric issues such as social behaviour in robots and the plausibility of building minds that are thoughtful, conscious, emotional and imaginative. Although imaginative, these chapters are as rigorously technical as the rest of the book.

Arkin's style is friendly and concise with suitable academic depth and tone. The chapters begin with epigrams that introduce the main issues in humorous, non-technical language. The limitations of planning for a mobile robot in a rapidly changing environment and the consequent need for reactive abilities are alluded to with a quote from Mike Tyson: "Everybody's got plans ... until they get hit."

The academic material within each chapter begins with a tight statement of objectives. These are met with accurate and well-motivated historical and contextual material, clear diagrams and photographs from primary texts, and crisp algorithms presented in readable pseudo-code and formulae. One of the book's strengths is that the main systems or robots that introduced a key idea, architecture or algorithm are always described, and numerous references to the original sources are included. The chapters conclude with short summaries, in the form of bullet points, that would make ideal lesson plans or revision aids.

Behaviour-Based Robotics fills a clear need for a main textbook for a ten-week course in this paradigm. It could also serve as a secondary text for courses on adaptive behaviour, artificial life or the philosophy of artificial intelligence. New research students should find the book an excellent primer and source of reference.

This book presents a technical history that is explained in a clear, well-illustrated fashion that should help a new generation of researchers understand why philosophical issues are as important as engineering theory when building robots to navigate the real world.

Simon Shurville has lectured in artificial intelligence and now researches writing, education and the world wide web, University of Brighton.