Fans of Britney Spears, Celine Dion or even Frank Sinatra will know that after the artist has been around for a while a collection of songs usually appears on one album under the title "The Best of...". This book by Rodney Brooks is exactly that. Although dated 1999 it represents a compilation of eight scientific papers by Brooks over the period 1985 to 1991, along with a dainty prelude in each case to explain how the paper came into being and a personal view of the academic scene at the time each paper appeared.
The field of behaviour-based robots can be undeniably traced back to the pioneering work of W. Gray Walter in Bristol in the early 1950s. There followed a dark (or at least shady) age when robots became commonplace in manufacturing environments, but only tended to move around by following guide wires. Brooks was one who was largely responsible for making behaviour-based robots fashionable again - not by producing extensive, elaborate computer simulations, but by actually building robots, robots that moved around, perceived the world and acted, in their way, on their perceptions.
Brooks's contribution to the field of behaviour-based robotics is considerable, something that is clearly apparent from this book. In the preface, he succinctly explains the approach he has taken, chiefly by reference back to a presentation he gave as a junior faculty member at Stanford University in 1983-84. Two transparencies from that time are reproduced. The first he describes as depicting the traditional Artificial Intelligence (AI) model indicating the cognition-action-world-perception-cognition cycle of life. The second, "new model", shows the perceptual and action subsystems as "all there really is", ie the cycle includes no cognition, which Brooks sees as a separate, observed entity. Despite the fact that I prefer the traditional model, the preface is a nice read and makes many of the arguments transparent.
The main body of the book is split into two nominal sections, Technology and Philosophy, each containing four reprinted papers. This seems strange as some of the best technological descriptions are given in the philosophical section, while the more appealing philosophical arguments spring, when least expected, from under a technological cloak.
Of the papers included, several appear to consist of the idle ramblings of the self-styled bad boy of American robotics, when given an invitation to a conference, but with nothing new to say. There is a lot of overlap and repetition between the papers, indeed some, almost identical descriptions appear four or five times. The whole book is also very Americacentric, giving any purely behavioural human readers a view that mobile robots worth their salt do not do their stuff outside the US. Brooks also appears, surprisingly, to adhere to the quaint but outdated view that "The holy grail of AI [is] namely human level equivalence", ie it is all about copying human intelligence - this really surprised me.
However, some of the papers fall into the must-read category as far as the topic of mobile robots is concerned. Indeed, the very first paper, "A robust layered control system for a mobile robot", is one of the most widely referenced papers in the field. The paper's beauty lies in the simplicity of the approach described, namely a subsumption architecture, something that has been much copied and is very much in evidence today. Many of our robots at Reading University contain a control system based on such a task-achieving behavioural approach.
When it appeared, this paper broke new ground and upset the AI establishment. I can remember Mike Brady on his transfer from MIT to Oxford University, where I was at the time, throwing the paper at me in disgust, warning me not to include it in a bibliography on mobile robots that we were putting together - the paper was most definitely included.
For me the sexiest paper appears as seventh in the book, entitled "Elephants don't play chess". Although aimed at answering some of the criticisms of behaviour-based robotics it actually contains a review of the robots built by Brooks's team at MIT. Included are Allen, their first robot; Tom and Jerry; Herbert; Squirt; Seymour; Toto; and, most famous of all, Genghis, a six-legged robot that walks under subsumption control. For anyone wishing to build mobile robots, this chapter should be seen as a necessary primer.
In paper three, entitled "Learning a distributed map representation based on navigation behaviours", the work of Maja Mataric, a student of Brooks for her masters' thesis, is described. Here, Toto, a three-wheeled robot, is put through its paces to learn a map, using landmark detection, to find suitable paths through a cluttered room and to optimise the paths found. Essentially, in the paper it was shown that the behaviour-based approach could be employed to make plans and deal with changing goals, by means of a distributed, rather than the traditional central, control structure. Brooks saw this as the nail in the coffin of the traditionalists in AI. While this might have been the case, unfortunately they still appear from time to time, particularly when there is a full moon.
If you are the type of person who has a general interest in gaining an insight into mobile robots, robot intelligence and the so-called behavioural control approach, this is not really the book for you. It is quite technical and although not overtly mathematical, it requires some knowledge of the field.
But for researchers interested in a compilation of Brooks's main works, without the need to pull together disparate journal articles, this book is a must. The scene-setting, chatty, personalised introductions to each paper give a unique insight into Brooks's feelings and thoughts as the paper was put together. They contain his fears about each paper's acceptance, counteracted by his "firing from the hip" attitude to robot intelligence and they paint a colourful picture of the mobile robotics and AI scene in the US in the late 1980s.
Kevin Warwick is professor of cybernetics, University of Reading.
Cambrian Intelligence: The Early History of the New Al
Author - Rodney A. Brooks
ISBN - 0 262 02468 3 and 52263 2
Publisher - MIT Press
Price - £43.75 and £17.95
Pages - 199