Slaves with sensors

Mike Brady tours the swinging, crawling, hopping world of robots.

The word "robot" was first coined by Czech playwright Karel Èapek in Rossum's Universal Robots in the early 1920s. The play is not a bundle of laughs. A factory introduces a new product, a set of artificial slaves - robots - who are intended to relieve human drudgery. But it all goes horribly wrong and the robots wipe out the human race. Since that time, the word "robot" has taken on two quite different meanings. First, it is the stuff of science fiction, from Asimov to R2D2 and beyond. Second, it is the vastly different reality of factory production lines.

Despite 25 years of intensive research aimed at the development of a robotics science, the statistics of actual robot use in industry have remained essentially constant: more than 90 per cent of industrial robots are used for spot welding and spray painting, mostly in the automobile industry with parts transfer and electronics assembly accounting for almost all of the rest. A far cry, indeed, from R2D2. But despite the general lack of interest in industry, robotics in university laboratories addresses a number of problems that severely limit current robots, experimental or otherwise. Such problems include development of novel sensors and their associated signal/ image processing algorithms, real-time sensor-guided control systems, algorithms for trajectory and task planning, and novel kinematic and dynamic structures such as multi-fingered "hands" and lightweight, flexible robot "arms".

An autonomous robot vehicle, for example, typically maintains in its memory a software representation of its environment, plans trajectories to attain goals such as "take this load from station three to station eight", and uses its sensing and control architecture to detect unexpected obstacles such as people and navigate smoothly around them. Hundreds of such vehicles have been installed around the world, providing an approach to parts transfer that is more flexible than the procrustean world of conveyor belts, that is able to deal with heavy loads and that, most important of all, is easily reprogrammable to cope with changes to the environment. Curiously, such robot vehicles are at least as technologically advanced as many of the systems described in this book, and, unlike many of the systems described, they work - day in and day out.

How might we define a robot? Various august bodies have tried, though their "definitions" have always soon been surpassed by the experimental systems developed in research laboratories. For this reason, my favourite "alternative definition", proposed by David Grossman, one-time leader of robotics at IBM's Thomas J. Watson Laboratories, is: "A robot is a surprisingly animate machine." This has the virtue of always being current, though what surprises today will not tomorrow.

This book is full of experimental robots that are, for the most part, surprisingly animate. It is an excellent sampler of the state of research, mainly in the United States and Japan, aimed at building systems that significantly extend the capabilities of robots. The authors have worked together several times previously. Peter Menzel is a photographer, and Faith D'Aluisio, a science journalist. The first, and most lasting, impact of the book is the photography, which is, quite simply, stunning. From the inside cover, which shows a multi-legged robot shaped like an ant crawling across Great Sand Dunes National Monument, through Toshio Fukuda's robot that swings like an ape from horizontal bar to bar in a laboratory, to the Honda P3 biped walking humanoid robot, almost every page contains a remarkable and exciting visual image.

The book is the result of a tour by the authors to many of the world's leading robotics laboratories, including Carnegie Mellon University (CMU), MIT, Stanford, the universities of Tokyo, Waseda and Nagoya, the Tokyo Institute of Technology and the German Air and Space Research Laboratory.

There is relatively poor coverage of Europe and other parts of the Pacific Rim. Nevertheless, the selection of projects is a very good introduction to the world of advanced experimental robotics. Most of the US projects are supported by government, primarily by the Department of Defense. Many of the project reports discuss possible military applications. Some, such as flying drone robots and autonomous tracked vehicles, are familiar. The Japanese projects are also supported principally by the government, though two remarkable demonstrators, the Honda P3 biped humanoid robot and the Sony dog robot, were developed in corporate laboratories.

The book comprises a mostly excellent introduction by Menzel, then six sections entitled electric dreams; robo sapiens; biological; remote possibilities; work mates; and serious fun. The journalistic flavour of the book is evident from these titles. The electric dreams section, for example, has a brief overview of Marc Raibert's running, hopping robots, the Cog humanoid project led by Rod Brooks at MIT and Kris Pister's dust robots at Berkeley. Similarly, the remote possibilities section includes Nasa's Sojourner robot (similar to the one that landed on Mars in 1997), various instrumented gloves and headsets to provide telepresence from a robot, and Red Whittaker's (CMU) planetary explorer mobile robots. Taken as a whole, the variety is fascinating and thought provoking.

Each of the six sections comprises a sequence of brief two to three-page overviews of a robot project followed by an interview by D'Aluisio with one or more of the project engineers. It soon becomes clear that the interviewer is as fascinated with the people who developed the robots as with the robots themselves. The interviewees' motivations are many. Some are talented engineers attacking difficult practical problems in innovative ways. Some are equally talented engineers building systems to walk, skip, fly, crawl or manipulate for the sheer fun of doing so (and strength to their collective elbows for getting funding to indulge their fantasies). Some are driven by the belief that building systems that emulate human abilities provides unrivalled testbeds for exploring issues in cognition, perception or motor control.

With few exceptions, the technical coverage is shallow, though this might be expected in a popular science book. The interested reader does not get any sense of how the system being described operates and in many cases does not get a sense of what its scope and limitations are. The interviews are largely uncritical of the technology. The latter is an important point.

Most robotics researchers would point to Marc Raibert's work on the dynamics of legged robots as an inspirational landmark in the development of the subject. Raibert recounts, with typical self-effacing honesty, the way in which one of the systems that he had developed, and which worked unerringly in the laboratory, failed miserably when it was taken to a film set in the desert. He stresses the need to build systems that work reliably in an unpredictably wide range of situations. Most of the systems described in the book are far less reliable than Raibert's.

Furthermore, the great Japanese researcher Shigeo Hirose, whose work has led to a series of "snake" robots that have not only found practical application but have shed new light on snake motion, denounces the popular trend towards building "humanoids". These are robots that resemble, in some crude sense, humans (typically biped locomotion, two arms and a head-like structure sporting a couple of television cameras). For Hirose, the quintessential engineer, one should concentrate on a specific task and build systems that solve that task, irrespective of whether or not it does so in a way that resembles a human. The interviews are also surprisingly uncritical of the ethical stance of some of the research, such as that involving attaching wires to insects or reptiles to exploit their sensory pathways in order to control some device.

To summarise so far: the photographs are wonderful, the selection of projects is fine and, although I have considerable reservations about many of the interviews, they are, on the whole, also fine. Would that I could end the review at this point.

Consider, however, the subtitle of the book: "Evolution of a New Species". Menzel’s introduction begins with a typically dry definition of "robot" from the Merriam-Webster Collegiate Dictionary (1999); but then offers an "admittedly hypothetical" definition of "Robo sapiens" from the future (2099) Microsoft Universal Dictionary : "1. A hybrid species of human and robot with intelligence vastly superior to that of purely biological mankind; began to emerge in the 21st century. 2. The dominant species in the solar system of Earth."

Of course, this is partly tongue-in-cheek. However, the subtitle, the selection of some of the people and projects covered in the book, as well as the flavours of many of the interviews strongly suggest that the authors - and several of the interviewees - really do believe that "Robo sapiens" will evolve from, and will soon surpass, the exquisite capabilities of Homo sapiens. The fun interviews, the beautiful pictures, and even the starry-eyed commitment of some of the publicity-conscious interviewees, do not constitute a serious scientific case for this wild conjecture. The cutting-edge robotic systems do indeed constitute "surprisingly animate machines"; but that is a far, far cry from a new species and the hints about evolution are completely specious.

Mike Brady is professor of information engineering, University of Oxford.