Machines that go it alone

The problem with attractive book titles is that they can be misleading - which is a shame, because Almost Human - Making Robots Think has much to commend it. However, what this book is not about is robots that are almost human or, perhaps even more disappointingly, the deeply interesting question of what processes might constitute robot thought. Perhaps the most accurate word in its title is "making", because this book is more about what roboticists do rather than what robots are.

The roboticists in question are the staff and students of the Robotics Institute of Carnegie Mellon University and this book tells not one story, but several. The two longest are about Zoe, a proof-of-concept Mars planetary rover, and Sandstorm, CMU's entry to the US Darpa (Defense Advanced Research Projects Agency) Grand Challenge and their two very different project leaders, David Wettergreen and William "Red" Whittaker. Both stories are fascinating.

Zoe was an ambitious attempt to build an autonomous planetary scientist, a robot capable of deciding which rocks look "interesting" and, as it were, looking under those rocks for signs of life. Sandstorm had a more prosaic but still technically challenging aim: to win a race against other unmanned vehicles. What these two robots have in common is the quest for "autonomy" - the Holy Grail of roboticists. And because they have to work in the real world, the old maxim "what can go wrong will go wrong" could not be more apparent in Lee Gutkind's drama of sleep-deprived students and researchers working around the clock against impossible deadlines.

The US Darpa Grand Challenge, launched in 2003, was a competition in which the aim was to race fully autonomous vehicles 250 miles through the Nevada desert. The rules on autonomy were extremely demanding - absolutely no direct or remote control was allowed by human or "other biological entity" (sic). Autonomous robot navigation in research labs, warehouses or museum corridors is relatively commonplace so the basic techniques required are well known, but three things made the Darpa competition uniquely challenging. First, the rugged and uncompromising terrain. Second, since this was a race the sedate 1mph-2mph of indoor robots would not do - more like 30mph would be required. And third, the complex and delicate sensors and computing equipment that comprise the subsystems for autonomous navigation would need to be flawlessly integrated and fast enough to react in time to unexpected obstacles or other racers, while rugged enough to cope with being bounced around in an off-road vehicle.

I recall reading the rules at the time and immediately seeing this was a serious and difficult undertaking that would almost certainly cost more than the $1 million prize money to mount an entry. To CMU's charismatic Whittaker, however, the challenge was irresistible. He approached his senior management, who initially refused support. Undaunted he set up a private entry and charmed corporate sponsors - while being careful not to fall out with his CMU managers so that there would be no embarrassment when they realised that university support could no longer be reasonably withheld.

Whittaker also saw that postgraduate students would be too expensive to employ in the team, so he came up with the ruse of mounting an undergraduate course on "Fundamentals of robot development" in which the group project would be to design, build and race an entry to the Grand Challenge. This, according to Gutkind, is a hallmark of robotics at CMU - the "rookie option". It's both a strength, because the rookies will be prepared to expend huge energy and enthusiasm tackling and sometimes surmounting impossible problems, and a weakness, because CMU were judged not to have sufficient experience to win the really big (Nasa) robotics research grants.

Three problems with this book made it hard going for me. First, it is written in idiomatic American English so I was, for instance, left wondering what a "soccer-mom van" is (thank goodness for Google). Second, there is no glossary or index to help with the technical acronyms. Third, and perhaps most surprising, there are very few pictures of the robots themselves. Despite these niggles, it is well worth persevering. The closing story of Zoë's triumph in January 2006, when she "finds" life in the Chilean Atacama desert for the first time on her own, is a real tribute to the dogged perseverance of Wettergreen's team and a compelling demonstration of the remarkable potential of autonomous robots.

Alan Winfield is Hewlett-Packard professor of electronic engineering, and associate dean, faculty of computing engineering and mathematics, University of the West of England.