The half-hour performances are intended not merely to stir the curiosity of the school kids who dominate the 500-seat auditorium, but also to advance the emerging field of evolutionary robotics. Noel Sharkey, a computer-science professor at Sheffield University, is using the contest as a Darwinian breeding ground, where the weak robots perish and the strong survive to perpetuate their kind. The idea is to build dumb robots with the capability to “evolve,” leave them alone and see how smart they become. Says Sharkey: “What we are after are really simple explanations for complex behavior.”
The experiment starts out simply enough. Each of the 11 robots in Sharkey’s menagerie has the same goal: to obtain energy. Six of them are prey robots, who “graze” beneath a tree of white light, replenishing their batteries through solar panels. Five of them are predators who, like Gaak, can derive energy only by draining it from the “preybots.” Each species recognizes the other by the amount of heat they give off, as detected through infrared sensors. The biorobots hunt and flee free of all human control, guided only by their simple “brains”–so-called neural networks that mimic the function of brain cells. To simulate evolution, Sharkey every so often picks out the most successful robots and transfers part of their brains to a new generation, who start the hunt all over again. Darwin would approve. Locked in a continuous battle to survive, preybots evolve new methods of escape, while the “predatorbots” devise new ways to entrap.
Sharkey, 53, has become a kind of latter-day Frankenstein. A school dropout at 15, he did stints as a dance-band musician and psychiatric nurse before earning a psychology degree at Exeter University and doing research at Harvard and Stanford. He is perhaps best known as a judge on the cult BBC TV show “Robot Wars,” where contestants–from school kids to engineering graduates–enter their own fighting ‘bots. His research has a slightly more practical bent. Unmanned missions to faraway planets might, for example, need different species of autonomous robots bred to support each other in different roles. Sharkey’s robots may also shed some light on behavior. What if preybots, say, begin banding together and acting as a herd for their own protection? This kind of adaptation might give scientists some insight into how human and animal proclivities arose. “By building artificial models we can get some evidence of how learning and evolution inter-react,” says Dylan Evans of Bath University.
Biorobots are also a step toward creating intelligent robot helpers endowed with something like common sense–a goal that has eluded artificial-intelligence rsearchers for decades. “Natural creatures are much more complex than anything that we have managed to produce so far,” says Inman Harvey, a researcher in robotics at Sussex University, “but they aren’t designed on the drawing board. So perhaps we can pick up a few basic ideas from Darwinian evolution.” Hopefully scientists will be able to catch on faster than nature, which took 4 billion years to hit upon the human brain.
