The crowd went wild, almost as if it were a World Cup match. Actually, it was RoboCup 2002. The annual robotic soccer tournament was held in Fukuoka, Japan, this past June as the World Cup was getting under way. The timing was no coincidence. ¿The goal of RoboCup is to develop a team of robots that can beat the human World Cup champions by 2050,¿ says Hiroaki Kitano, a Sony artificial-intelligence specialist who is also president of the RoboCup Federation.
The notion of robots taking on Brazil would be laughable if roboticists around the world were not so enthusiastically answering the call. Kitano and his collaborators started RoboCup in 1997 with hopes that a grand challenge would spur advances in robotics and artificial intelligence. The first year only a couple dozen groups competed with wheeled robots and simulations, in which ¿players¿ were simulated on separate computers and a server governed the interactions. All the matches took place in one afternoon in the ballroom of a Nagoya hotel.
This year there were almost 200 teams. The matches were held in the Fukuoka Dome sports stadium, stretched over four days, and drew 127,000 spectators. More significantly, humanoid robots took to the RoboCup pitch for the first time. The two-leggers are not running yet, so they competed at walking speed¿traversing a distance equal to five times their height, circling a pole and returning.
It was surprisingly entertaining, if not quite as exciting as the duel between Brazil¿s Ronaldo and Germany¿s Oliver Kahn. Some of the robots walked with the uncertainty of children taking their first steps; others, with the baby-step caution of the aged. Many of the robot makers hovered over their creations with outstretched hands, ready to catch a stumbler. (Human intervention netted a 30-second time penalty.) More soccerlike were the robot face-offs in penalty kicks. Because the robots¿ reactions take so much time, defense meant getting in front of the ball and hoping the kicks didn¿t go off at an angle.
The wheeled robots, which don¿t have to fritter precious computational power on balancing, can react in real time to moving balls. For pure efficiency, there would seem to be little reason to walk. So why bother with legs? That has been a perennial robotics question. ¿In the early 1980s there was a big debate in the U.S. over whether robots should look like humans or not,¿ explains Christopher G. Atkeson, a roboticist at Carnegie Mellon University. At the time, researchers¿and funding agencies¿wanted robots for factory automation. Atkeson says the thinking was, ¿If the goal is to make VCRs, there is no need for robots to look human.¿ He adds that the U.S. military wrote off the idea of robotic soldiers. ¿Work on humanoid robots in the U.S. has been slow to take off,¿ he says.
In Japan, opting for legs or wheels has long depended simply on the application. Masato Hirose, who led the development of Asimo, Honda¿s quasi-autonomous walking humanoid, explains that Japanese researchers want a robot to assist humans with daily activities. ¿The merit of a humanoid is that it can go every place a human can go,¿ he states, including up and down stairs and into confined spaces. He adds that robots that look human will also make interactions more natural.
Kitano notes that humanoid success is not the final word. Japanese teams have not fared well in the wheeled and simulated robot competitions, where researchers from the U.S. and Europe have developed new approaches to robotic teamwork, among other techniques that rely on artificial intelligence. These advances could be quickly applied to humanoids, once their hardware is up to it. Beating the World Cup champs might require worldwide cooperation.
Dennis Normile is based in Tokyo.