Robot soccer pits machines against machines
PITTSBURGH — When an IBM computer beat chess champion Garry Kasparov two years ago, its creators cheered. But they knew the limits of their victory.
Their Deep Blue computer couldn't talk, or walk, or even move the chess pieces. And outside the chess world, the machine was clueless.
That is why scientists from around the world have gathered in Sweden this week to challenge each other to another game: soccer. Or, more correctly, robot soccer.
The third annual RoboCup competition in Stockholm, which begins today and runs through Aug. 4, will showcase machines that can move, kick, and think (a little) all at the same time.
By testing them in a constantly changing environment, researchers hope to learn how to create machines that can adapt to the real world. If all goes well, the machines will also know how to work as a team.
"We view it as the next milestone after Deep Blue," says Wei-Min Shen, senior research scientist and head of the University of Southern California's DreamTeam, a 1997 RoboCup champ.
Of course, that's not the only reason the teams have gone to Sweden. Scientists love to compete, even when their players are, well, emotionally detached.
"It's so embarrassing," confesses Manuela Veloso, who heads the Carnegie Mellon University team, CMUnited. "I get extremely involved. You try to be rational, but you still scream at them."
Some 90 teams are vying for top honors in four different leagues: the small size (imagine a very thick piece of toast standing on edge), the middle size (printer boxes on wheels), the simulator (where software programs compete on screen), and the Sony legged robot (where teams battle using Sony's cute robot dog).
Last year, the Carnegie Mellon team won three of those competitions, which were held in Paris. A visit to the team's robo-soccer training camp here in Pittsburgh - actually a large office laid out with desks and soccer "fields" the size of Ping-Pong tables - suggests the team isn't letting up this year. "The challenge this year is to get a better goalie," Professor Veloso says.
Robot soccer brings together a variety of challenges that machines will face in the real world. For example, it's not enough for a robot to see the ball and move toward it because the ball is usually moving. Instead, the robot has to calculate the ball's speed and direction and where it should go to intercept it.
Once it has the ball, new challenges present themselves. The robot has to figure out where the goal is, where the opponents are, and whether it should try to score or pass the ball to a teammate. Even robots without the ball have to determine where to position themselves in relation to the opponents.
"In the games, you would see these little robots move dynamically to these open positions and it was beautiful," says Michael Bowling, a computer-science graduate student and the team's software guru. "We have them anticipating the needs of the team."
The different leagues pose different challenges. The small robots can play as a team because they're linked wirelessly to a central computer, which sees all the action from an overhead camera. Thus the computer knows where its team members are at all times and can coordinate their moves.
That's a far more difficult proposition for the larger robots, which carry their computers onboard and don't have an overall view of the action. These machines have to use their partial knowledge to figure out where they are on the field, where the ball is, and the direction of their opponent's goal.
All items are color-coded to help the robots' cameras distinguish them. Even so, glitches occur. The robots sometimes chase phantom balls. Carnegie Mellon lost a match to Portugal in the small-robot competition last year because security personnel were using the same frequency as the team's machines.
Professor Shen's DreamTeam, the 1997 champs in the mid-size competition, lost its crown last year when one of its own robots scored against the team in the last 20 seconds of a match.
Will a robot team ever beat, say, the US women's soccer team? Perhaps at some point far in the future, researchers say, but their immediate goals are far more modest. "My dream is that you build a machine that thinks like a kid and you put it in a new environment and it plays in the environment and learns something," Shen says.
"When I was a graduate student and before I had two kids, I used to think it was not too far away. But I see amazing things they [the children] do. And I can't imagine when computers will be able to do that."
"But every step we make towards that direction will bring many benefits to the society," he adds.
(c) Copyright 1999. The Christian Science Publishing Society