2003:CMU hosting first RoboCup American Open Interest grows in making robots work as teams
Monday, April 28, 2003
By Byron Spice, Post-Gazette Science Editor
The day when a team of robots can defeat a World Cup team of human soccer players is still a distant dream, but at least the robots have the victory dance down pat.
Members of Carnegie Mellon University's CMpack03 team "celebrate" by waving their forelegs after scoring a goal. (Pam Panchak, Post-Gazette)
SCHEDULE The International RoboCup Federation's first American Open, a five-day event, will begin Wednesday in the University Center at Carnegie Mellon University. It will be open to the public from 10 a.m. to 6 p.m. Friday and Saturday and from 9 a.m. to 2:30 p.m. on Sunday, with competition in the four-legged league, small robot league, junior league and simulation league, as well as demonstrations of urban search and rescue robots. For more information, visit the Web site at www.americanopen03.org.
When the four-legged, Sony Aibo robots on Carnegie Mellon University's team score a goal, they go into a "happy dance" -- simultaneously sitting back on their haunches and waving their front legs over their plastic heads.
It's a small step, but then the development of robotic soccer has been a series of very small steps.
"When we started this back in 1996, I never thought [the play] would become so good," said Manuela Veloso, a professor of computer science who heads CMU's robotic soccer program. In the beginning, when the only players were boxy, balky, wheeled robots, the level of play was akin to that of peewee leagues.
"With 6-year-olds, no matter how often the coach says, 'One to the ball,' they all go to the ball," Veloso said. Likewise, those early robots would swarm around the ball, hoping to somehow shove or bump it into the goal.
Today, the robotic players move quickly and nimbly, dribbling and passing the ball, shooting accurately and using strategy. Some robotic leagues still use wheeled robots, but there's also the four-legged league and, worldwide, about a dozen teams are fielding two-legged, humanoid robots.
Each year, the world's robotic soccer teams compete for the RoboCup, which has grown dramatically in its seven-year history. Last year, in Fukuoka, Japan, where the CMU team won the championship of the four-legged league, the RoboCup matches were held in a baseball park and more than 120,000 spectators attended during the four-day competition.
More than 300 teams have entered this year's RoboCup, which will be in Padua, Italy.
"Things are becoming so complex that you really need to practice before you go," Veloso said, so this year Carnegie Mellon will host the first RoboCup American Open, a competition that this week will attract more than 150 researchers from North and South America to CMU's campus in Oakland. Forty-eight college teams and 11 high school teams will compete in six leagues.
Japan and Germany have had similar competitions for years and Australia recently launched its own Open.
Though the competition can be fierce, robotic soccer is, more than anything else, a research effort, Veloso emphasized. So when each year's competition is over, each research group shares the computer code it developed for its team with all of the other researchers.
Though constructing a robot can be a mechanical challenge, controlling a robot -- getting it to think -- has proven to be the bigger challenge. That's especially so in the four-legged competition, where everyone uses the same mass-produced Aibo robot. The only thing differentiating the teams is the software.
So sharing software means the secrets that the CMU team used to win in Japan last summer will be known by every competitor that shows up in Italy for RoboCup this summer. Other teams can use CMU's software, if they choose.
"I think they're all going to be using our goalie," Veloso said. "He was really good."
But the software is complex and each team's code is unique, so most teams won't simply copy all of the winning team's software. Rather, they likely will use the winning team's software to program their opponents in scrimmages, so that they are able to produce superior software by testing it against the most successful software from the previous year.
Such sharing makes sense given the great challenge posed by robotic soccer -- getting robots to work together at a task. This ability to coordinate action will be key to the widespread use of mobile robots, whether for constructing buildings, cleaning up hazardous waste sites or stocking grocery shelves.
Robotic search and rescue, which will be demonstrated this week, is an example of robotic teamwork -- though the team in that case likely would involve robots and humans.
Much of the research on multi-robot teams is now focused on two issues: brittleness and collective perception.
Robots are brittle, Veloso explained, in the sense that they don't adjust easily to changes in environment. They are sensitive, for instance, to changes in lighting. Robots have a hard time seeing things when looking from a brightly lit area into a darker area, or vice versa.
Whereas people put on sunglasses or turn on lights so they can adjust to different light levels, robots have trouble recognizing that the lighting has changed, or that they are looking into an area where the lighting is different, so they can't readjust themselves. One of Veloso's students is working on software to help robots realize when their environment has changed.
Collective perception refers to sharing information between the robots -- with one robot, for instance, alerting teammates to something it sees but that cannot be seen by the others. Communication isn't a problem; while human players can yell at each other, robots can interact by radio link.
But, as with adjusting to changes in lighting, something that is second nature for humans can trip up robots. Picking out just the information that is important, such as the trajectory of a ball, and sharing that information in time for other players to benefit is a problem roboticists are still trying to solve.
Soccer is a good model for studying multi-robot teams because soccer players must not only cooperate, but also anticipate how an opposing team will react. Making it all extra hard is the fact that the ball can move very fast, so the robots must respond rapidly, Veloso said.
She has less interest in the game itself.
"I never watch a soccer game, ever," said Veloso, 45, noting both of her sons no longer play the game. "Only when I go home to Portugal might I see a game."
But Veloso devotes about two-thirds of her research into artificial intelligence on robotic soccer because of its usefulness as a model for robotic teamwork.
it seems to be more of a design and functionality forum than an end unto itself imo. not entirely unlike auto manufacturers sponsoring racing teams to test and advance engineering imo, and there's been almost a century of auto racing... so on a relative basis, that's really "old."
and if you would point me to an "old" windows-based bot soccer software simulator, it'd be spincerely appreciated.
News 
Market Data 
Discover 
