RadioBDC Logo
| Listen Live

Getting robots to play together

Team approach used to perform human tasks

Get Adobe Flash player
By Carolyn Y. Johnson
Globe Staff / March 18, 2010

E-mail this article

Invalid E-mail address
Invalid E-mail address

Sending your article

Your article has been sent.

Text size +

CAMBRIDGE — Two soccer teammates took the field on a recent Saturday afternoon: one made shot after shot while the other played goalie.

The shooter, wearing a distinctive uniform dotted with colorful spots, had just taken a powerful kick when something went wrong. The call went out to stop practice — and to find a hot glue gun.

“Something is hanging off,’’ Svilen Kanev, a Harvard University sophomore and member of Robotic Futbol Club Cambridge, said as fellow students picked up the cylindrical black robot to administer first aid.

When most people think about robotics and artificial intelligence what comes to mind are individual robots — whether it is one that vacuums the floor, a rover that explores Mars, or the computer that beats a human chess champion.

But at the cutting edge of robotics, many researchers are focused on multiple-robot systems in which it is not the individual that matters, but the group. These robots are being designed to work together and react to each other. Researchers envision that these could one day perform tasks dangerous or menial or that machines in combination might do better than humans — whether it is search and rescue or construction of a building.

It is likely to be many years before teams of robots become useful or commonplace, however. Chess can seem difficult to a human, but it is much easier for a robot than seemingly everyday tasks, like seeing and picking up a cinder block, say, or navigating a real-world environment and coordinating its actions with its robo-mates to achieve a goal.

Robot soccer might appear to be a whimsical hobby for engineers who do not want to grow up, but many real-world technical and scientific challenges must be solved for these robots to work together in dynamic scenarios that occur during the game. It turns out to be a fun way of learning to work with multiple-robot systems, and creating a robotic team that can beat the human World Cup soccer champions by 2050 has become a grand challenge in robotics.

“When we started this, the main research question was . . . how do you get multiple robots to coordinate,’’ said Manuela Veloso, a professor at Carnegie Mellon University and president of the International RoboCup Federation, which runs robot soccer competitions. The RFC Cambridge, a team of about 15 Harvard and MIT undergraduates, is working to build a five-member soccer team for RoboCup this summer in Singapore. In a recent session, the two robots played against each other as a test.

Researchers are also looking for inspiration from termites, bees, and ants — simple insects that work together to accomplish complex goals such as building an intricate nest.

“These collective behaviors are very powerful and arise from very simple individuals,’’ said Radhika Nagpal, an associate professor at Harvard University and a member of the Wyss Institute for Biologically Inspired Engineering. “One thing about termites that’s really hard to replicate is that they build skyscrapers.’’

Nagpal is interested in developing multirobot systems, built out of individual robots that might not seem intelligent but can accomplish complicated tasks when their efforts are massively multiplied with others.

In Nagpal’s lab, decorated with photos of a wasp’s nest and an elaborate, castle-like termite nest, Kirstin Hagelskjaer Petersen and Justin Werfel demonstrated an insect-like robot with a hybrid of wheels and legs (called “whegs’’) that can climb a stack of blocks while carrying a block on its head.

“The big idea is to take a system of five robots and building blocks and give it a picture of what you want to build,’’ Werfel said. The robots, he said, should work according to a picture, not by following a set of rigid, preset orders. If other robots put down a piece before them, they simply add on appropriately.

Nagpal is also part of a group working on creating a colony of robotic bees, an effort that received a $10 million economic stimulus grant last year from the National Science Foundation. The ultimate goal of the project is to create robotic bees that mimic those in nature — a colony that could pollinate crops, emulating bees’ ability to hover, fly from flower to flower, and work with others to scout or forage. The technology will require scientists to develop tiny, lightweight sensors and advanced computer programming to get a bunch of flying robots to coordinate movement.

At Rice University, James McLurkin is developing a swarm of robots that can seek out the boundaries of an area, which could be useful in exploration. He is working on more basic problems, as well, such as ensuring that robots move at speeds that will allow them to work effectively with others. If a robot moves too fast, for example, it may not receive a message being sent through the network by another robot or it may carry around stale information.

Ultimately, the goal of the soccer challenge is to create a team of robots in which each “thinks’’ for itself, but RFC Cambridge plays in a league in which soccer players are controlled by a central computer — and that was proving difficult enough for the students practicing recently with cylindrical black robots in a third-floor lounge of a Harvard computer science and engineering lab. The goalie was not having much success stopping the offensive player’s relentless shots.

The students have to create robots that can move without crashing into unpredictable foes and accurately aim and kick an orange golf ball, but they also must create a team that acts in concert.

Instead of having each robot rush for the ball, potentially blocking a teammate’s kick, the computer has to decide which robot can get to the ball fastest and which has the clearest shot.

“How are you defending — should you pass to the other guy or shoot? Or dribble around?’’ Kanev said, describing the split-second decisions the computer must make and then direct the robots to do. “A really big challenge is stitching them together.’’

Carolyn Y. Johnson can be reached at