Brenna Argall’s goal for the summer doesn’t seem out of the ordinary: To practice her soccer skills to “become a better teammate.”
But the Carnegie Mellon University grad student won’t be doing line sprints and if she falls a little behind on her practice, there’s little chance her teammates will complain. Because Argall’s team plays the beautiful game while mounted on Segway scooters, and her teammates – bearing cartoon-derived names like “Pinky” and “Dot” – are robots, who’ll chase the ball just as hard whether she’s practiced up or not.
Robot soccer has been an important part of robotics study for nearly ten years, since the founding of Robocup – an annual, international competition of robot soccer teams designed to spur creativity in the artificial intelligence design community. Robocup competition is divided into separate leagues, such as the dog-like four-legged competitors, the small-sized league using golf-ball sized balls, and the new Segway league in which Argall’s robots compete. It was Carnegie Mellon professor Manuel Veloso, who leads the Coral Research Group at CMU and is now vp of the RoboCup Federation, who was the founder of RoboCup in the U.S.
According to Dr. Brett Browning – systems scientist at the Robotics Institute at Carnegie Mellon, advisor to the Institute’s Segway-soccer project, and co-chair of the Robocup U.S. Segway league – Robocup isn’t just good fun, but has become one of the premier ways the AI community advances its work.
“The idea behind Robocup is that we’re trying to use soccer as a standardized domain,” says Browning. “It’s fun, but what we’re interested in is what intelligence is required to make a robot autonomous, make it play with teammates, in this challenging environment. A lot of those things required in robot soccer are the same things we need a robot to be able to do for it to be usable in a normal environment.”
Those requirements include building a robot that can make real-time, autonomous decisions, based not just on the robot’s own parameters but on the actions and reactions of its teammates. If, for example, one ’bot passes down the wing, its teammate must know to cut the ball off before it goes out of bounds. But in the Segway league, half of each team are Segway-mounted robots, while the other half are Segway-riding humans. And the interaction between human and robot teammates makes CMU's Segway team one of the most
unique projects in artificial intelligence.
Training fussy 'bots
Tucked into a garage-like corner of the Robotics Institute’s field robotics lab, which comprises the entire ground floor of a large building on CMU’s campus, Brenna Argall’s Segway-mounted robots stand next to a massive red Hummer-like vehicle. It must be inspirational and perhaps somewhat intimidating for a robotics student to work, literally, in the shadow of one of Red Team’s autonomous vehicles – the vehicles built by robotics legend Red Whittaker to compete in last year’s DARPA Grand Challenge, a race of autonomous vehicles across a 132-mile course. But in its own way, Argall’s Segway team may one day be just as important in the building of the kinds of everyday robots we think of inhabiting the late-21st Century.
“When robots really make it, there will be situations where robots need to work alone without human insight or other robot insight,” says Browning. “But more often than not, they’re going to be working in environments where they’re working around humans, or with other robots, or both. The challenge in Segway soccer is to make robots that are not only autonomous, but that can understand what humans are communicating, and can communicate back to the humans itself.”
As far as robots go, Pinky and Dot are no Data and C3PO. In fact, they’re a bit closer in looks to Rosie, the maid on The Jetsons: A two-wheeled Segway platform with an added-on ball trapper and kicker mechanism, topped off with a series of boxes containing the computers that serve as the robots’ brains. In attitude, they’re more like Marvin – the paranoid android from Hitchhiker’s Guide to the Galaxy: Temperamental, and prone to leaps of logic that constantly frustrate their creators. In a demonstration of the Segway bots, for example, Dot recognized a small box on the floor of the lab as the same size and color as the soccer ball, and refused to stop tracking it. (“If you don’t want me to track it,” Dot almost seemed to say, “then don’t ask me to.”) It’s the kind of problem that plagues robotics engineers, and takes up much of Argall’s time.
“Perception problems are probably our biggest limitations,” says Argall. “For the robot to build a model of its world with just vision – that’s very difficult.”
“It’s a problem that’s universal to robot soccer and robotics in general,” says Browning. “They’re meant to be making their own decisions, and the quality of those decisions is driven from their perception of the world. But in Segway soccer, because it’s a human-robot team playing another human-robot team, things are moving around very quickly, so they must make these perceptions very quickly.”
But if anyone is moving along with Segway soccer’s dream of hybrid teams in which human and robot players work on equal terms, without one set of players taking complete command of the match, it’s Carnegie Mellon. Until recently, according to Argall, the ’bots have been making all the decisions: calling for the ball, or for shots on goal. But Browning says that a new system of wearable computers allows for more equal exchange – human teammates can see their robot’s information, and give simple commands in response.
In fact, CMU might be just a little too far along for its own competitive good. While the league kicked off with one match at 2005’s Robocup U.S. Open, between CMU and San Diego’s Neurosciences Institute, NSI has since dropped out of competition for 2006. And with no other Segway teams ready for the U.S. Open April 20-23 in Atlanta, CMU will have to wait until later in the year, when Browning hopes to invite budding Segway teams from Universities of Texas and Alberta to compete in Pittsburgh.
Until then, CMU’s Segway team will have to hone its skills, tighten each bolt, refine each algorithm, until Pinky and Dot can be renamed Ronaldinho and Pele, even if the bicycle kick and flying header remain a little bit further down the road.
Justin Hopper is a freelance writer who last wrote about Mr. Small's for Pop City.
photo captions:
(top) Dr. Brett Browning
Yang Gu, Brenna Argall, Dr. Browning
Pinky and Dot
Brenna running Pinky
All photographs copyrighted by Jonathan Greene