From 1999 to 2005, Cornell’s Big Red robot soccer team dominated the international competition. Why it’s time now to call it quits.

By Beth Saulnier

Raff D'Andrea and friends

Nicole Schlegel was tired. Very, very tired.

She was exhausted from jet-lag after the flight from New York to Australia; from long nights of frantic work; from the stress of the upcoming competition. Schlegel was finally getting some sleep when there was a knock on the door of her Melbourne hotel room: she had to come right away. Still clad in her pajamas, she raced downstairs—and spent a sleepless night trying to stop a small robot from jumping around a miniature soccer field like the proverbial headless chicken.

The 2000 computer science grad was in charge of artificial intelligence for Cornell’s goalie in RoboCup, the international robot soccer competition in which the Big Red has long been the undisputed powerhouse in the small-size league. Her late-night adventure—she managed to correct a problem with how the goalie was processing information fed to it from an overhead camera and help the team to its second consecutive win—is just one of countless RoboCup war stories. There was also the time when mechanical and aerospace engineering professor Raffaello D’Andrea, the team’s adviser, was mugged after scoping out the 1998 contest—but managed to hold on to his video camera while whacking a pair of Parisian thugs with his laptop computer.

And when the team landed at the Stockholm airport for the ’99 competition, only to learn that the airline had lost most of their equipment.

And when they were tied 0-0 in the 2000 semi-finals against one of their arch-rivals, the game hinging on penalty kicks. Schlegel forgot to command the goalie to switch sides, costing Cornell the first point (and, quite likely, shaving a few years off the team’s collective lives).

And when the Cornellians had to stand idly by during a 2001 game in Seattle, teeth clenched, while another team pinned the ball against the wall and ran out the clock—skirting the rules against immobile players by having the robot twitch every twenty-nine seconds.

"It was quite a bit of sleepless nights, and it was very nerve-wracking at the end," sums up Bryan Audiffred, M.Eng. ’00, an electrical engineer and alumnus of the 2000 squad that won in Melbourne, "but it was definitely one of the greatest experiences of my life."

But after amassing the strongest record in RoboCup history—including four wins, a place, and a show—the Big Red is retiring; for a variety of reasons, D’Andrea has decided not to field a team for the foreseeable future. When the next tournament is held in Bremen, Germany, in June, the competition can breathe a little easier: Cornell is staying home. "In a way, it’s sad that we can’t carry on our legacy," says Schlegel, a Ph.D. candidate in earth and planetary sciences at the University of California, Berkeley, who keeps current with RoboCup news via the organization’s list-serve.

RoboCup took up a fair chunk of D’Andrea’s time, but that’s not his primary motivation for retiring the squad. Like the game itself, it’s complicated. One reason is a sense of frustration with RoboCup rules that don’t acknowledge the technical level of the competition: some rules needed modification as the teams improved; others were imposed before the competition was ready for them.

Then there’s a difference of opinion over how much the game ought to be faithful to the precepts of real-life soccer. Since the ultimate goal of the RoboCup competition is to field a team of person-size robots who can play against human athletes by the year 2050, organizers want the game to resemble soccer as much as possible. That prompted them, for instance, to remove the walls surrounding the pitch, but the change led to frequent stoppages when the ball went out of play. Schlegel and others found such decisions more than a little constricting. "With what they allow," she observes, "you can only do so much."

CEM cover from Summer 1999 featuring RoboCup

But D’Andrea’s principal reason for retiring the team was something else entirely: the fact that, essentially, RoboCup was a victim of its own success. For the past decade, some of the world’s best and brightest student minds have been puzzling out how best to play five-on-five soccer, with 15-centimeter-high robots battling over a neon-orange golf ball—and they pretty much figured it out. Although dramatic improvements were made in the first few years, lately the advances have been increasingly incremental.

"We were kind of like the Bell Labs of the competition," D’Andrea says. "We always showed up with new things, were always innovating—not necessarily polished, but something really pushing on the creative side. But as the competition matured, you couldn’t do that as much; everybody tended to converge to the same solution. This year in Japan, everybody’s robots looked just like ours—four-wheeled, omni-directional robots." He compares the situation to the design of automobiles, which all tend to have four tires and a steering wheel. "Most look pretty much the same, because we’ve had eighty or ninety years to converge to a solution that’s optimized," he says. "The constraints remain the same; the roads haven’t changed much. Therefore, everybody pretty much figures out the best way to do it."

Sherback, too, is a bit nostalgic for the early days of the competition. Although the robots were slower and less dexterous, he and his teammates were working in the kind of anything-goes atmosphere that no engineer could resist. "Back in the day, there was a lot of latitude," he says. "It was not at all clear what the best way to do things was in terms of hardware. There was a lot of freedom to have some wacky idea and see it through." To this day, Sherback believes that one of the dumbest decisions of his entire life was opting to go straight to work at Applied Materials after graduation, rather than accompanying the team to Melbourne. When the Big Red won, he says, "I was blown away. And I found out about it in an e-mail in my cube at work."

Aris Samad-Yahaya, a 1999 electrical engineering grad who runs his own software company in his native Malaysia, relished the project’s research challenges. "Doing something like RoboCup, you’re not using technology that’s fully resolved or fully in the literature," says Samad-Yahaya, who remembers wearing the same clothes for several days after the lost-luggage incident in Stockholm. "You look at what’s out there, and if nothing fits the application you want to do, you have to create a solution of your own. You have no idea what the outcome is going to be. You really need to solve a problem, to take things you learned in your undergraduate career and apply them in a very real way."

Although Cornell’s RoboCup days may be over, the team spirit lives on. Interdisciplinary projects with the opportunity for hands-on engineering are wildly popular among students. The engineering college is home to more than a dozen teams competing in everything from the design of a solar house to the creation of a concrete canoe.

Ask some of the 150 or so students who participated in the RoboCup over the years, and they’ll tell you: the games were fun, and winning was even better, but the greatest pleasure of the project was getting the chance to work with other engineers from across the college. "One person just couldn’t build it by himself," says Oliver Purwin, a Ph.D. candidate in mechanical engineering. "The whole team effort is amazing."

Purwin was an M.Eng. student on an exchange program from Germany when he worked on the team that finished third in 2001; he still gets steamed over that twitching robot. He helped improve the omni-directional drive—which allowed the robots to move in all degrees of freedom—and as a grad student he worked on trajectory generation, figuring out the best way to get a robot from Point A to Point B. His fellow RoboCup alums have tackled that issue and a wide variety of others: how to control the ball, interpret visual information, improve agility, or just plain kick harder. "It’s a really great feeling, how in the end it all comes together," he says. "The whole thing is just bigger than the sum of all the parts."

Purwin is sitting in Cornell’s RoboCup headquarters, a large room on the first floor of Rhodes Hall. The walls are decorated with posters and shirts from previous tournaments around the globe, but the dominating feature is the regulation RoboCup soccer pitch: a green-carpeted surface 2.8 meters long by 2.3 meters wide. Down the hall is a display case featuring the trophies the Big Red brought home every year.

Except one: in 2004, the team made an ambitious foray into distributed intelligence, but wound up being eliminated in the first round. In the space where a trophy might have been, instead sits a small wind-up toy robot, painted bright red but looking rather sad. (Last year, however, Big Red had an improved distributed platform that performed well, and the team came in second.)

"It’s great when you have a whole bunch of smart kids working toward a common goal," says D’Andrea. "They see that the stuff they learn in school is actually useful. You work for a year toward building something complex like this, and you go to a competition and you win—that gives you the confidence that, ‘Hey, I can do anything if I set my mind to it.’"

atching a RoboCup match can make you kind of dizzy: those little robots career around like lightning, and the ball moves so quickly it can be hard for your eyes to follow it. Despite the governing organization’s focus on imitating human soccer, a RoboCup match is much more like hockey. "It’s a fast game," D’Andrea says. "It’s entertaining. It’s exciting. When you watch some of the games it looks like they’re alive, like the robots have a mind of their own."

The present regulation pitch is roughly four times the size of the 1999 field; design improvements made the robots go so fast, they were constantly knocking into each other on the smaller surface. "If you watch the game in 1999, the robots were maybe moving at top speeds of one meter per second," D’Andrea says. "There was essentially no passing, just getting to the ball and aiming it toward the net. By the time of the competition in Japan last year, robots were routinely passing the ball to each other, they were moving at speeds up to 2.3 meters per second, kicking the ball at speeds up to ten meters per second."

On top of offering students a chance for teamwork and technical innovation, RoboCup and other such competitions have another advantage: they’re great topics for job interviews. When companies ask for examples of how they’ve solved problems or worked on a team, students can point to long nights trying to figure out how to whack a golf ball into a soccer goal or build an autonomous submarine. "A common complaint about engineers coming right out of college is that they have absolutely no idea about practical limitations and how to connect theory and practice," says Audiffred, now an instructor in electrical engineering at Louisiana State University. "This kind of project forces you to do that."

Sherback, who worked in Silicon Valley for three years between college and grad school, took the lessons of the RoboCup with him to the office. The experience, he says, may not have turned him into a computer scientist or an electrical engineer—but it gave him a working knowledge of what those colleagues confront when they’re solving problems. "Like any team thing, you learn not to charge in like a bull in a china shop and start barking complaints and orders at people," he says. "You can empathize and understand what kind of burdens you may be putting on them, and appreciate their efforts. It prepares students to go out into the so-called real world."