Robots Don’t Play Sports for Fun

MIAMI BEACH, FL - MAY 06:  Robots from the University of Miami, College of Arts and Sciences' Department of Computer Science play a game of soccer at the eMerge Americas Techweek held in the Miami Beach Convention Center on May 6, 2014 in Miami Beach, Florida.  The eMerge conference brought together more than 100 speakers from large technology companies and emerging ventures throughout South Florida and Latin America as well as gave people hands on experience with robots, video games and other high tech ventures.  (Photo by Joe Raedle/Getty Images)

Two weeks after the World Cup ended last July, a far more ambitious soccer tournament kicked off in Joao Pessoa, a city on Brazil’s northeastern coast. The fields were indoors, the balls were miniature, and the athletes were robots. These mechanical players were not on field just to kick the ball around. They were working toward a very serious goal, established when the first RoboCup was held in 1997: Building a robot soccer team able to defeat the humans that win the 2050 World Cup.  

This goal is a long way off: At RoboCup 2014, the players struggled just to stay upright. Like 4-year-olds swarming around a ball in a suburban park, these robots are comically inept. But, like kids, they’re also learning. Each year the scientists and engineers who build the RoboCup competitors debut nimbler, quicker-thinking robots that get a little closer to the ambitious 2050 goal.  

But the RoboCup teams aren't just playing a game — they're pushing robotics forward. Sports provide unique physical and artificial intelligence challenges to engineers and scientists trying to build robots that can move and think like people. “If we can have robots that are advanced enough to play soccer and beat humans, there will be a lot of theory and practical inventions” that go well beyond the playing field, says Andrew B. Williams, a professor of electrical and computer engineering at Marquette University.

At this stage in robot research, practical innovations that seem simple still count as significant leaps forward. Teaching a robot to play soccer on two legs was particularly vexing, says Williams. His students eventually settled on a gait that has its robot bend its legs and lower its center of gravity, providing better balance as it moves forward.

And even the most agile robot can’t play a sport without processing information about what’s happening around it. At RoboCup, robots must be aware of field boundary lines, goals, and other robot players. “It doesn’t have GPS for the soccer field,” Williams says of his team’s robot. “It can only use markers and information of where it’s been and where it’s at. There are some complex mathematical algorithms that are used for robot localization." German roboticist Dieter Fox, for instance, is a RoboCup veteran and collaborator of Sebastian Thrun, the Udacity CEO who led Google’s development of a driverless vehicle. “Through RoboCup,” Williams says, “they developed a lot of the localization algorithms that are used in autonomous driving cars.”

It’s not just soccer helping roboticists innovate. Increasingly, engineers and scientists are using the complex movements and quick processing necessary to play sports to refine robots. For robots (as with humans), balance is more difficult in sports like hockey and skiing — which is why a team at the University of Manitoba put its robot Jennifer on the ice. As researchers put it in a press release last month, “The changing nature of snowy ground, and the rapid control response required by alpine skiing, present significant challenges to gait design and dynamic balancing in Humanoid Robots.”

But not all sports are so challenging for robots. Some are already beating humans in games we invented. A robotic snake won a race across an Olympic-size swimming pool, and a trash-talking robot kept pace with Rory McIlroy in a golf accuracy contest. There's also a table-tennis-playing robotic arm and an industrial robot that's shot jumpers with San Antonio Spur Marco Belinelli.

Soccer, though, is a common language, providing a universal framework familiar to teams from around the world. “Everyone understands soccer,” says Williams, who serves as faculty adviser to a team of Marquette undergraduates who participated in RoboCup 2014 and have qualified for RoboCup 2015.

In a 2010 paper, Claude Sammut of the ARC Centre of Excellence for Autonomous Systems in Australia argued that soccer is uniquely useful in developing robot artificial intelligence “because it requires the robot to perceive its environment, to use its sensors to build a model of that environment and then use that data to reason and take appropriate actions. On a football pitch that environment is rapidly changing and unpredictable, requiring a robot to swiftly perceive, reason, act and interact accordingly.”

Sports aren’t only useful when it comes to developing new robotic systems. They’re also a practical and engaging way to show off what robots can already do. Take this badminton-playing robot built by the Flanders Mechatronics Technology Center in Belgium. As researcher Wim Symens says in the video, “We decided to build a badminton robot to demonstrate new technologies we developed t… [because] it’s a real eye catcher.” To put a finer point on it, robots that play sports get people to pay attention to advancements in robotics.

RoboCup is actually part of a weeklong extravaganza of robotic excess with exhibits, symposiums, and workshops, in addition to the robot-on-robot matches. Would it generate the same interest if it were a robotics competition that didn’t involve the most popular sport in the world? Williams is doubtful. Of all the other robotics competitions he can think of, none use sports as a way in, and perhaps consequently, he says, “none has grown to the scale and popularity of RoboCup.”