Robot Soccer Players Move the Ball, and Technology, Forward
They lurch and waddle. They fall, a lot. They cluster around the ball like a herd of mechanical toddlers at a Saturday morning scrimmage.
Soccer may be the beautiful game, but there is nothing elegant about watching robots play it, even as we come up on the 25th anniversary of the Robot World Cup, or “RoboCup.”
It may seem like child’s play, but for the researchers watching on the sidelines, this game offers valuable insight into the way robots relate to objects around them and navigate a fast-changing environment. Robot soccer requires technological advances that could someday help humans travel more safely, do household chores, or be rescued with less risk to first responders.
But first, they need to work on their passing game.
Why Make A Soccer Bot?
This idea – to design robots that could beat the team that wins the FIFA World Cup – was first written about in 1992 by Alan Mackworth, a professor at the University of British Columbia.
Independent of Mackworth, a group of Japanese artificial intelligence researchers started playing with the idea of robot soccer in the early-1990s to accelerate the development of AI and robotics for commercial and real-world applications.
Soccer presented complex, exciting challenges for hardware and software design – to act independently but in coordination – and it’s a global sport.
Within a year, an international cohort of researchers launched the Robot World Cup Initiative, “RoboCup” for short.
Robot soccer is a test case for algorithms used in all kinds of applications, like using cameras for reasoning, or predicting the movements of others, said RoboCup President Peter Stone, who is also the executive director of Sony AI America, a professor at the University of Texas at Austin, and director of Texas Robotics.
“These solutions can be deployed in the field to make a real-world difference,” said Stone.
The first game was in 1997, with five robots per team on a scaled-down green pitch. Now it’s an annual competition, averaging between 350 to 500 teams, from dozens of nations.
But Can They Play? Sort Of.
RoboCup has several leagues, each with different aims. There is Humanoid – the ones with the feet; Middle Size – like souped-up Roombas; Small Size – under 6 inches tall and using a golf ball for play; and Simulation – which focuses on AI strategy on a virtual field, played on a computer. There’s also Standard Platform, which focuses on coding finesse, as all teams use the same humanoid robot.
Once a human has placed each robot on the pitch, and a human referee blows the whistle, players kick-off following the classic robotics process: sense, plan, act.
That’s no small feat. The human players in a professional match—or for that matter, the tykes coming out for U.S. Youth Soccer games—are barely conscious of the endless tiny tasks and judgments the sport requires. Humans programming robots to play the same game, however, are painfully aware of them.
Each robot must do several things at once: assess its environment, locate the ball, distinguish teammate from opponent, and plan to pass, shoot, or defend. It also needs the mechanical dexterity to kick, block and move around the pitch. Or maybe just tip over for what looks like no apparent reason.
Each year, the RoboCup challenge is made harder. Initially, balls were orange, and the goals were blue and yellow, so color detection was an obvious parameter for successful robot design. Now the balls are black and white, and the goals are white, too, like in a professional game. Programming and hardware design must keep advancing. For example, the Standard Platform league recently added short-range sonar to the robots’ chests to detect other robots.
But even if they can avoid collisions – sometimes – they can’t run spontaneous plays. Which professional players can do. Often since childhood.
Instead, robots can generally be programmed to act as defenders, or outfielders, and to switch behaviors under certain conditions. And they can execute a limited series of moves during routine experiences, such as kick-offs.
But winning is also about heart and hustle. Which, perhaps unexpectedly, can be programmed. When he was a graduate student, Stone created a program to consider the score and how much time was left in the game.
“The robots would become more defensive if we were winning,” he said. “And if there was not much time left, or if we were losing, they would take more risks. But you have to program that in—they can’t do it by themselves.”
Beyond the Pitch
What started with soccer now includes challenges built around rescue assistance – RoboCupRescue; domestic chores – RoboCup@Home, workplace tasks – RoboCupIndustrial, and both soccer and rescue challenges for young students and undergraduates – RoboCupJunior.
The idea is that these challenges are similar enough to soccer programming that those working on a robot kicking a ball can parlay that into a robot climbing stairs, unloading groceries or carrying fire hoses.
“We’re just trying to create the fundamental technologies that companies will be able to take and run with,” Stone said.
For example, Stone used many algorithms and ideas he developed for robot soccer to make a vehicle to compete in the 2007 DARPA Urban Challenge, one of the earliest autonomous driving competitions, as did several other RoboCup competitors.
But designing an autonomous car is easier than making a soccer robot. For one thing, other cars aren’t typically trying to cause you problems.
“By contrast, in soccer, there are players that are actively trying to thwart you,” Stone points out.
But Stone, and his organization’s ambitions are big: by 2050, to see that “a team of fully autonomous humanoid robot soccer players shall win a soccer game, complying with the official rules of FIFA, against the winner of the most recent World Cup.”
It sounds like a daunting task to complete in 28 years, but Stone points to other advancements as reasons for modest optimism.
“From the Wright brothers’ first flight to landing a man on the moon was in the order of 60 or 70 years. From Watson and Crick discovering the double helix model and understanding what human DNA looks like, to sequencing the whole human genome, was 50 years.
“From the first computer, to a computer beating the human chess champion? Fifty years.”
Stone wants his robots to win and look good doing it.
“We want the robots to play the beautiful game,” he said.