UT computer scientist and neuroscientist develop robot soccer players

Jack Stenglein

Tune in this weekend to watch some UT soccer players compete against other college teams. The twist: they’re robots. 

UT computer science professor Peter Stone and neuroscience professor Michael Mauk are working together to develop agile robot soccer players. Stone has been developing these robots since 1997, and Mauk’s lab has been researching and creating a simulation of the cerebellum, a part of the brain that contributes to motor control, for more than 20 years. The two teamed up three and a half years ago and began feeding data from Stone’s virtual robots into the cerebellum simulation.

Stone said the robots currently perform basic functions consistently but fall over more often when performing tasks that require more agility.

“If we want them to walk without falling over, that’s easy to do,” Stone said. “In competitions, we want them to walk as fast as possible, to kick as hard as possible. We’re pushing their performance in various ways.”

Mauk said his lab helps maximize the performance of the robots by making sure they don’t fall over. 

“Our cerebellum improves our agility by predicting what to do next,” Mauk said. “It’s the same sort of thing (with the robots). If we can predict when a (robot will fall), there are a variety of things we could do to fix it.”

Mauk said it will take the researchers a couple more months to develop the fall prediction algorithms so the robots have a better chance of preventing falls. If this program is successful, the cerebellum simulation will then need to be scaled down to work with the smaller computing power of the robots. 

“Half of the neurons in your brain are in the cerebellum,” Mauk said. “We need specialized computers with parallel processing to run the simulations. Even our smallest simulations have a million simulated neurons. The robots don’t have that kind of power yet.”

The robots will play this weekend in the RoboCup US Open at the University of Miami. Three other teams are entered in the tournament: University of Pennsylvania, University of Miami and a mixed team between UPenn and UT. Stone said the mixed team will give him a chance to see how well the robots can work together.

“We’re trying to get the robots to be good at playing with players they’ve never worked with,” Stone said. “It’s kind of like playing a pick-up game.”

Stone said the ultimate goal of the RoboCup project is for a team of robots to beat the FIFA World Cup winners by the year 2050. However, Stone said his personal goal in working with these robots is more extensive.

“My long-term goal is to create autonomous agents that can interact in the real world,” Stone said. “The same tech that goes into soccer would have to go into self-driving cars. The research challenges that come up in robot soccer are very general. It’s a very good application for me.”

The RoboCup US Open matches will be live streamed on the Miami CS website.