By Rachel Robey
Thanks to their unique reinforcement learning-based approach, researchers from Wisconsin and Texas celebrate an impressive win at RoboCup 2024.
Meet WisTex United, winners of the RoboCup SPL “Challenge Shield” Division at RoboCup 2024, an international competition dubbed “the premier championship for robotics.”
A brand new collaboration between researchers from UW–Madison and UT-Austin, WisTex United claimed 1st place in their division in a convincing 7-0 shutout during the finals in Eindhoven, Netherlands on Sunday, July 21.
RoboCup’s mission remains the same as when it was officially founded in 1997: to spur research in artificial intelligence (AI) and robotics. In addition to the annual soccer tournament and symposium, RoboCup urges scientists toward a “publicly appealing, but formidable” long term goal: creating robots capable of playing soccer against the World Cup champions by the year 2050.
This year, the five-day competition attracted over 3,000 researchers from more than 45 countries. The Standard Platform League (SPL), which WisTex United competed in, is just one league within RoboCup. To compete, teams of identical but autonomous humanoid robots play against each other.
“This means the robots have to be able to navigate decisions on the soccer field, move the ball down the field, visually recognize their own team’s players and other objects and effectively beat an opposing team programmed separately for the same challenges,” explains UT-Austin’s College of Natural Sciences. Though fun, RoboCup is a serious scientific project designed to solve some of the biggest technical challenges facing robotics researchers.
Reinforcement learning a winning approach
“I’m very proud of the work the students did both before and during the competition to achieve a really good result,” says UW–Madison Department of Computer Sciences (CS) Professor and WisTex United team member Josiah Hanna. While most teams in RoboCup SPL use hand-coded rules to direct the robots in different situations, WisTex United took a different approach.
“We’ve instead based our robot’s behavior on reinforcement learning,” says Hanna. “That means that the robots learn through trial-and-error what the most effective way to play is.” Soccer, Hanna explains, makes for a highly dynamic environment requiring real-time decision-making; simply following directions won’t cut it.
Watch as WisTex United wins the final match 7-0.
From Wisconsin, PhD students Adam Labiosa and Will Cong, along with rising senior Chen Li, traveled to the competition with Hanna. From Texas, Professor Peter Stone, PhD students Zhihan Wang and Siddhant Aggarwal, and recent alumnus Josh Kelle rounded out the on-site team. Ahead of the competition, UW–Madison PhD student Abhinav Harish and undergraduate students Benjamin Hong, Zisen Shao, and Yuhao Li contributed to the work, alongside UT-Austin alumnus Geethika Hemkumar.
WisTex United already looking ahead
“This is the second time a team from Wisconsin participated in RoboCup. Last year, we finished third in the same division with the aim to largely base our decision-making on reinforcement learning,” says Hanna. “UT’s team was interested in adopting the same approach, which led to the joint effort this year.”
Thus, WisTex United was born.
“This year, we chose to compete in the lower division because we were forming a new collaborative team and didn’t know how well it would work,” says Stone, “but we intend to compete in the upper division next year.”
In 2025, the RoboCup tournament — and WisTex United — will be traveling to Brazil, where they will undoubtedly face stiffer competition. Yet if this year’s performance is any indication, we anticipate another trophy in their future.
Congratulations and well done to team WisTex United!
