An Inside Look at Robot Soccer

Presentation by Peter Stone of the Artificial Intelligence Research Department of AT&T Labs

In preparation for the Fifth Annual Robot Soccer World Cup, which is coming to the Washington State Convention and Trade Center in Seattle August 2 to 10, we offer a behind-the-scenes look at Robot Soccer.

This is the first time that the RoboCup has been held in North America. The RoboCup Symposium and Competition is co-located with the 17th International Joint Conference on Artificial Intelligence. The robot soccer competitions are open to the public for free, though registration fees are required for the conferences.

Abstract

This is a new machine learning approach in which teams of autonomous agents act in real-time, noisy, collaborative, and adversarial environments.

1. A flexible team structure allows agents to decompose the task space into roles and formations. Team organization is achieved by the introduction of a "locker-room agreement" as a collection of shared conventions, including team plans and pre-defined conditions for agents to switch roles in coordination.
2. Layered learning is a general-purpose paradigm for complex domains. It allows for learning at every level of a hierarchical task decomposition, with learning at each level directly affecting learning at the next higher level.
3. Team-partitioned, opaque-transition reinforcement learning (TPOT-RL) is a new multi-agent algorithm. It is designed for domains in which agents cannot necessarily observe the state changes when other team members act.

In simulated robotic soccer, these three contributions produce a winning team that learns in a real-time, noisy environment. The approach has been validated both in controlled experiments and in competition.

The talk will also report on results in the RoboCup international competitions. Dr. Stone's contributions were fully incorporated in the CMUnited-98 and CMUnited-99 simulator champion teams and partly in the CMUnited-97 small-robot RoboCup-97 champion team.

This research is published in book form as: "Layered Learning in Multiagent Systems: A Winning Approach to Robotic Soccer." Peter Stone, MIT Press, 2000. ISBN: 0262194384

About Dr. Peter Stone

Dr. Peter Stone is a Senior Technical Staff Member in the Artificial Intelligence Principles Research Department at AT&T Labs Research. He received his Ph.D. in 1998 and his M.S. in 1995 from Carnegie Mellon University, both in Computer Science. He received his B.S. in Mathematics from the University of Chicago in 1993. Peter's research interests include planning and machine learning, particularly in multiagent systems. His doctoral thesis research contributed a flexible multiagent team structure and multiagent machine learning techniques for teams operating in real-time noisy environments in the presence of both teammates and adversaries. He is currently continuing his investigation of multiagent learning at AT&T Labs.

About IDGA/Sputnik

IDGA/Sputnik is the Washington State Chapter of the Independent Game Developer's Association, Sputnik. More information is available at their website, http://www.seattlesputnik.com/.