Data Mining Techniques for RoboCup Soccer Agents

The paper describes an application of the data mining components with learning for RoboCup soccer agents. Data mining modules capable of on-line learning in dynamically changing environment are suggested. These modules provide for adaptive agent behavior in the form of the cognitive control systems that are able to self-tune in non-deterministic environment. Reinforcement learning is considered as the basic method for forming an agent behavior. A cognitive soccer agent for RoboCup Simulation League competitions capable of on-line learning has been developed and studied. Examples of shooting and teamwork are considered. The STEP team based on the cognitive agent succeeded in RoboCup German Open 2004 and was the winner of the RoboCup-2004 World Championship in simulation league.

[1]  Barbara Hayes-Roth,et al.  A Cognitive Model of Planning , 1979, Cogn. Sci..

[2]  Richard S. Sutton,et al.  Reinforcement Learning: An Introduction , 1998, IEEE Trans. Neural Networks.

[3]  Vladimir I. Gorodetski,et al.  Multi-agent technology for planning, scheduling, and resource allocation , 1998, Proceedings International Conference on Multi Agent Systems (Cat. No.98EX160).

[4]  Tom M. Mitchell.  Plan-then-compile architectures , 1991, SGAR.

[5]  Lev Stankevich.  A Cognitive Agent for Soccer Game , 1999, CEEMAS.

[6]  Richard S. Sutton,et al.  Introduction to Reinforcement Learning , 1998 .

[7]  David J. Israel,et al.  Plans and resource‐bounded practical reasoning , 1988, Comput. Intell..