next up previous
Next: Conclusion Up: Robotic Soccer Previous: MAS in Robotic Soccer

Machine Learning in Robotic Soccer

As well as addressing most of the issues inherent in MAS, robotic soccer is a great domain for multiagent Machine Learning. In another soccer simulator, Stone and Veloso use Memory-based Learning to allow a player to learn when to shoot and when to pass the ball [82]. They then use Neural Networks to teach a player to shoot a moving ball into the goal [84]. They use similar techniques in the soccerserver system as well, extending the learned behavior as a part of a hierarchical learning system [83]. Matsubara et al. also use a Neural Network to allow a player to learn when to shoot and when to pass in the soccerserver system [54]. Uchibe et al. have successfully combined RL modules for shooting and for avoiding opponents using real robots [88].

Once low-level behaviors have been developed, the opportunity to use ML techniques at the strategy level is particularly exciting. For example, Balch uses a behavioral diversity measure to encourage role learning in a RL framework, finding that providing a uniform reinforcement to the entire team is more effective than providing local reinforcements to individual players [6]. Luke et al. use genetic programming to evolve cooperative behaviors within a team of players [50].

Peter Stone
Wed Sep 24 11:54:14 EDT 1997