Adaptive Behavior Acquisition based on Modular Learning System and Scheduling in Multi-Agent Environment

The existing reinforcement learning approaches have been suering from the policy alternation of others in multiagent dynamic environments such as RoboCup competitions since other agent behaviors may cause sudden changes in state transition probabilities of which con-stancy is needed for the learning to converge. A modular learning approach would be able to solve this problem if we can assign each mod-ule to one situation in which the module can regard the state transition probabilities as con-stant. This paper presents a method of mod-ular learning in a multiagent environment, by which the learning agent can adapt its behav-iors to the situations as consequences of the other agent's behaviors. Scheduling for learn-ing is introduced to avoid the complexity in au-tonomous situation assignment.