State generalization method based on uncertainty minimization of behavior outcomes for reactive agents

Abstract

Autonomous state generalization is one of the key issues in the behavior acquisition problem of reactive agents. The paper proposes a novel state generalization method which discretizes the continuous sensor space based on entropy minimization of agents' behavior outcomes. This general framework unifies the heuristic criteria for state generalization used in conventional works. An experimental study in the latter part suggests that our method increases the adaptability of agents to the environment and improves the overall behavior performance.