Behavior Strategy Acquisition of Soccer Robot Used Fuzzy State Division-type Shaping Reinforcement Learning

Abstract

In this laboratory, we have already proposed the fuzzy state division type reinforcement learning to acquire high-level behavior selection rules of the hierarchical fuzzy behavior control by reinforcement learning automatically. In this research, we propose a method which we make learn to the robot efficiently by giving appropriate rewards by human. Furthermore, we define the weighted value of attack and defense for the robot by using the fuzzy rule and incorporate the method to let it learn the behavior strategy according to the situation. We performed the comparison experiment with the conventional reinforcement learning method by the soccer robot simulation.