Title: Policy Gradient Method Using Fuzzy Controller in Policies and Its Application

Year of Publication: Nov - 2014
Page Numbers: 167-174
Authors: Noor Imanina N.H. , Harukazu Igarashi
Conference Name: The International Conference on Artificial Intelligence and Pattern Recognition (AIPR2014)
- Malaysia


One of the reinforcement learning algorithms proposed by Igarashi and Ishihara is a combining method of policy gradient method and fuzzy control. In 2012, M. Sugimoto implemented the algorithm to the RoboCup Small Size League action decision system. The system received 30 scenes, taken from RoboCup Japan Open 2012 Competition to be learned. The purpose of this paper is to present the detailed analysis on the fuzzy rules in the policies taken from the system in order to find out the cause of the failure in the learning of 5 of the scenes received. A method was proposed to determine the rules that caused error in the learning of 5 scenes by evaluating the degree of contribution and divergence of each rule.