Siti Sendari, Shingo Mabu, Kotaro Hirasawa
This paper proposes Fuzzy Genetic Network Programming with Reinforcement Learning (Fuzzy GNP-RL). This method integrates fuzzy logic to the conventional GNP-RL. The new part of the proposed method is fuzzy judgment nodes. Fuzzy GNP-RL provides flexibility to determine the appropriate next node by the probabilistic transition instead of that by the threshold values on GNP-RL. The simulation of the wall following behavior of a Khepera robot is used to evaluate the performance of Fuzzy GNP-RL compared with that of GNP-RL. The result shows that Fuzzy GNP-RL is more robust than GNP-RL. © 2011 IEEE.
Graduate School of Information, Production and Systems, Waseda University, Wakamatsu-ku, Kitakyushu, Fukuoka 808-0135, Hibikino 2-7, Japan; Dept. of Electrical Engineering, Faculty of Engineering, State University of Malang, Malang 65145, Jl Semarang 5, Indonesia
