Comparison of Fuzzy Reinforcement Learning in Pursuit Problem of Real Number Environment

DOI
  • Umano Motohide
    Department of Mathematics and Information Sciences, Graduate School of Science, Osaka Prefecture University
  • Fujii Yasuaki
    Department of Mathematics and Information Sciences, Graduate School of Science, Osaka Prefecture University
  • Hosoya Yuu
    Department of Mathematics and Information Sciences, Graduate School of Science, Osaka Prefecture University
  • Seta Kazuhisa
    Department of Mathematics and Information Sciences, Graduate School of Science, Osaka Prefecture University

Bibliographic Information

Other Title
  • 実数値環境の追跡問題におけるファジィ強化学習の比較

Abstract

A pursuit problem is a multi-agents' benchmark problem, where four hunters pursue and capture the prey in a grid environment. We have extended a grid environment to a real number one and applied fuzzy Q-learning with the state of fuzzy sets of distance and direction to the prey. In fuzzy Q-learning, hunters learn state-action values of the previous one action. In this research, we applies fuzzy Profit Sharing, fuzzy Q(lambda) and two kinds of extended Q(lambda) to the pursuit problem of the real number environment. In these four methods, hunters learn state-action values of the actions already executed. Simulation results shows that fuzzy Q(lambda) and one kind of extended fuzzy Q(lambda) have the best performance.

Journal

Keywords

Details 詳細情報について

  • CRID
    1390282680649809792
  • NII Article ID
    130004591775
  • DOI
    10.14864/fss.27.0.18.0
  • Text Lang
    ja
  • Data Source
    • JaLC
    • CiNii Articles
  • Abstract License Flag
    Disallowed

Report a problem

Back to top