多重強化学習法に基づく群ロボットシステムにおける環境変化認識

概要

論文の詳細を見る
Reinforcement Learning has been used as one of methods for decision-making of autonomous robots. However, the algorithm for learning concludes a parameter, and the reaction speed to an unknown environment depends on value of the parameter. Because of the dependence, a robot with Reinforcement Learning becomes often late for adapting to environmental change. In this paper, we propose a method to reduce the adaptation delay. Our idea is to make it possible for the robots to know if environmental change happened. We found that if the robot uses two kinds of Reinforcement Learnings with different parameters simultaneously, the correlation between the Reinforcement Values becomes lower by environmental change. Thus, we can use the correlation as an index showing environmental change. We try to make a learning process after environmental change better for adaptation speed by recognizing an environmental change. The effect is proved through computer simulation.
2000-03-25