物体識別のためのロボットの行動学習の高速化

抄録

We have proposed active perception system which can learn an exploratory behavior using reinforcement learning．This system acquires an effective behavior which extracts features from objects．However，the learning method of the system requires the long time to complete learning． In this study，we improve previous method．We propose new learning method that can learn more efficient than the previous one．We apply our new method to a mobile robot simulation to observe its effectiveness．Results indicate that the new method accelerates the learning．