抄録
We propose a reinforcement learning system that is constructed autonomously according to the state of environment observed during trials and the efficient method of selecting action whose direction varies. Effectiveness is shown through mobile robot control.
