Advance Motion Acquisition of an Actual Robot by Reinforcement Learning using Reward Change
Ryota YAMASHINA, Haruhisa Motoyama, Mariko URAKAWA, Jian Huang, Tetsuro Yabuta · TRANSACTIONS OF THE JAPAN SOCIETY OF MECHANICAL ENGINEERS Series C · 2006
Generally, a reward is given as the fixed value for the Reinforcement Learning process. However, as for a human being's training process, the reward could be changed according to the improvement of the process. Therefore, it is very interesting to study the Q-Learning process under the reward change. In this paper, Q-Learning is applied to an actual robot in order for advance motion acquisition. Results show that the Q-Learning process changes according to the reward change. Moreover, this paper clarifies how the robot obtains its optimal motion form under the learning process.