Abstract
In recent years, the development of rescue robots has attracted attention to protect humans from the risk of the rescue work in disaster areas. Particularly, the development of the robots that can move through small space deeply is expected. We have paid our attention to peristalsis as a mechanism for such a robot, and developed a many segmental peristaltic crawling robot with a motor drive. Previously, by modeling a peristaltic crawling robot, we derived the movement pattern by using Q-learning, which is one of the reinforcement learning. Moreover, we were going to examine the movement pattern of the many segmental robots from improvement of fault tolerance and propulsive force. However, Q-learning was not able to be executed due to the insufficient memory when the number of the segments of the robot increases. In this paper, we examined the algorithm that can execute with low memory, and decided to use Actor-Critic. Moreover, we derived the movement pattern of a six segmental peristaltic crawling robot, and compared it with the conventional movement pattern using a regressive wave.
