Collision Avoidance with a Moving Object Based on Q-Learning.
Masashi Furukawa, Michiko WATANABE, M. Ikeda, Masahiro Kinoshita, Yukinori Kakazu · TRANSACTIONS OF THE JAPAN SOCIETY OF MECHANICAL ENGINEERS Series C · 2003
Knowledge acquisition on AGV collision avoidance with other AGV has been investigated for realizing AGV autonomous driving by use of Q-learning. A proposed method adopts an indirect approach to solve collision avoidance problem. Namely, the knowledge acquisition method on AGV collision with other AGV is proposed by using Q-learning. Q-value in Q-learning plays a role in representing knowledge and it is obtained by trial and error. The obtained Q-value for collision is inversed and applied to collision avoidance. Through numerical experiments, it is verified that Q- value realizing collision knowledge is obtained and collision avoidance is realized by using inversed Q-value. At the same time, additional learning on Q-learning is also examined by comparing with simultaneous learning. As a result, it is found out that additional learning is as effective as simultaneous learning.