Learning architecture for real robotic systems-extension of connectionist Q-learning for continuous robot control domain

This paper describes the overall architecture for complex motion learning of practical robotic systems and then proposes a method to extend reinforcement learning to the domain of continuous robot control problem in order to apply it to behavior learning of practical robotic systems. To represent continuous control variables, CMAC is employed for utility networks, and to fully utilize experiences, experience sequences are stored and replayed with priorities. As a testbed, the learning system is applied in simulation to the control of swing amplitude of a two-link brachiation robot which is hardly constrained with dynamics