H. Iwasaki, H. Ohki, and N. Sueda (Japan)
reinforcement learning, self-organizing map, software agent, robot, planning
A state space construction is a very important problem in the application of reinforcement learning to real tasks. A system for state space construction with self-organizing map is proposed in this paper. In this system, an agent con structs state space from its own experience autonomously. In the experimentations, the system verifies the agent’s ability to construct a suitable state space from any unknown situation. Subsequently, it can improve ability and steadi ness of learning, and robustness to noise. Furthermore, it is capable of reconstructing the state space to fit any change most in the environment.
Important Links:
Go Back
