A State Space Filter for Reinforcement Learning - Concept and a Design -
Masato Nagayoshi, Hajime Murao, Hisashi Tamaki · IEEJ Transactions on Electronics Information and Systems · 2006
Reinforcement Learning (RL) attracts much attention as a technique of realizing computational intelligence such as adaptive and autonomous decentralized systems. However, in general, it is not easy to put RL into practical use. This difficulty includes a problem of designing a reasonable state space of an agent, i.e., satisfying two requirements in trade-off: to reduce the search space for making a learning process be fast and to keep the characteristics of the search space for seeking better strategies. In this paper, in order to overcome the above difficulty, we propse a concept of a “state space filtering", and a method to adjust the search space adaptively by referring to an entropy. Then, through computational experiments by using a robot navigation problem with continuous state space, the validity and the potential of the proposed method have been comfirmed.