An Entropy-Guided Adaptive Co-construction Method of State and Action Spaces in Reinforcement Learning

Engineers and researchers are paying more attention to reinforcement learning (RL) as a key technique for realizing computational intelligence such as adaptive and autonomous decentralized systems. In general, it is not easy to put RL into practical use. In previous research, Nagayoshi et al. have proposed an adaptive co-construction method of state and action spaces. However, the co-construction method needs two parameters for sufficiency of the number of learning opportunities. These parameters are difficult to set. In this paper, first we propose an entropy-guided adaptive co-construction method with and index using the entropy instead of the parameters for sufficiency of the number of learning opportunities. Then, the performance of the proposed method and the efficiency of interactions between state and action spaces were confirmed through computational experiments.