Reinforcement learning(RL) attracts attention as a technique of realizing autonomous systems. However, in general, it is not easy to put RL into practical use. This difficulties include the problem of designing a reasonable state space of an agent contains two requirements in trade-off: to reduce the search space as much as possible in order to make a learning process be fast and to keep the characteristics of the search space as much as possible in order to seek better strategies. In this paper, it is considered to satisfy the above requirements. We newly introduced a concept of a "state space filtering". Then, we proposed an acquisition method which is to adjust the search space adaptively by referring to an entropy. The validity and the potential of the proposed method were comfirmed by computational experiments.