抄録

F-013
オリジナルゲームの強化学習型CPU

In this research, I make an original game and propose a reinforcement learning based CPU for a player in the original game. The Monte Carlo method was used as a structure of the reinforcement learning and a filtering mechanism was applied to solve huge state space. The reinforcement learning based CPU player fights against a rule base CPU player in the original game. As a result of 2,000 times of play, the winning rate of the reinforcement learning based CPU reached an average of 59%.