Integrating agent actions with genetic action sequence method

Man-Je Kim, Jun Suk Kim, Donghyeon Lee, Sungjin James Kim, Min-Jung Kim, Chang Wook Ahn

Abstract: Reinforcement learning in general is suitable for putting actions in a specific order within a short sequence, but in the long run its greedy nature leads to eventual incompetence. This paper presents a brief description and implementative analysis of Action Sequence which was designed to deal with such a "penny-wise and pound-foolish" problem. Based on a combination of genetic operations and Monte-Carlo tree search, our proposed method is expected to show improved computational efficiency especially on problems with high complexity in which situational difficulties are often troublesome to resolve with naive behaviors. We tested the method on a video game environment to validate its overall performance.