Deep Reinforcement Learning Using Parallel Actors and Prioritized Experience Replay for Puyo Puyo AI

Abstract

“Puyo Puyo AI” includes rule-based methods and those using correlation matrices, but both are inferior to human players. Additionally, “Puyo Puyo AI” using deep reinforcement learning have shown insufficient performance. This study focuses on improving the performance of “Puyo Puyo AI” using deep reinforcement learning and proposes a method employing parallel actors and prioritized experience replay. Experimental results demonstrate that the proposed method achieved an average maximum chain length of 6.243 and an average score of 33,114, surpassing previous studies utilizing deep reinforcement learning.