rovenr12 / td_mcts Goto Github PK
View Code? Open in Web Editor NEWMonte Carlo Tree Search (MCTS) is a popular technology in the field of real-time games and has good performance comparing with the other agents in pommerman. Recently, few studies introduce using the conceptsin reinforcement learning to enhance the performance in MCTS. By combining the concept of temporal difference (TD) in the tree policy, it gives a better result in the twoplayer game. This report implements the TD-UCT in Pommerman, being multiply agents’ game, and shows it does not have significant improvement. This report also enhances the opponent model and shown the opponent model can increase the performance.