Fuzzy Evaluation of Macroscopic Situation for Turn Based Strategic Games

摘要

Turn-based Stratege Game(TSG) has not been studied as Go and Shogi, because of huge game space computer AIs of TSG are usually weaker than human players. There are so many possible moves in one turn because of many combinations of order to move units and each actions. This study proposes M-UCT that the game tree is comprised of nodes of an action of the unit and F-UCT that searches using the perspective evaluation function by fuzzy sets based on the intuitive knowledge of human players. Intuitive knowledge is based on 1000 "good" situations evaluated by human players and generate fuzzy sets. In Monte-Carlo Tree Search, nodes are given evaluation value depending on menbership value. We examine the effectiveness of M-UCT and F-UCT by play experiments.