Applying Fuzzy Control in Fighting Game AI
○朱 晉賢,ラック ターウォンマット(立命館大)
Past researches in game AI have been focusing on algorithms for turn-based games, such as chess and poker, while application of AI in real-time, fast-pace video games is an emergent research field. To facilitate research of AI for real-time video game, our lab had developed a fighting game platform named FightingICE, based on which international competitions were held. Past research using FightingICE has proven that kNN method is effective in predicting opponent's action in fighting games, notwithstanding, AI relying on kNN to predict opponent's action suffers a disadvantage at the beginning of the game, when there is not enough opponent action data for prediction. To tackle this problem, this paper introduces fuzzy control, and experiments have shown that combining fuzzy control with kNN method is able to improve the performance of existing fighting game AI.

footer 情報処理学会 セキュリティ プライバシーポリシー 倫理綱領 著作権について