A Fighting Game AI with Evolutionary Strategy and Imitation Learning in Opportunity Maximization and Sensible Maneuvering Tactic
○陸 霏羽,チョイ チーケン,ラック ターウォンマット(立命館大)
In fighting games, human players usually prefer to play with other human players, the NPC controller used in fighting games utilize pre-scripted Artificial Intelligence (AI). This paper discusses simulating a part of the human player's thinking process to create a sensible human-like AI with an evolutionary strategy and imitation learning as sub-process in the fighting game "Fighting ICE". Based on simple cognitive process of human thinking when encountering opportunities, used evolutionary strategy for near-optimization in selecting a set of actions to maximize the situation's outcome. In common game terms, the set of actions is referred to as a combo. The algorithm generates unique sets of combos based on basic knowledge and experiences throughout the game.

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