Real-Time Imitation Based Learning for Commercial Fighting Games

DOI: 10.5176/978-981-08-3190-5_301

Authors: Sarayut Lueangrueangroj, Vishnu Kotrajaras

Abstract:

Ghost AI is an Artificial Intelligence (AI) for fighting games which is capable of observing and imitating any player’s style of play. By using Ghost AI, a player can play the game as if playing against another player. Although Ghost AI can simulate a human player, current research only applies it offline.

Therefore, a human player can easily learn to beat Ghost AI during a match. We propose a methodology that allows Ghost AI to learn in real-time. The real-time data is also used to vary the frequency of each imitated action, so that the action can only be executed in situations which favor the AI. Our experimental result shows that the enhanced AI appears more human-like and more intelligent than the unmodified Ghost AI and provides a more satisfying experience for human players.