Minimax guided reinforcement learning for turn-based strategy games

Abstract

Games are a good medium for artificial intelligence (AI) research, since they compare user and machine behavior directly. AI in games is required to imitate human behavior. Though nowadays the use of rule base scripting and hard code behavior is still dominant in commercial game, the use of learning algorithm can be an alternative because of its ability to adapt to changes and to achieve substantial result at a moderate time. This paper investigates the use of learning which is derived from dynamic scripting to provide action in a turn-based strategy game. The algorithm is then combined with the Minimax algorithm to achieve a better performance. The performance of the proposed algorithm is evaluated through a series of matches against a static manually designed AI. The result shows that the proposed algorithm is able to adapt the static AI at a shallow Minimax depth. The algorithm also shows the ability reducing calculation time and using less memory space. It is concluded that Minimax guided reinforcement learning can be applied to the turn based strategy genre.