Improving Adaptive Game Ai with Evolutionary Learning

Game AI is defined as the decision-making process of computercontrolled opponents in computer games. Adaptive game AI can improve the entertainment provided by computer games, by allowing the computer-controlled opponents to fix automatically weaknesses in the game AI, and to respond to changes in humanplayer tactics online, i.e., during gameplay. Successful adaptive game AI is based invariably on domain knowledge of the game it is used in. Dynamic scripting is an algorithm that implements adaptive game AI. The domain knowledge used by dynamic scripting is stored in a rulebase with manually designed rules. In this paper we propose the use of an offline evolutionary algorithm to enhance the performance of adaptive game AI, by evolving new domain knowledge. We empirically validate our proposal, using dynamic scripting as adaptive game AI in a real-time-strategy (RTS) game, in three steps: (1) we implement and test dynamic scripting in an RTS game; (2) we use an offline evolutionary algorithm to evolve new tactics that are able to deal with optimised tactics, which dynamic scripting cannot defeat using its original rulebase; (3) we translate the evolved tactics to rules in the rulebase, and test dynamic scripting with the revised rulebase. The empirical validation shows that the revised rulebase yields a significantly improved performance of dynamic scripting compared to the original rulebase. We therefore conclude that offline evolutionary learning can be used to improve the performance of