Generation of rule-based adaptive strategies for games

Abstract

Real Time Strategy Games (RTSG) are a strong test bed for AI research, particularly on the subject of Unsupervised Learning. They offer a challenging, dynamic environment with complex problems that often have no perfect solutions. Learning Classifier Systems are rule-based machine learning techniques that may rely on Genetic Algorithms to discover a knowledge map used to classify an input space into a set of actions. This work focuses on the use of Accuracy-based Learning Classifier System (XCS) as the learning mechanism for generating adaptive strategies in a Strategy Game. The performance and adaptability of the strategies and tactics developed with the XCS is analyzed by facing these against scripted opponents on an open source game called Wargus. Our results indicate that Genetic Algorithms can be used effectively to enhance Real Time Strategy Games.