Building reactive characters for dynamic gaming environments

Interactive computer games are widely seen as a killer application domain for Artificial Intelligence (AI) [8]. Quite apart from the significant size of the games market in terms of revenue [3], computer games provide complex, dynamic, uncertain and competitive environments that are perfect for developing, testing and deploying AI technologies. While many researchers currently focus on enhancing games with sophisticated AI, most overlook the role that AI has to play in the development of the games themselves. In this paper we present an approach to building non-player characters (NPCs) for a well-known computer game, Unreal Tournament1. Specifically, we use decision trees induced from human player strategies to define how an NPC in the game performs in a highly dynamic environment. The benefits of this approach are twofold. Firstly, it provides a basis for building competitive AI-based NPCs for interactive computer games. Secondly, this approach eases the development overhead of such characters. Our empirical evaluation demonstrates that the NPCs we create are very competitive against hand-crafted ones in a number of simulated gaming sessions.