In-game tactic adaptation for interactive computer games

Computer controlled characters in games typically lack the ability to adapt to situations that were not envisaged by the game artificial intelligence (AI) programmer. Consequently game characters tend to be lifeless and susceptible to repetitive and predictable behaviour. The focus of this work is to provide practical approaches to in-game learning and thus produce NPCs that are more believable and responsive to the human player. In this paper, we present an approach to online tactic learning and adaptation for autonomous agents in virtual environments. The design of the tactic adaptation architecture was influenced by commercial and theoretical approaches to the organisation and control of NPC behaviour in FPS games. The system is applied to Unreal Tournament using the GameBots API and evaluated in FPS game scenarios. The results show that the architecture is capable of successfully adapting NPC tactic selection and illustrate how a multi instance approach to organising tactic libraries can enhance adaptation.