An architecture for integrating plan-based behavior generation with interactive game environments

One central challenge for game developers is the need to create compelling behavior for a game's characters and objects. Most approaches to behavior generation have either used scripting or finite-state approaches. Both of these approaches are restricted the requirement to anticipate game state at design time. Research in artificial intelligence has developed a number of techniques for automatic plan generation that create novel action sequences that are highly sensitive to run-time context. In this paper, An architecture for integrating plan-based behavior generation with interactive game environments, we describe an architecture called Mimesis, designed to integrate a range of intelligent components with conventional game engines. The architecture is designed to bridge the gap between game engine design and development and much of the work in artificial intelligence that focuses on the automatic creation of novel and effective action sequences. Users of the system construct two parallel models of the game world, one using extensions to existing game engine code, the other using techniques for explicit modeling of actions in terms of their requirements for execution and their effects on the game world. When integrated with a game engine, Mimesis acts as a run-time behavior generator, responsible for both generating plans – coherent action sequences that achieve a specific set of in-game goals – and maintaining the coherence of those plans as they execute in the face of unanticipated user activity. In this paper, we describe the architecture, its main components and the APIs available for integrating it into existing or new game engines.