As the complexity of modern games increases, there has been a movement away from traditional artificial intelligence (AI) technologies as developers search for AI that is more scalable, flexible and provides diverse behaviour in non-player characters (NPCs). Recent innovations in the field of AI Planning have seen the first Goal Oriented Action Planner (GOAP) developed for a commercial game. GOAP has been hailed as the next step in evolving AI in games and can offer several advantages over traditional AI techniques. This dissertation investigated, analysed and compared a traditional AI technique, finite state machines, with Goal Oriented Action Planning when applied in games to discover whether GOAP is indeed an improvement over FSMs and what the relative merits of each system are.
As part of the dissertation, a GOAP and FSM system was developed and placed in a game scenario where experiments were carried out between the two. Comparisons were made on three levels, the first was based on the results of the games i.e. how the systems performed against one another in the Domination, Capture the Flag, Deathmatch and Last Man Standing game modes. The second comparison examined the two systems from a technical perspective and investigated how each system performed from a memory management, CPU usage and efficiency standpoint. The final comparison considers the merits of the two systems with respect to ease of management, flexibility and re-usability.
The results of the experiments clearly highlighted that GOAP is a superior AI system for many reasons however there are certain drawbacks associated with it which may require some consideration before choosing GOAP as a primary AI technique for a commercial game title.
No other previous piece of work has taken a GOAP system and compared it with another AI technique so the findings of this dissertation add to the limited pool of knowledge associated with AI planning in games.