Modeling and Transformation for Non-Player Characters

This master thesis explores the possibility of modeling and transformation of AI (Artificial Intelligence) for non-player characters. Non-player characters are people or objects in a virtual world that are not directly controlled by the user. Currently, the behavior of non-player characters (NPCs from here on) is described textually, this makes AI for a particular NPC difficult to re-use for another NPC whose behavior might be only slightly different. Imagine now, that this behavior is described by a model, consisting of different sub-models for different aspects of the NPC's thought pattern. It should then be easy enough to swap a particular sub-model for another one, resulting in different NPC behavior for minimal developer effort. Transformation might also be applied, by performing some random transformations on models new behaviour can emerge, which may or may not be desired, depending on the desired characteristics for the NPC. This enables developers to focus on implementing more complex and interesting AI, instead of fiddling with a large code base that gets increasingly difficult to maintain.

In order to explore and ultimately demonstrate this new approach to AI development, the thesis will consist of three main parts:

• Researching and implementing a framework to enable AI modeling and transformation in the statechart formalism.
• Research how transformations affect modeled AI and create a mechanism to test the quality (with resp. to some specified heuristic) of modeled and transformed AI
• Finally, use this work to implement a Proof Of Concept in a commercial Game Engine, showing that modeling (and transformation) of AI is feasible in commercial games.