Abstract

This thesis concerns our innovation in game AI techniques, mainly game agents' modeling, planning and learning. The research topic involves the development of a game design software --- Gameme. Our work mainly focus on the development of the core AI module.

In this thesis, after discussing the system design of Gameme, we explain our contributions in two parts: off-line design and real-time processing. In off-line design, we present goal-oriented behaviour design and related modeling methodology for game agents. The goal-oriented design provides not only an intuitive behaviour design methodology for non-professional game designers but also efficient support for real-time behaviour control. In particular, the goal-oriented design can be used in modeling agents in different games.

The real-time processing component includes planning and learning mechanisms for game agents. These mechanisms are placed in a layered architecture. Basically, a procedural planning mechanism allows game agents to have the ability of fast reaction to their environment. Then, the creative transfer and adaptive learning mechanism trains game agents to learn from their experience and cooperate in teamwork. Furthermore, the unique emergent learning mechanism can allow game agents to have the ability to analyze different PCs' behaviour patterns and to find the suitable strategy to defeat PCs in real-time.

Most of the experiments in this thesis are performed in fighting scenarios. We connected the core AI module with a 3D graphics engine in order to have visual testing results. All test cases show that our goal-oriented behaviour design along with planning and learning mechanisms can provide fast, autonomous, collaborative and adaptive behaviour instructions for game agent in real-time game play.