Decision Making AI for Games
This project aims to inspire a paradigm shift in the algorithms used to generate agent behaviour in commercial games. As game worlds grow in complexity and scale, the current reliance of the industry on designing agent behaviour by fixed rules will hit the same limits that expert systems hit in other AI applications. Most notably, it is infeasible to design rules that capture all situations an agent may end up in and need to act within. Search algorithms and machine learning can overcome these issues by autonomously deciding how to act. However, the common perception within the industry is that this approach makes the design process harder than deploying agents following fixed behaviours. This project will explore the causes of this negative perception, use this to inform the design of new algorithms for non-player character behaviour, and make these available to the industry.