Building a Believable Agent for a 3D Boxing Simulation Game

This paper describes an approach used to build a practical AI solution for a 3D boxing simulation game. The features of the designed AI agent are based on our deliberate concentration on believability, i.e. human-likeness of agent's behavior. We show how learning by observation and case-based reasoning techniques can be used to create an AI decision-making system for an industrial-level computer game. The chosen AI design principles support high usability and maintainability, which is important for game developers. We prove experimentally that our AI system provides both believable and effective behavior.