Building a believable and effective agent for a 3D boxing simulation game

This paper describes an approach used to build and optimize a practical AI solution for a 3D boxing simulation game. The two main features of the designed AI agent are believability (human-likeness of agent's behavior) and effectiveness (agent's capability to reach own goals). We show how learning by observation and case-based reasoning techniques are used to create believable behavior. Then we employ reinforcement learning to optimize agent's behavior, turning the agent into a strong opponent, acting in a commercial-level game environment. The used knowledge representation scheme supports high maintainability, important for game developers.