Machine learning techniques for FPS in Q3

Abstract

This paper presents a First Person Shooter Artificial Intelligence system that makes use of machine learning capabilities to achieve more human-like behavior and strategies. The AI is trained with a supervised learning paradigm using example recorded during the observation of expert human players. The Machine Learning section of the AI is based on various Feed Forward Multi-layer Neural Networks trained by Genetic Algorithms. The AI system is developed and tested in the Quake 3 Arena game engine. The system is able to learn certain behaviors but still lack on some others. The results are evaluated and possible improvements are proposed.