Bayesian imitation learning in game characters

Abstract

Imitation learning is a powerful mechanism applied by primates and humans. It allows for a straightforward acquisition of behaviours that, through observation, are known to solve everyday tasks. Recently, a Bayesian formulation has been proposed that provides a mathematical model of imitation learning. In this paper, we apply this framework to the problem of programming believable computer games characters. We will present experiments in imitation learning from the network traffic of multi-player online games. Our results underline that this indeed produces agents that behave more human-like than characters controlled by common game AI techniques.