Imitation learning in uncertain environments

Abstract

This paper describes a new evolutionary learning method to handle the fast adaptation of a group of agents in an uncertain environment. The method is a result of our research towards the generation of intelligent agents in computer games and is inspired by the idea of social learning or cultural evolution. Thus, the agents try to adapt by the exchange of information about advantageous behaviours within the population. This paper evaluates the new approach by addressing the generation of competitive artificial players in a real-time action game. © 2008 Springer-Verlag Berlin Heidelberg.