Learning Human Behavior from Analyzing Activities in Virtual Environments

Abstract

Present day multiplayer video games offer an interesting perspective for researching artificial cognitive systems. In this contribution, we focus on the problem of learning believable behavior models for artificial characters. Recordings of the network traffic of modern games allow for applying machine learning techniques to realize artificial agents that act more human-like than conventional current game characters. We detail an imitation learning approach and present the results of an extensive believability study that was carried out on the Internet.