PUB - Publikationen an der Universität Bielefeld

Modern FPS games do not only demand fast reactions from the participating human players but also require them to think tactically and strategically. While artifical players usually surpass their human counterparts in reaction time and accuracy, they are hopelessly lost if it comes to tactical and strategical gameplay. In this paper, we propose Bayesian imitation learning in conjunction with tactical waypoint maps as a means to learn tactical behaviors suited for "1 vs. 1" game situations. The empirical basis for our approach is given by the data contained in recordings of matches between human players. Experimental results underline the usability of tactical waypointmaps and bayesian imitation learning for the acquisition and generation of new behaviors in artificial players.