Decision networks for integrating the behaviors of virtual agents and avatars

We examine the use of decision networks in animating virtual agents. We have developed a system that allows the realization of multiple, parallel behaviors for an agent. The networks we utilize, called PaT-Nets, are used both to represent individual behaviors and also to encode rules of engagement between agents. The multiple networks simultaneously attached to an individual agent are used to control locomotion, planning, visual attention and decision-making strategy. We discuss how human players may be substituted for autonomous players and still operate under the represented behaviors in the PaT-Nets.