Real-time Interaction in VR with a Distributed Multi-Agent System

We describe a Virtual Reality system that allows users at different locations to interact with a distributed multiagent system. We use RoboCup (robot soccer) as a case study. A human player who is immersed in a CAVE can interact with the RoboCup simulation in its natural domain, by playing along with a virtual soccer game. The system supports distributed collaboration by allowing humans at different geographic locations to participate and interact in real time. The most difficult problem we address is how to deal with the latency that is induced by the multi-agent simulation and by the wide-area network between different CAVEs. Our navigation software anticipates the movements of the human player and optimizes the interaction (navigation, kicking). Also, it sends a minimal amount of state information over the wide-area network.