In this paper we show how symbolic Artificial Intelligence (AI) techniques
can be used to develop intelligent Virtual Reality (VR) environments. We ha
ve developed an expert system architecture for building intelligent agents
that respond to voice and gesture commands in virtual environments. To demo
nstrate the utility of this we present a simple application that allows a u
ser to drive a virtual robot (the VirBot) around a virtual environment with
multimodal commands. This software agent is able to accept spoken commands
from the user, interpret them using natural language understanding and int
eract with and modify the virtual environment. Command recognition is furth
er improved by the use of contextual information which reduces recognition
error rates by 31%. Context is encoded using scripts, making it easy to com
pose inferences about events for which there is incomplete or fuzzy informa
tion. (C) 1998 Elsevier Science Ltd. All rights reserved.