Realizing Emotional Autonomous Virtual Agents in a Multi-user Virtual Environment

Abstract

Most of the Multi-user Virtual Environment Systems (MVE) today do not allow computer-controlled virtual agents to interact with real users in an emotional manner. In this paper, we aim to design autonomous virtual agents that can generate emotional behaviors through textual dialogs, route selection, and body gestures when interacting with other avatars controlled by real users or by the computer. These behaviors are expressed via interaction events delivered through the network. The interaction events are generated according to the mental state of the agent based on a psychological model including emotion, memory, and personal traits. We have also designed an update mechanism to interpret an interaction event and update the mental state of the virtual agent accordingly. This model has been implemented and tested in an experimental MVE platform based on VRML/X3D to evaluate the effectiveness of the design. With the mechanism of autonomous virtual agent proposed in this paper, we hope to enhance MVE with virtual agents possessing emotional behaviors that can be easily authored by regular users.