Modeling accessibility of embodied agents for multi-modal dialogue in complex virtual worlds

Virtual humans are an important part of immersive virtual worlds, where they interact with human users in the roles of mentors, guides, teammates, companions or adversaries. A good dialogue model is essential for achieving realistic interaction between humans and agents. Any such model requires modeling accessibility of individuals, so that agents know which individuals are accessible for communication, by what modality (e.g. speech, gestures) and at what degree they can see or hear each other. This work presents a computational model of accessibility that is domain independent and capable of handling multiple individuals inhabiting a complex virtual world.