Punit Kunjam

I worked on a system comparing different hazard indication and avoidance methods for users immersed in a virtual environment. Specifically, I developed a novel system for collision avoidance with people who may or may not be engaged in VR. I implemented this system using a machine learning-based human pose estimation framework, which tracks users in the simulated environment and offers visual and auditory feedback following different hazard indication and avoidance techniques. I used the redirected walking technique, which manipulates the scene around the users, causing them to reorient themselves in the virtual world, thus avoiding the collision. Video shows an experiment where a VR user tends to collect a specific item in the scene meanwhile the non-VR user (animated model) deliberately gets in the way of the VR user. The scene compensates itself by rotating the world so that the VR user will reposition itself accordingly in the 3D space, driving both VR and non-VR users away from each other.