Thesis Description & Objectives
Socially immersive Virtual Reality environments allow users to interact through avatars, yet current systems often animate only the head and hands, while the rest of the body relies on basic Inverse Kinematics (IK). This can result in unnatural gestures and physical interactions, reducing social presence, engagement, and the overall quality of multi-user experiences.
This thesis aims to enhance avatar-based social interaction animation by making movements more realistic, expressive, and responsive to interpersonal exchanges. The goal is to support natural social gestures such as handshakes, high-fives, hugs, or other forms of embodied communication, improving the sense of presence and fostering richer collaborative and social experiences in VR.
Machine Learning techniques may be used to refine avatar animations beyond standard IK solutions, but the primary focus is on the quality and naturalness of social interactions, rather than on the algorithms themselves. Other key aspects include adaptive management of avatar morphology, real-time collision handling, and the development of a multi-user VR system optimized for seamless, context-aware social interactions.
The expected outcome is a more immersive and socially engaging VR experience, where users can interact naturally through avatars that behave intuitively and expressively, enhancing both social presence and collaboration without compromising system performance.
References:
1. Visconti, Alessandro; Macaluso, Roberta; Di Bartolomei, Gabriele; Calandra, Davide; … (2026)
Improving fidelity of close social interaction animations in social VR with a machine learning-based refinement framework. In: 38th International Conference on Computer Animation and Social Agents (CASA 2025), https://dx.doi.org/10.1007/978-981-95-0100-7_13
