Hand2Any: Real-time Avatar Manipulation through Hand-to-Any Motion Mapping with Few-Shot User Adaptation

We propose user-agnostic and user-specific mapping techniques for avatar control using hand gestures in VR. The user-agnostic mapping method relies on common conventions, while the user-specific mapping adapts to individual user preferences through few-shot learning. Both methods utilize supervised learning with paired motion data to achieve accurate and intuitive avatar control. Our evaluation includes quantitative assessments demonstrating the effectiveness of user-specific mapping techniques with avatars of diverse joint structures. The results indicate that user-specific mapping provides finer control.