Adaptive safety bubbles in mixed reality: optical feedback with machine learning optimisation

This study presents an adaptive safety bubble system in mixed reality (MR), using optical feedback and machine learning (ML) to enhance user safety in social interactions. The system dynamically adjusts visual cues, including colour and transparency, based on real-time proximity data, enabling more intuitive personal space boundaries. To improve feedback accuracy, a supervised ML classifier is trained on the number of avatars, stopping distance, and avatar velocity. The trained model predicts feedback states in real time, generating responsive visual indicators that help users maintain a comfortable social experience. This system provides a fundamental framework for socially intelligent MR interfaces, aiming to mitigate harassment risks and enhance social comfort through adaptive feedback.