Edge-Cloud Collaborative Privacy-Preserving Motion-Based Authentication for XR Systems

Current extended reality (XR) systems predominantly rely on conventional authentication methods such as password entry or iris recognition. However, password-based login in immersive environments is cumbersome due to the difficulty of operating virtual keyboards, while iris recognition often fails for users wearing head-mounted displays or glasses. Inspired by the emerging use of companion devices for realtime authentication, we propose a novel edge-cloud collaborative motion-based authentication framework tailored for XR systems. In our approach, the edge device first employs a lightweight machine learning model to extract a user's action sequence in the form of SMPL-X parameters, which are then encoded into a compact motion-password representation. To ensure both privacy and robustness against motion variability, the encoded sequence is further protected using error correction coding (ECC), transforming it into a secure binary representation without exposing raw biometric features. Finally, the cloud server verifies the ECCprotected motion-password against stored references, enabling scalable, secure, and low-latency authentication for XR applications. We implement a prototype show superior performance to FMCode and MetaFL: 97.3% accuracy, 0.4% EER, stronger low-FPR behavior, and improved scalability. We also include a security analysis under practical threat models.