A Privacy-Preserving Extended Reality Framework for Immersive Learning and Collaborative Virtual Environments

Abstract

Extended reality (XR) platforms can support situated instruction, embodied practice, and synchronous collaboration, but continuous collection of gaze, motion, voice, hand pose, and spatial mapping information creates substantial privacy risks. This paper proposes a privacy-preserving extended reality framework (PPXR) combining purpose-limited sensing, on-device representation learning, federated optimization, differential privacy, secure aggregation, attribute-based authorization, and auditable collaborative session management. A threat model distinguishes curious service operators, malicious participants, compromised endpoints, and inference attacks on shared updates. The framework optimizes pedagogical utility, interactive latency, communication overhead, and privacy exposure subject to access and safety constraints. A detailed pipeline and mathematical formulation describe data minimization, noisy aggregation, trust-aware access, and adaptive quality selection. Because no original user study or deployed system measurements were supplied, numerical results are explicitly labeled as illustrative scenario calculations, not experimental observations. Under transparent assumed configurations, the privacy-first design reduces modeled centralized raw-data retention and improves a normalized privacy-risk score while imposing latency and accuracy trade-offs. The paper concludes with a falsifiable experimental protocol, limitations, and deployment considerations for responsible educational XR.