Adaptive Virtual Reality Museum: A Closed-Loop Framewor for Engagement-Aware Cultural Heritage

A team of researchers has introduced a novel 'Adaptive Virtual Reality Museum' framework, published on arXiv, that tackles a core problem in digital cultural heritage: static content that often bores or overwhelms visitors. The system creates a closed-loop where it continuously monitors a user's implicit behavior—specifically gaze dwell time, head kinematics, and locomotion—through sensors on consumer VR hardware. A transparent, rule-based classifier analyzes this data in under a millisecond to infer real-time engagement levels.

This engagement signal is then fed to a Large Language Model (LLM), which acts as a dynamic content engine. Instead of presenting a fixed script, the LLM modulates the complexity and depth of exhibit explanations on the fly, tailoring the information to the visitor's apparent interest without interrupting their exploration. A proof-of-concept deployed at the Berat Ethnographic Museum with 16 participants showed compelling results: those using the adaptive system demonstrated 2-3x increases in reading engagement and total exploration time compared to a static VR tour, while maintaining high usability with a SUS score of 84.3.

The work, presented as a discussion-sparking prototype, highlights the potential of implicit, AI-driven personalization in immersive education. However, the authors also note it raises important questions for future large-scale study, including how to best validate engagement metrics, ensure AI transparency to users, and thoughtfully integrate generative models into sensitive cultural contexts.