PAVR: Personalized Adaptive Virtual Reality for Early Cybersickness Prevention with Minimal Experience Disruption

Abstract

Cybersickness remains a practical barrier to sustained virtual-reality use because mitigation often begins only after discomfort is reported, while aggressive countermeasures can themselves degrade presence and task performance. This paper presents PAVR, a closed-loop adaptive virtual-reality architecture that reframes mitigation as a constrained control problem: estimate an early-warning cybersickness-risk score from continuously available interaction telemetry, then apply the least disruptive intervention expected to lower that risk. The implemented system uses head angular velocity, head acceleration, translational velocity, controller motion, turning frequency, jerk, task errors, pauses, exposure duration, and frame-performance information to compute a personalized early-warning score. A minimum-cost controller selects among speed reduction, rotation reduction, dynamic field-of-view restriction, motion stabilization, teleportation, or no intervention, while explicitly penalizing presence cost, task cost, and intervention burden. The complete software includes deterministic simulation, three policy baselines, automated tests, a Streamlit research application, one-command reproduction, and continuous integration. In 240 controlled synthetic sessions of 120 steps each, PAVR reduced mean risk from 0.5072 under Fixed VR to 0.3379, eliminated the high-risk fraction observed in the baseline, and used 33.33% less intervention burden than an aggressive threshold-adaptive policy. These results establish software-level computational consistency only; human efficacy remains an empirical question. PAVR therefore provides a reproducible engineering foundation for future real-time, participant-level evaluation of minimally disruptive cybersickness prevention.