Harnessing Foveated Rendering and AI to Tackle VR Cybersickness: A Feature-Centric Perspective

Abstract

As virtual reality becomes increasingly immersive, issues related to cybersickness pose a major challenge. This review investigates how foveated rendering techniques, powered by artificial intelligence, are transforming our response to this topic. We analyze the primary factors that lead to cybersickness, including latency, field of view, vergence-accommodation mismatch, and unnatural locomotion, while demonstrating how adaptive visual strategies can significantly alleviate user discomfort. By considering individual traits like age, previous virtual reality experience, and real-time physiological indicators, including heart rate and skin conductance, modern rendering systems are evolving to be more intelligent and user-specific. We emphasize the role of advanced machine learning models, from interpretable symbolic frameworks to deep neural networks, along with gaze prediction systems that enable real-time adjustments through predictive rendering and user-context-specific optimization. Our findings highlight the promise of closed-loop rendering systems, which aim to preserve visual fidelity while enhancing comfort and engagement, steering to toward safer, more personalized virtual reality experiences.