A machine learning-driven adaptive adjustment framework for real-time cybersickness mitigation in virtual reality
Cybersickness remains one of the most persistent barriers to the widespread adoption of Virtual reality (VR), often causing nausea, dizziness, and visual discomfort that limit user engagement. This study proposes a machine learning-based adaptive adjustment framework that predicts and mitigates cybersickness in real time using only non-intrusive head-tracking data (HTD). The system dynamically adjusts the field of view (FoV) and fixed foveated rendering (FFR) parameters based on predicted discomfort levels, striking a balance between visual quality, system performance, and user comfort. A within-subjects experiment with 28 participants compared a baseline VR configuration with the adaptive system. A Gradient Boosting regressor achieved an average normalized prediction error of 1.84% on a six-participant held-out set, supporting preliminary feasibility of estimating cybersickness scores from kinematic features. With the adaptive module active, mean VRSQ scores were 27% lower and perceived cognitive workload 28% lower (both statistically significant), though the fixed, non-counterbalanced condition order means order or habituation effects may contribute. Average frame rates decreased modestly, indicating a trade-off between rendering performance and comfort. Overall, the results support real-time, sensor-driven adaptation as a lightweight approach for improving comfort in consumer VR systems without additional physiological sensors, pending counterbalanced, participant-disjoint validation.
Authors
- Hassan Reza (ORCID: https://orcid.org/0000-0002-7593-9624)
- Ananth N. Ramaseri-Chandra (ORCID: https://orcid.org/0000-0002-4753-5964)
Institutions
- University of North Dakota (US)
- Turtle Mountain Community College (US)
Publication Details
- Journal
- Virtual Reality
- Published
- 2026-11-15
- DOI
- https://doi.org/10.1007/s10055-026-01524-x
- Citations
- 1
- Primary Topic
- Virtual Reality Applications and Impacts
- Type
- article
- Field-Weighted Citation Impact
- 2.72