CyberGATE: Multiscale State-Space Fusion for Ocular and Vestibular Cybersickness Forecasting
J. N. Zhang · Advanced Electromagnetics · 2026
Cybersickness is still a major usability and safety issue for virtual reality because discomfort may emerge gradually during exposure or appear suddenly when scene motion changes. Questionnaire-based evaluation is difficult to use as a continuous monitoring tool. This study presents CyberGATE, a multimodal time-series framework that estimates cybersickness from eye-tracking and head-motion signals collected by an HMD. The framework separates long-context representation and short-event modeling into two coordinated branches. The global branch uses multi-view embedding, Mamba sequence modeling, and KAN-based sparse experts to learn persistent temporal patterns, whereas the local branch applies decomposition-aware patch encoding and window attention to detect transient disturbances. Experiments on two public datasets show that CyberGATE reduces prediction error compared with recurrent, Transformer, and existing multimodal baselines. Component removal and sensitivity analyses further confirm that global dynamics, local responses, and adaptive routing all improve performance. These results indicate that ordinary HMD behavioral signals can support objective, real-time, and individualized cybersickness warning in practical VR applications.