Predicting Motion Sickness Caused by Mismatched VR Visual Motion and Physical Rotation Using Resting-State EEG and EEGPT
Abstract
This study presents a new analysis of existing resting-state EEG datasets collected under Type I motion sickness, where visual motion conflicts with physical rotation. Unlike prior studies that focused on task-evoked responses or restricted EEG features to vestibular regions, we extend the feature space to include midline cognitive electrodes and apply the pre-trained newly-proposed EEGPT model to evaluate the interpretability of its outputs for predicting motion sickness susceptibility. The novelty of this work lies not in the model itself, but in demonstrating how its feature-level explanations can validate known mechanisms while revealing new insights into Type I motion sickness. Our results show that EEGPT confirms established associations by identifying Beta oscillations as key predictors of overall severity, reinforcing their role as biomarkers of visual-vestibular conflict. Also, EEGPT highlights novel contributions: domain-wise analysis shows that the posterior midline cognitive (Pz) and right PIVC (P4) are the most informative regions, while frequency-symptom mapping reveals that Beta oscillations at Pz provide the strongest prediction of oculomotor disturbances, yielding the highest explained variance and lowest error across all single-channel analyses. Together, these findings indicate that interpretable deep learning applied to resting-state EEG can both strengthen prior evidence and unveil new frequency- and domain-specific biomarkers, advancing the understanding of neural mechanisms underlying Type I motion sickness and paving the way for pre-exposure risk prediction and preventive countermeasures.
Authors: Haerin Byeon, Gang Li, Frank Earl Pollick, Sung-Bae Cho
Published in: IEEE International Conference on Artificial Intelligence and eXtended and Virtual Reality (AIxVR) (2026)