Predicting Cybersickness Trend and Extent Based on FMS Labeled Dataset.

Area of Science:

• Virtual Reality
• Human-Computer Interaction
• Neuroscience

Background:

• Cybersickness impedes virtual reality (VR) adoption.
• Existing predictive models often use post-experience labels and physiological data, limiting real-time application.
• Dynamic nature of cybersickness necessitates continuous monitoring and prediction.

Purpose of the Study:

• To develop a practical dataset for real-time cybersickness prediction.
• To enable reliable prediction models without complex physiological sensors.
• To improve the accuracy of cybersickness prediction by incorporating user-specific factors.

Main Methods:

• Created a publicly available dataset with dense, on-demand sickness level annotations (every 0.5 seconds) using the Fast Motion Sickness (FMS) scale.
• Excluded difficult-to-collect physiological signals, focusing on content motion profiles and FMS data.
• Trained predictive models using the novel dataset, including user-specific parameters like age and gender.

Main Results:

• Predictive models trained on the new dataset achieved reliable cybersickness prediction.
• Models incorporating user-specific parameters demonstrated improved prediction accuracy.
• The dataset's design facilitates practical, real-time application in VR environments.

Conclusions:

• A practical dataset enabling real-time cybersickness prediction is now available.
• Content motion profiles and FMS data are sufficient for reliable prediction, especially with user personalization.
• This work addresses key limitations in current cybersickness research, paving the way for smoother VR experiences.