Toward Accurate Cybersickness Prediction in Virtual Reality: A Multimodal Physiological Modeling Approach.

Area of Science:

• Physiological computing
• Human-computer interaction
• Virtual reality

Background:

• Cybersickness is a major barrier to virtual reality (VR) adoption, negatively impacting user experience and performance.
• Objective assessment of cybersickness severity is crucial for developing effective countermeasures.
• Existing methods often rely on subjective reporting, which can be unreliable.

Purpose of the Study:

• To develop and validate a physiological modeling approach for objective cybersickness severity assessment.
• To investigate the predictive capabilities of electrodermal activity (EDA) and electrocardiogram (ECG) signals for cybersickness.
• To identify key physiological features for accurate cybersickness prediction.

Main Methods:

• An interactive VR experiment was designed to induce varying levels of cybersickness.
• Physiological signals, including EDA and ECG, were continuously recorded during VR tasks.
• Machine learning regression models were built using extracted physiological features for cybersickness prediction.

Main Results:

• EDA-based models demonstrated superior predictive accuracy (R² = 0.98 with Ensemble Learning) compared to ECG-based models (R² = 0.53).
• EDA features, such as skin conductance metrics, were identified as the most significant predictors of cybersickness.
• ECG features provided modest contributions, indicating a limited complementary role.

Conclusions:

• Low-burden physiological signals, particularly EDA, can accurately and interpretably predict cybersickness severity.
• The findings support the development of lightweight, real-time monitoring systems for VR applications.
• This approach offers practical advantages for enhancing VR user experience and safety.