A Deep Motion Sickness Predictor Induced by Visual Stimuli in Virtual Reality.

Area of Science:

• Virtual Reality (VR) and Human-Computer Interaction
• Neuroscience and Cognitive Science
• Computer Vision and Machine Learning

Background:

• Virtual reality (VR) environments often cause cybersickness, specifically visually induced motion sickness (VIMS), due to sensory conflicts.
• Current methods for VIMS assessment are limited in scope and predictive accuracy.
• Understanding the neurological and spatiotemporal factors contributing to VIMS is crucial for improving VR experiences.

Purpose of the Study:

• To develop a novel computational framework for simultaneously estimating VIMS scores and calculating VIMS sensitivity from VR content.
• To propose a deep learning architecture that models the neurological and spatiotemporal aspects of motion sickness.
• To provide a tool for predicting and potentially mitigating VIMS in VR applications.

Main Methods:

• A novel two-stage deep learning architecture was proposed, comprising neurological and spatiotemporal representation networks.
• The first network learns the neurological mechanisms of motion sickness, while the second expresses spatiotemporal features from generated frames.
• A weakly supervised approach was used to calculate VIMS sensitivity for unannotated temporal VIMS scores, and a large VR content database was released.

Main Results:

• The proposed framework demonstrated superior performance in VIMS score prediction compared to existing feature engineering and deep learning methods.
• The model successfully calculated VIMS sensitivity for individual frames within VR content.
• A method for visualizing cognitive responses to visual stimuli was developed, showing tendencies similar to clinical findings.

Conclusions:

• The developed framework offers a robust and accurate method for predicting VIMS in VR.
• This approach can aid in the design of more comfortable and immersive VR experiences by identifying and addressing potential sickness triggers.
• The findings contribute to a better understanding of the cognitive and physiological responses to self-motion simulation in virtual environments.