Cybersickness is a major barrier to comfortable VR use, and reliable prediction often requires multiple physiological and behavioral sensors. However, existing models typically assume a fixed set of modalities, making them vulnerable when sensors are unavailable and difficult to generalize across datasets with different sensor configurations. We adapt a transformer-based model that handles arbitrary sensor combinations without imputing missing data. Anchor-based contrastive learning maps heterogeneous sensor embeddings into a shared latent space using eye tracking as the anchor modality, while masked self-attention excludes unavailable sensors during fusion. We evaluate our approach on three publicly available cybersickness datasets: VR.net, SIM21, and Blur-Rollercoaster, collected using different VR systems and sensor configurations. Across single-sensor dropout conditions, prediction error changes by less than 0.09 MAE. Joint training across all datasets achieves 0.51 MAE and 0.63 Pearson correlation, substantially outperforming single-dataset models. These results demonstrate robust and generalizable cybersickness prediction across diverse VR systems.
Purnata Saha, Rifatul Islam, Qiliang He, John Quarles
原文摘要
Cybersickness is often assessed during VR exposure through discrete overall sickness, which cannot capture specific symptoms or their changes over time. Post-exposure questionnaires such as the Simulator Sickness Questionnaire provide symptom-specific measures but cannot be collected during immersion. We investigate whether separating cybersickness into three symptom-specific dimensions -- nausea, oculomotor discomfort, and disorientation -- during immersion provides more informative continuous measurements. We conducted a between-subjects study with 52 participants in a VR-rollercoaster, comparing a three-slider interface with a single-slider interface. Concurrently, participants also completed an auditory Simon task to measure attention. We additionally collected multimodal set of physiological signals, including GSR, ECG, PPG, eye and head tracking. Results show that each slider correlated most strongly with its corresponding SSQ subscale, whereas the single slider correlated with all SSQ subscales. Using three sliders did not significantly affect usability, workload, or attention. We publicly release the multimodal dataset to support future cybersickness research.