Users would experience individually different sickness symptoms during or after navigating through an immersive virtual environment, generally known as cybersickness. Previous studies have predicted the severity of cybersickness based on physiological and/or kinematic data. However, compared with kinematic data, physiological data rely heavily on biosensors during the collection, which is inconvenient and limited to a few affordable VR devices. In this work, we proposed a deep neural network to predict cybersickness through kinematic data. We introduced the encoded physiological representation to characterize the individual susceptibility; therefore, the predictor could predict cybersickness only based on a user's kinematic data without counting on biosensors. Fifty-three participants were recruited to attend the user study to collect multimodal data, including kinematic data (navigation speed, head tracking), physiological signals (e.g., electrodermal activity, heart rate), and Simulator Sickness Questionnaire (SSQ). The predictor achieved an accuracy of 97.8\% for cybersickness prediction by involving the pre-computed physiological representation to characterize individual differences, providing much convenience for the current cybersickness measurement.

翻译：用户在沉浸式虚拟环境中导航时或之后，会经历个体差异显著的病症，通常称为晕动症。已有研究基于生理和/或运动数据预测晕动症严重程度。然而，与运动数据相比，生理数据在采集过程中严重依赖生物传感器，这不仅不便，且仅限于少数可负担的VR设备。本文提出了一种通过运动数据预测晕动症的深度神经网络。我们引入编码后的生理表征来刻画个体易感性，从而使预测器仅基于用户的运动数据即可预测晕动症，无需依赖生物传感器。我们招募了53名参与者进行用户研究，收集了多模态数据，包括运动数据（导航速度、头部跟踪）、生理信号（如皮肤电活动、心率）以及模拟器病问卷（SSQ）。该预测器通过引入预计算的生理表征来刻画个体差异，实现了97.8%的晕动症预测准确率，为当前晕动症测量提供了极大便利。