摘要
Real-time measurement of physiological discomfort states during virtual reality (VR) exposure presents significant instrumentation challenges due to the need for noninvasive, wearable sensing solutions with low latency and high accuracy. This study presents a closed-loop measurement and intervention system integrating inertial measurement unit (IMU)-based multimodal sensing, deep learning signal processing, and vibrotactile feedback instrumentation for continuous cybersickness monitoring. The measurement system employs triaxial accelerometers and gyroscopes positioned at the head and bilateral foot locations to capture kinematic parameters across three postures (sitting, standing, and walking). A Mamba-based deep learning architecture processes Kalman-filtered IMU signals, extracting features including root mean square, standard deviation (STD), jerk, and frequency-domain characteristics to estimate cybersickness severity with 30-s temporal resolution. Stage I experimental validation with 20 participants established posture-specific measurement models, achieving coefficients of determination (R2) values of 0.75, 0.79, and 0.85 for sitting, standing, and walking conditions, respectively, with a mean squared error (mse) of 0.41 ± 0.13 for the walking model. Stage II validation with five participants demonstrated closed-loop system performance, where message queuing telemetry transport (MQTT) protocol-based communication (average latency: 0.1335 ± 0.0385 s) enabled real-time triggering of a Raspberry Pi-controlled vibrotactile device operating at 150 Hz. Statistical analysis using the Wilcoxon signed-rank test confirmed significant reductions in simulator sickness questionnaire (SSQ) scores for sitting (p = 0.043) and standing (p = 0.043) postures following measurement-driven intervention. The instrumentation system demonstrates that IMU sensor fusion combined with real-time signal processing enables accurate physiological state estimation for closed-loop measurement applications in wearable monitoring systems.