The ride comfort of autonomous vehicles is affected by motion sickness. This study quantifies the thresholds of vehicle dynamics parameters that induce motion sickness in both curved and straight-road scenarios, and constructs a predictive model. The results show that on S-shaped curves, the lateral acceleration thresholds (Δay) for moderate motion sickness (M2) and severe motion sickness (M3) are 0.398 m/s2 and 0.419 m/s2, respectively, while the Z-axis angular velocity thresholds (Gyroz_mean) are 6.876°/s and 8.022°/s. In straight-road scenarios, the Δay thresholds for M2 and M3 are 0.394 m/s2 and 0.648 m/s2, respectively, and the maximum longitudinal velocity (vx_max) reaches 13.961 m/s and 18.492 m/s. The proposed model achieves an accuracy of 86% for M2 and 82% for M3. Real-vehicle validation demonstrated that dynamically controlling vehicle motion states to maintain lateral acceleration, angular velocity, and longitudinal velocity below the specified thresholds reduced overall motion sickness risk by 39.7%.