The pre-crash model predicts occupant motion during the moments preceding a collision, caused by vehicle accelerations from braking or evasive maneuvers. The surrogate model takes time-dependent inputs in the longitudinal, lateral, and yaw directions, and approximates the displacement of a high-dimensional human body mesh using model order reduction. This reduction maps complex motion into a low-dimensional feature space, where a regressor learns the dynamic behavior. The predicted motion is then projected back to the original space, enabling real-time visualization of occupant movement. Currently, the full high-fidelity model is too computationally demanding to run on the server; therefore, a simplified interactive example is provided below.