arXiv:2611.08117v1 Announce Type: new Abstract: Longitudinal velocity is a core input for vehicle control, driver-assistance, and automated driving functions. Production wheel-speed sensors deliver only a few pulses per revolution, so they lose resolution below 1 km/h, the regime in which automated parking and traffic-jam assistance operate. The motor speed from the electric machine's resolver stays finely resolved down to standstill but reaches the wheels only through a compliant drivetrain. We therefore estimate the longitudinal velocity with an unscented Kalman filter (UKF) on a double-track model adapted to low velocities and incorporate the motor speed along two routes, one purely physical and one hybrid. The first augments the model with an explicit torsional spring-damper drivetrain, making the motor speed a state of the estimator. The second feeds the output of a neural network-based virtual wheel-speed sensor into the UKF as a pseudo-measurement. Both routes are compared against two baselines, a wheel-speed-only configuration and one that adds the motor-speed sensor but treats the driveline as rigid. Because the model is numerically stiff near standstill, the comparison rests on a systematic study of explicit and semi-implicit discretization schemes. All four configurations are evaluated against an inertial reference in real-world test drives with two distinct series-production electric vehicles, anonymized as Vehicle A and Vehicle B for confidentiality. Below 1 km/h, both proposed estimators lower the velocity root-mean-square error by at least 54% and 45% against both baselines on Vehicle A. Carried over zero-shot to Vehicle B without re-tuning on target-vehicle data, they lower it by at least 41% and 51%. Both estimators therefore carry across vehicles without a new identification campaign.