Modeling Motion Compensated Radar Doppler Uncertainty for Autonomous Driving

Automotive radars play a prominent role in enhancing the safety of vehicles by providing advanced driver assistance capabilities. An automotive radar measures the range, angle, and Doppler of a radio reflection point in the sensor frame, while autonomy applications such as tracking and sensor fusion are processed in the vehicle frame. To account for representation mismatch, a spatial transformation is necessary to convert measured quantities like Doppler. Ego vehicle compensation is an essential step to accurately extract motion information from a target, as the Doppler is a relative measurement. While there are existing studies on removing the vehicle's contribution, uncertainties from both the vehicle and radar sensors have yet to be explored. This work closes the gap by extensively studying and providing sources of uncertainties in the compensated Doppler. Our approach involves deriving uncertainty formulas and proposing approximation methods that take into account computing constraints. The accuracy and efficiency of our approximations are validated with Monte-Carlo simulations under different autonomous driving scenarios. Our approximation results are close to the 99% of true values.