Generating Odometry Measurements from Automotive Radar Doppler Measurements

In order to obtain odometry measurements, we use off the shelf automotive radar sensors. To this end, we use Doppler measurements of static objects to estimate the yaw rate and the speed of a vehicle. We integrate those measurements with a nearly constant turn rate and velocity (CTRV) model in order to obtain dead reckoning position and yaw estimates. This can be performed in a single time frame, if at least two suitable measurements are available. However, it is crucial that only measurements from static objects are used. We leverage the information provided by previous time steps’ estimates in order to determine which measurements to use. We present results with simulated and real world radar data. In a sparse static object scenario, the proposed approach is more accurate than a baseline random-sample-consensus-based (RANSAC) selection of static objects. With dense static objects in the environment, the results from both approaches are equally good. In each case, our approach does require less computational resources.