Calibration of Front Wheel Odometry Model

Accurate and cost-effective state estimation is needed to reach self-driving. The well-known GNSS and IMU fusion can be improved by the integration of wheel odometry. The robustness of this type of odometry is increased if both the rear and front wheels are utilized. Furthermore, the method is cost-effective, but for accurate motion estimation, the vehicle model parameters have to be calibrated. This paper presents the whole calibration task, such as input estimation, filtering of reference outputs, and parameter identification. The proposed estimation method is a unique version of the Gauss-Newton method, to mitigate the distortion effect of pose initialization. The effectiveness of the proposed calibration process is illustrated through vehicle test experiments. The validation demonstrates that the calibration results in below than 1% relative estimation error, thus the front-odometry can be integrated into the state estimation layer of a self-driving vehicle.