Parameter Identification of the Nonlinear Wheel Odometry Model with Batch Least Squares Method

Abstract

The wheel odometry can be applied to expand the state estimation performance of an autonomous vehicle, since this type of motion estimation is robust in the cases when the generally utilized GNSS and IMU-based methods fail, e.g. near high-rise buildings or in low-speed maneuvering. This type of odometry is cost-effective as well, but the estimation accuracy is limited by the parameter uncertainty. Due to the nonlinear behavior and the noises, the calibration with high precision remains an open question in the context of autonomous vehicles. This paper proposes an identification method that applies batch mode of the nonlinear least squares fitting to mitigate the impact of noises. The method is validated through vehicle test experiments which demonstrate that the bias in the parameter identification is reduced, and the calibrated wheel odometry can be utilized in the state estimation layer of an autonomous vehicle.