Dynamics parameter estimation for AGVs: A Levenberg–Marquardt-optimization and least-squares-method framework

Dynamics parameter estimation is of vital importance to establish the accurate dynamics model for autonomous ground vehicles (AGVs). In this paper, a Levenberg–Marquardt-optimization and least-squares-method (LMO-LSM) framework is proposed to estimate vehicle dynamics parameters requiring only conventional sensors. This innovative LMO-LSM framework incorporates the simplified Pacejka magic formula tire model alongside the vehicle lateral dynamics model and is composed of two phases to estimate the twelve parameters. The first phase is to estimate the distances from the vehicle center of gravity to the front and rear axles, the Pacejka parameter calculation coefficients and the Pacejka parameters through Levenberg–Marquardt-optimization, ensuring the predicted lateral acceleration sequence closely aligns with the real lateral acceleration sequence. The second phase is to estimate the yaw moment of inertia through least-squares-method by minimizing the discrepancy between the predicted yaw moment sequence and the real yaw moment sequence. Furthermore, the proposed LMO-LSM framework is tested in the high-fidelity MATLAB/Simulink-CarSim co-simulation and real-world field experiments, validating the effectiveness and practicality of the LMO-LSM framework.