Vehicle State and Bias Estimation Based on Unscented Kalman Filter with Vehicle Hybrid Kinematics and Dynamics Models

Abstract

In recent years, vehicle state estimation methods incorporating different vehicle models have received extensive attention. When the vehicle is disturbed by external forces not considered in traditional vehicle models (for example, a certain slope, or wind resistance different from theoretical calculation), the problem of model mismatch will occur, which leads to the inaccurate estimation of the vehicle states. To solve this problem, an Unscented Kalman Filter (UKF) algorithm is used to fuse inertial navigation data with the vehicle hybrid model in this paper. The hybrid model introduces a switching strategy that fuses the vehicle kinematics and the dynamics models while augmenting biases that need to be estimated in the vehicle states. The switching strategy resolves the integration divergence problem of vehicle dynamics models at low speeds and the inaccurate estimation of vehicle kinematics models at high speeds. Simulation experiments demonstrate that the proposed method can accurately estimate biases induced by external forces, enhancing the accuracy and confidence of states by eliminating errors caused by these biases. The robustness of the method is validated in vehicle verification platform experiments, where errors in vehicle lateral speed and yaw rate are reduced by 9.7 cm/s and 0.012 °/s, respectively, under large curvature maneuvers, and 9.6 cm/s and 0.004 °/s under quarter-turn maneuvers. The proposed method significantly improves lateral speed and vehicle position accuracies.