Physics-Data-Driven Hybrid State Estimation for Articulated Heavy Vehicles in Vehicle-Road-Cloud Cooperative Framework

Owing to their multi-body configuration, the handling and stability of articulated heavy vehicles have received increasing attention in recent years. Accurate estimation of the trailer states and the articulated angle is essential for the active control of these vehicle systems. This paper proposes a hybrid state estimation approach for articulated heavy vehicles in the vehicle-road-cloud cooperative framework. A dynamic model is established to describe the yaw motion of both the tractor and the trailer. A trailer state network based on long short-term memory layers is designed to estimate the trailer state and articulated angle based on the information from the vehicle-road-cloud cooperative framework. An adaptive hybrid extended Kalman filter is proposed to integrate the data-driven estimator with the dynamics model of articulated heavy vehicles. Compared with traditional model-based state estimators, simulation results demonstrate that the proposed hybrid estimation method achieves better accuracy, with up to a 23.4% reduction in the mean absolute error of the semi-trailer yaw rate estimation during a double lane change maneuver on a low-adhesion road surface. Furthermore, the adaptive strategy enables the proposed method to remain accurate against significant outliers in the pseudo-measurements of trailer state network.