PTV Longitudinal-Lateral State Estimation Considering Unknown Control Inputs and Uncertain Model Parameters

The mixed driving of human-driven and automated vehicles is believed to be the most likely future transportation pattern, and this work studies longitudinal-lateral state estimation for human-driven preceding target vehicles (PTVs) in such situations, as this information cannot be directly obtained by the host vehicle (HV). For longitudinal state estimation, the constant turn rate and acceleration (CTRA) model is adopted and the unscented Kalman filter (UKF) is used for estimator construction. For lateral state estimation, a PTV lateral motion model constructed based on the bicycle model and PTV lateral motion relative to the road reference is used. However, the steering angle of the conventional human-driven PTV is unknown, and PTV parameters such as cornering stiffness are uncertain. To provide more accurate PTV lateral states, the design of a state estimator that handles both unknown control inputs and uncertain model parameters is investigated in this study. First, considering that the model parameters are unobservable with the conventional iteration-based method, a system model-based method that measures uncertain parameters via estimated states is proposed. With this method, the observability of uncertain model parameters is naturally established, as the system model fully characterizes the relationship between system states and model parameters. Then, the sensor measurements and estimated states are combined as measurements, and a state-input-parameter mixed Kalman filter (SIPMKF) is proposed to cope with the coexistence of uncertain model parameters and unknown control inputs for PTV lateral state estimation. The effectiveness of the proposed PTV longitudinal-lateral state estimation architecture is verified via simulations and experiments.