Preceding Vehicle Longitudinal And Lateral State Estimation With Measurement Delay Based On Moving Horizon Estimation

Recently,with the increase of vehicle ownership and the development of information and intelligent technologies,many advanced driver assistance systems,autonomous driving systems and the other vehicle automotive technologies has been studied and applied.Information perception system serving as a pioneer in the whole system,it is supposed to supply the information support to the planning,decision and controlling systems.Thus,it is critical to the system performance.When a vehicle is driving,it is necessary to derive the preceding vehicle motion states,due to the motion of preceding vehicle has a significant influence to the host vehicle.It is limited by the current sensor technologies,some critical information can not be measured directly.To solve these problems,the estimators are designed to derive the unmeasurable information.To derive the longitudinal and lateral motion states of preceding vehicle under the car following scenario,we designed two estimators for the longitudinal and lateral states estimation,respectively.Firstly,two dynamic models are presented for the longitudinal and lateral motion,respectively.Then two MHE based estimators are designed for the longitudinal and lateral states estimation,respectively.Due to the MHE estimators are solved based on optimization,thus it may have heavy computational burden,which is not suitable for the vehicular applications.To accelerate the computation,multiple shooting method was introduced in this thesis.Firstly,to depict the longitudinal motion of preceding vehicle,it is necessary to construct a dynamic model.In this thesis,an acceleration model is constructed under the car following scenario based on Helly’s driver acceleration model.Besides,different drivers may have different driving habits,a driver’s aggressiveness parameter is introduced to depict different driving habits.To depict the lateral motion states of preceding vehicle,a 2DOF bicycle model was adopted in this thesis.The tire forces are calculated by a nonlinear tire model for a higher accuracy.However,when a vehicle is driving,the motion states are not only related to the driver’s manipulation,but also has a close relationship with the expect path.To take the path information into account,the Serret-Frenet equations are adopted to depict the vehicle dynamics about reference path.Then the 2DOF bicycle model is combined with Serret-Frenet equations to constructed a dynamic model which the vehicle dynamics and path information can be considered simultaneously.The effectiveness of the presented model is tested by Carsim MATLAB joint simulations.Then,based on the theory of MHE and dynamic models,the longitudinal and lateral estimators are designed,respectively.Due to the longitudinal dynamic model is linear model,a linear MHE estimator is designed for the preceding vehicle’s longitudinal motion states estimation.The optimization problem of linear MHE estimator can be transformed and solved by QP.In this thesis,a nonlinear dynamic model is presented to depict the lateral motion states of preceding vehicle for a higher accuracy.Thus,a nonlinear MHE estimator is designed to estimate the lateral states of preceding vehicle.By the solution of optimization problem,the lateral speed and yaw rate of preceding vehicle can be derived.A series of joint simulations based on MATLAB/Simulink and Carsim are conducted to verify the effectiveness of the proposed estimators.In the real applications,there may be some conflicts between the higher state update rate and lower measurement sampling rate.Besides,the measurements are also influenced by the random delays.To solve these problems,a multi-rate model is presented based on the aforementioned lateral motion dynamic model.Then we introduced a new random variable so that the random measurement delays can be taken into account.Then the multi-rate system model is combined with the random delays,so that the multi-rate character and measurement delays can be taken into account simultaneously.Then the MHE estimator was redesigned based on the multi-rate and delayed system dynamic model.Due to the nonlinear MHE problems are transformed to NLP problems,there may be a high computational burden which is unacceptable for the vehicular applications.To solve this problem,the multiple shooting method is introduced.By the use of multiple shooting method,the optimization process was reformulated,the computational burden was also released,so that the real-time performance can satisfy the vehicular application requirements.At the end of this thesis,the effectiveness of aforementioned algorithms are tested by the joint simulation of MATLAB/Simulink and Carsim.