Vehicle Trajectory Planning And Control Based On Model Predictive Control

Autonomous driving technology has brought numerous new opportunities for the development of the automobile industry and its correlation technique.Various auxiliary driving and even autonomous driving functions have been equipped on a large number of automobile products.With the development of the industry,the realization of a higher level of automatic driving still faces many challenges.In the autonomous driving system,the planning and control module is important.Since Model Predictive Control(MPC)was proposed in the 1970s,it has been widely used in many fields.MPC uses the model of the controlled system to realize automatic control according to the ideas of model prediction,feedback correction,and rolling optimization,which brings new ideas to realize the planning and control functions of autonomous driving.According to the above statements,the following contributions are provided in cooperations with MPC algorithms to improve the planning and control technique in autonomous driving:·A Main and Residual Model-based MPC(M&R-MPC)algorithm is proposed to improve the control performance.The algorithm effectively improves the control performance by collecting and learning the deviation between the used main model and the controlled system to form a residual model.This paper presents the recursive feasibility and stability analysis of the M&R-MPC algorithm.In addition,the paper respectively presents the optimization problem form when the main model is a state-space model or a transfer function model.·In the vehicle yaw stability control(YSC),aiming at the problem that there is a large deviation between the model used and the actual vehicle dynamics especially in high-speed,turning and other scenarios,this paper proposes a YSC algorithm based on two-layer learning MPC.In the double-layer YSC framework,the upper controller adopts the M&R-MPC algorithm and introduces Gaussian Process Regression(GPR)as the residual model.As for the slip angle constraint,the probability constraint is converted to the deterministic constraint.In addition,this paper also designs the weight distribution rules of the lower-level controller optimization problem.The combination of the two controllers finally improves the system’s overall performance.·To prevent the potential collision risk in parking trajectory planning,this paper proposes a guaranteed collision-free autonomous parking trajectory planning algorithm.In this study,the collision avoidance constraints between vehicles and obstacles are proposed from the perspective of geometric relations.When discretizing the constrained optimal control problem,collision avoidance constraints between adjacent states are added.In addition,in order to integrate the trajectory planning and tracking,a reduced time horizon MPC algorithm is proposed to track the parking trajectory,and a series of simulations and experiments demonstrate the effect of the integrated structure.·To coordinate trajectory planning and tracking in overtaking tasks,an overtaking trajectory planning approach based on MPC is proposed.The algorithm utilizes the vehicle kinematics model,which makes the generated trajectory more in line with the vehicle dynamic characteristics than the trajectory obtained by the planning method based on the point mass model.In order to avoid collision and reduce the complexity of the optimization problem when using the vehicle kinematics model,a two-stage collision avoidance constraint condition is proposed.The predicted trajectory of the vehicle is divided into the trajectory in the predicted horizon and the trajectory outside the predicted horizon.Different constraint conditions are proposed for them,and the trajectory planning in a longer horizon is realized without significantly increasing the calculation time.