Research On Trajectory Tracking Control Method For Autonomous Vehicles Based On Fast Model Predictive Control

Trajectory tracking control is one of the key technologies of autonomous vehicles,which can make the vehicle reach the destination safely,accurately and quickly by controlling the steering system of the vehicle.This work was supported in part by the Electric intelligent Vehicle Innovation Team of the Science and Technology Department of Jilin Province under Grant 20180519022 JH,according to the project requirements and tasks,the research on trajectory tracking control of self-driving vehicles,the design of the controller,and for the real-time performance of the controller improvements are made,and the main contents are as follows.Vehicle kinematic,dynamic and tire modeling.In order to design the vehicle state prediction model in the model predictive control,the kinematic model and the3-DOF dynamic model were established for different vehicle speeds based on the kinematic relationship and dynamic characteristics of the vehicle,respectively.In order to accurately describe the tire force in the dynamic model,two classical tire models,Magic Formula and Gim,were established respectively.By comparing with the tire lateral and longitudinal force data in Carsim,the fitting accuracy,coefficient determination method and generality of the two models were analyzed and compared,and finally the Gim tire model was selected as the tire model of the vehicle dynamics model.The design of automatic driving vehicle trajectory tracking controller based on model predictive control.According to the principle of model predictive control,the established vehicle kinematic model and the 3-DOF dynamic model were linearized and discretized,in which the linearization was performed by using Taylor expansion and retain the first-order terms and the discretization was performed by using the Forward Euler method.To ensure that the vehicle can execute the commands given by the controller as well as the ride comfort,the constraints of the controller were designed.The design of the controller was completed by combining the constraints with an objective function based on the trajectory deviation and converting it into the standard form of quadratic programming.The double lane change trajectory tracking test was carried out by Simulink and Carsim co-simulation,and the test results showed that the designed controller can effectively control the vehicle to track the reference trajectory accurately.Improvement of the trajectory tracking controller based on model predictive control for real-time performance.The advantages and disadvantages of three methods to improve the real-time of model predictive control were analyzed and compared,and the input blocking strategy was selected to improve the trajectory tracking controller based on the vehicle 3-DOF dynamic model.Based on the principle of input blocking strategy and preliminary experiments,the input blocking matrix was designed and the dynamic model predictive control was improved.The real-time performance and tracking accuracy of the controller were compared with the original controller through Simulink and Carsim co-simulation tests.The computation time of the controller was reduced significantly.Collaborative controller design based on kinematic and dynamic model predictive control.To further reduce the computation time and improve the real-time performance of the controller,the effectiveness conditions of the vehicle kinematic model were analyzed,in other word the effectiveness of the vehicle kinematic model is better when the lateral acceleration of the vehicle is less than 5 μg,and the validity of the effectiveness conditions was verified.The switching strategies of the kinematic and dynamic model predictive control were designed according to the effectiveness conditions,and in order to prevent the step output during the controller switching,the switching strategy based on the excessive working conditions of lateral acceleration with variable weight coefficients was designed,and the involved co-controllers were verified by Simulink and Carsim co-simulation tests,and compared with the improved dynamic model predictive controller based on the input blocking strategy.The results showed that the kinematic and dynamic model predictive co-controller can control the vehicle tracking reference trajectory with better accuracy and further improve the real-time performance.