Design And FPGA Implementation Of Parallel Predictive Controller For Vehicle Path Tracking

Recently,along with the development of automotive industry,the research and application of intelligent vehicles has gained widely concern.As a key part of the intelligent vehicle system,the path tracking control of the vehicle has become the focus and hotspot of research.Aiming at the problems of nonlinearity,strong coupling,multi-constraint and multi-variable in the path tracking control system,a nonlinear model predictive controller is designed by lateral and longitudinal dynamic coupling vehicle model to achieve the integrated control,thereby improving the control performance of the system.Since the nonlinear system model of vehicle contains nonlinear dynamics,online iterative calculation of system equations is required,which will result in a heavy burden of online calculation and the poor real-time performance.This paper improves the online calculation speed of nonlinear model predictive control(NMPC)from the perspective of parallel acceleration.The parallel Newton optimization method is used to solve the NMPC problem to make the variables in the predictive time domain decoupled and independent.Then,by using the parallel computing features of FPGA,the hardware acceleration of the algorithm is realized through the optimization design such as loop unrolling and array partitioning.To verify the effectiveness and quickness of the designed NMPC controller,a real-time experiment of vehicle path tracking control based on the lateral and longitudinal dynamic coupling model is carried out,and a good control performance is achieved.This paper mainly includes the following four parts:1.Considering the control requirements of the vehicle path tracking system comprehensively,a nonlinear model predictive controller is designed and the offline simulation verification is completed.In order to improve the control accuracy,a nonlinear vehicle model with lateral and longitudinal dynamic coupling is established,and the tire dynamic modeling is completed by "magic formula" tire model.Then,according to the control requirements,the nonlinear model predictive controller of the vehicle path tracking is designed,and the sequential quadratic programming(SQP)method is used to solve the nonlinear programming(NLP)problem of the controller.Finally,the closed-loop system of "NMPC controller-vehicle" is built in MATLAB environment,and the vehicle path tracking control simulation experiments under lane changing and double lane shifting conditions verify the effectiveness of the NMPC controller based on the coupling of lateral and longitudinal dynamics designed in this paper.2.For higher on-line computing speed of NMPC controller for vehicle path tracking,a parallel Newton optimization method is proposed.Firstly,the controller model is discretized by backward Euler method and transformed into an NLP problem,and then using KKT condition to construct a set of nonlinear equations for optimal solution.Then,by analyzing the coupling relationship between the variables to be solved in the adjacent prediction time domain,the variables are approximated to make the coupling equation decoupled and independent,so as to realize parallel computing.Finally,a simulation experiment is carried out under the lane changing condition and compared with the experimental results based on the SQP method,which verifies the effectiveness and quickness of the parallel Newton optimization method for vehicle path tracking nonlinear model predictive controller.3.From the perspective of practical application,aiming at the requirements of miniaturization,low cost and quickness of vehicle system controller,and for achieving the parallel acceleration of hardware for parallel Newton optimization method,a hardware implementation scheme based on FPGA is proposed.First,convert the controller designed in m language on the MATLAB platform to C/C++ language and perform fixed-point data structure design and open-loop verification,and then transplant all the code to ARM processor on SDSo C platform to complete the board-level verification.Then,the hardware acceleration of the control algorithm is carried out,and the parallel optimization module and the whole NMPC module are transplanted to FPGA respectively.It is verified that the acceleration multiples of the two schemes are basically the same.The acceleration scheme of transplanting parallel optimization module to FPGA is further optimized.In order to give full play to the parallel feature of the solution method,the loop unrolling optimization design is carried out to realize parallel independent calculation of decoupling equation,and the array partition optimization design is carried out to increase the data access bandwidth.After FPGA acceleration,the millisecond level calculation of the controller is realized.Finally,the overall test is carried out to verify the real-time performance of the controller.4.To verify the overall performance of the system,the vehicle path tracking hardwarein-loop experiment is carried out.First,build a hardware-in-loop experiment platform,use Micro Auto Box to run a 14-degree-of-freedom vehicle model to simulate the actual vehicle system,and select the Zynq(ARM+FPGA)development board to run the nonlinear model predictive controller.The two parts interact with each other through Ethernet to realize the closed loop.The vehicle path tracking control experiment is completed on the built experimental bench,and the results verify the validity and quickness.