Real-time planning and control methods for autonomous vehicles

Autonomous driving is a complex and large-scale system, where trajectory planning and trajectory tracking of vehicles are key technologies. This paper primarily focuses on structured road scenarios and the real-time handling of various dynamic obstacles. Based on the idea of decoupling speed and path, the 3D optimization problem, which includes time-speed and time-position information, is decomposed into two 2D problems. Firstly, vehicle behavior decisions are made using dynamic programing, followed by optimization to determine the path and speed. Secondly, we design SL and ST iterative frameworks to handle obstacles. Then we compare DP-QP with NMPC to validate the efficiency of the algorithms. Thirdly, a two dof error dynamics model predictive tracking controller is designed for path tracking. A nonlinear longitudinal dynamics controller is developed using the Gauss pseudospectral method to achieve velocity tracking. Compared to the nonlinear model predictive controller, improvements are achieved in both speed calculation and control accuracy. Finally, the proposed methods are validated using a joint simulation platform consisting of PreScan, MATLAB/Simulink, and CarSim. The results demonstrate that the proposed approach can generate reasonable and smooth trajectories while successfully avoiding obstacles. The execution frequencies of the planner and controller meet the requirements for autonomous driving.