On the Optimal Path Following for an Autonomous Vehicle via Nonlinear Model Predictive Control

An optimal path tracking controller for autonomous vehicles is presented to coordinate longitudinal and lateral vehicle dynamics. With the nonlinear model predictive controller (NMPC), the vehicle could follow an arbitrary reference path at high speed while maintaining stability. To achieve this, we first transform the Cartesian coordinates of the reference path to curvilinear coordinates, which enables us to establish decoupled heading error and lateral error dynamics. The nonlinear tire model is used to construct a third-order vehicle dynamics that accurately predicts the vehicle's state. Furthermore, we propose a fifth-order (NMPC) that combines both path and vehicle dynamics. The cost function considers previewed path information and future vehicle dynamics within a moving horizon. By solving nonlinear optimization problems, the optimal steering angle and the desired longitudinal acceleration command can be obtained. The lower level controller distributes the acceleration command as the rear driving torques and/or the four-wheel braking torques. The simulation results in CarSim demonstrate that the vehicle can stably follow the planned path at an average speed around 85 km/h while keeping the tracking error within a small range.