Path tracking control of intelligent vehicles via a speed-adaptive MPC for a curved lane with varying curvature

Abstract

When driving on a curved lane with varying curvature, human drivers usually make a small range of longitudinal speed adjustments to maintain a yaw response that feels more stable. Taking this type of response as a reference, intelligent vehicles must also make adjustments to their longitudinal speed within a small range during the path tracking process with curvature changes. The current variable curvature path tracking algorithm using model predictive control (MPC) basically assumes that the vehicle moves at a constant speed, which does not match the small-range adjustment of the longitudinal speed and thereby affects the accuracy of path tracking. In this paper, considering the small range of speed variation in the path tracking process, the path and speed decoupling control in Frenet coordinates are used to replace the longitudinal-lateral-yaw complex coupling dynamics control. Meanwhile, considering the problem that the steady-state error of the MPC controller caused by curvature variation in the path tracking process cannot be eliminated, the adaptive weight control (AWC) and adaptive feedforward (AFF) models based on BP neural network (BPNN) data learning are designed to dynamically adjust the lateral error weight and feedforward factors of the MPC controller. As a result, a more accurate path tracking effect is achieved. Simulation results in the joint CarSim-Simulink environment show that the proposed algorithm significantly improves the adaptive capability of the linear MPC controller in response to time-varying conditions and has a higher tracking accuracy.