Research on Path Tracking Control Based on Optimal Look-Ahead Points

Abstract

Pure pursuit tracking algorithms are a popular control method in the field of autonomous navigation, where the selection of a look-ahead point plays a crucial role in tracking performance. However, the computation of the look-ahead point involves issues that are challenging to describe precisely using mathematics. To enhance the tracking precision of vehicles on curved trajectories, we propose an improved optimal look-ahead point path tracking algorithm. This algorithm primarily seeks the optimal look-ahead point by considering both longitudinal look-ahead distance and lateral position offset. To begin, we employ the Deep Deterministic Policy Gradient (DDPG) algorithm to train vehicles to determine the optimal longitudinal look-ahead distance under various constant curvature and velocity conditions. Subsequently, by utilizing the optimal longitudinal look-ahead distance and the front-wheel steering angle, we construct a lateral deviation search region. Finally, we use an evaluation function to search for the optimal look-ahead point within this region. Simulation tests demonstrate that the proposed algorithm significantly improves tracking accuracy under varying curvature trajectory conditions.