Adaptive Look-ahead distance for Pure Pursuit Controller with Deep Reinforcement Learning Techniques

The trajectory generated by the motion planner depends on various factors such as road curvatures, static and dynamic obstacles, passenger comfort, and road conditions. Geometric path tracking is one of the most critical aspects of path tracking in Autonomous Vehicles. The lateral and longitudinal controllers aim to track the path as closely as possible and minimize along-track and cross-track errors considering the system’s dynamics and latencies present in it. Pure pursuit is one of the simplest geometric tracking controllers having only one parameter to tune, i.e., the look-ahead distance which handles steering aggressiveness. In this paper we propose a novel lateral controller which tunes the look-ahead distance of pure pursuit controller with the help of Deep Deterministic Policy Gradient(DDPG).