Path tracking is a crucial task for autonomous vehicles. Traditional control methods often struggle with the nonlinear and coupled characteristics of systems, making it difficult to establish accurate models. By contrast, deep reinforcement learning approaches eliminate the need for complex mathematical models by learning path tracking strategies from sensing data, although they suffer from poor interpretability. Therefore, solely relying on either traditional control methods or deep reinforcement learning poses inherent limitations. This study combines the traditional pure pursuit (PP) method with the proximal policy optimization (PPO) algorithm from deep reinforcement learning to construct a truck controller architecture collaboratively. The PP method serves as the foundational control strategy, while the PPO algorithm is utilized to optimize control adjustments, enhancing control accuracy. In the state design, a look-ahead mechanism and its influence on speed were considered. Furthermore, a PreScan-Simulink-ROS software-in-the-loop testing platform was established to simulate the dynamics of trucks. Extensive simulation experiments were conducted, and the results demonstrated the effectiveness of our approach.
A Truck Path Tracking Controller Based on Combined Pure Pursuit and Deep Reinforcement Learning
2024-09-24
704653 byte
Conference paper
Electronic Resource
English
Deep reinforcement learning based path tracking controller for autonomous vehicle
SAGE Publications | 2021
|