Adaptive Path Tracking Using a Dynamic PID Controller Enhanced by Proximal Policy Optimization

This paper introduces a novel adaptive path tracking controller that integrates the Proximal Policy Optimization (PPO) algorithm with a Proportional-Integral-Derivative (PID) control framework. Designed for challenging autonomous driving scenarios, this hybrid approach addresses the limitations of traditional controllers, which lack adaptability, and typical Reinforcement Learning (RL) methods, which often face training inefficiency and stability issues. Key innovations include real-time PID gain optimization via PPO, an adaptive error fusion mechanism with dynamic weighting of lateral and heading deviations, and a speed-adaptive preview mechanism. Notably, this framework does not require an accurate vehicle dynamics model. Experimental results demonstrate that our approach significantly outperforms both conventional and learning-based methods across various high-curvature driving conditions, including S-curves and right-angle turns, achieving superior tracking accuracy and robust stability, thus validating its practical effectiveness.