Utilizing Both DDPG and DQN Algorithms for Path-Following Control in Autonomous Vehicles

Abstract

Enhancing autonomous vehicles (AVs) can significantly improve the reliability and safety of transportation systems. To achieve full autonomy (SAE Level 5), AVs must handle complex and unpredictable traffic conditions. A critical aspect of automated driving is path-following, which ensures the vehicle accurately tracks a predefined trajectory. Traditional path-following techniques often rely on parameter tuning or rule-based systems, which may struggle in highly dynamic environments. Reinforcement learning (RL) offers a promising alternative by enabling control strategies to be learned through experience. This study investigates the performance of both Deep Deterministic Policy Gradient (DDPG) and Deep Q Network (DQN) algorithms in managing AV acceleration and steering for path-following. The results show that DDPG and DQN algorithms converge quickly, enabling stable and efficient trajectory tracking while ensuring smooth control inputs. These findings underscore the potential of hybrid DDPG and DQN algorithms in advancing autonomous driving technology.