Path-following control for autonomous vehicles utilizing both DDPG and DQN algorithms

Enhancing autonomous vehicles (AVs) ensures a safe and reliable transportation system. Achieving level ۵ autonomy, as per the Society of Automotive Engineers (SAE) classification, requires AVs to navigate through complex and unconventional traffic environments. Path-following, a key aspect of automated driving, involves guiding a vehicle accurately and safely along a predefined path. Traditional path-following methods often rely on parameter adjustments or rule-based approaches, which may not be suitable for dynamic or intricate environments. Reinforcement learning (RL) has emerged as a promising technique capable of learning effective control strategies from an agent's experiences. This study investigates the effectiveness of the Deep Deterministic Policy Gradient (DDPG) method for controlling acceleration and the Deep-Q Network (DQN) technique for controlling steering in AV path-following. The combination of the DDPG and DQN algorithms together demonstrates rapid convergence, allowing the agent to achieve stable and efficient path-following while maintaining smooth control without excessive actions. The results indicate the efficiency of the new approach, suggesting its potential contribution to the advancement of automated driving technology.

کلیدواژه ها:

Autonomous vehicles ، RL method ، path-following ، DDPG ، DQN ، automated driving