A Data-Driven Control Framework Using Deep Reinforcement Learning for Autonomous Driving

This paper presents a data-driven control framework for autonomous driving based on deep reinforcement learning (DRL). The proposed framework addresses lane-following and car-following using a dual-actor DDPG architecture with taskspecific rewards. The proposed approach improves safety, efficiency, and robustness, as demonstrated through extensive simulation, hardware-in-the-loop (HIL), and real-world testing. Comparative results show significant advantages over adaptive model predictive control (AMPC), highlighting its potential for real-world autonomous vehicle deployment.