Enhancing Autonomous Driving: A Scalable Multi-Layered DRL Approach for Decision and Control

Autonomous driving technology is rapidly progressing, promising significant societal benefits. This paper presents an advanced DRL-based framework for end-to-end autonomous vehicle decision-making and control. Our technical proposal is articulated through a scalable, multi-layered architecture that integrates sensory data aggregation, trajectory prediction, and dynamic control layers. The framework is distinctive for its Actor-Critic approach, where the Critic network evaluates potential trajectories and the Actor network implements the control commands, ensuring safety and compliance with traffic norms. The adaptability and robustness of our system are validated in extensive simulations and real-world trials, demonstrating its efficacy across various sensory system configurations. This research contributes to the autonomous driving domain by providing a versatile and reliable framework for intelligent vehicle navigation. The methodologies and systems presented in this paper have undergone rigorous validation through simulation experiments and practical real-vehicle tests, confirming the viability and effectiveness of our advanced DRL-based framework for autonomous vehicle decision-making and control.