Autonomous vehicles, a rapidly advancing technology in transportation, pose intricate operational challenges. A pivotal aspect requiring in-depth understanding and continuous enhancement is the decision-making architecture steering the self-navigational capabilities of these vehicles. It encompasses various elements such as vehicle and environmental representations, trajectory planning, risk-uncertainty assessment, and human behavior simulation \cite{1}. Categorizing and analyzing these components based on scenario types, action-space utilization, and time-related parameters aids in comprehending the complex decision-making system.
Within autonomous traffic navigation, Adaptive Cruise Control (ACC) stands out as a significant tool influencing decision-making \cite{2}. An effective ACC system ensures safe navigation, adapts to stop-and-go traffic, and minimizes driver intervention. However, prevalent model-based and Deep Reinforcement Learning (DRL)-based ACCs encounter challenges, including model complexity limitations and poor performance in uncertain conditions.
Addressing these challenges requires innovative solutions, such as employing a DRL-based ACC system with a novel multi-objective reward function. This approach, tested in car-following scenarios, exhibits promise in enhancing performance, improving adaptability in diverse traffic conditions, and potentially preventing rear-end collisions. In conclusion, ongoing research aims to refine and optimize decision-making processes in autonomous vehicles by overcoming current limitations and enhancing adaptability across diverse traffic scenarios.