Deep Reinforcement Learning for Adaptive Cruise Control in Connected Autonomous Vehicles

Deep Reinforcement Learning (DRL) has emerged as a viable methodology for the development of Adaptive Cruise Control (ACC) systems in Connected Autonomous Vehicles (CAVs), providing improved flexibility and efficiency. This work investigates the use of DRL in ACC, highlighting the enhancement of speed regulation and the preservation of safe inter-vehicle spacing in fluctuating traffic conditions. A DRL agent, informed by synthesizing simulated and empirical data, can modulate throttle and brake responses to optimize safety, comfort, and fuel economy. The quantitative findings indicate the efficacy of the proposed method using DRL-based ACC system realizes a 15% decrease in fuel consumption relative to conventional rule-based ACC systems, concurrently achieving a 20% enhancement in tracking accuracy at target following distances. The device decreases sudden acceleration and deceleration occurrences by 30%, improving passenger comfort. In safety-critical situations, such as abrupt halts, the DRL-based ACC demonstrates a 25% quicker reaction time, significantly reducing accident hazards. These findings underscore the capability of DRL to develop more adaptable, efficient, and secure ACC systems, facilitating the development of more intelligent and dependable autonomous driving technologies.