Gap Setting Control of ACC‐Equipped Vehicles for Enhancing Traffic Flow: Deep Reinforcement Learning Approach

The longitudinal behaviour of commercial vehicles equipped with adaptive cruise control (ACC) systems varies significantly based on their gap settings, which determine how closely they follow a lead vehicle. Short gap settings could increase road capacity but may decrease traffic flow stability. Leveraging these variations, this study proposes a double deep Q network (DDQN)‐based gap‐setting control strategy to improve traffic flow efficiency and stability on freeways. We assume that connected and automated vehicles (CAVs) have two adjustable gap settings, short and long, calibrated using trajectory datasets that include commercial ACC‐equipped vehicles. The DDQN agent selects either a short or long gap setting for CAVs based on their state at every gap setting control interval. The proposed strategy is evaluated in a loop environment under single‐CAV and multiple‐CAV control scenarios, and its performance is compared with (i) a uniform short gap setting approach and (ii) a uniform long gap setting approach. In the single‐CAV control scenario, the proposed strategy enhanced both traffic flow efficiency, evaluated by average travel time rate, and stability, evaluated by average acceleration noise, reducing them by up to 5.4% and 22%, respectively, compared with baseline approaches. In the multiple‐CAV control scenario, the DDQN agent balanced efficiency and stability across all market penetration rates, maintaining an average travel time rate comparable to the uniform short gap setting approach while lowering acceleration noise by up to 58.7%. Overall, this study presents a novel practical approach to improving traffic flow by leveraging existing manufacturer‐developed ACC systems.