Resource-Efficient Dynamic Platooning Control through Deep Q-Network-Based Event-Triggered Scheduling

In modern transportation systems, dynamic platooning is a promising approach, leveraging adaptable vehicle formations to optimize traffic flow and fuel efficiency. However, relying on periodic broadcasts to preserve string stability during platoon maneuvers poses a challenge, demanding substantial resources within a constrained spectrum. This paper presents an event-triggered message scheduling that uses deep reinforcement learning specifically tailored for a dynamic platoon of autonomous vehicles on highways. In the proposed approach, the platoon leader intelligently triggers communication instances only when platoon speed variations threaten string stability. Furthermore, this study introduces a categorization system for triggered messages, distinguishing between safety and non-safety messages, ensuring their efficient storage and prioritized transmission. Simulation results demonstrate the efficacy of the trained leader in maintaining string stability, even amidst joining and leaving maneuvers, achieved through reduced communication frequency. Additionally, shorter waiting time for safety messages within the priority buffer guarantees reliable and expedited communication of critical information.