Decentralized coordination of platoons – A conceptual approach using deep reinforcement learning

In this paper, a conceptional two-stage approach for decentralized coordination of platoons using deep reinforcement learning is developed. For this purpose, the relevant existing approaches for the decentralized and centralized coordination of platoons are identified first with the help of a systematic literature review. Then the advantages and disadvantages of the different approaches are analyzed, and the most important results are described. Derived from the findings, a conceptional two-stage approach is developed, which consists of a dispositive and operational level. At the dispositive level, the individual vehicle routing of a trucking company is performed. All feasible routes between the origin and destination nodes under consideration of time window restrictions have to be identified first. The truck then decides at the operational level, based on changing environmental parameters, which tour to choose and whether or not to form a platoon. With the help of decentralized coordination of platoons, it will thus be possible to achieve a cost optimum for platoon-capable trucks.