A reinforcement learning approach to vehicle coordination for structured advanced air mobility

Advanced Air Mobility (AAM) has emerged as a pioneering concept designed to optimize the efficacy and ecological sustainability of air transportation. Its core objective is to provide highly automated air transportation services for passengers or cargo, operating at low altitudes within urban, suburban, and rural regions. AAM seeks to enhance the efficiency and environmental viability of the aviation sector by revolutionizing the way air travel is conducted. In a complex aviation environment, traffic management and control are essential technologies for safe and effective AAM operations. One of the most difficult obstacles in the envisioned AAM systems is vehicle coordination at merging points and intersections. The escalating demand for air mobility services, particularly within urban areas, poses significant complexities to the execution of such missions. In this study, we propose a novel multi-agent reinforcement learning (MARL) approach to efficiently manage high-density AAM operations in structured airspace. Our approach provides effective guidance to AAM vehicles, ensuring conflict avoidance, mitigating traffic congestion, reducing travel time, and maintaining safe separation. Specifically, intelligent learning-based algorithms are developed to provide speed guidance for each AAM vehicle, ensuring secure merging into air corridors and safe passage through intersections. To validate the effectiveness of our proposed model, we conduct training and evaluation using BlueSky, an open-source air traffic control simulation environment. Through the simulation of thousands of aircraft and the integration of real-world data, our study demonstrates the promising potential of MARL in enabling safe and efficient AAM operations. The simulation results validate the efficacy of our approach and its ability to achieve the desired outcomes.