UAM Autonomous Guidance Control Command Generation Through Deep-learning-based Aircraft Marshalling Recognition

Urban air mobility (UAM) introduces an innovative paradigm shift in urban transportation, heralding a revolutionary approach to aviation mobility. In its current nascent stage, UAM operations involve manned aircraft guided by skilled pilots. However, the ultimate vision for UAM is the deployment of autonomous unmanned aerial vehicles (UAVs), which is the focus of ongoing research and development. Although specific operational regulations for UAM have not been precisely defined, installation of vertiports within urban areas is anticipated. In this study, a deep-learning-based software is designed and proposed to guide UAM vehicles through aircraft marshalling, especially during critical maneuvers, such as takeoff, landing, and ground movements. A bounding-box-based object detection algorithm is utilized to differentiate effectively between pedestrians and aircraft security personnel, thereby recognizing objects of interest. Upon detection of aircraft marshalling, the proposed algorithm identifies and interprets the marshalling signals. Then, the information is converted into control commands that are transmitted to the UAM vehicle. To validate the efficacy of the proposed algorithm, experiments are conducted using the software-in-the-loop simulation. The analytical results confirm that the proposed framework can deliver suitable real-time performance in a dynamic image processing environment.