Autonomous UAV Landing on a Moving UAV using Machine Learning

摘要

View Video Presentation: https://doi.org/10.2514/6.2022-0789.vid Autonomous landing of a UAV(agent UAV) on another UAV(base UAV) is critical to extend UAV manipulation but has not been studied well. The objective of this research was to examine the possibility of the autonomous landing of a UAV on another UAV. The whole landing process was conducted in a simulated environment (AirSim). The Single Shot MultiBox Detector (SSD)network was trained with a customized dataset and embedded into the agent UAV to recognize the base UAV, and landing was manipulated with a proportional-integral-derivative controller with the aid of robot operating system (ROS). The base UAV was operated at the velocities of 1,2, 3, and 5 km/h and the trajectories of straight line, square, and ellipse. The results showed the SSD network detected and localized the base UAV with over 99% accuracy. The landing was successfully achieved with the four different velocities and three different trajectories. Such results demonstrate great potential of the autonomous and movable landing for future aerial robot applications