Direct Neural Network Based UAV Control Algorithm for Moving Landing Site Approach Trajectory Following Using Computer-Vision Based Tracking

摘要

Landing site tracking using on-board UAV computer vision system provides angular coordinates of the landing site. However, the traditional on-board approach course control system assumes multiple steps of consequent computations necessary to issue the final directives for aircraft actuators that include current position estimation, intercept trajectory generation and finally low-level UAV control to follow the computed trajectory. Direct neural network-based algorithm is proposed to convert 2D angular measurements of the target and INS readings into control signals for fixed-wing UAV actuators that would lead the aircraft to landing site using a specified approach pattern without any intermediate computation steps. The algorithm is robust, fast and simple and easy to implement in hardware.