Vision-based Autonomous UAS Landing on a Stochastically Moving Platform
Abhishek Shastry, Inderjit Chopra, Anubhav Datta · 2021
This paper describes the development of avionics and control systems, followed by flight demonstration for vision-based landing of a multirotor UAS on a moving platform replicating a ship-deck. The UAS is a quadrotor designed and fabricated in-house with an integrated avionics package developed for autonomous vision-based navigation. A monocular camera is used for detection and pose estimation of a moving platform. The platform replicates ship-deck motion using six degrees of freedom capable of oscillatory motions upto 10 cm in linear and 30 degrees in angular amplitude with speeds upto 50 cm/s. Ship-deck motions of progressively increasing intensity were obtained from ONR and simulated using the moving platform to understand the problems involved in estimation and tracking of large stochastic motions. Results were verified with an independent Vicon system. The key conclusion is that it is possible to track a moving ship-deck with vision alone. Indeed vision based ship-deck position estimation was found to be near perfect with errors of 1-2 cm. Ship-deck tracking shows a phase lag error that increases with increasing frequency of ship-deck motion. It was found that this error can be decreased by estimating ship-deck's velocity and using it as a feedback in tracking controller.