Toward Vision-Based Autonomous Landings on Stochastic Ship Decks

This work is a study of stochastic ship deck motions and how to land on them. The emphasis is on the minimum tools that are necessary and sufficient to carry out the task. A U.S. Navy–provided dataset is used to develop a ship-deck dynamics model whose frequency spectrum and statistical behavior match those of a generic DDG-51-class ship. The model drives a Stewart platform that reproduces this motion in the laboratory. Drawing on the operating principle of the beartrap recovery system, it is shown that only the surge and sway motions of the deck need to be tracked for landing. A fiducial-based monocular vision system is then developed that can estimate the deck position to within 1.2 cm. An in-house-fabricated quadrotor aircraft fitted with this vision system is used to demonstrate that a single camera (with a cooperative fiducial on deck) and a 0.15 Hz bandwidth control system are all that is needed to land on stochastic ship decks. This is a proof-of-concept work in a laboratory environment; validation on a real-ship deck is deferred to future work.