Scaled Experiments in Vision-Based Approach and Landing In High Sea States

Landings on ship decks in higher sea states pose significant challenges to UAV operators. This paper describes the development and testing of a self-contained vision-based autonomous deck landing system. A single monocular smart camera was used for detection and 6DOF pose estimation of a recursive AprilTag marker array. The fiducial marker array was detected and localized at 48 Hz from distances up to 5 m. The scalable fiducial marker system and wide field of view camera used were found to improve deck observability and the quality of deck state estimates over a wider range of distances compared to non-scalable visual aids. Fusion of vision and inertial sensor data was performed using an Unscented Kalman Filter for relative deck state estimation. Tau trajectories were generated and followed using an explicit model following controller created from identified vehicle dynamic models. Performance of the vision system and estimator was measured using two separate motion capture systems for ground truth in hovering and landing flight tests. Fifteen successful autonomous landings were performed on the model ship deck in scaled sea states as high as six.