Current aerial manipulation and grasping with flying robots lack versatility and agility. This paper presents the first example of a soft aerial manipulator with a fully onboard perception pipeline, capable of robustly localizing and grasping diverse objects. The system uses a novel passively closed tendon-actuated soft gripper for fast closure, compensating for position errors and dampening reaction forces. An onboard perception pipeline combines a neural-network-based semantic keypoint detector, a 3D object pose estimator, and a fixed-lag smoother. Experiments demonstrate dynamic, high-speed, and versatile grasping of various objects in indoor and outdoor environments, achieving speeds up to 2.0 m/s – the fastest vision-based grasp reported. Motion-capture-based grasps of moving targets (up to 0.3 m/s) were also demonstrated.

Samuel Ubellacker, Aaron Ray, James M. Bern, Jared Strader, Luca Carlone