1 Motion-Aware Underwater Robotic Grasping

Grasping moving objects underwater introduces an entirely new set of challenges in contrast with structured terrestrial or industrial environments. Underwater environments are highly unstructured with waves causing the water particles to move in circular orbits that diminishes with depth which results the object to oscillate in a unique way. This causes a typically high confident and stable grasp to become an unstable and of F-target grasp when the object pose changes, rendering the existing grasping methods useless. Dynamic grasp generation needs to be adaptive, real time, robust to noise, and have low computing needs. We present a novel dynamic grasping framework that processes a streaming point cloud generated from depth video in real time to generate 6- DO F 4D grasps. By tracking the successive position of point cloud, we predict the current motion of objects. Using the relative position and motion predictor, we decide on the most confident grasp. We recorded the motion of multiple objects under three different wave conditions at the O.H. Hinsdale Wave Research Laboratory at Oregon State University, USA, and tested them in a simulation environment using pyBullet. Our method achieved an average success rate of 70% across four different object configurations under three different wave conditions.