End-to-End Learning to Grasp via Sampling From Object Point Clouds

The ability to grasp objects is an essential skill that enables many robotic manipulation tasks. Recent works have studied point cloud-based methods for object grasping by starting from simulated datasets and shown promising performance in real-world scenarios. Nevertheless, of them still rely on ad-hoc geometric heuristics generate candidates, which fail generalize with significantly different shapes respect those observed during training. Several approaches exploit complex multi-stage learning strategies local neighborhood feature extraction while ignoring semantic global information. Furthermore, they are inefficient terms number training samples time required inference. In this letter, we propose end-to-end solution 6-DOF parallel-jaw grasps the 3D partial view object. Our Learning Grasp (L2G) method gathers information input cloud through a new procedure combines differentiable sampling strategy identify visible contact points, encoder leverages cues. Overall, L2G guided multi-task objective generates diverse set optimizing sampling, regression, classification. With thorough experimental analysis, show effectiveness as well its robustness generalization abilities.