Grasp it Like a Pro 2.0: A Data-Driven Approach Exploiting Basic Shapes Decomposition and Human Data for Grasping Unknown Objects

With the improvements in the computational and physical intelligence of robots, they are now capable of operating in real-world environments. However, manipulation and grasping capabilities are still areas that require significant improvements. To address this, we propose a novel data-driven grasp planning algorithm called Grasp it Like a Pro 2.0.
Our algorithm leverages a small set of human demonstrations to transfer the skills of a human operator to a robot for grasping of previously unseen objects. By decomposing objects into basic shapes, our algorithm generates candidate grasps that can generalize to different object’s geometry. The algorithm selects the grasp to execute based on a selection policy that maximizes a novel grasp quality metric introduced in this work. This metric considers the complex interdependencies between the predicted grasp, the local approximation produced by the basic shape decomposition, and the gripper used for the grasp to rank the generated candidate grasps.
We compare the method with several baselines with different grippers and unknown objects. Results demonstrate that our method can generate and select high-quality and robust grasps, achieving a success rate of 94.0% on 150 grasps of 30 different objects with a soft underactuated robotic hand, and an 85.0% success rate on 80 grasps of 16 different objects with a rigid gripper.