A Billion Ways to Grasp: An Evaluation of Grasp Sampling Schemes on a Dense, Physics-Based Grasp Data Set

Robot grasping is often formulated as a learning problem. With the increasing speed and quality of physics simulations, generating large-scale data sets that feed algorithms becoming more popular. An overlooked question how to generate grasps make up these sets. In this paper, we review, classify, compare different grasp sampling strategies. Our evaluation based on fine-grained discretization SE(3) uses physics-based simulation evaluate robustness corresponding parallel-jaw grasps. Specifically, consider than 1 billion for each 21 objects from YCB set. This dense set lets us existing schemes w.r.t. their bias efficiency. experiments show some popular contain significant do not cover all possible ways an object can be grasped. The available at https://bit.ly/2HWEI2r .