Grasp hypothesis generation for parametric object 3D point cloud models

Abstract

Grasping by Components (GBC) is a very important component of any scalable and holistic grasping system that abstracts point cloud object data to work with arbitrary shapes with no apriori data. Superquadric representation of point cloud data is a suitable parametric method for representing and manipulating point cloud data. Most Superquadrics based grasp hypotheses generation methods perform the step of classifying the parametric shapes into one of different simple shapes with apriori established grasp hypotheses. Such a method is suitable for simple scenarios. But for a holistic and scalable grasping system, direct grasp hypothesis generation from Superquadric representation is crucial. In this paper, we present an algorithm to directly estimate grasp points and approach vectors from Superquadric parameters. We also present results for a number of complex Superquadric shapes and show that the results are in line with grasp hypotheses conventionally generated by humans.