Automatic Layered 3D Reconstruction of Simplified Object Models for Grasping.

Most grasping systems require complete CAD models as input, but building these models automatically is not straightforward. We propose a strategy for approximating complete models of previously unknown objects from a single 3D scan of a cluttered scene. We achieve this by exploiting common symmetries in objects of daily use, and by handling over- and under-segmentation issues. The idea is to use 3D shape primitives by breaking up objects into parts than can be approximated by boxes and cylinders. To validate the reconstruction, we present a set of experimental results using real-world data-sets containing a large number of objects from different scan-points.