Shape Warping and Analysis

When a robot system encounters objects in its environment for which no specific sensory, geometric, and semantic model knowledge are available, its perceptual capabilities are strongly challenged. This situation would be very common in the ordinary human living environment, that is, in private homes as well as at public places or at many work places.

Sensible interaction with unknown objects, in particular their correct manipulation, can be guided through relations established between these objects and already known exemplars of a similar kind. The critical question is: which parts or points of a new object correspond to which parts or points of a known, prototypical object? It is answered through computation of a suitable warping, driven by global and local shape similarities, of the prototype shape onto the new shape.

Object parts with particular functional significance can be detected through a suitable warping of prototypes onto new objects. Even detailed relations to a known exemplar can be uncovered. In the image, estimated point correspondences between the handles of two differing detergent bottles are marked with equal color; the color gradient codes the Cartesian coordinates in the prototype system.

Functionally correct handling, e.g. of an unknown spray bottle shown in the image, is enabled through transfer of contact points (b) of a functional example grasp (a) made on a known object of the same kind onto the new case (c). Based on the new contacts, an analogous functional grasp can then be planned (d).   Statistics of the degree of all local deformations between a prototype object and an unknown object quantifies their shape similarity. Based on this shape similarity measure, the unknown object can be determined to belong to either the same or a different category as the prototype. The image shows a tabletop scene (top) along with data points from its depth image (bottom) where all cup-like objects found in this way are marked in red.

Selected Publications

Th. Stouraitis, U. Hillenbrand, and M. A. Roa. Functional power grasps transferred through warping and replanning. Proceedings IEEE International Conference on Robotics and Automation -- ICRA 2015.

U. Hillenbrand. Detecting objects of a category in range data by comparing to a single geometric prototype. Proceedings IEEE/RSJ International Conference on Intelligent Robots and Systems -- IROS 2013.

U. Hillenbrand and M. A. Roa. Transferring functional grasps through contact warping and local replanning. Proceedings IEEE/RSJ International Conference on Intelligent Robots and Systems -- IROS 2012.

Dr. Ulrich Hillenbrand German Aerospace Center Institute of Robotics and Mechatronics, Perception and Cognition Oberpfaffenhofen-Weßling Tel.: +49 8153 28-3501 Fax: +49 8153 28-1134