Announcing Grasp Package based on Topographic Features

From David Fischinger via ros-users@

We are pleased to announce a new package for grasp calculation on unknown and known objects.

This package receives a point cloud representing objects and identifies where to best place the gripper. The algorithm does not require segmentation or a-priori knowledge about the objects. It has already been employed on various platforms, including a PR2 [4], a Kuka LWR [5], a Schunk arm [6] and the service robot Hobbit [7].

More details, a scientific foundation and evaluation results can be found in an IJRR journal publication from August 2015 [1], a more technical description and a simple getting started guide can be found at [2]. Code is available on GitHub [3].

Currently Indigo is supported.

Links:

1. http://ijr.sagepub.com/content/34/9/1167.full.pdf+html ? IJRR publication 2015 or http://users.acin.tuwien.ac.at/dfischinger/files/IJRR_FinalRevision.pdf (final revision)
2. http://wiki.ros.org/haf_grasping ? Technical description, Getting started
3. https://github.com/davidfischinger/haf_grasping ? Code on GitHub

Videos:

PR2, unknown objects

Kuka arm, known object

Schunk arm, unknown objects in box

Service robot Hobbit