The Sliced Pineapple Grid Feature for Predicting Grasping Affordances

Mikkel Tang Thomsen, Dirk Kraft, Norbert Krüger

Publikation: Kapitel i bog/rapport/konference-proceedingKonferencebidrag i proceedingsForskningpeer review

Abstrakt

The problem of grasping unknown objects utilising vision is addressed in this work by introducing a novel feature, the Sliced Pineapple Grid Feature (SPGF). The SPGF encode semi-local surfaces and allows for distinguishing structures such as “walls”,“edges” and “rims”. These structures are shown to be important when learning successful grasping affordance predictions. The SPGF feature is used in combination with two different grasp affordance learning methods and achieve grasp success-rates of up to 87% for a combined varied object set. For specific object classes within the object set, success-rates of up to 96% is achieved. The results also show how two different grasp types can complement each other and allow grasping of objects that are not graspable by one of the types.
OriginalsprogEngelsk
TitelInternational Joint Conference on Computer Vision, Imaging and Computer Graphics
RedaktørerJosé Braz, Nadia Magnenat-Thalmann, Paul Richard, Lars Linsen, Alexandru Telea, Sebastian Battiato, Francisco Imai
ForlagSpringer
Publikationsdato2017
Sider418-438
DOI
StatusUdgivet - 2017
Begivenhed11th International Joint Conference on Computer Vision, Imaging and Computer Graphics - Rome, Italien
Varighed: 27. feb. 201627. feb. 2016
Konferencens nummer: 11

Konference

Konference11th International Joint Conference on Computer Vision, Imaging and Computer Graphics
Nummer11
LandItalien
ByRome
Periode27/02/201627/02/2016
NavnCommunications in Computer and Information Science
Vol/bind693
ISSN1865-0929

Citationsformater