Plane-based Humanoid Robot Navigation and Object Model Construction for Grasping

Pavel Gritsenko, Igor Gritsenko, Askar Seidakhmet, Bogdan Kwolek; Proceedings of the European Conference on Computer Vision (ECCV) Workshops, 2018, pp. 0-0

Abstract


In this work we present an approach to humanoid robot navigation and object model construction for grasping using only RGB-D data from an onboard depth sensor. A plane-based representation is used to provide a high-level model of the workspace, to estimate both the global robot pose and pose with respect to the object, and to determine the object pose as well as its dimensions. A visual feedback is used to achieve the desired robot pose for grasping. In the pre–grasping pose the robot determines the object pose as well as its dimensions. In such a local grasping approach, a simulator with our high-level scene representation and a virtual camera is used to fine-tune the motion controllers as well as to simulate and validate the process of grasping. We present experimental results that were obtained in simulations with virtual camera and robot as well as with real humanoid robot equipped with RGB-D camera, which performed object grasping in low-texture layouts.

Related Material


[pdf]
[bibtex]
@InProceedings{Gritsenko_2018_ECCV_Workshops,
author = {Gritsenko, Pavel and Gritsenko, Igor and Seidakhmet, Askar and Kwolek, Bogdan},
title = {Plane-based Humanoid Robot Navigation and Object Model Construction for Grasping},
booktitle = {Proceedings of the European Conference on Computer Vision (ECCV) Workshops},
month = {September},
year = {2018}
}