Stereo Vision For Grasping by Humanoid Robot

A NAO robot grasps a cup - it needs some coffee!

A NAO robot grasps a cup - it needs some coffee!

(Reference: DFKI GmbH)

The project aims at improving the quality of interaction between a human and a domestic robot. One of the most important tasks of such a robot will be to grasp an object and to hold it out to the person or to grasp an object that is being held by the user. A stereovision head for the humanoid robot NAO will be developed. Based on this head, a functionality allowing the robot to grasp objects autonomously from the user’s hand will be realized. The innovations addressed in the GRASPY project are the development of an embedded stereovision system for a small humanoid robot, the avoidance of collisions with the fingers while grasping, and the detection of exactly the correct moment to release the object when it is being grasped by the user.

Duration: Oct 1, 2010 - Jun 30, 2012
Partners: Aldebaran Robotics
Research area: Smart Living  
Grant number: FP7-231143

The cute little NAO, barely 60 cm tall, is produced by the French manufacturer Aldebaran Robotics and is anything but a toy. It is, instead, considered to be a robot research platform, developed primarily for scientific experiments. Since 2007, NAO has been the robot used on the teams in the RoboCup Standard Platform League, an international robot soccer competition.

GRASPY is a joint research project of DFKI and Aldebaran Robotics with the aim of developing the programs that give NAO the ability to accept an object offered by a human and, as requested, to give it back. A stereo camera system developed by the manufacturer enables NAO to recognize an object that is within his grasping range, perhaps, a pen or the handle of a plastic cup. DFKI researchers implemented visual object recognition as well as the design and execution of the grasping movements for the small robot. All calculations run on an embedded PC in NAO.

The system works with gray scale images and is largely independent of lighting conditions. When the object it recognized, the motion generator selects one of the hands to do the grasping, plans the path of the hand to a safe gripping position and executes the action. In view of the fact that the object is to be accepted from an outstretched human hand, NAO must continuously recalculate a new plan and be capable of reacting to the possibility of a moving target. A performance control system decides when the object is to be gripped, returned, or released. This involves an assessment of the recognized user inputs (voice or touch) and the object’s current location.

GRASPY is funded as an experiment under the framework of the EU ECHORD project (European Clearing House for Open Robotics Development), where the aim is to strengthen the cooperation between research institutes and industrial enterprises with a focus on robotics.

last updated 08/02/2017
to top