Documentation: https://berkeleyautomation.github.io/gqcnn
The gqcnn Python package is for training and analysis of Grasp Quality Convolutional Neural Networks (GQ-CNNs).
This package is part of the Dexterity Network (Dex-Net) project: https://berkeleyautomation.github.io/dex-net
Created and maintained by the AUTOLAB at UC Berkeley: https://autolab.berkeley.edu
See the website at https://berkeleyautomation.github.io/gqcnn for installation instructions and API Documentation.
As of Feb. 1, 2018, the code is licensed according to the UC Berkeley Copyright and Disclaimer Notice. The code is available for educational, research, and not-for-profit purposes (for full details, see LICENSE). If you use this code in a publication, please cite:
Mahler, Jeffrey, Jacky Liang, Sherdil Niyaz, Michael Laskey, Richard Doan, Xinyu Liu, Juan Aparicio Ojea, and Ken Goldberg. "Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics." Robotics: Science and Systems (2017). Boston, MA.
Our GQ-CNN training datasets and trained models can be downloaded from this link.
We developed a ROS service for grasp planning with GQ-CNNs. The service takes as input a color image, depth image, camera info topic, and bounding box for the object in image space, and returns a parallel-jaw gripper pose relative to the camer along with a predicted probability of success. This has been tested on our setup with ROS Jade on Ubuntu 14.04
To illustrate using our ROS service, we've shared the ROS node that we use to plan grasps for and control an ABB YuMi on our local setup. This file should be considered READ-ONLY as it uses parameters specific to our setup. If you have interest in replicating this functionality on your own robot, please contact Jeff Mahler (jmahler@berkeley.edu) with the subject line: "Interested in GQ-CNN ROS Service".