Documentation: https://berkeleyautomation.github.io/gqcnn.
We're excited to announce version 1.0, which brings the GQ-CNN package up to date with recent research in the Dexterity-Network (Dex-Net) project. Version 1.0 introduces support for:
- Dex-Net 4.0: Composite policies that decide whether to use a suction cup or parallel-jaw gripper.
- Fully Convolutional GQ-CNNs: Fully convolutional architectures that efficiently evaluate millions of grasps faster than prior GQ-CNNs.
Version 1.0 also provide a more robust ROS grasp planning service that includes built-in pre-processing.
- Support for training GQ-CNNs on Dex-Net 4.0 parallel jaw and suction datasets.
- Support for faster Fully Convolutional GQ-CNNs (FC-GQ-CNNs).
- More robust ROS policy with integrated pre-processing.
- Improved interface for training GQ-CNNs and evaluating policies.
- Faster training due to improved parallelism in data prefetch/pre-processing.
- Easy-to-use shell scripts for replication of published results from Dex-Net {2.0,2.1,3.0,4.0} and FC-GQ-CNN.
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.
Our GQ-CNN training datasets and trained models can be downloaded from this link.
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 any part of this code in a publication, please cite the appropriate Dex-Net publication.