The tutorial covers the main use cases of the dex-net package: Database Manipulation through the dex-net Command Line Interface (CLI).

Click on the link or scroll down to get started!

Running Python Scripts

For best performance, we recommend running all scripts from the dex-net root directory:

cd /path/to/your/dex-net
python apps/dexnet_cli.py

Database Manipulation

The Dex-Net CLI includes functionality for opening, reading, and writing HDF5 databases of 3D object models, parallel-jaw grasps, and grasp robustness metrics.

The tutorial walks through the Dex-Net Command Line Interface dexnet_cli.py which is included in the dex-net repository under the apps/ directory. The tutorial is hosted on Google Docs with support for comments to help us improve it.

Point Cloud Dataset Generation

Robust grasping policies based on Grasp Quality Convolutional Neural Networks (GQ-CNNs) may be useful for planning grasps on novel objects with a physical robot. This requires training GQ-CNNs on a dataset of synthetic point clouds, grasps, and grasp robustness metrics such as the Dex-Net 2.0 dataset.

It may be beneficial to train GQ-CNNs on custom robot grippers and datasets of 3D object models based on the application. We are working toward making this functionality publicly available. If you are interested in this, please email Jeff Mahler (jmahler@berkeley.edu) with the subject line: “Interested in GQ-CNN Dataset Generation.”