Grasp'D: Differentiable Contact-rich Grasp Synthesis for Multi-fingered Hands

Prerequisites

• Python 3.6 or higher
• PyTorch 1.4.0 or higher
• Pixar USD lib (optional, for advanced visualization)

Pre-built USD Python libraries can be downloaded from https://developer.nvidia.com/usd, once they are downloaded you should follow the corresponding instructions to add them to your PYTHONPATH environment variable. Besides using the provided basic visualizer implemented using pyvista, Grasp'D can generate USD files for rendering, e.g., in NVIDIA Omniverse™ or usdview.

Environment setup

Using anaconda set up a python environment with conda env create -vv -f grasping/docker/environment.yml or use grasping/docker/Dockerfile to build a docker image.

Download dataset

You can download discretized SDFs for the YCB objects here. Unzip and copy these to grasping/data/ycb. The folder structure should look like:

grasping/data/ycb
    |-- 002_master_chef_can
    |-- 003_cracker_box
    |-- ...

Getting started

To collect grasps for the YCB banana, try running python grasping/scripts/collect_grasps.py collector_config=011a_banana. See grasping/conf/collect_grasps for configuration options.