/grasp_diffusion

Pytorch implementation of diffusion models on Lie Groups for 6D grasp pose generation https://sites.google.com/view/se3dif/home

Primary LanguagePythonMIT LicenseMIT

Conda Environment Setup

conda create -n analogical_grasping python=3.10;
conda activate analogical_grasping;
conda install -c conda-forge scikit-sparse;
pip install -e .;

git clone git@github.com:TheCamusean/mesh_to_sdf.git;
cd mesh_to_sdf; pip install -e .; cd ..;

Download Data and Trained Models

Refer to original Readme to download data and trained models.

Train

Train pointcloud conditioned model

python scripts/train/train_pointcloud_6d_grasp_diffusion.py

Train partial pointcloud conditioned model

python scripts/train/train_partial_pointcloud_6d_grasp_diffusion.py

If you are training on a machine with a display, you can add --summary 1 to the commands above to log visualizations of the generated grasps and SDF during training.

To overfit a single object for debugging, you can add --overfit_one_object 1.