AutoURDF: Unsupervised Robot Modeling from Point Cloud Frames Using Cluster Registration

Project page | Paper

This repository contains the official implementation associated with the paper "AutoURDF: Unsupervised Robot Modeling from Point Cloud Frames Using Cluster Registration".

Run

Environment

# setup a vitural environment
conda create -n autourdf python=3.9
conda activate autourdf

We used pytorch3D, pybullet, open3d, pyvista.

Prepare

For pytorch and pytroch3D installation, please follow this instruction: pytorch3D installation.

On my computer

Ubuntu 22.04 & CUDA 12.4
python==3.9
torch==2.4.1
torchvision==0.19.1
pytorch3d==0.7.7

Other packages

pip install -r requirements.txt

Data Collection

Collect point cloud sequences, by default wx200_5, 5 sequences

bash scripts/dataset.sh

Train Registration Model

Registration, by default wx200_5, run 5 sequences

bash scripts/registration.sh

URDF Results

Output URDF, by default wx200_5, run 1 or 5 sequences, unknown DoF infomation with 5 sequences, 50 frames

python PointCloud/coord_map.py --robot wx200_5 --unknown_dof

with only 1 sequence, 10 frames

python PointCloud/coord_map.py --robot wx200_5 --end_video 1 --unknown_dof

Robot URDF Demos

Acknowledgments

We sincerely thank Changxi Zheng and Ruoshi Liu for their invaluable feedback.

BibTex

@article{lin2024autourdf,
  title={AutoURDF: Unsupervised Robot Modeling from Point Cloud Frames Using Cluster Registration},
  author={Lin, Jiong and Zhang, Lechen and Lee, Kwansoo and Ning, Jialong and Goldfeder, Judah and Lipson, Hod},
  journal={arXiv preprint arXiv:2412.05507},
  year={2024}
}

Website Template

Our website template is based on:

• BundleSDF
• Which in turn was based on Nerfies

Website License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.