🚀CGDF: Constrained Grasp Diffusion Fields

IROS 2024

Gaurav Singh* ¹, Sanket Kalwar* ¹, Md Faizal Karim¹, Bipasha Sen², Nagamanikandan Govindan¹, Srinath Sridhar³, K Madhava Krishna¹

*denotes equal contribution, ¹ International Institute of Information Technology Hyderabad, ² MIT CSAIL, ³ Brown University

This is the official implementation of the paper "Constrained 6-DoF Grasp Generation on Complex Shapes for Improved Dual-Arm Manipulation" accepted at IROS 2024

Installation

Clone the repository

The pretrained checkpoint is in the demo/data/models/cgdf_v1 directory. Please use git lfs to clone the repository as follows:

git lfs install
git clone https://github.com/constrained-grasp-diffusion/constrained-grasp-diffusion.git

CREATING THE ENVIRONMENT

conda create --name cgdf -y python=3.8
conda activate cgdf
bash build_env.sh

Running the Demo

# Uniform grasp generation
CUDA_VISIBLE_DEVICES=0 python scripts/sample/generate_6d_grasp_poses.py --n_grasps 300 --model cgdf_v1 --input demo/data/meshes/15847850d132460f1fb05d58f51ec4fa.obj  

# Part-constrained grasp generation
CUDA_VISIBLE_DEVICES=0 python scripts/sample/generate_part_guided_6d_grasp_poses.py --n_grasps 300 --model cgdf_v1 --input demo/data/meshes/15847850d132460f1fb05d58f51ec4fa.obj

The generated grasps are saved as a mesh containing the object and gripper markers in output_mesh.obj.

👏 Acknowledgement

This repository is heavily based on grasp_diffusion and also borrows code from Convolutional Occupancy Networks. We thank the authors for releasing their code.

📜 BibTeX

If you find our work useful, please consider citing us!

@article{singh2024constrained,
  title={Constrained 6-DoF Grasp Generation on Complex Shapes for Improved Dual-Arm Manipulation},
  author={Singh, Gaurav and Kalwar, Sanket and Karim, Md Faizal and Sen, Bipasha and Govindan, Nagamanikandan and Sridhar, Srinath and Krishna, K Madhava},
  journal={IROS 2024},
  year={2024}
}