README.md

---
license: mit
pipeline_tag: robotics
tags:
- robotics
- grasping
- learning-from-humans
- dexterous-manipulation
---

# Human Universal Grasping (HUG)

HUG is a flow-matching model that generates diverse human grasps for any user-specified object in a single RGB-D image. By learning from a large-scale egocentric dataset of human grasps (1M-HUGs), the model can predict human-like grasps that can be retargeted to various robot hands for zero-shot manipulation.

- 📄 **Paper**: [Human Universal Grasping](https://arxiv.org/abs/2606.17054)
- 🌐 **Website**: [grasping.io](https://grasping.io)
- 💻 **Code**: [github.com/kevinywu/hug](https://github.com/kevinywu/hug)

## Installation

The codebase is tested on Ubuntu 22.04/24.04, CUDA 12.8, PyTorch 2.9.1, and Python 3.10.

```bash
# 1) Environment setup
conda env create -f environment.yaml && conda activate hug
pip install torch==2.9.1 torchvision==0.24.1 torchaudio==2.9.1 --index-url https://download.pytorch.org/whl/cu128
pip install torch-cluster -f https://data.pyg.org/whl/torch-2.9.1+cu128.html
pip install --no-build-isolation git+https://github.com/mattloper/chumpy.git@580566e
pip install -e .
```

Please refer to the [official repository](https://github.com/kevinywu/hug) for instructions on downloading required assets like MANO models.

## Usage

### Download Weights

Download the full model weights (`.safetensors`) using the `huggingface-cli`:

```bash
hf download kevinywu/hug hug_full.safetensors --local-dir checkpoints/
```

### Inference

HUG predicts human grasps in MANO form. You can run the interactive application to predict grasps for objects in the camera frame:

```bash
CKPT=checkpoints/hug_full.safetensors
DATA=data/hug_bench/

# Launch the app: click an object to predict a grasp
python -m hug.app --checkpoint-path "$CKPT" --dataset-path "$DATA" --save-pred

# Visualize saved predictions
python -m hug.visualize_predictions --dataset-path "$DATA"
```

## Citation

```bibtex
@article{wu2026hug,
  title={Human Universal Grasping},
  author={Kevin Yuanbo Wu and Tianxing Zhou and Isaac Tu and Billy Yan and Irmak Guzey and David Fouhey and Dandan Shan and Lerrel Pinto},
  journal={arXiv preprint arXiv:2606.17054},
  year={2026}
}
```