---
license: mit
task_categories:
- robotics
tags:
- robotics
- manipulation
- in-hand-rotation
- allegro-hand
- reinforcement-learning
---

# Allegro Hand In-Hand Rotation - Grasp Pose Dataset

This dataset contains pre-generated grasp poses for the **Allegro Hand V4** platform, designed for in-hand object rotation tasks using the HORA (In-Hand Object Rotation via Rapid Motor Adaptation) algorithm.

## Dataset Description

This dataset provides reliable initial grasp configurations for various object geometries (Ball, Cylinder, Cube, etc.) used in training and evaluating in-hand manipulation policies with the Allegro Hand.

### Dataset Contents

The dataset consists of grasp pose cache files stored as NumPy arrays (`.npy` format):

- Pre-computed stable grasp configurations for multiple object types
- Optimized for both **right** and **left** Allegro Hand V4 configurations
- Generated through simulation-based grasp quality evaluation

### File Structure

```
cache/
├── [object_type]_[hand_type]_grasp_poses.npy
└── ...
```

## Usage

> **⚠️ Important:** This dataset is designed to be used with the main repository. For complete setup, training, and deployment instructions, please refer to:
>
> **👉 [Wonikrobotics-git/allegro_inhand_rotation](https://github.com/Wonikrobotics-git/allegro_inhand_rotation)**

### Quick Setup

1. Clone the main repository:
   ```bash
   git clone https://github.com/Wonikrobotics-git/allegro_inhand_rotation.git
   cd allegro_inhand_rotation
   ```

2. Download this dataset and extract the `cache/` folder into the project root:
   ```
   allegro_inhand_rotation/
   ├── cache/              # This dataset
   ├── configs/
   ├── hora/
   └── ...
   ```

3. Follow the [main repository README](https://github.com/Wonikrobotics-git/allegro_inhand_rotation) for:
   - Environment setup (Isaac Gym, ROS 2)
   - Training procedures
   - Simulation testing
   - Real-world deployment

### Generating Custom Grasp Poses

You can also generate grasp poses from scratch. Refer to the **"Generate Grasping Poses"** section in the [main repository](https://github.com/Wonikrobotics-git/allegro_inhand_rotation#generate-grasping-poses) for detailed instructions.

## Related Resources

- **Main Repository:** [allegro_inhand_rotation](https://github.com/Wonikrobotics-git/allegro_inhand_rotation)
- **Hardware Controller:** [allegro_hand_ros2](https://github.com/WonikRobotics-git/allegro_hand_ros2)
- **Original Algorithm:** [HORA by Haozhi Qi](https://haozhi.io/hora/)

## System Requirements

- Compatible with Allegro Hand V4 (Right and Left)
- Designed for use with Isaac Gym 4.0 simulator
- Supports ROS 2 Humble for real-world deployment

## Citation

If you use this dataset in your research, please cite:

```bibtex
@misc{allegro_inhand_rotation_dataset,
  title={Allegro Hand In-Hand Rotation Grasp Pose Dataset},
  author={Wonik Robotics},
  year={2025},
  howpublished={\url{https://huggingface.co/datasets/Wonik-Robotics/allegro_inhand_rotation}}
}
```

And the original HORA work:

```bibtex
@article{qi2022hora,
  title={In-Hand Object Rotation via Rapid Motor Adaptation},
  author={Qi, Haozhi and Kumar, Ashish and Calandra, Roberto and Ma, Yi and Malik, Jitendra},
  journal={arXiv preprint arXiv:2210.04887},
  year={2022}
}
```

## License

This dataset is released under the MIT License.

- Dataset by [**Wonik Robotics**](https://github.com/Wonikrobotics-git) (© 2025)

## Contact & Support

For questions or issues:
- Open an issue in the [main repository](https://github.com/Wonikrobotics-git/allegro_inhand_rotation)
- Visit [Wonik Robotics](https://github.com/Wonikrobotics-git)