README.md

# TTP: Human-Centric Transferable Tactile Pre-Training for Dexterous Robotic Manipulation

[![Blog](https://img.shields.io/badge/Blog-TTP-green)](https://research.beingbeyond.com/ttp)
[![Paper](https://img.shields.io/badge/arXiv-Paper-b31b1b.svg)](https://arxiv.org/pdf/2607.01067)
[![Models](https://img.shields.io/badge/🤗%20Hugging%20Face-Models-yellow)](https://huggingface.co/BeingBeyond/TTP)
[![Data](https://img.shields.io/badge/🤗%20Hugging%20Face-Data-orange)](https://huggingface.co/datasets/BeingBeyond/H-Tac_Sample)
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](./LICENSE)

## Model Checkpoints and Datasets

Download models from Hugging Face:

| Model Type | Model Name | Description |
|------------|------------|-------------|
| **VLA Pre-trained** | [TTP-pretrained](https://huggingface.co/BeingBeyond/TTP/tree/main/pretrained) | Base vision-language-action model (preview) |
| **VLA Post-trained** | [TTP-LIBERO](https://huggingface.co/BeingBeyond/TTP/tree/main/libero) | Post-trained on LIBERO benchmark |
| **VLA Post-trained** | [TTP-LIBERO-plus](https://huggingface.co/BeingBeyond/TTP/tree/main/libero-plus) | Post-trained on LIBERO, zero-shot evaluated on LIBERO-plus |
| **VLA Post-trained** | [TTP-RoboCasa](https://huggingface.co/BeingBeyond/TTP/tree/main/robocasa) | Post-trained on RoboCasa 24 tasks |

The whole datasets of H-Tac will be coming soon. For now, you can download a small subset of H-Tac from Hugging Face for testing and evaluation: [H-Tac_Sample](https://huggingface.co/datasets/BeingBeyond/H-Tac_Sample)

## Setup

### Clone repository

```bash
git clone https://github.com/BeingBeyond/TTP.git
cd TTP
```

### Create environment

```bash
conda create -n beingh python=3.10
conda activate beingh
```

### Install package

```bash
pip install -r requirements.txt
pip install flash-attn --no-build-isolation
```

## Pre-Training

```bash
bash scripts/train/train_human_tactile.sh
```

## Post-Training

```bash
bash scripts/train/train_libero_example.sh
bash scripts/train/train_robocasa.sh
```

## Evaluation

```bash
bash scripts/eval/eval-libero-fast.sh
bash scripts/eval/eval-libero-plus.sh
bash scripts/eval/eval-robocasa-multiprocess.sh
```

## Acknowledgments

TTP builds on the following excellent open-source projects:

- [InternVL](https://github.com/OpenGVLab/InternVL): Vision-Language model backbone
- [Bagel](https://github.com/ByteDance-Seed/Bagel): Training framework
- [Qwen](https://github.com/QwenLM/Qwen): Language model and MoE expert
- [LIBERO](https://github.com/Lifelong-Robot-Learning/LIBERO): Benchmark for lifelong robot learning
- [RoboCasa](https://github.com/robocasa/robocasa): Large-scale simulation benchmark for everyday tasks
- [Being-H0.5](https://github.com/BeingBeyond/Being-H): Scaling Human-Centric Robot Learning for Cross-Embodiment Generalization

We thank the authors for their contributions to the robotics and machine learning communities.

## Citation

If you find this work useful in your research, please consider citing us!

```bibtex
@article{beingbeyond2026ttp,
  title={Human-Centric Transferable Tactile Pre-Training for Dexterous Robotic Manipulation},
  author={Zhang, Chi and Cai, Penglin and Xi, Ziheng and Yuan, Haoqi and Luo, Hao and Zhang, Wanpeng and Zheng, Sipeng and Xu, Chaoyi and Lu, Zongqing},
  journal={arXiv preprint arXiv:2607.01067},
  year={2026}
}
```