Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -17,6 +17,146 @@ task_categories:
|
|
17 |
|
18 |
<div align="center">
|
19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
[](https://beingbeyond.github.io/Being-H0)
|
21 |
[](https://arxiv.org/abs/2507.15597)
|
22 |
[](https://github.com/BeingBeyond/Being-H0)
|
|
|
17 |
|
18 |
<div align="center">
|
19 |
|
20 |
+
[](https://beingbeyond.github.io/Being-H0)
|
21 |
+
[](https://arxiv.org/abs/2507.15597)
|
22 |
+
[](https://huggingface.co/BeingBeyond/Being-H0)
|
23 |
+
[](./LICENSE)
|
24 |
+
|
25 |
+
</div>
|
26 |
+
|
27 |
+
<p align="center">
|
28 |
+
<img src="https://raw.githubusercontent.com/BeingBeyond/Being-H0/refs/heads/main/docs/assets/image/overview.png"/>
|
29 |
+
<p>
|
30 |
+
|
31 |
+
|
32 |
+
We introduce **Being-H0**, the first dexterous Vision-Language-Action model pretrained from large-scale human videos via explicit hand motion modeling.
|
33 |
+
|
34 |
+
## News
|
35 |
+
|
36 |
+
- **[2025-08-02]**: We release the **Being-H0** codebase and pretrained models! Check our [Hugging Face Model Hub](https://huggingface.co/BeingBeyond/Being-H0) for more details. π₯π₯π₯
|
37 |
+
- **[2025-07-21]**: We publish **Being-H0**! Check our paper [here](https://arxiv.org/abs/2507.15597). πππ
|
38 |
+
|
39 |
+
## Model Checkpoints
|
40 |
+
|
41 |
+
Download pre-trained models from Hugging Face:
|
42 |
+
|
43 |
+
| Model Type | Model Name | Parameters | Description |
|
44 |
+
|------------|------------|------------|-------------|
|
45 |
+
| **Motion Model** | [Being-H0-GRVQ-8K](https://huggingface.co/BeingBeyond/Being-H0-GRVQ-8K) | - | Motion tokenizer |
|
46 |
+
| **VLA Pre-trained** | [Being-H0-1B-2508](https://huggingface.co/BeingBeyond/Being-H0-1B-2508) | 1B | Base vision-language-action model |
|
47 |
+
| **VLA Pre-trained** | [Being-H0-8B-2508](https://huggingface.co/BeingBeyond/Being-H0-8B-2508) | 8B | Base vision-language-action model |
|
48 |
+
| **VLA Pre-trained** | [Being-H0-14B-2508](https://huggingface.co/BeingBeyond/Being-H0-14B-2508) | 14B | Base vision-language-action model |
|
49 |
+
| **VLA Post-trained** | [Being-H0-8B-Align-2508](https://huggingface.co/BeingBeyond/Being-H0-8B-Align-2508) | 8B | Fine-tuned for robot alignment |
|
50 |
+
|
51 |
+
## Dataset
|
52 |
+
|
53 |
+
We have provided the dataset for post-training the VLA model. The dataset is available in Hugging Face:
|
54 |
+
|
55 |
+
| Dataset Type | Dataset Name | Description |
|
56 |
+
|--------------|--------------|-------------|
|
57 |
+
| **VLA Post-training** | [h0_post_train_db_2508](https://huggingface.co/datasets/BeingBeyond/h0_post_train_db_2508) | Post-training dataset for pretrained Being-H0 VLA model |
|
58 |
+
|
59 |
+
## Setup
|
60 |
+
|
61 |
+
### Clone repository
|
62 |
+
|
63 |
+
```bash
|
64 |
+
git clone https://github.com/BeingBeyond/Being-H0.git
|
65 |
+
cd Being-H0
|
66 |
+
```
|
67 |
+
|
68 |
+
### Create environment
|
69 |
+
```bash
|
70 |
+
conda env create -f environment.yml
|
71 |
+
conda activate beingvla
|
72 |
+
```
|
73 |
+
|
74 |
+
### Install package
|
75 |
+
```bash
|
76 |
+
pip install flash-attn --no-build-isolation
|
77 |
+
pip install git+https://github.com/lixiny/manotorch.git
|
78 |
+
pip install git+https://github.com/mattloper/chumpy.git
|
79 |
+
```
|
80 |
+
|
81 |
+
### Download MANO package
|
82 |
+
|
83 |
+
- Visit [MANO website](http://mano.is.tue.mpg.de/)
|
84 |
+
- Create an account by clicking _Sign Up_ and provide your information
|
85 |
+
- Download Models and Code (the downloaded file should have the format `mano_v*_*.zip`). Note that all code and data from this download falls under the [MANO license](http://mano.is.tue.mpg.de/license).
|
86 |
+
- Unzip and copy the contents in `mano_v*_*/` folder to the `beingvla/models/motion/mano/` folder
|
87 |
+
|
88 |
+
## Inference
|
89 |
+
|
90 |
+
### Motion Generation
|
91 |
+
|
92 |
+
- To generate hand motion tokens and render the motion, you should use the Motion Model (`Being-H0-GRVQ-8K`) and the pretrained VLA model (`Being-H0-{1B,8B,14B}-2508`).
|
93 |
+
- You can use the following command to inference. For the `--motion_code_path`, you should use a `+` symbol to jointly specify the wrist and finger motion code paths, e.g., `--motion_code_path "/path/to/Being-H0-GRVQ-8K/wrist/+/path/to/Being-H0-GRVQ-8K/finger/"`.
|
94 |
+
- The `--hand_mode` can be set to `left`, `right`, or `both` to specify which hand to use for the task.
|
95 |
+
|
96 |
+
```bash
|
97 |
+
python -m beingvla.inference.vla_internvl_inference \
|
98 |
+
--model_path /path/to/Being-H0-XXX \
|
99 |
+
--motion_code_path "/path/to/Being-H0-GRVQ-8K/wrist/+/path/to/Being-H0-GRVQ-8K/finger/" \
|
100 |
+
--input_image ./playground/unplug_airpods.jpg \
|
101 |
+
--task_description "unplug the charging cable from the AirPods" \
|
102 |
+
--hand_mode both \
|
103 |
+
--num_samples 3 \
|
104 |
+
--num_seconds 4 \
|
105 |
+
--enable_render true \
|
106 |
+
--gpu_device 0 \
|
107 |
+
--output_dir ./work_dirs/
|
108 |
+
```
|
109 |
+
|
110 |
+
- **To inference on your own photos**: See [Camera Intrinsics Guide](https://github.com/BeingBeyond/Being-H0/blob/main/docs/camera_intrinsics.md) for how to estimate camera intrinsics and input them for custom inference.
|
111 |
+
|
112 |
+
### Evaluation
|
113 |
+
|
114 |
+
- You can use our pretrained VLA model to post-train on real robot data. When you get your post-trained model (e.g., `Being-H0-8B-Align-2508`), you can use the following commands to communicate with real robot, or evaluate the model on a robot task.
|
115 |
+
|
116 |
+
- Setup robot communication:
|
117 |
+
|
118 |
+
```bash
|
119 |
+
python -m beingvla.models.motion.m2m.aligner.run_server \
|
120 |
+
--model-path /path/to/Being-H0-XXX-Align \
|
121 |
+
--port 12305 \
|
122 |
+
--action-chunk-length 16
|
123 |
+
```
|
124 |
+
- Run evaluation on robot task:
|
125 |
+
|
126 |
+
```bash
|
127 |
+
python -m beingvla.models.motion.m2m.aligner.eval_policy \
|
128 |
+
--model-path /path/to/Being-H0-XXX-Align \
|
129 |
+
--zarr-path /path/to/real-robot/data \
|
130 |
+
--task_description "Put the little white duck into the cup." \
|
131 |
+
--action-chunk-length 16
|
132 |
+
```
|
133 |
+
|
134 |
+
## Contributing and Building on Being-H0
|
135 |
+
|
136 |
+
We encourage researchers and practitioners to leverage Being-H0 as a foundation for their own creative experiments and applications. Whether you're adapting Being-H0 to new robotic platforms, exploring novel hand manipulation tasks, or extending the model to new domains, our modular codebase is designed to support your innovations. We welcome contributions of all kinds - from bug fixes and documentation improvements to new features and model architectures. By building on Being-H0 together, we can advance the field of dexterous vision-language-action modeling and enable robots to understand and replicate the rich complexity of human hand movements. Join us in making robotic manipulation more intuitive, capable, and accessible to all.
|
137 |
+
|
138 |
+
## Citation
|
139 |
+
If you find our work useful, please consider citing us and give a star to our repository! πππ
|
140 |
+
|
141 |
+
**Being-H0**
|
142 |
+
|
143 |
+
```bibtex
|
144 |
+
@article{beingbeyond2025beingh0,
|
145 |
+
title={Being-H0: Vision-Language-Action Pretraining from Large-Scale Human Videos},
|
146 |
+
author={Luo, Hao and Feng, Yicheng and Zhang, Wanpeng and Zheng, Sipeng and Wang, Ye and Yuan, Haoqi and Liu, Jiazheng and Xu, Chaoyi and Jin, Qin and Lu, Zongqing},
|
147 |
+
journal={arXiv preprint arXiv:2507.15597},
|
148 |
+
year={2025}
|
149 |
+
}
|
150 |
+
```
|
151 |
+
|
152 |
+
# Being-H0: Vision-Language-Action Pretraining from Large-Scale Human Videos
|
153 |
+
|
154 |
+
<p align="center">
|
155 |
+
<img src="https://raw.githubusercontent.com/BeingBeyond/Being-H0/refs/heads/main/docs/assets/image/being-h0-black.png" width="300"/>
|
156 |
+
<p>
|
157 |
+
|
158 |
+
<div align="center">
|
159 |
+
|
160 |
[](https://beingbeyond.github.io/Being-H0)
|
161 |
[](https://arxiv.org/abs/2507.15597)
|
162 |
[](https://github.com/BeingBeyond/Being-H0)
|