---
extra_gated_prompt: "You agree to not use the model to conduct experiments that cause harm to human subjects."
extra_gated_fields:
  Company/Organization: text
  Country: country

license: apache-2.0
library_name: transformers
pipeline_tag: robotics
---
# RoboBrain-X-Dataset
The RoboBrain-X Dataset is a comprehensive embodied robotics dataset featuring two dual-arm robots, Agilex and R1lite, across diverse scenarios with detailed annotations.

# Get started 🔥    
## Download the Dataset 
To download the full dataset, you can use the following code. If you encounter any issues, please refer to the official Hugging Face documentation.
```
# Make sure you have git-lfs installed (https://git-lfs.com)
git lfs install
# When prompted for a password, use an access token with write permissions.
# Generate one from your settings: https://huggingface.co/settings/tokens
git clone https://huggingface.co/datasets/BAAI/RoboBrain-X-Dataset
```

## Dataset Structure
### Folder hierarchy

```
data
├── RoboBrain_Robotic_AGXCompany
│   ├── annotation
│   │   ├── train.json
│   │   ├── ...
│   ├── video
│   │   ├── 0
│   │   ├── 1
│   │   ├── ...
├── ...
├── agilex_company_annotations
│   │   ├── agilex_company_annotations.json
│   │   ├── ...
```

The total trajetory of annotations are in ``train.json'' and the details of subtask annotations can be found in ``agilex_company_annotations.json''.
### [train.json](train.json):
```json
[
  "Pick up the beef on the table": [
    {
      "raw_task": "Put_the_two_slices_of_bread_and_the_beef_patty_together_on_the_table_to_make_a_hamburger",
      "task": "Pick up the beef on the table",
      "frame": {
        "all": 573,
        "start_frame": 51,
        "end_frame": 115
      },
      "data": {
        "high": "./RoboBrain_Robotic_AGX/videos/train/2773/2773_cam_high.mp4",
        "left": "./RoboBrain_Robotic_AGX/videos/train/2773/2773_cam_left_wrist.mp4",
        "right": "./RoboBrain_Robotic_AGX/videos/train/2773/2773_cam_right_wrist.mp4",
        "state": "./RoboBrain_Robotic_AGX/videos/train/2773/2773.hdf5"
      }
    },
    ...
]
```
### key format of features in hdf5 files
```
|-- qpos (N * 14) (6DoF Right Joint+ Right Gripper (0 for close) + 6DoF Left Joint + Left Gripper (0 for close))
|-- action (N * 14) (Right EEPose (xyz, rpy)+ Right Gripper (0 for close) + Leftt EEPose (xyz, rpy) + Left Gripper (0 for close))
```

# 📅 TODO List 
- [ ] **RoboBrain-X-Dataset**: ~50,000 simulation demonstrations with rich human annotations. 
- [ ] **Data process for training format**: data processing scripts for transfering raw data into FlagScale training data.