# A Multi-embodiments Benchmark
<p align="center">
  <img src="assets/benchmark.png" width="100%">
</p>

Thanks to [GM-100](https://www.rhos.ai/research/gm-100), which
provides 100 carefully designed tasks, we collect data across three robotic platforms to enable a systematic evaluation. 

The dataset is stored in the Lerobot2.1 format.

## Dataset Description

- **Robot Platform**: AgileX Cobot Magic, Galaxea R1Pro, AgiBot G1
- **Data Format**: Lerobot2.1

## Dataset Structure

```text
.
├── AgiBotG1/                  # Robot Platform
│   |── BM-001                 # Task 1
│   │   ├── meta/             
│   │   │   ├── info.json
│   │   │   ├── tasks.jsonl
│   │   │   └── ...
│   │   ├── videos/
│   │   │   └── chunk-000/
│   │   │       ├── observation.images.camera_top/
│   │   │       │   ├── episode_000000.mp4
│   │   │       │   ├── episode_000001.mp4
│   │   │       │   └── ...
│   │   │       ├── observation.images.camera_wrist_left/
│   │   │       │   ├── episode_000000.mp4
│   │   │       │   ├── episode_000001.mp4
│   │   │       │   └── ...
│   │   │       └── observation.images.camera_wrist_right/
│   │   │           ├── episode_000000.mp4
│   │   │           ├── episode_000001.mp4
│   │   │           └── ...           
│   │   └── data/                    
│   │       └── chunk-000/
│   │           ├── episode_000000.parquet
│   │           ├── episode_000001.parquet
│   │           └── ...
│   |── BM-002                 # Task 2
│   ├── ...
├── AgileX/                
│   ├── ...
├── GalaxeaR1Pro/                
│   ├── ...
└── ...
```

## License Agreement
This project is licensed under the [CC BY-NC-SA 4.0](LICENSE).