Traly/RoboMIND-lerobot

---
license: cc-by-nc-4.0
tags:
- lerobot
- robotics
- manipulation
- RoboMIND
configs:
- config_name: agilex_3rgb
data_files:
- split: train
path: agilex_3rgb/data/**/*.parquet
metadata:
- split: train
path: agilex_3rgb/meta/**
- config_name: franka_1rgb
data_files:
- split: train
path: franka_1rgb/data/**/*.parquet
metadata:
- split: train
path: franka_1rgb/meta/**
- config_name: franka_fr3_dual
data_files:
- split: train
path: franka_fr3_dual/data/**/*.parquet
metadata:
- split: train
path: franka_fr3_dual/meta/**
- config_name: ur_1rgb
data_files:
- split: train
path: ur_1rgb/data/**/*.parquet
metadata:
- split: train
path: ur_1rgb/meta/**
---

# RoboMIND-lerobot (Unofficial Partial Conversion)

> **This is an unofficial conversion** of a subset of the [RoboMIND](https://huggingface.co/datasets/x-humanoid-robomind/RoboMIND) dataset into [LeRobot v3.0](https://github.com/huggingface/lerobot) format.
>
> **This repository does NOT contain the full RoboMIND dataset.** Only a selection of embodiments has been converted. For the complete official dataset, please visit: [x-humanoid-robomind/RoboMIND](https://huggingface.co/datasets/x-humanoid-robomind/RoboMIND).

## Included Subsets

| Config Name | Robot | Episodes | Frames | Tasks | Cameras | Arm Config |
|---|---|---|---|---|---|---|
| `agilex_3rgb` | AgileX | 8,638 | 5,431,759 | 53 | front, left_wrist, right_wrist | Dual-arm (left + right) |
| `franka_1rgb` | Franka Panda | 5,333 | 943,842 | 2 | top | Single-arm |
| `franka_fr3_dual` | Franka FR3 | 1,774 | 375,807 | 6 | front, top, left, right | Dual-arm (left + right) |
| `ur_1rgb` | Universal Robots | 25,002 | 3,766,292 | 114 | top | Single-arm |

**Total: 40,747 episodes / 10,517,700 frames across 4 embodiments**

## Data Format

All subsets follow LeRobot v3.0 format:
- **Video codec:** AV1 (yuv420p)
- **FPS:** 30
- **Data:** Parquet files with chunked storage

### Features per subset

Each subset provides:
- `observation.images.*` — RGB video streams from onboard cameras
- `observation.states.end_effector` — End-effector pose (cartesian position + orientation)
- `observation.states.joint_position` — Joint angles + gripper position
- `observation.states.gripper` — Gripper state (extracted from joint position)
- `actions.end_effector` — Action in end-effector space
- `actions.joint_position` — Action in joint space
- `actions.gripper` — Gripper action
- `timestamp`, `frame_index`, `episode_index`, `index`, `task_index`

> Note: Dual-arm subsets (`agilex_3rgb`, `franka_fr3_dual`) split states/actions into left/right components.

## Usage

```python
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

# Load a specific subset
dataset = LeRobotDataset("Traly/RoboMIND-lerobot", subset="ur_1rgb")
```

## Citation

If you use this data, please cite the original RoboMIND paper:

```bibtex
@article{robomind2024,
title={RoboMIND: Benchmark on Multi-embodiment Intelligence Normative Data for Robot Manipulation},
author={Yan, Kun Wu and Fanqi, Lin and Haoran, Geng and Yong, Jia Li and Jia, Zeng and Jia, Jie Pan and Gang, Zeng and Wen, Hui Liang and Hao, Dong},
journal={arXiv preprint arXiv:2412.13877},
year={2024}
}
```

## Disclaimer

This is a community conversion for research convenience. The original data is collected and owned by the RoboMIND team. Please refer to the [official repository](https://huggingface.co/datasets/x-humanoid-robomind/RoboMIND) for the authoritative source and full dataset.