andreaskoepf/robotwin_dk1_realcam_randomized_100

---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- RoboTwin
- bimanual
- simulation
- multi-task
configs:
- config_name: default
data_files: data/*/*.parquet
---

This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).

<a href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=andreaskoepf/robotwin_dk1_realcam_randomized_100">
<img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/>
<img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/>
</a>

## Dataset Description

Multi-task bimanual manipulation dataset combining **50 tasks** (5000 episodes total),
generated in the [RoboTwin](https://github.com/TianxingChen/RoboTwin) simulator with domain randomization.
Episodes are shuffled across tasks for mixed-task training.

- **Robot:** TRLC DK1 bimanual (`bi_dk1_follower`) - 2x 6-DOF arms with parallel grippers
- **Tasks:** 50 distinct manipulation tasks
- **Episodes:** 5000 (shuffled across tasks)
- **Total frames:** 772,264 (757.1 min @ 17 fps)

### Cameras

| Camera | Resolution | Codec | FPS |
|---|---|---|---|
| `observation.images.head` | 640x360 | av1 | 17 |
| `observation.images.left_wrist` | 640x360 | av1 | 17 |
| `observation.images.right_wrist` | 640x360 | av1 | 17 |

### Action & State Space

- **`action`**: float32[14] - joint position targets (6 joints + 1 gripper per arm)
- **`observation.state`**: float32[14] - current joint positions + gripper state

Each episode has a natural language task description generated from RoboTwin templates.

### Task Distribution

| Task | Episodes |
|---|---|
| Adjust Bottle | 100 |
| Beat Block Hammer | 100 |
| Blocks Ranking Rgb | 100 |
| Blocks Ranking Size | 100 |
| Click Alarmclock | 100 |
| Click Bell | 100 |
| Dump Bin Bigbin | 100 |
| Grab Roller | 100 |
| Handover Block | 100 |
| Handover Mic | 100 |
| Hanging Mug | 100 |
| Lift Pot | 100 |
| Move Can Pot | 100 |
| Move Pillbottle Pad | 100 |
| Move Playingcard Away | 100 |
| Move Stapler Pad | 100 |
| Open Laptop | 100 |
| Open Microwave | 100 |
| Pick Diverse Bottles | 100 |
| Pick Dual Bottles | 100 |
| Place A2B Left | 100 |
| Place A2B Right | 100 |
| Place Bread Basket | 100 |
| Place Bread Skillet | 100 |
| Place Burger Fries | 100 |
| Place Can Basket | 100 |
| Place Cans Plasticbox | 100 |
| Place Container Plate | 100 |
| Place Dual Shoes | 100 |
| Place Empty Cup | 100 |
| Place Fan | 100 |
| Place Mouse Pad | 100 |
| Place Object Basket | 100 |
| Place Object Scale | 100 |
| Place Object Stand | 100 |
| Place Phone Stand | 100 |
| Place Shoe | 100 |
| Press Stapler | 100 |
| Put Bottles Dustbin | 100 |
| Put Object Cabinet | 100 |
| Rotate Qrcode | 100 |
| Scan Object | 100 |
| Shake Bottle | 100 |
| Shake Bottle Horizontally | 100 |
| Stack Blocks Three | 100 |
| Stack Blocks Two | 100 |
| Stack Bowls Three | 100 |
| Stack Bowls Two | 100 |
| Stamp Seal | 100 |
| Turn Switch | 100 |

### Episode Manifest

The file `manifest.json` maps each dataset episode index to its source task and original episode number:
```json
[
{"dataset_episode": 0, "task": "shake_bottle", "original_episode": 42},
{"dataset_episode": 1, "task": "place_bread_basket", "original_episode": 7},
...
]
```

## Loading the Dataset

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

dataset = LeRobotDataset("andreaskoepf/robotwin_dk1_realcam_randomized_100")
frame = dataset[0]
print(frame["observation.state"].shape) # torch.Size([14])
print(frame["action"].shape) # torch.Size([14])
```

## Citation

```bibtex
@misc{robotwin2025,
title = {RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins},
author = {Chen, Yao Mu et al.},
year = {2025},
url = {https://github.com/TianxingChen/RoboTwin}
}
```