Traly/Robotwin2.0-lerobot

---
license: apache-2.0
task_categories:
- robotics
tags:
- lerobot
- robotwin
- bimanual
- multi-embodiment
- simulation
pretty_name: RoboTwin2.0 LeRobot
size_categories:
- 100K<n<1M
---

# RoboTwin2.0 - LeRobot v3.0

[RoboTwin2.0](https://huggingface.co/datasets/TianxingChen/RoboTwin2.0) converted to [LeRobot v3.0](https://github.com/huggingface/lerobot) format. Contains **both joint-space and end-effector (EE) pose** data for bimanual manipulation across 5 robot embodiments.

## Dataset Statistics

| | |
|---|---|
| **Episodes** | 114,001 |
| **Frames** | 18,709,206 |
| **Tasks** | 50 |
| **Embodiments** | 5 |
| **FPS** | 30 |
| **Format** | LeRobot v3.0 (Parquet + MP4) |
| **Size** | ~20 GB |

## Embodiments

| Embodiment | Episodes | DOF (per arm) |
|---|---|---|
| Franka | 22,501 | 7+1 |
| ARX-X5 | 24,500 | 6+1 |
| Aloha-AgileX | 25,000 | 6+1 |
| Piper | 18,500 | 6+1 |
| UR5 | 23,500 | 6+1 |

All robots are bimanual (dual-arm). Each arm has `joint_dim + gripper(1)` DOF.

## Features

| Feature | Shape | Description |
|---|---|---|
| `observation.state` | `[16]` | Joint state: `left_joint(7) + left_grip(1) + right_joint(7) + right_grip(1)` |
| `observation.state_ee` | `[16]` | EE pose: `left_xyz(3) + left_quat(4) + left_grip(1) + right(8)` |
| `observation.embodiment_id` | `[5]` | One-hot embodiment vector |
| `action` | `[16]` | Next-frame joint state (absolute) |
| `action_ee` | `[16]` | Next-frame EE pose (absolute) |
| `observation.images.head` | `[240, 320, 3]` | Head camera (video) |
| `observation.images.left_wrist` | `[240, 320, 3]` | Left wrist camera (video) |
| `observation.images.right_wrist` | `[240, 320, 3]` | Right wrist camera (video) |
| `observation.images.front` | `[240, 320, 3]` | Front camera (video, not all embodiments) |

### Conventions

- **Joint state**: Padded to 7 DOF per arm (zero-padded for 6-DOF robots)
- **EE quaternion**: `[x, y, z, w]` format
- **Gripper**: `[0, 1]`, where `1 = open`

## Tasks (50)

<details>
<summary>Click to expand full task list</summary>

adjust_bottle, beat_block_hammer, blocks_ranking_rgb, blocks_ranking_size,
click_alarmclock, click_bell, dump_bin_bigbin, grab_roller,
handover_block, handover_mic, hanging_mug, lift_pot,
move_can_pot, move_pillbottle_pad, move_playingcard_away, move_stapler_pad,
open_laptop, open_microwave, pick_diverse_bottles, pick_dual_bottles,
place_a2b_left, place_a2b_right, place_bread_basket, place_bread_skillet,
place_burger_fries, place_can_basket, place_cans_plasticbox, place_container_plate,
place_dual_shoes, place_empty_cup, place_fan, place_mouse_pad,
place_object_basket, place_object_scale, place_object_stand, place_phone_stand,
place_shoe, press_stapler, put_bottles_dustbin, put_object_cabinet,
rotate_qrcode, scan_object, shake_bottle, shake_bottle_horizontally,
stack_blocks_three, stack_blocks_two, stack_bowls_three, stack_bowls_two,
stamp_seal, turn_switch

</details>

Each task uses the `randomized_500` variant (domain-randomized, up to 500 episodes per embodiment).

## Source

Converted from [TianxingChen/RoboTwin2.0](https://huggingface.co/datasets/TianxingChen/RoboTwin2.0) HDF5 data using parallel conversion with per-task checkpointing.

## Citation

```bibtex
@article{mu2024robotwin,
title={RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins},
author={Mu, Yao and Chen, Tianxing and Peng, Shijia and Chen, Zanxin and Gao, Zeyu and Zou, Yude and Lin, Lunkai and Xie, Zhiqiang and Luo, Ping},
journal={arXiv preprint arXiv:2409.02920},
year={2024}
}
```