---
license: cc-by-4.0
---

<p align="center">
  <img src="assets/assemble_parts_overview.gif" alt="RoCo Task Board Assembly overview" width="100%">
</p>

<h1 align="center">RoCo Task Board Assembly Demonstrations</h1>

<p align="center">
  <b>Real-world LeRobot demonstrations for contact-rich task-board assembly.</b>
</p>

<p align="center">
  <a href="https://www.sharpa.com/"><img src="https://img.shields.io/badge/Website-sharpa.com-555555?style=flat-square" alt="Website"></a>
  <a href="https://github.com/sharpa-robotics"><img src="https://img.shields.io/badge/GitHub-Sharpa-24292f?style=flat-square&logo=github&logoColor=white" alt="GitHub"></a>
  <a href="https://www.linkedin.com/company/sharpa-robotics"><img src="https://img.shields.io/badge/LinkedIn-Sharpa-0a66c2?style=flat-square&logo=linkedin&logoColor=white" alt="LinkedIn"></a>
  <a href="https://www.youtube.com/@sharpa-robotics"><img src="https://img.shields.io/badge/YouTube-Sharpa-ff0000?style=flat-square&logo=youtube&logoColor=white" alt="YouTube"></a>
  <a href="https://x.com/SharpaRobotics"><img src="https://img.shields.io/badge/X-%40SharpaRobotics-000000?style=flat-square&logo=x&logoColor=white" alt="X"></a>
  <a href="https://rocochallenge.github.io/RoCo-IROS2026/"><img src="https://img.shields.io/badge/Competition-RoCo%20IROS%202026-0077b6?style=flat-square" alt="Competition Website"></a>
  <a href="https://forms.gle/d2NKNAE7dqSfYZB87"><img src="https://img.shields.io/badge/Register-Google%20Form-34a853?style=flat-square" alt="Registration"></a>
  <a href="https://huggingface.co/datasets/SharpaIT/RoCo_TaskBoardAssembly"><img src="https://img.shields.io/badge/Dataset-Hugging%20Face-f3b000?style=flat-square" alt="Dataset"></a>
</p>

## Overview

**RoCo Task Board Assembly Demonstrations** is a real-world robot manipulation dataset for assembling parts on a task board. It is released by **Sharpa** in a **LeRobot-compatible** format for imitation learning, visuomotor policy learning, visual-tactile representation learning, and contact-rich manipulation research.

Task-board assembly requires precise part localization, fine contact timing, bimanual coordination, and robust perception under hand-object occlusion. The demonstrations include synchronized multi-view video, tactile observations, proprioceptive state, torque signals, and action targets.

This dataset supports the [RoCo IROS 2026 Challenge](https://rocochallenge.github.io/RoCo-IROS2026/). Teams can use it to develop, train, and evaluate policies for the task-board assembly track. Interested participants should visit the official competition website and complete the [registration form](https://forms.gle/d2NKNAE7dqSfYZB87) to receive challenge updates and participation details.

<table>
  <tr>
    <td><b>Task</b><br>Task-board part assembly</td>
    <td><b>Format</b><br>LeRobot v3.0 / v2.1</td>
    <td><b>Scale</b><br>30 seasons / 562 episodes</td>
    <td><b>Frequency</b><br>30 FPS</td>
  </tr>
  <tr>
    <td><b>Video</b><br>6 synchronized streams</td>
    <td><b>State / Action</b><br>65D joint space</td>
    <td><b>Tactile</b><br>60D signal + tactile video</td>
    <td><b>Use</b><br>Training and policy development</td>
  </tr>
</table>

## Competition and Registration

The RoCo IROS 2026 Challenge provides a shared benchmark for real-world robotic assembly, focusing on contact-rich manipulation, bimanual coordination, visual-tactile perception, and robust policy execution. The task-board assembly dataset is released as training data for teams participating in the challenge and for researchers working on related manipulation problems.

- Competition website: [https://rocochallenge.github.io/RoCo-IROS2026/](https://rocochallenge.github.io/RoCo-IROS2026/)
- Registration form: [https://forms.gle/d2NKNAE7dqSfYZB87](https://forms.gle/d2NKNAE7dqSfYZB87)

Please refer to the competition website for the latest schedule, rules, evaluation details, and participation instructions.

## Dataset Capabilities

| Capability | Dataset Support |
| --- | --- |
| Contact-rich assembly learning | Real-world demonstrations for assembling task-board parts |
| Multi-view visuomotor policies | Synchronized head-camera and wrist-camera observations |
| Visual-tactile learning | High-resolution tactile videos, synchronized raw tactile camera views, and 10-fingertip 6-axis tactile signals |
| Joint-space control | 65D synchronized state and action for two arms, two dexterous hands, and torso/motor-related joints |
| LeRobot ecosystem | `lerobot3.0` and `lerobotv2.1` exports for every released season |

## Example Views

The demonstrations include synchronized head, wrist, and tactile video streams. Each preview below uses a representative window from the same `lerobotv2.1` episode, played at 10x speed. GIF previews render directly in Markdown; click any preview to open the MP4 version. Tactile previews preserve their full wide-frame layout.

<table>
  <tr>
    <td width="50%">
      <b>Head Left</b><br>
      <a href="assets/example_views/head_left.mp4">
        <img src="assets/example_views/head_left.gif" alt="Head Left" width="100%">
      </a>
    </td>
    <td width="50%">
      <b>Head Right</b><br>
      <a href="assets/example_views/head_right.mp4">
        <img src="assets/example_views/head_right.gif" alt="Head Right" width="100%">
      </a>
    </td>
  </tr>
  <tr>
    <td width="50%">
      <b>Wrist Left</b><br>
      <a href="assets/example_views/wrist_left.mp4">
        <img src="assets/example_views/wrist_left.gif" alt="Wrist Left" width="100%">
      </a>
    </td>
    <td width="50%">
      <b>Wrist Right</b><br>
      <a href="assets/example_views/wrist_right.mp4">
        <img src="assets/example_views/wrist_right.gif" alt="Wrist Right" width="100%">
      </a>
    </td>
  </tr>
  <tr>
    <td width="50%">
      <b>Tactile Deformation</b><br>
      <a href="assets/example_views/tactile_deform.mp4">
        <img src="assets/example_views/tactile_deform.gif" alt="Tactile Deformation" width="100%">
      </a>
    </td>
    <td width="50%">
      <b>Raw Tactile</b><br>
      <a href="assets/example_views/tactile_raw.mp4">
        <img src="assets/example_views/tactile_raw.gif" alt="Raw Tactile" width="100%">
      </a>
    </td>
  </tr>
</table>

## Dataset Statistics

| Item | Value |
| --- | ---: |
| Total collection seasons | 30 |
| `lerobot3.0` seasons | 30 |
| `lerobot3.0` episodes | 562 |
| `lerobot3.0` frames | 2,461,024 |
| `lerobotv2.1` seasons | 30 |
| `lerobotv2.1` episodes | 562 |
| `lerobotv2.1` frames | 2,461,024 |
| FPS | 30 |
| Video streams | 6 |
| State/action dimension | 65 |
| Tactile signal dimension | 60 |
| Approximate data size | 324.3 GB |

For new users, we recommend starting from `lerobot3.0`. The `lerobotv2.1` export is included for compatibility with pipelines that still depend on the older LeRobot layout.

## Get Started

### Download The Dataset

Make sure Git LFS is installed before cloning from Hugging Face.

```bash
git lfs install
git clone https://huggingface.co/datasets/SharpaIT/RoCo_TaskBoardAssembly
```

If you want to clone only metadata first and fetch large files later:

```bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/SharpaIT/RoCo_TaskBoardAssembly
```

If you only want a specific season, use sparse checkout:

```bash
git init RoCo_TaskBoardAssembly
cd RoCo_TaskBoardAssembly
git remote add origin https://huggingface.co/datasets/SharpaIT/RoCo_TaskBoardAssembly
git sparse-checkout init
git sparse-checkout set season_POC22061_2026_06_11_14_29_08_train README.md
git pull origin main
```

### Quick Inspection

Each season contains both `lerobot3.0` and `lerobotv2.1` exports. Inspect `meta/info.json` first to understand the exact schema and file templates.

```python
import json
from pathlib import Path

dataset_root = Path("RoCo_TaskBoardAssembly")
episode_root = dataset_root / "season_POC22061_2026_06_11_14_29_08_train" / "lerobot3.0"

with open(episode_root / "meta" / "info.json", "r") as f:
    info = json.load(f)

print(info["total_episodes"])
print(info["total_frames"])
print(info["features"].keys())
```

## Dataset Structure

The dataset is organized by collection season. Each season contains a `lerobot3.0` export and a `lerobotv2.1` export.

```text
RoCo_TaskBoardAssembly/
├── README.md
├── season_POC22061_2026_06_11_14_29_08_train/
│   ├── lerobot3.0/
│   │   ├── meta/
│   │   │   ├── info.json
│   │   │   ├── modality.json
│   │   │   ├── episodes/
│   │   │   └── tasks.parquet
│   │   ├── data/
│   │   │   └── chunk-000/
│   │   └── videos/
│   │       ├── observation.images.head_left/
│   │       ├── observation.images.head_right/
│   │       ├── observation.images.wrist_left/
│   │       ├── observation.images.wrist_right/
│   │       ├── observation.images.tactile_deform/
│   │       └── observation.images.tactile_raw/
│   └── lerobotv2.1/
│       ├── meta/
│       ├── data/
│       └── videos/
└── season_.../
```

### LeRobot Storage Layout

| Part | Description |
| --- | --- |
| `meta/` | Dataset metadata, feature schema, task metadata, and path templates |
| `data/` | Episode frame data stored as Apache Parquet files |
| `videos/` | Per-camera MP4 videos |

The most important metadata file is `meta/info.json`. It defines `total_episodes`, `total_frames`, `fps`, `splits`, `data_path`, `video_path`, and `features`.

### File Path Templates

LeRobot v3.0 uses templates similar to:

```text
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
```

LeRobot v2.1 uses templates similar to:

```text
data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4
```

## Features Schema

### Main Feature Groups

| Feature | Type | Shape | Description |
| --- | --- | ---: | --- |
| `observation.state` | float32 | 65 | Joint-space robot state |
| `action` | float32 | 65 | Joint-space action |
| `observation.state.joint_torque` | float32 | 65 | Joint torque signal |
| `observation.tactile` | float32 | 60 | Tactile force/torque signal |
| `observation.images.*` | video | varies | Multi-view visual observations |
| `timestamp` | float32 | 1 | Timestamp |
| `frame_index` | int64 | 1 | Frame index within an episode |
| `episode_index` | int64 | 1 | Episode index |
| `task_index` | int64 | 1 | Task index |

### Proprioceptive State

The 65D `observation.state` and `action` vectors are ordered as follows:

| Range | Names | Meaning |
| --- | --- | --- |
| 0 - 6 | `left_arm_j0` to `left_arm_j6` | Left arm joints |
| 7 - 28 | `left_hand_j0` to `left_hand_j21` | Left dexterous hand joints |
| 29 - 35 | `right_arm_j0` to `right_arm_j6` | Right arm joints |
| 36 - 57 | `right_hand_j0` to `right_hand_j21` | Right dexterous hand joints |
| 58 - 64 | `motor_j0` to `motor_j6` | Torso / motor-related joints |

### Video Streams

The complete visual observation set contains six camera streams:

| Feature key | Description | Shape |
| --- | --- | --- |
| `observation.images.head_left` | Left head camera | 480 x 480 x 3 |
| `observation.images.head_right` | Right head camera | 480 x 480 x 3 |
| `observation.images.wrist_left` | Left wrist camera | 480 x 480 x 3 |
| `observation.images.wrist_right` | Right wrist camera | 480 x 480 x 3 |
| `observation.images.tactile_deform` | Tactile deformation video | 480 x 1200 x 3 |
| `observation.images.tactile_raw` | Raw tactile video | 480 x 1600 x 3 |

Video files are MP4 without audio. Codec may differ across exports and seasons. `lerobot3.0` is primarily AV1, while `lerobotv2.1` is primarily H.264.

## Tactile Modality

The dataset includes tactile observations as both compact numeric signals and high-resolution video streams. These modalities are synchronized with the robot state, action, and visual camera streams at 30 FPS, making them suitable for contact-rich policy learning and visual-tactile representation learning.

| Tactile feature | Type | Shape | Description |
| --- | --- | --- | --- |
| `observation.tactile` | float32 | 60 | Per-frame tactile force/torque signal: 10 fingertips x 6 axes |
| `observation.images.tactile_deform` | video | 480 x 1200 x 3 | Deformation-oriented tactile video stream that visualizes contact-induced surface changes |
| `observation.images.tactile_raw` | video | 480 x 1600 x 3 | Raw tactile camera stream preserving the full tactile sensor image layout |

The `observation.tactile` vector provides a compact force/torque representation for each frame. It contains 10 fingertip groups: left and right `thumb`, `index`, `middle`, `ring`, and `little`. Each fingertip contributes six values, ordered as `fx`, `fy`, `fz`, `tx`, `ty`, and `tz`, for a total of 60 dimensions.

The two tactile video streams provide complementary image-based tactile observations: `tactile_deform` emphasizes deformation patterns caused by contact, while `tactile_raw` preserves the raw tactile image for users who want to build their own visual-tactile preprocessing or representation learning pipeline. For downstream experiments, users can start with `observation.tactile` as a lightweight contact signal, then add one or both tactile video streams when the model architecture can handle the additional spatial resolution and bandwidth.

## Season List

| Season | `lerobot3.0` | `lerobotv2.1` |
| --- | ---: | ---: |
| `season_POC22061_2026_06_11_14_29_08_train` | 3.39 GB | 3.40 GB |
| `season_POC22061_2026_06_11_19_10_57_train` | 3.45 GB | 3.46 GB |
| `season_POC22061_2026_06_11_20_21_30_train` | 4.76 GB | 4.78 GB |
| `season_POC22061_2026_06_14_10_25_39_train` | 3.82 GB | 3.83 GB |
| `season_POC22061_2026_06_14_15_44_09_train` | 7.40 GB | 7.43 GB |
| `season_POC22061_2026_06_15_10_15_11_train` | 4.07 GB | 4.08 GB |
| `season_POC22061_2026_06_15_11_15_27_train` | 2.91 GB | 2.92 GB |
| `season_POC22061_2026_06_15_15_56_02_train` | 9.96 GB | 10.00 GB |
| `season_POC22061_2026_06_15_19_24_12_train` | 6.47 GB | 6.49 GB |
| `season_POC22061_2026_06_16_10_36_10_train` | 2.47 GB | 2.48 GB |
| `season_POC22061_2026_06_16_13_39_56_train` | 7.77 GB | 7.79 GB |
| `season_POC22061_2026_06_16_15_59_28_train` | 6.01 GB | 6.03 GB |
| `season_POC22061_2026_06_16_19_09_01_train` | 5.15 GB | 5.17 GB |
| `season_POC22061_2026_06_17_10_36_58_train` | 6.48 GB | 6.50 GB |
| `season_POC22061_2026_06_17_13_36_51_train` | 6.02 GB | 6.04 GB |
| `season_POC22061_2026_06_17_15_33_05_train` | 7.65 GB | 7.68 GB |
| `season_POC22061_2026_06_17_19_15_29_train` | 5.62 GB | 5.64 GB |
| `season_POC22061_2026_06_18_10_08_09_train` | 6.11 GB | 6.13 GB |
| `season_POC22061_2026_06_18_13_41_46_train` | 5.31 GB | 5.33 GB |
| `season_POC22061_2026_06_18_19_11_08_train` | 6.76 GB | 6.78 GB |
| `season_POC22061_2026_06_18_20_41_26_train` | 3.66 GB | 3.68 GB |
| `season_POC22061_2026_06_19_10_02_47_train` | 3.16 GB | 3.17 GB |
| `season_POC22061_2026_06_19_10_45_34_train` | 3.53 GB | 3.54 GB |
| `season_POC22061_2026_06_19_13_51_05_train` | 5.85 GB | 5.87 GB |
| `season_POC22061_2026_06_19_16_00_30_train` | 5.91 GB | 5.93 GB |
| `season_POC22061_2026_06_19_19_14_55_train` | 5.61 GB | 5.63 GB |
| `season_POC22061_2026_06_20_10_09_11_train` | 6.21 GB | 6.23 GB |
| `season_POC22061_2026_06_20_13_43_19_train` | 5.80 GB | 5.82 GB |
| `season_POC22061_2026_06_20_15_57_28_train` | 5.60 GB | 5.62 GB |
| `season_POC22061_2026_06_20_19_18_15_train` | 4.94 GB | 4.96 GB |

## Usage Recommendations

This release is provided as training data. For local experiments, users may split by season to avoid mixing demonstrations from the same collection session across train and evaluation sets.

For policy learning, a typical setup is:

- Visual observations: one or more `observation.images.*` streams
- Proprioception: `observation.state`
- Optional tactile signal: `observation.tactile`
- Supervision target: `action`

For multi-view policies, start with:

```text
observation.images.head_left
observation.images.head_right
observation.images.wrist_left
observation.images.wrist_right
```

Then add tactile video streams if your model can use high-resolution tactile observations:

```text
observation.images.tactile_deform
observation.images.tactile_raw
```

## Dataset Notes

- This repository contains only the LeRobot exports prepared for release.
- Raw POC recording folders and intermediate HDF5 folders are intentionally excluded from this release.
- The dataset is released by season, not as a single flattened LeRobot root.
- Every released season includes both `lerobot3.0` and `lerobotv2.1`.
- `motor_j0` to `motor_j6` are the torso/motor-related dimensions in `observation.state` and `action`.
- The task is contact-rich and precision-sensitive: policies should expect object contacts, hand-part occlusions, tactile events, and fine pose adjustments.

## License and Terms

This dataset is released under the [Creative Commons Attribution 4.0 International License (CC-BY-4.0)](https://creativecommons.org/licenses/by/4.0/). You may use, share, and adapt the dataset, including for commercial purposes, provided that you give appropriate attribution.

If you use the dataset for RoCo IROS 2026 Challenge participation, please also follow the official competition rules and evaluation protocol.

## Citation

If this dataset contributes to your research, please cite or acknowledge the dataset.

```bibtex
@misc{roco_task_board_assembly_2026,
  title        = {RoCo Task Board Assembly LeRobot Dataset},
  howpublished = {\url{https://huggingface.co/datasets/SharpaIT/RoCo_TaskBoardAssembly}},
  year         = {2026}
}
```