CosData-Set-0001 / README.md
WenlongHuang's picture
Point the showcase clips at the renamed demos repo
23c33e3 verified
|
Raw History Blame Contribute Delete
4.88 kB
---
license: cc-by-nc-4.0
task_categories:
- robotics
tags:
- lerobot
- embodied-ai
- manipulation
- human-demonstration
- stereo-vision
- grasping
language:
- en
size_categories:
- 100K<n<1M
---
<div align="center">
# DeskOrganizationV1
A 600-episode desktop manipulation dataset
Recorded from human demonstrations · LeRobot v3.0 · 60 fps stereo
TRICOS.AI — embodied-AI data infrastructure
</div>
---
<h3 align="center">600 episodes · 12 tasks · 199,675 frames · 60 fps · stereo 1920×1200</h3>
Every episode carries synchronized stereo video, head and wrist poses, and a
per-frame gripper state: tissue, clear tape, glue stick, pen, bowl and desktop
items, each placed into a left or right box.
## Our capture capability
The clips below show the capture stack behind this dataset: precise hand
trajectories, stable tracking through visual occlusion and rapid head motion.
They illustrate how we collect data, and are not part of the dataset itself.
<table>
<tr>
<td width="50%"><video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/overview_1.mp4"></video><p align="center"><em>Capture overview</em></p></td>
<td width="50%"><video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/overview_2.mp4"></video><p align="center"><em>Capture overview (2)</em></p></td>
</tr>
<tr>
<td width="50%"><video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/high_precision.mp4"></video><p align="center"><em>High-accuracy trajectory</em></p></td>
<td width="50%"><video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/rapid_head_motion.mp4"></video><p align="center"><em>Rapid head motion</em></p></td>
</tr>
<tr>
<td width="50%"><video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/occlusion_1.mp4"></video><p align="center"><em>Visual occlusion · 1</em></p></td>
<td width="50%"><video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/occlusion_2.mp4"></video><p align="center"><em>Visual occlusion · 2</em></p></td>
</tr>
</table>
## Inside the dataset
Two sample episodes from the dataset. Camera view on the left with the
per-frame gripper state (LEFT / RIGHT, 1 = grasping, 0 = open); synchronized
3D trajectory on the right.
**Task: Place the desktop items into the left box**
<video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/fusion_desktop_items_left_box.mp4"></video>
**Task: Place the desktop items into the right box**
<video controls muted autoplay loop playsinline preload="metadata" width="100%" src="https://huggingface.co/datasets/TRICOS/CosData-Set-0001-demos/resolve/main/demo/fusion_desktop_items_right_box.mp4"></video>
## Access
This dataset is **gated**. Request access on this page and, once approved,
download the full videos, metadata and parquet data.
## Collaboration
<div align="center">
**100,000 hours every month**
**Under ¥100 per hour · open for collaboration**
</div>
We work with robotics and embodied-AI teams on data for real deployment.
- **Custom capture tasks** — tasks, objects and scenes designed for your robot
- **Free trial dataset** — a sample release to evaluate data quality first
- **Adjustable data format** — LeRobot v3.0 by default, other schemas on request
- **Model training support** — our VLA team helps with data preparation before
training: world-to-camera frame conversion, cross-embodiment mapping and
related tooling
Tell us what you are building and we will design the capture around it.
## Contact
**contact@tricosai.com**
---
## Citation
```bibtex
@misc{tricos_deskorganizationv1,
title = {TRICOS.AI DeskOrganizationV1: A Desktop Manipulation Demonstration Dataset},
author = {{TRICOS.AI}},
year = {2026},
howpublished = {Hugging Face Datasets},
note = {600 episodes, 12 language-conditioned tasks, LeRobot v3.0}
}
```
## License
Released under the Creative Commons Attribution-NonCommercial 4.0
International License (CC BY-NC 4.0). Commercial licensing: contact TRICOS.AI.
---
<p align="center">contact@tricosai.com</p>
---
Technical documentation: [DATASET.md](DATASET.md) · Video orientation and
projection: [VIDEO_ROTATION.md](VIDEO_ROTATION.md)