File size: 3,560 Bytes
88f6314 22ba390 ecc0f5f 22ba390 ecc0f5f 22ba390 ecc0f5f 22ba390 20adb80 22ba390 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 | ---
license: cc-by-4.0
pretty_name: RoboMoCap
tags:
- robotics
- humanoid
- robot-learning
- egocentric
- whole-body-motion
- 3d
---
<div align="center">
<video
width="100%"
autoplay
muted
loop
playsinline
controls
preload="metadata"
>
<source
src="https://bitrobot-foundation.github.io/robomocap/media/hero-film-v10.mp4"
type="video/mp4"
/>
</video>
</div>
<h1 align="center">RoboMoCap</h1>
<p align="center">A Large-Scale Whole-Body Motion Capture Dataset with Egocentric Vision for Robot Learning</p>
<p align="center">
<a href="https://bitrobot-foundation.github.io/robomocap/"><strong>[Website]</strong></a>
<a href="https://bitrobot-foundation.github.io/robomocap/RoboMoCap.pdf"><strong>[Paper]</strong></a>
<strong>[Code]</strong> (coming soon)
</p>
**Egocentric vision. High-fidelity whole-body motion.**
RoboMoCap pairs multimodal first-person observations with optical whole-body motion capture for humanoid robot learning. With six head-mounted and two wrist-mounted cameras, the dataset captures diverse locomotion and loco-manipulation behaviours, from everyday walking to dynamic movements and complex object interactions.
**Beyond what egocentric cameras can see.** While many egocentric datasets focus primarily on hand-object interactions, RoboMoCap emphasises locomotion and loco-manipulation. By pairing first-person observations with high-fidelity optical motion capture, it provides accurate supervision for lower-body movement, foot placement, and whole-body coordination during navigation and object interactions.
## Dataset Overview
- **200+ hours** of recording
- **14** participants
- **5TB+** of data
- **290+** captured actions
## Data Modalities
| Modality | Description |
| --- | --- |
| **Egocentric sensing ([RoboCap](https://robocap.io/))** | Six head-mounted cameras provide forward, downward, and side views, alongside IMU, magnetometer measurements, and WAV audio. |
| **Wrist sensing ([RoboWrist](https://robocap.io/))** | Left and right wrist cameras record close-range hand–object interactions, alongside IMU and magnetometer measurements. |
| **Whole-body motion** | A 53-marker [NOKOV](https://en.nokov.com/) optical mocap setup records whole-body motion at 120 Hz, including leg and foot movement outside the wearable camera views. |
| **External video** | [NOKOV](https://en.nokov.com/)'s third-person video records the participant, manipulated objects, and the surrounding space. |
| **Object motion** | Reflective markers track manipulated objects alongside the participant's body motion. |
| **Object geometry** | Textured 3D scans preserve the shape and appearance of objects used in the recordings. |
| **Hand pose** | Hand pose complements wrist video to describe hand movements during interaction. |
## License
This dataset is licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
## Citation
If you use this dataset, please cite:
```bibtex
@misc{deshmukh2026robomocap,
title = {{RoboMoCap}: A Large-Scale Whole-Body Motion Capture Dataset with Egocentric Vision for Robot Learning},
author = {Deshmukh, Mayur and Tung, Aaron and Li, Junyi and Bhardwaj, Nikash and Adeniji, Ademi and Liu, Vincent and Cho, Michael and Duan, Jiafei},
year = {2026},
eprint = {2611.10994},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
doi = {10.48550/arXiv.2611.10994},
url = {https://arxiv.org/abs/2611.10994}
}
``` |