README.md

---
title: README
emoji: 🦾
colorFrom: blue
colorTo: indigo
sdk: static
pinned: false
---

<p align="center">
  <b>Leverage Physical Superintelligence, with real-world data.</b>
</p>

<p align="center">
  <a href="https://humaid.co">humaid.co</a> ·
  <a href="https://www.linkedin.com/company/humaidphysicalai/">LinkedIn</a> ·
  <a href="https://github.com/humaidtech">GitHub</a> ·
  <a href="mailto:info@humaid.co">info@humaid.co</a>
</p>

---

We collect **real-world egocentric demonstration data for robotics** — across
manufacturing, warehouse, hospitality and food-service environments. Trained operators wear
calibrated multi-sensor rigs on real jobs, and we run the whole pipeline from capture through
annotation, quality control and delivery.

Not lab data. Not scripted. People doing their actual work.

## What we capture

| | |
|---|---|
| **Egocentric** | Stereo RGB-D, global shutter, 1920×1200 @ 30 fps |
| **Wrists** | Dual wrist cameras, synchronized |
| **Depth** | 16-bit, losslessly compressed |
| **IMU** | ~200 Hz, head and both wrists |
| **Hands** | 21-keypoint 3D model per hand |
| **Labels** | Temporal action segmentation + natural-language descriptions |
| **Format** | MCAP, every signal on a shared clock |

## Open datasets

### 🎬 [EgoViz-120](https://huggingface.co/datasets/humaidtech/EgoViz-120)

**120 hours · 5,392 clips · 7.25 TB · CC-BY-4.0**

Synchronized egocentric demonstrations of real cleaning, cooking and facility work. Stereo
ego video, two wrist cameras, depth, three IMUs, 3D hand pose and frame-level action labels,
in a single self-contained MCAP per clip.

## Working with us

We build custom datasets against a spec: environments, tasks, sensor configuration,
annotation schema and volume. 50+ companies served, 12+ countries, 55,000+ operators across
our partner networks.

If you are training manipulation or humanoid policies and the bottleneck is real-world data,
get in touch — [info@humaid.co](mailto:info@humaid.co).