Egocentric Skilled Work
First-person footage of trained professionals in commercial environments.
- Target hours
- 5,000
- Workplace types
- 14
- Partner sites
- 60+
Illustrative rendering of annotation layers.
01Overview
Head-mounted video of skilled labor in commercial kitchens, warehouses, retail back-of-house and light assembly. Captured on-site with partner businesses, annotated with procedural steps, tool use and hand–object contact.
Modalities
- RGB video
- IMU
- Hand–object contact
- Procedures
Use cases
- Humanoid deployment in commercial settings
- Procedure and step recognition
- Tool-use and contact modeling
02Specification
Capture
- Device
- Head-mounted RGB camera with 6-axis IMU
- Resolution
- 1920 × 1080 @ 30 fps
- Sites
- Commercial kitchens, warehouses, retail, assembly
Annotation
- Procedures
- Hierarchical step labels per standard operating procedure
- Contact
- Hand–object contact state and tool identity
03Schema
| Field | Type | Description |
|---|---|---|
| clip_id | string | Unique clip identifier. |
| video | mp4 (H.264) | 1080p30 head-mounted video. |
| steps | list<Step> | Procedure steps with timestamps. |
| contacts | parquet | Per-frame hand–object contact states. |
04File structure
egocentric-skilled-work/
├── manifest.parquet
└── clips/<clip_id>/
├── video.mp4
├── imu.parquet
├── steps.json
└── contacts.parquet05Sample
Sample on request.
- Pilot clips from 3 workplace types
- Size
- Available on request
- Formats
- MP4, Parquet, JSON
06License
Graviti Commercial Data License
Commercial training rights. Exclusive slices available.
How we license data →Related datasets
Egocentric Household Activities
First-person video of real people doing real chores in real homes.
- Hours of video
- 10,000+
- Contributors
- 2,400+
Bimanual Teleoperation Trajectories
Synchronized multi-view robot demonstrations with language instructions.
- Episodes
- 120,000+
- Task families
- 180
Get started
Need more than Egocentric Skilled Work?
We extend catalog datasets with new environments, tasks, languages and annotation layers — or collect from scratch to your spec.