Egocentric Household Activities
First-person video of real people doing real chores in real homes.
- Hours of video
- 10,000+
- Contributors
- 2,400+
- Countries
- 38
- Task taxonomy
- 600+ tasks
Illustrative rendering of annotation layers.
01Overview
Head-mounted video of everyday household tasks — cooking, cleaning, laundry, tidying — captured by consented contributors in their own homes across dozens of countries. Every clip ships with dense action segments, 3D hand pose and natural-language narration, ready for training vision-language-action models and humanoid policies.
Modalities
- RGB video
- IMU
- 3D hand pose
- Narration
Use cases
- Pre-training vision-language-action (VLA) models
- Learning manipulation priors from human hands
- Task segmentation and long-horizon planning
- World models of domestic environments
02Specification
Capture
- Device
- Head-mounted RGB camera with 6-axis IMU
- Field of view
- 120° diagonal, fisheye
- Resolution
- 1920 × 1080 (1080p)
- Frame rate
- 30 fps
- IMU rate
- 200 Hz, time-synchronized
- Clip length
- 2–20 min, median 6 min
Annotation
- Action segments
- Verb–noun labels with start/end timestamps
- Hand pose
- 21 keypoints per hand, 3D, every frame
- Objects
- Bounding boxes for manipulated objects, 2 fps
- Narration
- Timestamped free-form English narration
- QA
- Two-pass human review, ≥ 98% segment accuracy
Coverage
- Environments
- Private homes: kitchens, living rooms, bedrooms, laundry
- Geography
- 38 countries across 6 continents
- Lighting
- Daylight, artificial, mixed, low-light
- Demographics
- Balanced age and gender; left- and right-handed
Compliance
- Consent
- Written, revocable contributor consent for AI training
- Privacy
- Faces, screens and documents blurred; no audio by default
- Provenance
- Per-clip capture metadata and consent record ID
03Schema
| Field | Type | Description |
|---|---|---|
| clip_id | string | Globally unique clip identifier. |
| video | mp4 (H.264) | 1080p30 head-mounted video. |
| imu | parquet | Accelerometer and gyroscope at 200 Hz. |
| hands | float32[T, 2, 21, 3] | Per-frame 3D keypoints for left and right hand, camera frame, meters. |
| segments | list<Segment> | Action segments with verb, noun, start_s and end_s. |
| narration | list<Utterance> | Timestamped English narration of what the contributor is doing. |
| meta | json | Country, room type, lighting, device ID and consent record ID. |
04File structure
egocentric-household/
├── manifest.parquet
├── taxonomy/
│ ├── verbs.json
│ └── nouns.json
└── clips/
└── 0001f3a9c2/
├── video.mp4
├── imu.parquet
├── hands.npz
├── segments.json
├── narration.json
└── meta.json05Sample
Evaluate before you license.
- 50 clips across 12 homes and 9 countries
- Full annotations for every sample clip
- Task taxonomy and data card
- Size
- ≈ 4.8 GB
- Formats
- MP4, Parquet, NPZ, JSON
06License
Graviti Commercial Data License
Perpetual, worldwide license to train and commercially deploy models. Redistribution of raw data is not permitted.
How we license data →Related datasets
Bimanual Teleoperation Trajectories
Synchronized multi-view robot demonstrations with language instructions.
- Episodes
- 120,000+
- Task families
- 180
Egocentric Skilled Work
First-person footage of trained professionals in commercial environments.
- Target hours
- 5,000
- Workplace types
- 14
Get started
Need more than Egocentric Household Activities?
We extend catalog datasets with new environments, tasks, languages and annotation layers — or collect from scratch to your spec.