Robotics data

Real-world demonstrations, built to your brief.

First-person episodes of everyday tasks from homes around the world, labelled for robot learning and collected against the tasks you need.

18
task types at launch
1080p
first-person video
42
hand keypoints a frame
2×
checks on every clip
episode.jsonSample schema
{
  "episode_id": "zeth_ep_000001",
  "mission": "Z/01 fold_laundry",
  "video": { "res": "1920x1080", "fps": 30, "view": "head" },
  "context": {
    "country": "KE",
    "setting": "home/bedroom",
    "objects": ["t-shirt", "towel"]
  },
  "steps": [
    { "t": [0.0, 4.2], "text": "pick up shirt by shoulders" },
    { "t": [4.2, 9.8], "text": "fold sleeves to center" }
  ],
  "hand_pose": { "format": "21kp_xyz", "hands": 2 },
  "qa": { "auto": "pass", "human": "pass" }
}

Illustrative format. The final schema is shaped with launch partners.

The episode

Footage plus everything around it.

Each accepted clip ships as an episode with the labels and context your pipeline needs, in one consistent format.

Read the data format docs →
Video
1080p at 30 fps, first-person, both hands in view
Hand pose
21 keypoints per hand on every frame
Language
Timestamped, step-by-step task descriptions
Context
Country, home type, objects and surfaces
Quality
On-device checks and human review on every clip
Coverage

The home tasks robots struggle with most.

  • KitchenWashing up, food prep and loading dishes, at counter height.6 tasks
  • LaundryFolding, sorting and handling deformable fabric.6 tasks
  • TidyingSorting, organising and wiping, with precise hand movements.6 tasks

Need something else? Workshops, warehouses and retail are available as custom missions.

Washing up and food prep, at counter height, from the trainer's own eyes.

Slicing a tomato
Dicing cucumber
Rinsing a glass
Chopping greens
Rinsing a bowl
Cutting tomato

Folding and handling fabric. Deformable objects are among the hardest things for robots to learn.

Folding a shirt
Filling a pillowcase
Smoothing a sheet
Folding cloth
Sorting a basket
Unloading a washer

Sorting, organising and wiping down. The precise, everyday movements home robots need most.

Sorting small parts
Organising a drawer
Filling a basket
Arranging a drawer
Spray and wipe
Wiping a table

Need a task that isn't covered? Describe it and we open a mission for it.

Example request

Restocking shelves in small retail stores. 200 hours, Southeast Asia, both hands visible.

Request a mission

Example footage from stock libraries, shown until launch.

Custom collection

From brief to episodes, without the fleet.

  1. 01

    Brief

    Tell us the task, setting, regions and hours. We write the mission brief with you.

  2. 02

    Collect

    Trainers in your chosen regions record the task from their own point of view.

  3. 03

    Review

    Every clip is checked on the trainer’s phone, then reviewed by a person against your brief.

  4. 04

    Deliver

    Blurred, labelled episodes arrive in a consistent format as they're accepted.

Why Zeth

Data that looks like the real world.

Diversity

Every kind of home

Real kitchens, bedrooms and living rooms across many countries, with the clutter and lighting robots will actually meet.

Fit

Collected to your brief

Missions are written around your tasks, objects and settings, so you're not filtering out 90% of a bulk dump.

Privacy

Clean to use

Audio removed, faces and screens blurred, and every trainer licensed their footage under a clear agreement.

Contact sales

Tell us what you're training.

Share the tasks, settings and volume you need. We'll reply with sample episodes, timelines and pricing.

  • Sample episodes before you commit
  • Custom missions for tasks not yet covered
  • Pricing based on task, regions and volume

Contact sales

Tell us what you need and our team will reply by email.

By submitting, you agree to be contacted about Zeth data products. See our privacy policy.