Human demonstrationFrame 0184
Hand · L
Hand · R
Plate
Physical AI · Data services
The real world isn’t easy.
HumanSignal creates novel datasets around the exact environments, actions, sensors, and edge cases your robots need to learn.
HumanSignal Data Services
Specialists in novel datasets.
We turn human demonstrations into structured training examples, with the action boundaries, hand-object interactions, and taxonomy your pipeline expects.
Robot actionCorrespondence 0.94
Gripper · L
Gripper · R
Plate
From capture to training data
Precise data capture paired with precise annotation.
As the creators of Label Studio Enterprise, we build advanced annotation interfaces and custom rubrics that turn complex requirements into consistent, high-quality labels.
Data collection
Capture exactly the data you need. Two ways.
Use purpose-built environments in our secure facilities for precise control, or collect across diverse settings through our global contributor network.
Controlled collection
Controlled capture in our secure facilities.
We build and run purpose-built capture environments around your hardware, participant criteria, recording setup, and acceptance tests.
Purpose-built environments
Prototype hardware
Controlled light and audio
Synchronized sensors
Participant recruitment
Repeatable protocols
Real-world collection
Diverse capture across real-world environments.
Our global contributor network collects across the homes, layouts, objects, people, and task styles your system will encounter outside the lab.
27
Household activities
1080p
Head-mounted POV
30fps
Continuous capture
Captured POVHousehold 014
- Nigeria
- Kenya
- India
- United States
- Philippines
- South Africa
- Pakistan
- Indonesia
- Vietnam
- Romania
- Brazil
- Canada
- Colombia
- United Kingdom
- Turkey
- Italy
- Bangladesh
- Morocco
- Thailand
- Argentina
- Germany
- Spain
- Poland
- Ghana
- Egypt
- Mexico
- Malaysia
- Ethiopia
- Rwanda
- Peru
- Finland
- Sri Lanka
- Chile
- Uganda
- Tunisia
- Netherlands
- Tanzania
- Greece
- France
- Ecuador
- Portugal
- Japan
- Norway
- South Korea
- Nepal
- Cameroon
- Ireland
- Zambia
- Zimbabwe
- Hungary
- Austria
- Serbia
- United Arab Emirates
- Australia
- New Zealand
- Algeria
- Switzerland
- Saudi Arabia
- Jamaica
- Moldova
- Trinidad and Tobago
- Singapore
- Slovenia
- Costa Rica
- Czech Republic
- North Macedonia
- Hong Kong
- Iceland
- Panama
- Dominican Republic
- Mozambique
- El Salvador
- Slovakia
- Uzbekistan
- Sweden
- Georgia
- China
- Honduras
- Botswana
- Bosnia and Herzegovina
- Jordan
- Kazakhstan
- Ivory Coast
- Mongolia
- Nicaragua
- Israel
- Bulgaria
- Denmark
- Cyprus
- Guatemala
- Croatia
EgoChores sample dataset
Download a sample dataset.
Request a free one-hour sample of novel first-person household video, captured and annotated for dexterous hand-object manipulation.
01
Raw video
Judge framing, lighting, household variation, and continuous task coverage.
02
Standard annotation
Inspect temporal boundaries, hand-object interaction, and the published object taxonomy.
03
Custom annotation
Talk to us about how your own object classes, edge cases, or proprietary schema can be applied to captured datasets.
Request the sample
Request received
Tell us where to send the EgoChores sample and which annotation depth you want to evaluate.
Physical AI · Data services
Physical AI demands novel data creation.
Tell us what your robots need to learn. We will define the capture, annotation, and quality plan to build it.