Blog
The data layer for physical-world AI.
Practical, researched writing on how humanoids and embodied agents actually learn — and the real-world human data that gets them there.
Diversity over volume
Research·11 min read
Do scaling laws hold for humanoid learning data?
May 22, 2026
Coverage over volume
Research·9 min read
Why diverse home data matters more than volume for manipulation
May 18, 2026
Teleop or human video?
Engineering·10 min read
Teleoperation vs Human Video for Robot Manipulation Data
May 14, 2026
Raw or labelled?
Guide·10 min read
Raw vs labelled data: which does your robot policy need first?
May 10, 2026
Learning, first-person
Research·10 min read
Egocentric video as the training substrate for embodied AI
May 6, 2026
The long tail of chores
Research·10 min read
The long tail of home tasks (and why it is hard for robots)
May 2, 2026
Cleared on every clip
Trust·10 min read
How We Clear Rights on Every Clip of Training Data
April 29, 2026
The reality gap
Engineering·10 min read
Closing the sim-to-real gap with real human data
April 24, 2026
Labels built to last
Guide·11 min read
How to design an action-labeling taxonomy for production
April 18, 2026
What is a world model?
Research·10 min read
What Is a World Model, and What Data Does It Require?
April 12, 2026
Two hands, real data
Engineering·11 min read
Bimanual manipulation requires bimanual training data
April 8, 2026
Consent-first by design
Trust·11 min read
Building a Consent-First Data Supply Chain for AI
March 30, 2026
One rig, every clip
Engineering·10 min read
Standardising the capture rig: why consistent sensors make robot datasets usable
March 22, 2026