Notes on usable egocentric stereo data.
Buyer-facing guides on acceptance criteria, time bases, and what “accepted hours” should mean — grounded in a public sample you can inspect.
How to Choose an Egocentric Robot-Data Supplier
Supplier types, seven criteria, pilots, FAQ.
10KH-Ego Stereo vs Ego4D
Scale, purpose, license, custom collection.
10KH-Ego Stereo vs EgoDex
Scale, capture, license, custom collection.
How to Write Acceptance Criteria for Egocentric Stereo Human-Demonstration Data
Precise, measurable acceptance criteria for egocentric stereo demos — so “accepted hours” means the same thing to both sides before recording starts.
Why "Timestamps Are Recorded" vs "FPS Backfill" Is Not the Same Thing for VLA Training
A dataset that advertises “30 fps stereo” can still have a fabricated timebase. Recorded timestamps and FPS backfill are not the same product.
What Calibration Files Must Ship with Egocentric Stereo Data
Stereo depth and fusion need more than “calibration available on request.” Name the matrices, ship the version, and attach residual evidence.
Quantifying Scene Coverage of Human Demonstration Data
“More diversity” is not an acceptance metric. Separate the 89 clips / 8 scenes library from EP001/EP002 our-device proof and accepted-hour capacity.
LeRobot v3 vs MCAP — When to Choose Which for Egocentric Stereo Demos
Capture format is not delivery format. Default learning delivery to LeRobot v3; keep MCAP where runtime already depends on it.
On-Site Consent and Compliance for Egocentric Capture
Perfect stereo residuals cannot unblock a site without consent. Write compliance as a first-class acceptance column before capture day.