Why egocentric data companies are suddenly everywhere
Trying to understand the influx of human data companies — and what they promise.
A growing number of human/egocentric data companies have been emerging lately, and I've been trying to understand why.
Sharing what I've found so far, though I haven't reached a conclusion yet.
Most of these companies are chasing a similar objective: scaling human data collection with better hardware (gloves, wearable cameras, iPhone head mounts) that captures people doing tasks with their own hands.
Why now, exactly?
Part of the answer seems to be early signs of scaling laws and adaptability. Research commonly cited includes EgoMimic (2024), EgoZero (2025), and EgoScale (2026), all with promising results.
Here, scaling laws refers to the general trend of improvement in model performance as the data scales, and adaptability refers to needing less downstream data to adapt a model to a new environment.
Teleoperation has been the predominant way to collect training data for robotics models, but the interface has been a constraint.
Teleoperating a robot to do something like electrical cabling is nothing like doing it with your own hands. Fine dexterous work isn't yet as intuitive through a teleop setup, so collecting that data can be prohibitively expensive.
Whether human data actually translates into model capabilities that unlock new robot tasks in the real world is the part I'm still figuring out.
For now, the approaches I see deployment teams prefer are large, diverse robot datasets and, increasingly, UMI-type data.
Will share more as I learn.