A million videos doesn't automatically make a robot smarter.
That's one thing I find interesting about what
@axisrobotics is doing with Physical AI.
Humans can watch a video of someone picking up a cup and understand what happened.
A robot needs much more.
It needs to understand the action.
Where the hand moved.
How the object was positioned.
What happened when the grip was wrong.
How the action changes when the object, environment or conditions change.
That's why structured trajectories matter.
Axis isn't just trying to collect more content.
It is creating tasks where people can control simulated robots, generate demonstrations, process those trajectories and turn them into data that can actually be used for training.
The goal isn't:
more data.
It's:
more useful data, with more diversity.
And when you can generate that at scale, the training loop starts looking very different.
That's the part of Axis I'm paying attention to.