Why robots need data from skilled labor
The tasks we most want robots to do are the ones that are hardest to learn from the internet.
Most of the data used to train AI models comes from the internet. That works well for language and images, but it says very little about how a skilled person actually wires a panel, welds a joint or kits parts on a line.
Physical skill lives in hands, tools and small decisions made in the moment. It shows up in the angle of a wrist, the order of steps, and the checks a professional makes without thinking. None of that is written down, and very little of it is filmed in a way a model can learn from.
That is why we capture from the people who do the work, on the sites where it happens. First-person video shows the task the way the worker sees it. Third-person views, depth and hand tracking add the geometry. Narration and step labels add the intent.
The result is data that teaches a model not just what the end state looks like, but how a skilled person gets there.