Robot training data is scarce enough that Encord is putting EEG headsets on the people generating it

Encord and the German neuroscience startup Zander Labs are running a trial in which workers wear EEG headsets while performing manipulation tasks such as Jenga, tagging the video with brain-derived signals for error, intent and surprise. Encord’s Vineeth Velmurugan puts densely annotated data at roughly 100 times the value of raw ego-centric footage while costing about 20 times more to collect, against an estimated requirement of a corpus around five times the size of YouTube’s video library. TechCrunch reported the trial on 26 July 2026; the multiples are the companies’ own estimates, not independent measurements.

What it means Physical AI has no internet-sized corpus to scrape, so the unit economics of annotation — not model architecture — is what decides how fast robots get good.