Brain Waves as Training Data: The Next Frontier for Physical AI
Original: Are brain waves the next unlock for physical AI?
Why This Matters
Neurological data annotation could become a key differentiator in the race to build scalable physical AI training pipelines.
Encord, a data tooling startup, is trialing EEG headsets from Zander Labs to capture human brain wave data during robotic training tasks. The goal is to tag physical AI training data with mental state signals — such as error, intent, and surprise — to improve robot model performance.
At a warehouse in San Leandro, California, Encord is running a pilot program that may reshape how robots learn. Human trainers, called 'pilots,' perform physical tasks like Jenga while wearing headsets from German neuroscience startup Zander Labs that record brain wave activity alongside egocentric video. Zander neuroscientist Lucas Gehrke says brain activity levels during a task can signal to model builders when high-effort inference is needed — effectively annotating difficulty and intent in real time.
Encord was originally founded to help machine-vision companies annotate and manage data. But as customers began applying end-to-end learning to robotic manipulation, the company determined the training data simply did not exist — and pivoted to manufacturing it. Vineeth Velmurugan, Encord's head of robot learning and a former OpenAI robotics and Berkshire Grey veteran, estimates it would take roughly five times YouTube's entire video corpus to meaningfully advance physical AI training at scale.
The brain wave integration is currently a trial: Encord plans to build an initial tagged dataset, evaluate its effect on customer robotics models, and decide whether to scale. The project sits at the 'bleeding edge' of efforts to solve the robotics data bottleneck, which has become a core business problem rather than a research curiosity.