Encord Trials Brain-Wave Headset for Robot Data as Physical AI Hunts 5x YouTube Scale
Updated
Updated · TechCrunch · Jul 27
Encord Trials Brain-Wave Headset for Robot Data as Physical AI Hunts 5x YouTube Scale
1 articles · Updated · TechCrunch · Jul 27
Summary
In San Leandro, Encord is testing a Zander Labs headset that records brain waves while human trainers perform tasks like Jenga, aiming to build an initial brain-tagged dataset for robotics models.
The trial targets what Encord sees as robotics' main bottleneck: scarce real-world training data. It plans to run the dataset through customer models first, then decide whether the approach improves performance enough to scale.
Encord has expanded from managing vision-training data to manufacturing it, collecting egocentric video in factories, using leader-follower robot rigs, and developing forearm sensors to infer hand motion in 3D.
Vineeth Velmurugan says dense, physically annotated data can be worth 100 times more than low-quality ego video while costing 20 times more, underscoring why physical-AI training data has become a business problem.
The broader constraint is scale: Encord estimates a breakthrough may require a dataset roughly five times YouTube's video corpus, making data generation—not just model design—the frontier for humanoid and warehouse robotics.