Encord Tests Brain-Wave Data to Train Physical AI

Humanoid and warehouse robots increasingly face a data bottleneck rather than a shortage of model ideas. Internet video rarely provides the synchronized viewpoints, action trajectories and contextual labels needed to teach reliable physical manipulation, while real-world demonstrations are costly to collect. Data infrastructure startup Encord is working with German neuroscience company Zander Labs to test whether EEG readings can capture an operator’s intent, surprise, cognitive effort and recognition of errors, potentially making each training example more useful.

TechCrunch reported on July 26, 2026, that Encord had begun a trial at its San Leandro, California, warehouse, where robot pilots wear Zander Labs sensors while completing tasks including Jenga manipulation. The companies plan to create an initial brain-wave-tagged dataset, test it on customer robotics models and measure any performance gains before scaling. Encord estimates densely annotated task data can be 100 times more valuable than low-quality egocentric footage while costing about 20 times as much to produce. Financial terms of the collaboration were not disclosed.