Encord and Zander Labs bet brain-wave data could solve robotics’ physical AI bottleneck

The next frontier for physical artificial intelligence may not be a new algorithm or a bigger compute cluster, but a headset that reads the electrical activity of a human brain. In a warehouse in San Leandro, California, data-tooling company Encord is running a trial with German neuroscience startup Zander Labs to see whether brain-wave readings can produce more useful training data for robots.

The experiment is part of a growing recognition that the scarcest resource for humanoid and warehouse robotics is not model architecture but real-world physical training data. Encord, which originally built tools for managing and annotating machine-vision data, has shifted to manufacturing the data its customers cannot find anywhere else.

“The data simply does not exist,” Vineeth Velmurugan, Encord’s head of robot learning, told TechCrunch during a visit to the facility. Velmurugan, a former OpenAI robotics lab and Berkshire Grey veteran, joined Encord to build its internal data-creation team.

The brain-wave headset was developed by Zander Labs. It measures neural activity to deduce mental states such as error, intent and surprise. Lucas Gehrke, a Zander neuroscientist supervising the trial, says the amount of brain activity at any point during a task offers clues about when a model should deploy its highest-effort reasoning.

Encord’s work with Zander is currently a trial run. The company plans to build an initial brain-wave-tagged dataset, run it through customer robotics models, and evaluate whether performance improves before deciding to scale the approach.

The data bottleneck

Frontier AI labs have trained large language models on the text of the entire internet, but no equivalent repository exists for physical manipulation. Velmurugan estimates that breaking through the physical AI data barrier will require a dataset “something like five times the size of YouTube’s video corpus.” That scale explains why data-generation has become a business in itself, not just a research problem.

Encord’s San Leandro facility houses a dozen pilots who collect data using two main methods. The first is egocentric video captured by workers wearing head-mounted cameras, often augmented with additional camera angles and other metrics. The second involves collecting data from robots operated remotely using leader-follower rigs, where a human controls one robotic arm and a second mimics its movements.

During TechCrunch’s visit, pilots were training tasks such as pouring coffee from a pot into mugs and stacking poker chips. Storage racks held cartons of fake flowers, books, plastic vegetables, kitty litter trays and scoops – the stock in trade for training household manipulation.

“Every humanoid company has asked us for these pieces,” Velmurugan said.

Modalities beyond video

Encord is also experimenting with a set of sensors strapped to the forearm to detect electrical signals in muscles. Video of human hands manipulating objects often does not capture the entire hand, but Velmurugan hopes to reconstruct a 3D depiction of hand position based on arm sensors, giving models a more robust understanding.

The datasets are annotated with physical descriptions such as “right hand tightens bolt” to help LLM-based models understand what is happening. Velmurugan says dense annotation is worth “100 times as much as junky ego data” for training specific tasks, and costs only 20 times more to produce – a favourable trade-off on paper, but still real money.

Economics of manufactured data

The comparison with LLM training breaks down at the cost level. Scraping text from the internet cost frontier labs next to nothing. Generating physical training data does not, and that changes the economics of building these models.

Encord draws egocentric data from several factories around the globe, while its San Leandro facility serves as a testbed for new modalities such as brain waves. Velmurugan says he can see across the industry which data techniques are gaining traction, and that vantage point is part of Encord’s pitch to robotics companies.

Both pilots at the facility, Sofia Infante and Andrew Ceja, previously worked at Scale, another AI data annotation firm, before joining Encord. Ceja had worked at a waste management company before moving into data work.

What this means for Indian AI

For Indian AI startups, robotics firms and data annotation companies, the Encord-Zander trial highlights the increasing premium on high-quality, real-world physical data. Indian enterprises that build automation solutions for logistics, manufacturing or warehouse management may need to invest in similar data-generation pipelines rather than rely on open datasets.

The trial also signals that brain-wave data, while still experimental, is being taken seriously by major robotics firms. If the approach scales, it could create new opportunities for Indian companies that develop neuroscience hardware, biomedical sensors or annotation services tailored to physical AI tasks.

Source: TechCrunch — https://techcrunch.com/2026/07/26/are-brain-waves-the-next-unlock-for-physical-ai/

Datos clave

Punto	Detalle
Fuente	TechCrunch AI
Fecha	2026-07-27T00:19:14+00:00
Tema	Are brain waves the next unlock for physical AI?