In a warehouse in San Leandro, California, a worker named Andrew Ceja pulls blocks from a Jenga tower while wearing a headset that tracks his eye movements and brain activity. Ceja works for Encord, a company that builds data tooling for AI models. The headset, made by German startup Zander Labs, measures brain waves to deduce mental states like surprise or intent. The idea: such signals could create richer training data for physical AI systems.
Encord is betting that the biggest bottleneck for humanoid and warehouse robots isn’t model architecture but the sheer shortage of real-world training data. “The data simply does not exist,” says Vineeth Velmurugan, Encord’s head of robot learning. Rather than just managing existing data, Encord is manufacturing it—using cameras, sensors, and now brain-wave readings. The Zander trial aims to see if neural activity tagging improves model performance before scaling. Velmurugan estimates that building a dataset for physical AI may require five times the video content on YouTube, making data generation a business, not just a research problem.
Encord also collects egocentric video from workers wearing head cameras, and runs remote-operated robotic arms for fine-tuning tasks like pouring coffee or plugging Ethernet cables. Another modality: armband sensors that detect muscle signals to reconstruct hand position in 3D. All footage is annotated with dense physical descriptions—like “right hand tightens bolt”—which Velmurugan says is worth 100 times more than basic video for training, though it costs 20 times more to produce. That cost gap highlights the challenge: unlike scraping text for LLMs, physical training data must be manufactured, not just collected, shifting the economics of building these models.
Encord sits between many robotics companies, giving it a unique view of which data techniques gain traction industry-wide. That vantage point keeps its dozen pilots busy. Ceja, who previously worked at Scale and now trains robots, says the work offers constant novelty. As his Jenga tower finally collapses, he smiles: “It’s something new every day.”
Source: TechCrunch