Innodata opens motion-capture lab to help humanoids move more like people

A company has opened a laboratory that uses high-precision motion-capture technology to record how people move, creating training data for humanoid robots and other physical AI systems. The lab addresses a critical shortage of real-world interaction data needed to train these robots, which unlike language models cannot rely on vast amounts of existing data from the internet. The facility captures detailed 3D movement data directly from bodies rather than inferring it from video, achieving sub-millimeter accuracy that helps ensure robots learn movements precisely enough to perform tasks safely and effectively. This approach matters because robots trained on approximate data will themselves be approximate, potentially leading to mistakes in real-world tasks like picking up objects in different ways depending on context.