Vision Lab

Vision Lab describes itself as the industrial data layer for robotics. It supplies egocentric video footage and annotations captured in real industrial workplaces, intended for training robotics and embodied AI models that need examples of skilled physical work. The vendor says its footage spans more than 50 industries, including electronics, automotive, laboratory work, cryogenics, pharma, machine shops, chemicals, furniture, jewelry, auto repair, printing, rubber, textiles and food processing, with process knowledge at the level of standard operating procedures. Its named customer engagements cover atomic action data with clip level task diversity across more than 1,000 environments, frame accurate dense annotation with hand pose extraction, expert validated atomic action labels, and industrial procedure capture validated by engineers. Labelling is done with proprietary vision language models and human review in the loop. Customers are robotics model developers and AI laboratories. The vendor says four of the top frontier labs are customers, with identities withheld under NDA, and that it has announced a $6 million funding round. Pricing is not published. Datasets and annotation work are arranged by contacting the company directly through the site.