Addvista

Berlin-based Addvista captures physical AI data from real construction sites: tradespeople record their normal work from a helmet camera or smart glasses, and the platform turns that footage into verified training demonstrations — each clip checked, human-confirmed, with faces, plans and screens blurred, and proof of rights attached. Robotics labs and robot makers then license the result non-exclusively or commission capture to spec.

Robots are learning from human video, but skilled construction work is almost absent: a robot model needs hundreds of thousands of hours of demonstrations, and construction has only a few hundred today. Addvista is built to close that gap while skilled labour shortages make the data more valuable than ever.

The company says the core insight is that robots are trained on human video yet the trades — some of the hardest physical work to automate — are missing from the data. Addvista's answer is a marketplace where contractors get a free team tool, tradespeople are paid per verified clip and participate voluntarily, and labs get data they can't source elsewhere.

Addvista is led by founder David Ban, building a three-sided model that has to win trust with contractors, tradespeople and robotics labs at once. With the product starting narrow — electrical and building services in Germany — the team is choosing focus over breadth, which is the right instinct for a hard data problem.