One real robot demonstration, thousands of simulated copies — and the robot then works without ever touching real data
World Labs, Fei-Fei Li's company, published the first results from the real-to-sim-to-real engine it acquired with the startup SceniX in July. The system takes a single recorded real-world robot task and turns it into thousands of controlled variations, changing lighting, object placement, physics properties and camera angles, then trains control policies entirely inside those simulations. World Labs says policies trained with no real-world data transferred straight onto hardware and ran autonomously for an hour across four further robot platforms with no human intervention. For evaluation it compared 2,000 simulated runs against 100 real runs per checkpoint and reports that model rankings stayed largely the same in simulation as on real hardware. The reference platform used, ALOHA, is open source with published blueprints.
Related stories
Nevada just cleared up to 8,000 driverless taxis for Las Vegas — Tesla's ceiling went from 10 to 5,000 in five weeks2026-08-21Apple Music will tell you when a song was made by a machine — but the uploader decides whether to say so
2026-08-21Stripe has just paid $7.5 billion for a model router. Days later Ramp built one and is giving it away until January.
2026-08-21Meta's assistant is now a Mac app that reads your screen and types into any window — and what it sees can train the model
2026-08-21The doctors and lawyers who grade AI answers are now a half-billion-dollar business — and this one grew fivefold in eight months
2026-08-21