World Labs says R2S2R can make robot training and evaluation far cheaper

drfeifei · x · 2026-07-29

World Labs says real-to-sim-to-real can speed up robot training and evaluation

World Labs argues that the simulator is the linchpin in world models for robotics: it is where agents can act, learn, and be evaluated.

The company says its R2S2R engine can turn one physical task into many reusable virtual worlds.
That would let robotics teams train policies faster, catch failures earlier, and reduce costly hardware-bound iteration.
The post ties the effort to SceniX, a robotics and simulation company that joined World Labs on July 21.
World Labs says SceniX has been building systems that convert real robots, environments, and interactions into simulations for policy training and evaluation.
The goal is to make robotic evaluation more scalable by reproducing not just the task, but the failure conditions that lead a policy to succeed or fail in both virtual and physical worlds.

More from Embodied