World Labs has shown a simulation engine that turns one real-world robot task into thousands of controlled variations for training, The Decoder reports. The startup, founded by Fei-Fei Li, says the resulting controllers ran for one hour each on five robot platforms without human intervention.
The practical appeal is data efficiency. Robotics teams often struggle to collect enough real-world examples because hardware time is expensive and failures can be slow or unsafe. Simulation can multiply scenarios while keeping conditions controlled.
The important limit is transfer. A robot that succeeds in generated environments and short controlled tests may still fail in messy homes, warehouses or public spaces where objects, lighting and people change unpredictably.
The announcement is strongest as a training workflow update rather than proof of general-purpose robots. It suggests that better world simulation could make robotics data collection less manual, but real deployment will require broader and longer tests.