World Labs turns one real-world robot task into thousands of simulated variations for training

Aug 15, 2026 — THE DECODER

World Labs, the startup founded by AI pioneer Fei-Fei Li, has unveiled a simulation engine that trains robot control systems entirely in virtual environments. The models then run reliably for hours on real hardware.

The company’s “Real-to-Sim-to-Real” (R2S2R) engine turns real-world robot tasks into simulations for training and evaluating control models, cutting out expensive tests on actual hardware. The technology comes from SceniX, a startup World Labs acquired in July.

Bottleneck: experience volume, not architecture

The main bottleneck in robot deployment isn’t model architecture, World Labs says, but the sheer volume of experience a robot needs to operate reliably. Real-world data is expensive and hard to control.

One task → thousands of variations

The engine captures robots, sensors, the environment, and task demos, then rebuilds them as an interactive virtual world that behaves the same way physically. World Labs combines generative world models with task-oriented robot simulation.

From a single real-world task, the system generates thousands of variations by changing lighting, object position and count, environment, friction, and camera angle. Accuracy is validated by running the same action sequence in simulation and reality side by side.

Examples include cable routing, inserting an elastic cable end into a hole, and packing a box with both hands (rigid, movable, and deformable objects).

Transfer to real robots without real-world training

Control models train in simulation and transfer to real robots. On ALOHA (Stanford dual-arm open-source platform), models ran one hour across four additional robot platforms without human intervention — tasks included wrapping a power cord around a refrigerator, repositioning test tubes, and separating thin objects from dense jumbles.

The system is not tied to a specific control model or robot type; reconstructed worlds can be reused for new models and robots.

Simulation as policy evaluation proxy

World Labs argues simulation doesn’t need identical success rates to reality — it must preserve model rankings. On ALOHA two-handed cube handoff, simulation reproduced borderline grasps and matching failures. Across GR00T N1.6 and π₀.₅, model rankings in simulation and reality stayed largely consistent for known and unseen cube positions (2,000 simulated + 100 real runs per checkpoint).

Strategic context

World Labs was founded in 2024 by Fei-Fei Li for spatial intelligence. It raised $1B in venture capital to extend world models into robotics and science. R2S2R is the first concrete robotics application of that vision.