World Labs Unveils Real-to-Sim Platform for Robot Training

World Labs reports early results after SceniX joined, advancing its spatial intelligence work with a real-to-sim-to-real engine. Generative world models convert physical robot tasks into many controllable virtual environments that preserve dynamics and interactions. This approach addresses the robotics data bottleneck by enabling scalable training, earlier failure detection, and reduced hardware trials compared with language model data abundance. Company posts and team statements confirm the integration supports policy evaluation and reusable simulation worlds, with side-by-side examples shared publicly.