World models use visual data to generate simulated environments that reproduce physics and can train robots and autonomous cars. Companies are also collecting real-world video to build them.
Odyssey CEO Oliver Cameron describes them as a way to generate simulated environments that can serve as digital training grounds.
Niantic Spatial CTO Brian McClendon says robots could practice in 3D models made from photos of their intended operating spaces. He says the software could run millions of simulations, though the article does not report independent testing of this approach.
Wayve uses a generative world model, partly powered by dashcam footage, to recreate hazardous roads and other conditions that are impractical to test in real life.