Procedural Generation

Algorithmically creating diverse simulation environments, objects, and scenarios at training time rather than manually designing each one. Procedural generation is used to create large-scale training datasets with varied layouts, object geometries, textures, and task configurations. It is a form of domain randomization applied to scene structure rather than physics parameters.