NeoWorld-Pro: Programming Interactive Scenes from Monocular Images for Embodied Simulation
NeoWorld-Pro reformulates scene reconstruction by converting images into executable programs that define 3D objects and their physics properties. It uses Large Multimodal Models (MLLMs) to generate these programs, followed by a two-level physics-in-the-loop refinement process. This results in interactive, physically plausible 3D environments ready for robot simulation.