Episode Details

Plug-and-play framework that teaches multimodal LLMs spatial reasoning by extracting implicit 3D priors from video diffusion models, supporting geometric scene understanding and embodied decision-making without explicit 3D supervision.