None defined yet.
Omni-Modal LLMs Know More Than They Say
Learning Transferable World-Action Models from Task-Paired Human Videos