NVIDIA Cosmos 3: One World Model for Video Generation, Understanding and Robot Action

In an interview with Machine Learning Street Talk, NVIDIA's Ming-Yu Liu explained that Cosmos 3 is a single world model that generates video, understands video, and outputs robot actions, allowing robots to rehearse and validate policies in simulation before acting in the real world.

2026-09-16 ~ 2026-09-16 · 3 related posts

• NVIDIA's Ming-Yu Liu on Cosmos 3: one model for video, understanding and robot actions — Machine Learning Street Talk · 2026-09-16
• NVIDIA's Cosmos 3 world model lets robots rehearse actions in simulation before moving — Machine Learning Street Talk · 2026-09-16
• NVIDIA Cosmos 3 turns video into robot actions for sim-first policy testing — heghbalz · 2026-09-16