Introducing NVIDIA Cosmos Policy for Advanced Robot Control

NVIDIA has introduced Cosmos Policy, a state-of-the-art robot control policy for advanced manipulation tasks. It is created by post-training the Cosmos Predict world foundation model, which is designed to predict future video frames. The model uniquely encodes robot actions, physical states, and success scores as latent frames, allowing it to inherit the base model's understanding of physics and temporal dynamics through a unified diffusion process. This single model can predict action chunks, future observations, and expected returns, achieving superior performance on robotics benchmarks like LIBERO and RoboCasa.