Nvidia has unveiled the Cosmos platform aimed at developing and training physical AI models used for video analytics, autonomous vehicles, and robotics.
Cosmos Platform and Its Capabilities
Nvidia announced the Cosmos platform which allows developers to create tailored AI models based on pretrained models. The company claims that Cosmos utilizes spatiotemporal understanding of the physical world for preparing data that improves decision-making in robotics and autonomous vehicles.
Cosmos Functional Models
The Cosmos platform features three main functional models: Predict, Transfer, and Reason. Predict allows for generating videos up to 30 seconds long based on multimodal input data. Transfer simulates different environments, while Reason provides video analytics to assist in decision-making in urban and industrial settings.
Creating Digital Twins with Nvidia
Nvidia has also introduced the Omniverse libraries, enabling developers to create physically accurate digital twins. These libraries leverage RTX PRO Servers and DGX Cloud to generate synthetic data from the real world, which is essential for training physical AI models.
In conclusion, the launch of the Cosmos platform and Omniverse libraries highlights Nvidia's commitment to supporting developers in creating innovative AI solutions across various industries.
