NVIDIA has introduced a new platform, NVIDIA Cosmos, designed to enhance the development of physical AI systems, including autonomous vehicles and robotics. The platform features generative world foundation models (WFMs), advanced tokenizers, guardrails, and an accelerated video processing pipeline.

The Cosmos platform allows developers to efficiently create synthetic data for training AI agents, mitigating the high costs associated with building physical AI models. With the release of the open model license, developers can customize Cosmos WFMs to suit their applications.

Major robotics and automotive companies, such as Uber, XPENG, and Waabi, are among the first to implement Cosmos technology. According to Jensen Huang, NVIDIA's founder and CEO, the platform aims to democratize access to physical AI, reducing the barriers for developers with limited resources.

The Cosmos WFMs can generate realistic, physics-based videos from various inputs, enhancing model training while lessening the need for expensive real-world data collection. The platform has been designed to streamline data processing, requiring significantly less time compared to traditional methods.

Industry leaders are leveraging Cosmos for diverse applications, including searching for driving scenarios or developing high-fidelity testing environments. Huang emphasized that just as large language models have transformed text-based AI, world foundation models hold similar potential for robotics and AV development.

NVIDIA Cosmos is now available for developers, with models accessible through the NVIDIA NGC catalog and Hugging Face, providing support for customization and deployment in various environments.