AI chipmaker Nvidia has unveiled Nvidia Cosmos, a platform designed to advance the development of physical artificial intelligence (AI) systems, including robots and autonomous vehicles (AVs).

The platform integrates “state-of-the-art” generative world foundation models, advanced tokenisers, guardrails, and a high-performance video processing pipeline.

Developing physical AI models is claimed to be a complex and costly process, requiring vast amounts of real-world data and extensive testing, the company noted.

Cosmos World Foundation Models (WFMs) provide developers with a simplified approach to generating massive volumes of photorealistic, physics-based synthetic data.

This approach enables them to train and evaluate their existing models more efficiently.

Developers can also customise their models by fine-tuning Cosmos WFMs.

To accelerate the progress of the robotics and AV sectors, Cosmos models will be made available under an open model licence.

Developers can preview the initial models via the Nvidia API catalogue or download the full set of models and fine-tuning framework from Nvidia NGC or Hugging Face.

Among the early adopters of Cosmos are leading robotics and automotive companies such as 1X, Agile Robots, Agility and ridesharing platform Uber.

Nvidia founder and CEO Jensen Huang said: “The ChatGPT moment for robotics is coming. Just as large language models are transforming AI, world foundation models will be pivotal to the evolution of robots and AVs.

“However, not all developers have the expertise or resources to train their own models. We created Cosmos to democratise physical AI and make general robotics accessible to every developer.”

In December 2024, Nvidia completed the acquisition of Israeli AI company Run:ai after undergoing antitrust scrutiny in a deal valued at $700m.