Nvidia has introduced Cosmos, a platform designed to help developers create and refine advanced AI systems for robotics and autonomous vehicles (AVs). The platform aims to reduce costs and risks associated with building physical AI models by offering tools to generate, process, and test data in simulated environments, according to an official press release by Nvidia.

Cosmos includes world foundation models (WFMs) that allow developers to generate photorealistic, physics-based synthetic data for training AI systems. These models can simulate real-world scenarios, such as various road conditions or warehouse environments, enabling developers to test and improve their AI without relying heavily on real-world data collection.

The platform also features tools like the Cosmos Tokenizer, which processes video and image data significantly faster and more efficiently than current methods, and the Nemo Curator, which prepares massive datasets for AI training in a fraction of the usual time, says Nvidia. These tools are intended to make it easier to develop and evaluate AI systems for complex physical tasks.

According to the leading chip company, companies such as Agility Robotics, XPENG, and Uber are already adopting Cosmos. Uber, for instance, is using Cosmos to combine its extensive driving data with Nvidia's tools to accelerate the development of autonomous driving technology. Other robotics firms are leveraging Cosmos to simulate diverse scenarios for training robots to operate in real-world environments, states the press release.

Cosmos is available under an open model license, with resources accessible on platforms like Hugging Face and the Nvidia NGC catalogue. Nvidia has also included safeguards in the platform, such as watermarks in AI-generated videos to prevent misinformation and ensure responsible use.