NVIDIA is transforming autonomous vehicle (AV) development through cutting-edge simulated driving environments, which enable safe and cost-effective training, testing, and validation of AV systems. These environments are built using neural reconstruction of real-world fleet data or generated through world foundation models (WFMs)—AI systems that understand physics and real-world behaviors. At the recent GTC Paris and CVPR conferences, NVIDIA introduced major advancements in WFMs, enhancing its Cosmos platform, which combines generative WFMs, advanced tokenizers, guardrails, and accelerated data processing. Innovations such as Cosmos Predict-2, the Cosmos Transfer-1 NIM microservice preview, and Cosmos Reason allow AV developers to create highly realistic synthetic datasets and scalable simulation environments. Central to this process is Universal Scene Description (OpenUSD), a standardized framework that ensures seamless integration of 3D assets across development workflows. NVIDIA Omniverse, a suite of APIs, SDKs, and services built on OpenUSD, supports world-scale simulations using WFMs and neural reconstruction. Industry leaders like Foretellix, Mcity, Oxa, Parallel Domain, Plus AI, and Uber are already leveraging Cosmos models to drive the future of AV development.
