NVIDIA Launches Cosmos World Foundation Model Platform to Accelerate Physical AI Development

NVIDIA Cosmos Platform Overview - NVIDIA announced NVIDIA Cosmos™, a platform designed to advance the development of physical AI systems such as autonomous vehicles (AVs) and robots [1] - The platform includes state-of-the-art generative world foundation models, advanced tokenizers, guardrails, and an accelerated video processing pipeline [1] Key Features of Cosmos - Cosmos world foundation models (WFMs) enable developers to generate massive amounts of photoreal, physics-based synthetic data for training and evaluating models [2] - Developers can fine-tune Cosmos WFMs to build custom models [2] - Cosmos models are available under an open model license, accessible via the NVIDIA API catalog, NVIDIA NGC™ catalog, or Hugging Face [3] Industry Adoption - Leading robotics and automotive companies, including 1X, Agile Robots, Agility, Figure AI, Foretellix, Fourier, Galbot, Hillbot, IntBot, Neura Robotics, Skild AI, Virtual Incision, Waabi, and XPENG, along with Uber, are among the first to adopt Cosmos [4][11] - 1X launched the 1X World Model Challenge dataset using Cosmos Tokenizer, while XPENG is using Cosmos to accelerate humanoid robot development [11] - Hillbot and Skild AI are leveraging Cosmos to fast-track general-purpose robot development [11] Applications of Cosmos - Cosmos WFMs are purpose-built for physical AI research, generating physics-based videos from text, image, video, robot sensor, or motion data [6] - The platform supports advanced world model development tools, including an NVIDIA AI and CUDA®-accelerated data processing pipeline that can process 20 million hours of videos in 14 days using the NVIDIA Blackwell platform [7][10] - Cosmos enables video search and understanding, physics-based photoreal synthetic data generation, physical AI model development and evaluation, and foresight simulation [9] Partnerships and Use Cases - Uber is partnering with NVIDIA to accelerate autonomous mobility, leveraging rich driving datasets and Cosmos platform features [17] - Waabi is evaluating Cosmos for AV software development and simulation, while Wayve is using it to search for edge and corner case driving scenarios [17] - Foretellix is using Cosmos alongside NVIDIA Omniverse Sensor RTX APIs to generate high-fidelity testing scenarios and training data at scale [17] Open and Trustworthy AI - Cosmos was developed in line with NVIDIA's trustworthy AI principles, prioritizing privacy, safety, security, transparency, and reducing bias [13] - The platform includes guardrails to mitigate harmful content and features invisible watermarks on AI-generated videos to reduce misinformation [14] Availability and Additional Tools - Cosmos WFMs are available under NVIDIA's open model license on Hugging Face and the NVIDIA NGC catalog [15] - NVIDIA also announced new NVIDIA Llama Nemotron large language models and NVIDIA Cosmos Nemotron vision language models for enterprise AI use cases [16] - The NVIDIA NeMo framework is available for efficient model training, customization, and optimization [16] Industry Impact - Cosmos is seen as a democratizing force for physical AI, enabling developers to build general robotics without requiring extensive expertise or resources [5] - The platform is expected to accelerate the development of autonomous vehicles and robotics, addressing challenges like data scarcity and variability [12]