NVIDIA met à disposition des ressources open-source essentielles pour quiconque développe des systèmes d'IA physique, notamment pour la robotique et les véhicules autonomes.
Ces publications open-source de NVIDIA sont cruciales car elles démocratisent l'accès à des outils et des données de pointe, permettant aux développeurs d'accélérer la recherche et le développement dans des domaines critiques comme la robotique avancée et les véhicules autonomes, en fournissant une base solide pour l'innovation et la standardisation.
Lors de sa conférence GTC 2025, NVIDIA a dévoilé des avancées significatives pour l'IA physique en rendant disponibles de nouveaux outils et ressources open-source. Ces innovations incluent un modèle de fondation mondiale pour la génération de scènes virtuelles contrôlables, un vaste ensemble de données pour l'entraînement de modèles d'IA physique, et un modèle spécifiquement conçu pour le raisonnement et les compétences des robots humanoïdes, visant à accélérer le développement de la robotique et des véhicules autonomes.
Impact concret
Les développeurs peuvent dès maintenant utiliser Cosmos Transfer pour générer des données synthétiques contrôlables à grande échelle, le Physical AI Dataset pour entraîner leurs modèles avec des données de qualité commerciale, et Isaac GR00T N1 pour doter les robots humanoïdes de capacités de manipulation complexes et de raisonnement multimodal dans des tâches comme la manutention ou l'inspection.
Applications concrètes
Cosmos Transfer permet la création de séquences vidéo photoréalistes avec un contrôle précis du placement des objets et du mouvement pour la simulation de véhicules autonomes. Le Physical AI Dataset fournit des trajectoires et des actifs OpenUSD pour l'entraînement de modèles de fondation en robotique. Isaac GR00T N1 peut être utilisé pour des tâches de manipulation d'objets avec un ou deux bras, le transfert d'objets entre bras, et des tâches multi-étapes dans la manutention, l'emballage et l'inspection.
Back to Articles NVIDIA's GTC 2025 Announcement for Physical AI Developers: New Open Models and Datasets Published March 18, 2025 Update on GitHub Upvote 42 +36 Ming-Yu Liu mingyuliutw Follow nvidia Hanzi Mao hannamao Follow nvidia Jinwei Gu jwgu Follow nvidia Pranjali Joshi PranjaliJoshi Follow nvidia Asawaree asawareeb Follow nvidia NVIDIA Isaac GR00T N1 used in object manipulation. New World Foundation Model - Cosmos Transfer How it works Open Physical AI Dataset Purpose Built Model for Humanoids - NVIDIA Isaac GR00T N1 Path Forward At its annual GTC conference, NVIDIA has unveiled a trio of groundbreaking open-source releases aimed at accelerating physical AI development. Release of a new suite of world foundation models (WFMs) with multicontrols called Cosmos Transfer , a highly curated Physical AI Dataset , and the first open model for general humanoid reasoning called NVIDIA Isaac GR00T N1 - represent a significant leap forward in physical AI technology, offering developers powerful tools and resources to advance robotics systems, and enhance autonomous vehicle technology. New World Foundation Model - Cosmos Transfer Cosmos Transfer , the latest addition to NVIDIA's CosmosTM world foundation models (WFMs), introduces a new level of control and accuracy in generating virtual world scenes. Available in 7 billion parameter size, the model utilizes multicontrols to guide the generation of high-fidelity world scenes from structural inputs, ensuring precise spatial alignment and scene composition. How it works The model is built by training individual ControlNets separately for each sensor modality used to capture the simulated world. Input types include 3D bounding box map, Trajectory map, Depth map, Segmentation map. At inference time, developers can use various input types, including structured visual or geometric data such as segmentation maps, depth maps, edge maps, human motion keypoints, LiDAR scans, trajectories, HD maps, and 3D bounding boxes to guide the output. The control signals from each control branch are multiplied by their corresponding adaptive spatiotemporal control maps and then summed before being added to the transformer blocks of the base model. The generated output is photorealistic video sequences with controlled layout, object placement, and motion. Developers can control the output in multiple ways, such as preserving structure and appearance or allowing appearance variations while maintaining structure. Outputs from Cosmos Transfer varying environments and weather conditions. Cosmos Transfer coupled with the NVIDIA Omniverse platform is driving controllable synthetic data generation for robotics and autonomous vehicle development at scale. Find more Cosmos Transfer Examples on GitHub. Cosmos Transfer samples built using post-training base model are also available for autonomous vehicles. Open Physical AI Dataset NVIDIA has also released Physical AI Dataset , an open-source dataset on Hugging Face for developing physical AI. This commercial-grade, pre-validated dataset consists of 15 terabytes of data representing more than 320,000 trajectories for robotics training, plus up to 1,000 Universal Scene Description (OpenUSD) assets, including a SimReady collection. The dataset is designed for post-training foundation models like Cosmos Predict world foundation models, providing developers with high-quality, diverse data to enhance their AI models. Purpose Built Model for Humanoids - NVIDIA Isaac GR00T N1 Another exciting announcement is the release of NVIDIA Isaac GR00T N1 , the world's first open foundation model for generalized humanoid robot reasoning and skills. This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments. The NVIDIA Isaac GR00T-N1-2B model is available on Hugging Face . Isaac GR00T N1 was trained on an expansive humanoid dataset, consisting of real captured data, synthetic data generated using components of the NVIDIA Isaac GR00T Blueprint , and internet-scale video data. It is adaptable through post-training for specific embodiments, tasks and environments. Isaac GR00T N1 uses a single model and set of weights to enable manipulation behaviors on various humanoid robots, such as the Fourier GR-1 and 1X Neo . It demonstrates robust generalization across a range of tasks, including grasping and manipulating objects with one or both arms, as well as transferring items between arms. It can also execute complex, multi-step tasks that require sustained contextual understanding and the integration of diverse skills. These capabilities make it well-suited for applications in material handling, packaging, and inspection. Isaac GR00T N1 features a dual-system architecture inspired by human cognition, consisting of the following complementary components: Vision-Language Model (System 2) : This methodical thinking system is based on NVIDIA-Eagle with SmolLM-1.7B . It interprets the environment through vision and language instructions, enabling robots to reason about their environment and instructions, and plan the right actions. Diffusion Transformer (System 1) : This action model generates continuous actions to control the robot's movements, translating the action plan made by System 2 into precise, continuous robot movements. Path Forward Post-training is the path forward to advancing autonomous systems, creating specialized models for downstream physical AI tasks. Check out GitHub for Cosmos Predict and Cosmos Transfer inference scripts. Explore the Cosmos Transfer research paper for more details. The NVIDIA Isaac GR00T-N1-2B model is available on Hugging Face . Sample datasets and PyTorch scripts for post-training using custom user datasets, which is compatible with the Hugging Face LeRobot format are available on GitHub . For more information about the Isaac GR00T N1 model, see the research paper . Follow NVIDIA on Hugging Face for more updates. Models mentioned in this article 3 HuggingFaceTB/SmolLM-1.7B Text Generation • 2B • Updated Oct 16, 2024 • 70.9k • 181 nvidia/Cosmos-Transfer1-7B-Sample-AV Updated Apr 9, 2025 • 297 • 22 nvidia/GR00T-N1-2B Robotics • 2B • Updated Sep 2, 2025 • 204 • 356 Collections mentioned in this article 2 Cosmos-Transfer1 Collection ⚠️ This collection is archived. 👉 https://huggingface.co/collections/nvidia/cosmos3 • 5 items • Updated 14 days ago • 31 Physical AI Collection Collection of open, commercial-grade datasets for physical AI developers • 53 items • Updated 14 days ago • 166 More Articles from our Blog robotics ai community LeRobot goes to driving school: World’s largest open-source self-driving dataset 107 March 11, 2025 robotics ai community π0 and π0-FAST: Vision-Language-Action Models for General Robot Control 192 February 4, 2025 Community clem Mar 19, 2025 very cool! 👍 1 1 + Reply Edit Preview Upload images, audio, and videos by dragging in the text input, pasting, or clicking here . Tap or paste here to upload images Comment · Sign up or log in to comment Upvote 42 +30 Models mentioned in this article 3 HuggingFaceTB/SmolLM-1.7B Text Generation • 2B • Updated Oct 16, 2024 • 70.9k • 181 nvidia/Cosmos-Transfer1-7B-Sample-AV Updated Apr 9, 2025 • 297 • 22 nvidia/GR00T-N1-2B Robotics • 2B • Updated Sep 2, 2025 • 204 • 356 Collections mentioned in this article 2 Cosmos-Transfer1 Collection ⚠️ This collection is archived. 👉 https://huggingface.co/collections/nvidia/cosmos3 • 5 items • Updated 14 days ago • 31 Physical AI Collection Collection of open, commercial-grade datasets for physical AI developers • 53 items • Updated 14 days ago • 166 HuggingFaceTB/SmolLM-1.7B Text Generation • 2B • Updated Oct 16, 2024 • 70.9k • 181 nvidia/Cosmos-Transfer1-7B-Sample-AV Updated Apr 9, 2025 • 297 • 22 nvidia/GR00T-N1-2B Robotics • 2B • Updated Sep 2, 2025 • 204 • 356 Cosmos-Transfer1 Collection ⚠️ This collection is archived. 👉 https://huggingface.co/collections/nvidia/cosmos3 • 5 items • Updated 14 days ago • 31 Physical AI Collection Collection of open, commercial-grade datasets for physical AI developers • 53 items • Updated 14 days ago • 166 HuggingFaceTB/SmolLM-1.7B Text Generation • 2B • Updated Oct 16, 2024 • 70.9k • 181 nvidia/Cosmos-Transfer1-7B-Sample-AV Updated Apr 9, 2025 • 297 • 22 nvidia/GR00T-N1-2B Robotics • 2B • Updated Sep 2, 2025 • 204 • 356 Cosmos-Transfer1 Collection ⚠️ This collection is archived. 👉 https://huggingface.co/collections/nvidia/cosmos3 • 5 items • Updated 14 days ago • 31 Physical AI Collection Collection of open, commercial-grade datasets for physical AI developers • 53 items • Updated 14 days ago • 166