en.Wedoany.com Reported - On July 7, U.S. chip and AI computing company NVIDIA announced a partnership with U.S. AI open-source platform company Hugging Face to integrate NVIDIA Isaac GR00T 1.7 and NVIDIA Isaac Teleop into the LeRobot open-source robotics library, with plans to later introduce the NVIDIA Cosmos 3 model. This integration targets data collection, model training, post-training, deployment, and workflow management for robot developers, aiming to expand the availability of open physical AI tools.
LeRobot is an open-source robotics library launched by Hugging Face for training, running, and sharing robot datasets, models, policies, and workflows. It serves as a unified access layer in the robot development process, enabling developers to collaborate on data management, model training, policy evaluation, and deployment workflows. By integrating Isaac GR00T 1.7 and Isaac Teleop into LeRobot, NVIDIA allows robot developers to use relevant models, teleoperation tools, and data formats in a more standardized environment, reducing adaptation costs across different toolchains.
NVIDIA Isaac GR00T 1.7 is a foundation model for robotics introduced by NVIDIA, helping developers obtain high-quality human demonstrations via external devices and access LeRobot in a standardized, interoperable format. Robot learning heavily relies on real-world operational data, especially in scenarios such as grasping, movement, dual-arm collaboration, and complex task execution. Relying solely on simulation data or manually written rules is insufficient to cover variations in real environments. With GR00T 1.7 integrated into LeRobot, developers can connect demonstration data, post-training workflows, and robot policy development, advancing model adaptation in a more unified manner.
Isaac Teleop serves as a teleoperation framework. Developers can use teleoperation to collect human demonstration data during robot task execution, then perform post-training and deployment through LeRobot workflows. For robots to learn real-world actions, a large amount of high-quality human demonstrations is essential. Teleoperation tools make it easier for developers to collect data across different robots, tasks, and scenarios. By integrating Isaac Teleop into an open-source workflow, data collection is no longer confined to closed systems but can enter the broader robot development community.
The subsequent integration of NVIDIA Cosmos 3 will bring physical AI world foundation model capabilities into the LeRobot ecosystem. Robot development is often constrained by the high cost of real-world data collection; many scenarios cannot be repeatedly deployed for data acquisition, nor are they suitable for high-risk on-site trial and error. Cosmos 3 can generate and enhance robot data, build simulation scenarios, and assist policy development, enabling developers to advance training and evaluation even when real data is insufficient. Combining real data, teleoperation data, and synthetic data enriches the training materials for robot models and makes it easier to cover long-tail tasks.
This partnership also connects the developer ecosystems of both companies. NVIDIA has approximately 3 million robot developers, Hugging Face has about 16 million AI developers, and LeRobot provides an open-source community collaboration entry point. Robot development has traditionally been fragmented across hardware vendors, simulation platforms, lab codebases, and specialized software environments, creating toolchain isolation between AI developers and robot developers. The connection of GR00T, Teleop, Cosmos, and LeRobot will allow more AI developers to access robot datasets, policy training, and physical AI workflows, while also enabling robot developers to more easily use open-source models and collaboration methods from the Hugging Face ecosystem.
LeRobot's integrated NVIDIA resources also include a set of open-source physical AI datasets. These datasets have accumulated over 15 million downloads, containing more than 350,000 real and simulated trajectories and 57 million grasp data points, supporting the initiation of robot workflows. For teams new to robot development, the hardest part to obtain at the start is often not model code, but usable data, reproducible experimental workflows, and standardized evaluation resources. With datasets, foundation models, and teleoperation tools all entering LeRobot, developers can more quickly build operational robot training pipelines.
This collaboration between NVIDIA and Hugging Face places several key components of physical AI development on the same open-source path: human demonstration data collected via Isaac Teleop, robot foundation models accessed through Isaac GR00T 1.7, subsequent synthetic and enhanced data supplemented by Cosmos 3, and training, evaluation, and sharing workflows hosted by LeRobot. As related models and tools are gradually released, robot developers can complete data preparation, model adaptation, policy training, and deployment validation around a unified open-source workflow.
