Nvidia and Hugging Face entered into a new collaboration for open source robotics, integrating the chipmaker’s models onto LeRobot, Hugging Face’s open source library for physical AI developers.
Under the deal, unveiled on July 6, the companies added Nvidia’s Isaac GR00T 1.7 open reasoning vision-language-action (VLA) models for humanoid robotics and the Isaac Teleop framework for robotics data collection into LeRobot, with the Cosmos 3 frontier robotics model planned for future integration.
The integrations build on resources already connected to LeRobot. They include an open source physical AI data set with more than 350,000 real and simulated trajectories and 57 million grasps; the Isaac Sim and Isaac Lab simulation frameworks; and Jetson Thor (an AI supercomputer designed for robotics) integration with LeRobot’s Reachy 2 humanoid robot platform to support deployment of VLA models on open source humanoids.
The moves give developers an accessible workflow for collecting robot data, training and fine-tuning models, evaluating performance and deploying models.
The alliance between Nvidia and the AI collaboration platform appears to be a significant contribution to the developer community, but it also reflects a deliberate commercial strategy on Nvidia’s part, according to Jan Liphardt, founder of robotics software company OpenMind.
“How do we sell chips? We make those chips incredibly useful,” he said. “How do we make them incredibly useful? We open source the data, the tooling, the telemetry, and the simulation environments to build compelling solutions for a particular vertical.”
Liphardt drew a direct parallel with Nvidia’s approach to automotive AI, in which the company built out an open ecosystem of tools and data that became the de facto standard, running, naturally, on Nvidia hardware.
“The Nvidia-Hugging Face announcement maps that same strategy onto physical AI,” he said. “You’re taking a large amount of very high-quality data, open sourcing it, working with industry partners, and all of this runs best on Nvidia chips.”
The Open Source Strategy
The joint effort also reinforces Nvidia’s commitment to open source software but preferably running on its chips.
“The collaboration is a signal of Nvidia’s commitment to promote robotics software accessibility that is optimized for GPU,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget. “It has been very open in pushing for open source software solutions.”
Yet beyond business logic, Liphardt argued that the collaboration carries broader significance for how the robotics industry develops and who gets to participate in it.
“You don’t want a situation where two or three companies have extreme first-mover advantage and, simply by virtue of having moved first, have almost complete ownership and control of many different verticals,” he said. “You want alternative stacks and open data and tooling that allows other companies, developers, kids, educators, whoever, to also be able to build stuff.”
That concern is not abstract. As physical AI accelerates, the risk of the stack becoming opaque and concentrated is real. Open-source tooling, Liphardt argued, offers developers the option to read the documentation, inspect the data, and understand how a machine makes decisions.
“This opens a path to developing advanced robots using software which is much more intelligible to normal developers and normal people,” he said. “And that’s why the Hugging Face and Nvidia thing is very good for physical AI.”
A Competitive Landscape
The collaboration also comes at a time of increased pressure on Nvidia’s dominance in robotics.
“The company is facing some challenges from Chinese vendors due to the rise of humanoid robotics vendors in China,” Su noted.
In addition, Qualcomm, with its Dragonwing IQ10 and Intel, with its OpenVINO robotics offerings, have recently entered the fold to challenge Nvidia’s position.
Deepening its open source ecosystem, then, may serve as much to entrench developers’ loyalty to Nvidia as it does to advance the field.
