Hugging Face and NVIDIA Open Source Code for Robot Thinking

When academics began releasing open hardware for robotics, an entire generation of engineers freed itself from tedious infrastructure work. Now the next phase begins: teaching robots to think independently. And this process too is becoming open.

ROS—The Linux of Robotics

Since the 1990s, open source code existed in robotics, but projects were fragmented. Everything changed in 2007 when the Robot Operating System (ROS) emerged—a software platform that unified tools and became the de facto standard. Despite its name, ROS is not an operating system but a framework built on top of Linux.

ROS changed how roboticists work. Previously, each laboratory wrote its own infrastructure—moving data between components, interacting with hardware, building maps, planning paths. This took a year or two.

Brian Gerkey, one of ROS's creators and current CTO of Google Robotics, says:

"I'm a tool builder, and I love sharing everything as openly as

possible because it has the maximum effect."

In parallel, the AI community followed the same path—opening up research, models, and data. The field developed faster than all forecasts predicted. Now these approaches are coming to robotics as well.

Open Models for Robots

Computer vision, which was once a difficult task, has made significant advances in recent years. Spencer Huang, NVIDIA's director of robotics, notes that what once required a specialist can now be done in a few lines of code. Tools that were once accessible only in specialized laboratories are now open to everyone.

NVIDIA released a complete stack of open tools:

• Cosmos—generates synthetic data and simulates the physical environment
• GR00T—gives robots the ability to reason and perform complex tasks
• Isaac—orchestrates training, simulation, and deployment

All NVIDIA models live on Hugging Face—the platform that has become the standard for model sharing. Hugging Face launched LeRobot in May 2024. In a year, the number of robotics datasets grew from 1,145 to 58,000—the largest category on the platform.

Hugging Face also acquired Pollen Robotics, acknowledging that software alone is not enough.

Alibaba released RynnBrain at the beginning of 2025—an open foundation model for physical AI that, according to the company, outperforms Google and NVIDIA's counterparts on benchmarks.

When Researchers Become Competitors

This is where things get complicated. ROS was created primarily by academics who had no commercial interests. Today, the main contributors are companies that want people to build on their platforms. This isn't bad, says Bill Smart, a professor at Oregon State and a pioneer of open robotics. But the incentives need to be visible.

Smart has another concern: lowering the barrier to entry has a downside. Researchers coming from AI without robotics experience sometimes re-solve already solved problems. A beginner might spend a week training a neural network to control a robot's movement without knowing that the same thing can be done with classical code in a few lines.

But Smart is not without hope. Whatever the motives behind open source code, the effect is real: there are more people in the field, tools are simpler, and the community is larger and more diverse.

What This Means

The barriers to entry in robotics are falling at the same pace as they once did for creating AI applications. This means more experiments, more creativity, more solutions. But it's also worth remembering concerns about privacy and control—a world where household robots are controlled by a few Silicon Valley companies is far less appealing than a world of open alternatives.