Nvidia is putting a broader AI bet behind humanoid robotics with Groot N1, an open source foundation model designed for robots that can operate across different shapes and tasks. Announced at GTC 202 5 in San Jose, the model extends the company’s earlier Project Groot work and places more emphasis on generalist humanoid robots.
What Nvidia released
Groot N1 is described by Nvidia as an AI foundation model for humanoid robotics. The company calls it a "generalist" model, and says it was trained on both synthetic and real data.
The release matters because humanoid robots are often discussed as machines that should be able to handle varied environments, instructions, and physical tasks. A generalist model is aimed at that broader ambition rather than a single narrow robot behavior.
Nvidia is also releasing simulation frameworks and blueprints for generating synthetic training data alongside Groot N1. That pairing is important because robotics AI needs data connected to movement, perception, and object interaction. Synthetic training data can be part of that workflow when real-world data is limited, difficult to gather, or hard to repeat consistently.
The model is available in open source. That gives Groot N1 a different role from a closed demonstration: it can be examined, used, and adapted by others working in humanoid robotics, at least within the limits of what Nvidia has released.
How Groot N1 is meant to work
In a video introducing Groot N1, Nvidia says the model uses a "dual-system architecture" for "thinking fast and slow," inspired by human cognitive processes. The basic idea is that a robot needs both deliberation and execution.
According to Nvidia, the slow-thinking system lets a robot perceive and reason about its environment and instructions. It then plans the right actions to take. This is the part of the model associated with understanding what is happening and deciding what should happen next.
The fast-thinking system takes that plan and converts it into robotic actions. Nvidia says those actions include multi-step object manipulation. In practical terms, that is the difference between forming a plan and physically carrying it out through a robot body.
That split reflects a central challenge in humanoid robotics: a robot must connect language, perception, planning, and movement. A useful robot cannot only identify an object or produce a plan in isolation. It must turn those outputs into actions that work in the physical world.
From Project Groot to broader humanoid robots
Groot N1 is an evolution of Project Groot, which Nvidia launched at its GTC conference last year. Project Groot was aimed at industrial use cases. Groot N1 broadens that focus to humanoid robots in a range of different form factors.
That shift is significant because humanoid robotics is not limited to a single machine design. A model intended for multiple form factors has to support a wider view of how robots may be built and used. The source article does not specify the exact robot bodies or use cases covered, but it does make clear that Groot N1 is not presented only as an industrial system.
Nvidia CEO Jensen Huang framed the release in sweeping terms. "The age of generalist robotics is here," he said in a statement.
The statement captures the ambition behind the release: moving robotics from task-specific systems toward machines that can handle a wider set of instructions and environments. Groot N1 is Nvidia’s model for that direction, combining real data, synthetic data, simulation tools, and a two-part reasoning-and-action design.
Why humanoid robotics is drawing attention
Humanoid robots have attracted heavy publicity in recent years. Companies like X1 and Figure are attempting to create general-purpose robots that move more or less like humans.
The appeal is easy to understand. A robot built around a human-like form factor is often imagined as a machine that can operate in spaces and around objects designed for people. The source article does not claim that this goal has already been reached, and it makes clear that the challenges remain formidable.
Those companies claim that technology has reached the point where mass-produced humanoid robotic systems are a realistic near-term goal. That claim sits at the center of the current debate around humanoid robots: whether recent progress in AI and robotics is enough to move the field from impressive demonstrations toward reliable systems at scale.
Groot N1 fits into that debate as enabling technology rather than a finished robot. It is a model and a set of supporting resources, not proof that general-purpose humanoid robots are already ready for broad deployment.
The caution behind the excitement
The source article also points to the many disappointments in recent robotics history. That context matters because robotics has repeatedly proven harder than software alone. Physical systems have to deal with messy surroundings, uncertain movement, objects that behave unpredictably, and real-world failure modes.
For Groot N1, the key promise is not simply that it can reason, but that it can help connect reasoning to action. The slow-thinking system plans; the fast-thinking system turns plans into movement. If that connection works well, it addresses one of the central problems in humanoid robotics.
Still, the broader conclusion is measured. Nvidia is releasing an open source foundation model, simulation frameworks, and blueprints for synthetic training data. Humanoid robotics companies are pursuing general-purpose robots. The field is receiving major attention. But the article’s caution remains important: turning that attention into mass-produced robotic systems will be easier said than achieved.