In January 2025, Nvidia released a new blueprint aimed at humanoid robotics developers, marking a significant expansion of its physical AI ecosystem. The announcement, made at the company’s GTC event, introduced several new components: Cosmos 3, updated Isaac simulation tools, and Isaac GR00T humanoid models. These are open models designed to help developers build, train, and deploy next-generation intelligent robots.
The release also included Halos, a full-stack safety system for robotics. According to the source material, Halos is part of Nvidia’s initiative to power production-scale physical AI. The safety system is available in two configurations: NVIDIA Halos Core for NVIDIA IGX is in early access for registered developers, offered in Linux and Linux plus QNX configurations. Additionally, the open-source NVIDIA Halos Outside-In Safety Blueprint, which is part of the Halos Applications layer of Halos OS, is now available in early access on GitHub.
Agility Robotics, a humanoid robotics developer, is the first company to use NVIDIA Halos for Robotics. The company is building safety into its humanoids that work in factories, warehouses, and logistics operations for customers including Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada.
The collaboration with Hugging Face is a key part of this announcement. The two organisations have expanded their partnership to bring new AI models, robotics frameworks, and development tools to the open-source LeRobot platform. LeRobot is Hugging Face’s open-source robotics library for developing, training, and sharing robot datasets, models, and workflows. The goal is to make robot development more accessible for the wider robotics community.
The collaboration also extends to deployment. There is support for Nvidia Jetson Thor on Hugging Face’s Reachy 2 humanoid robot, which enables developers to run vision-language-action models on open-source humanoid robotics platforms.
Nvidia says the expanded partnership combines its community of more than three million robotics developers with Hugging Face’s 16 million AI developers, broadening access to physical AI technologies.
The company also announced a Physical AI Data Factory Blueprint. This is designed to help companies move from development to real-world deployment across industrial and commercial environments. Nvidia said the platform is designed to unify computing, models, and software frameworks for this purpose.
More than 110 robotics developers are now working on Nvidia’s platform, according to the company. The broader push into physical AI includes expanded integrations with companies such as ABB, Fanuc, and Hexagon Robotics.
Why it matters for European robot service
For the European robotics community, this announcement carries several implications that extend beyond the immediate product releases. The expansion of open-source tools and models has the potential to lower barriers to entry for smaller companies and research institutions across Europe that may not have the resources to build proprietary robotics stacks from scratch.
The combination of Nvidia’s three million robotics developers with Hugging Face’s 16 million AI developers creates a substantial pool of talent and expertise. For European service providers, integrators, and end users, this means a larger ecosystem of developers who can contribute to, troubleshoot, and improve the tools that are becoming available. The open-source nature of LeRobot and the Halos Outside-In Safety Blueprint means that European companies can examine, modify, and adapt these tools to their specific needs without being locked into a proprietary vendor relationship.
The safety aspect is particularly relevant for European operators. The Halos system is described as a full-stack safety solution, which suggests it addresses safety across the entire software stack rather than just at the application level. For European companies operating humanoids in factories, warehouses, and logistics environments, safety certification and compliance are critical concerns. The fact that Agility Robotics is using Halos for its humanoids in customer deployments with Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada provides a reference point for what safety-focused humanoid deployment might look like in practice.
However, it is important to note that the source material does not disclose specific safety certifications, compliance standards, or audit results for Halos. European operators will need to verify whether the system meets the specific regulatory requirements of their jurisdictions, particularly given the European Union’s evolving framework for AI and robotics safety.
The support for Nvidia Jetson Thor on Hugging Face’s Reachy 2 humanoid robot is another development worth noting for European service providers. Reachy 2 is an open-source humanoid platform, and the ability to run vision-language-action models on it could enable more flexible and adaptable robot behaviours. For European companies that are exploring humanoid robots for tasks such as inspection, maintenance, or customer service, this could open up new possibilities for customisation and capability development.
The Physical AI Data Factory Blueprint is also relevant for European operators. The ability to generate, manage, and use data for training robot models is a critical component of successful deployment. The blueprint is designed to help companies move from development to real-world deployment, which is a common challenge for organisations that have piloted robots but struggled to scale them to production.
It is worth noting that the source material does not provide specific details about the cost structure, licensing terms, or support arrangements for these tools. European companies that are considering adopting these technologies will need to engage directly with Nvidia, Hugging Face, or their partners to understand the commercial implications.
The fact that more than 110 robotics developers are now working on Nvidia’s platform suggests growing momentum, but the source does not break down this number by region. It is unclear how many of these developers are based in Europe or how active the European community is within this ecosystem.
What buyers and operators should know
For buyers and operators who are evaluating humanoid robotics solutions, there are several practical considerations to keep in mind based on the source material.
First, the availability of open models such as Cosmos 3 and Isaac GR00T means that developers have more options for building and training robots. These are described as open models, which suggests that they can be accessed and modified by the wider community. However, the source does not specify the exact licensing terms, the size of the models, the training data used, or the performance benchmarks. Buyers who are evaluating these models for specific applications will need to obtain additional technical documentation from Nvidia.
Second, the updated Isaac simulation tools are designed to help developers test and validate robot behaviours in simulated environments before deploying them in the physical world. This is a standard approach in robotics development, and the availability of updated tools could reduce the time and cost associated with testing. However, the source does not provide specific details about the new features or improvements in the updated Isaac tools.
Third, the Halos safety system is available in early access. This means that it is not yet a fully mature, production-ready product. Early access typically implies that there may be bugs, incomplete features, or limited documentation. Buyers who are considering Halos for production deployments should factor in the additional testing and validation that may be required.
Fourth, the collaboration with Hugging Face on LeRobot is significant because it brings together two large developer communities. The combination of three million robotics developers and 16 million AI developers creates a substantial ecosystem. For buyers, this could mean a wider range of pre-trained models, shared datasets, and community-contributed tools that can accelerate development. However, the source does not provide specific numbers on how many models or datasets are available on LeRobot, nor does it disclose the quality or coverage of these resources.
Fifth, the support for Nvidia Jetson Thor on Reachy 2 is a specific integration that enables vision-language-action models on an open-source humanoid platform. For operators who are considering Reachy 2 or similar platforms, this could provide a pathway to more advanced robot capabilities. However, the source does not specify the performance characteristics, power consumption, or computational requirements of Jetson Thor in this configuration.
Sixth, the Physical AI Data Factory Blueprint is designed to help companies manage the data lifecycle for robot development. This includes generating synthetic data, collecting real-world data, and organising it for model training. For operators who are struggling with data management, this blueprint could provide a structured approach. However, the source does not provide details about the specific components of the blueprint or how it integrates with existing data infrastructure.
Seventh, the involvement of Agility Robotics as the first user of Halos is a notable reference. Agility is deploying humanoids in factories, warehouses, and logistics operations for customers including Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada. This suggests that the safety system is being tested in real-world industrial environments. However, the source does not disclose the duration of these deployments, the number of robots in operation, or any safety incident data.
Eighth, the expanded integrations with ABB, Fanuc, and Hexagon Robotics indicate that Nvidia is working with established industrial automation companies. This could be relevant for buyers who are looking to integrate humanoid robots with existing automation infrastructure. However, the source does not specify the nature or scope of these integrations.
Finally, it is important to note that the source material does not disclose several key details that buyers and operators typically need for procurement decisions. These include pricing, delivery timelines, warranty terms, support service levels, spare part availability, and training requirements. The source also does not provide any performance data, reliability statistics, or total cost of ownership figures. Buyers who are considering these technologies should request this information directly from Nvidia, Hugging Face, or their authorised partners.
The source material also does not specify a specific day for the announcement, so it is reported here at month-level precision as January 2025. Similarly, the source mentions that robot orders increased in the second quarter of 2026, but this appears to be a separate data point from the main announcement and is not directly related to the Nvidia blueprint release.
In summary, the Nvidia announcement represents a significant expansion of the tools and models available for humanoid robotics development. The combination of open models, simulation tools, a safety system, and a data factory blueprint provides a comprehensive stack for developers. The collaboration with Hugging Face broadens the ecosystem further. However, buyers and operators should be aware that many commercial and technical details are not disclosed in the source material and will need to be obtained through direct engagement with the vendors.
Sources
Nvidia releases new blueprint for humanoid robotics developers
Published by Vigla Media OÜ (Estonia).