Nvidia is pushing humanoid and industrial robots toward safer operation around people with Halos for Robotics, a full-stack safety system spanning compute, software, sensors, and third-party inspection. The company unveiled it on June 22, 2026, at the Automate trade show in Chicago, calling it the industry's first full-stack, open safety system for physical AI. The push matters because humanoid robots are moving from controlled demos into warehouses and factories where they work beside people. Nvidia built Halos to give these machines a way to sense, decide, and act safely in that environment, rather than relying on fences or stop-on-contact sensors alone.
Table of Contents
- What Halos for Robotics Actually Is
- How the Certification Piece Works
- What's Available Now Versus What's Still Coming
- Who Is Already Using It
- What to Check Before Trusting a Safety Claim
What Halos for Robotics Actually Is
Halos for Robotics is not a single product but a stack of layered components. It combines Nvidia's IGX Thor compute platform with a Holoscan Sensor Bridge for processing sensor data, a software layer called Halos OS, and a new inspection program, according to Robotics 24/7's coverage of the Automate announcement. The system extends safety architecture Nvidia originally built for self-driving cars.
Nvidia says that automotive safety work drew on more than 18,600 engineering-years of development, and Halos for Robotics adapts that foundation for machines that move and act around people rather than just on roads, per Nvidia's press release. The rationale, as Bloomberg reports, is straightforward: humanoid robots need to make split-second safety decisions before they can work in close physical proximity to, or in direct contact with, human workers. That capability is the specific problem Halos is designed to solve, rather than a general efficiency or performance upgrade.
How the Certification Piece Works
A central part of Halos is the Halos AI Systems Inspection Lab, which Robotics 24/7 describes as the first ANAB-accredited program built specifically for physical-AI functional and AI safety. The lab is designed to help robot makers prepare for review by outside certifiers. Those certifiers include TÜV Rheinland, TÜV SÜD, UL Solutions, exida, SGS, and CertX — established names in industrial safety certification.
It's worth being precise about what the lab does and doesn't do: the lab itself does not issue certification; it readies a robot maker's systems for one of those outside bodies to certify. That distinction matters for anyone evaluating a Halos-equipped robot. Passing through Nvidia's inspection lab is a preparation step, not a substitute for the independent certification that actually governs deployment in regulated industrial settings.
What's Available Now Versus What's Still Coming
Parts of Halos are already accessible to developers, while other pieces remain ahead of general release. Halos Core for Nvidia IGX is in early access for registered developers, and Nvidia has also published an open-source Halos Outside-In Safety Blueprint in early access on GitHub, according to Robotics 24/7 and Nvidia's release. On the hardware side, most humanoid platforms run on Nvidia's Jetson Thor compute module, which pairs a Blackwell GPU with an integrated functional-safety processor and a high-performance CPU cluster.
Boston Dynamics has said it is integrating Jetson Thor into its electric Atlas robot and co-defining functional-safety and security architecture with Nvidia directly. A newer variant, the Jetson Thor T3000, is reported to deliver up to 865 TFLOPS of FP4 AI compute at 70 watts and to support Halos robotics safety technologies. That chip has a planned launch in Q1 2027, according to DigitalCitizen's reporting — a reminder that some of the strongest safety hardware tied to Halos is still ahead of shipping, not available today.
Who Is Already Using It
Agility Robotics is the first company to adopt Halos, applying it to humanoid robots that work in factories, warehouses, and logistics operations. Nvidia's release names the operations affected by this rollout: Agility's customers Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada, as reported by Robotics 24/7.
These are high-traffic industrial environments where robots and warehouse staff already share floor space. Deploying Halos there puts the safety stack to work in exactly the close-proximity scenarios Nvidia built it for, rather than in a lab or demo setting. For companies watching the humanoid robotics space, Agility's adoption is a useful signal of how Halos moves from announcement to practice: through an existing robot maker's fleet and its named industrial customers, not a standalone product purchase.
What to Check Before Trusting a Safety Claim
Not every voice in the industry is taking Nvidia's announcement at face value. A third-party commentary on the launch cautions that operators still need independent testing and should not rely on vendor safety claims alone — a practical limitation for any company evaluating Halos-equipped humanoids for deployment.
That caution lines up with how Halos itself is structured: the inspection lab prepares a system for review, but an outside certifier — TÜV Rheinland, TÜV SÜD, UL Solutions, exida, SGS, or CertX — has to sign off before the safety claim carries real weight. Anyone evaluating a Halos-based robot for deployment should ask three concrete questions:.
- Has the robot maker completed outside certification, or only Nvidia's inspection lab preparation step?
- Which Halos components are running — Halos Core, the Holoscan Sensor Bridge, Halos OS — and which are still early access?
- Is the compute platform the shipping Jetson Thor, or a future variant like the T3000 that isn't available until Q1 2027?
