Nvidia has extended its Halos safety architecture from self-driving cars to robots, pitching the full-stack system as the missing piece for deploying humanoids, mobile robots and surgical machines around people. The expansion, announced as Nvidia Halos for Robotics in June 2026, builds on a Halos release in 2025 that targeted autonomous vehicles. CEO Jensen Huang told the All-In Podcast in March 2026 that physical AI is already generating nearly $10 billion in annual revenue for Nvidia.
Amit Goel, head of robotics ecosystem and edge computing at Nvidia, told Ars Technica that model and hardware progress has pushed safety to the front of the queue. The company is positioning Halos as the layer that lets robotics firms ship into warehouses, factories and hospitals without bespoke safety infrastructure around every machine.
The stack starts with the Nvidia IGX Thor computing module, which carries an independent processor dedicated to safety workloads. The Halos operating system continuously monitors hardware blocks and software libraries for failures and isolates safety-critical workloads from the rest of the system.
Key facts
- 01Nvidia CEO Jensen Huang said physical AI is already driving nearly $10 billion in annual revenue, during a March 2026 appearance on the All-In Podcast.
- 02Nvidia announced Halos for Robotics in June 2026, extending a safety stack it first launched in 2025 for autonomous vehicles.
- 03Agility Robotics' Digit 5 is the first humanoid to incorporate Halos, moving safety sensors on-board instead of relying on external work-cell infrastructure.
- 04Boston Dynamics joined the Halos accreditation program covering Spot, Stretch and Atlas, with Atlas slated to deploy in parent Hyundai's automotive factories by 2028.
- 05KION Group and LG are also integrating Halos, with LG building a humanoid on Nvidia's Isaac GROOT foundation model.
Nvidia's Holoscan Sensor Bridge links sensor streams to safety processing in a way designed to flag corrupted data, and can be embedded on a microcontroller or FPGA. The Halos package also bundles simulation environments for virtual testing and an inspection lab program that gives robotics partners feedback on safety artifacts during development.
“In the context of physical AI, you need to have your functional system and your safety system all running on the same silicon.”— Amit Goel, Head of robotics ecosystem and edge computing at Nvidia
Porting the approach from vehicles to robots forced a redesign. Functional safety definitions for autonomous driving are relatively consistent across carmakers and jurisdictions, Goel said, while a robot vacuum and a loaded forklift have little in common. Nvidia rebuilt Halos to let developers write custom safety functions without breaking the underlying platform.
Agility Robotics is the first humanoid maker to integrate Halos, using it in the Digit 5 so the robot can work near people without isolated workstations or physical barriers. Prior Digit deployments depended on external sensors mounted around the work cell; the new design moves that hardware onto the robot itself.
“Now the robot is literally unchained—the safety goes with the robot wherever it goes.”— Amit Goel, Head of robotics ecosystem and edge computing at Nvidia
Boston Dynamics has joined the Halos safety accreditation program and is applying it across Spot, the wheeled Stretch and the Atlas humanoid. Atlas could start deploying in the automotive factories of parent Hyundai by 2028. Germany's KION Group, which makes self-driving forklifts and other autonomous mobile robots, is also integrating Halos, as is LG, which is building a humanoid on Nvidia's Isaac GROOT foundation model.
Goel framed the shift as a move away from fixed-station automation toward generalized robot bodies and skills operating across multiple environments. In his telling, the only way to get there in production is to raise the safety floor — which is the business Nvidia is now selling into alongside its chips.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
