NVIDIA Halos for Robotics: The Full-Stack Safety System Bringing Humanoid Robots to the Factory Floor

As humanoid robots and autonomous machines move from research labs into active warehouses and factories, the question of safety has become urgent. Robots that work alongside people cannot rely on ad-hoc safety measures; they need a systematic, certifiable safety architecture built in from the ground up.

On June 22, 2026, at the Automate conference in Chicago, NVIDIA introduced Halos for Robotics, which it calls the industry's first full-stack, open robotics safety system . The system extends the Halos safety framework originally developed for autonomous vehicles into the robotics domain, unifying AI compute, system software, sensor data, safety applications, and robot inspection into a single standardized architecture .

The system is built around three primary technology layers that span the entire development and deployment lifecycle .

Hardware Layer: NVIDIA IGX Thor and Holoscan Sensor Bridge

At the foundation is the NVIDIA IGX Thor industrial-grade edge computer, based on the NVIDIA Blackwell GPU architecture. It provides AI compute with built-in safety extensions, including fault detection and functional safety features . A Safety Extension Package (SEP) adds further safety foundations . The Holoscan Sensor Bridge handles real-time sensor connectivity, routing data for both robot operations and safety functions . The IGX platform also includes a Thor Functional Safety Island (FSI)—an independent, redundant set of processors—and supports a hypervisor that can run a Linux VM alongside a QNX VM for safety-critical tasks .

Software Layer: Halos OS and the Outside-In Safety Blueprint

Halos OS is the safety software stack that oversees safety operations. Adapted from NVIDIA's proven autonomous-vehicle safety stack, it delivers a functional safety foundation for robots, from the OS layer up . A key component is the Outside-In Safety Blueprint, which uses external cameras and AI agents to monitor robot behavior in real time from outside the robot's own system—a form of independent, external oversight . This blueprint is available as open-source software on GitHub in early access .

Validation Layer: AI Systems Inspection Lab

The NVIDIA Halos AI Systems Inspection Lab is the world's first program accredited by the ANSI National Accreditation Board (ANAB) for functional and AI safety assessment in physical AI . It helps partners prepare their Halos integrations for third-party safety certification from leading bodies including TÜV Rheinland, TÜV SÜD, UL Solutions, SGS, and CertX, which recognize the lab as part of their certification protocols . The lab is accredited as an ISO/IEC 17020 inspection body, enforcing impartiality and independence .

Agility Robotics: First Production Robot with Halos

Agility Robotics is the first company to adopt Halos for Robotics . Its humanoid robot Digit will be the first production robot shipping with NVIDIA Halos OS, using NVIDIA IGX Thor and Halos Core to safely navigate and work alongside people in real warehouse environments . Digit is designed for logistics, manufacturing, and warehouse work and already has a customer list that includes Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada . The robot is approximately 5'9" tall, weighs around 140 lbs, and is already deployed in live operations: GXO runs Digit fleets in a Georgia fulfillment center under an industry-first Robots-as-a-Service agreement .

Expanding Ecosystem

NVIDIA Halos for Robotics is backed by a growing ecosystem of over 40 companies spanning safety certification bodies, semiconductor firms, software vendors, and robotics partners . Named participants include:

• Certification organizations: TÜV Rheinland, TÜV SÜD, UL Solutions, SGS, CertX
• Semiconductor and hardware: Infineon, NXP Semiconductors, STMicroelectronics, Texas Instruments
• Software partners: Acontis, FreeRTOS, QNX
• Embedded systems: Advantech, NexCobot
• Additional partners: FORT Robotics (integrating its Trust Layer with the Outside-In Safety Blueprint), Flex, John Deere, Lattice Semiconductor, Novanta

Availability and Next Steps

• NVIDIA Halos Core (for IGX) is now available in early access for registered developers, supporting Linux and Linux+QNX OS for Safety 8.0 configurations .
• The Outside-In Safety Blueprint is available as open source on GitHub in early access . (The exact repository URL was not explicitly cited in the available sources; a direct GitHub search for "NVIDIA Halos Outside-In Safety Blueprint" is recommended.)

NVIDIA Halos for Robotics represents a significant step toward making physical AI safe enough for widespread industrial deployment—unifying hardware, software, and certification into a single open framework that companies like Agility Robotics are already putting to work alongside human crews.