NVIDIA Launches Halos for Robotics, Porting Its Autonomous-Vehicle Safety Stack to Humanoid and Industrial Robots

Overview

NVIDIA used the Automate 2026 trade show in Chicago to introduce Halos for Robotics, which it describes as the industry’s first full-stack, comprehensive safety system for robotics and physical AI that unifies AI compute and safety. According to NVIDIA, the system was announced on June 22, 2026, and adapts the functional-safety foundation the company built for self-driving cars to the humanoid and mobile robots now entering factories and warehouses.

The pitch is that as robots move out of cages and begin working alongside people, the industry lacks a standardized way to demonstrate that an AI-driven machine is safe. NVIDIA is offering its own answer drawing on, in its words, 18,600+ engineering years of autonomous vehicle safety development.

What We Know

Halos for Robotics is organized as a stack. NVIDIA says the system spans the key layers needed for robot safety, with NVIDIA IGX Thor and NVIDIA Holoscan Sensor Bridge providing industrial-grade AI compute, built-in safety and sensor connectivity for real-time robotics workloads. A software layer, NVIDIA Halos OS, supplies the stack for robotics safety, including a component called Halos Core and the NVIDIA Halos Outside-In Safety Blueprint, which the company says extends robot perception using external cameras and AI agents to dynamically control robot behavior in industrial settings.

The third layer is a certification path. NVIDIA says the NVIDIA Halos AI Systems Inspection Lab is the world’s first ANSI National Accreditation Board (ANAB)-accredited program for functional and AI safety for physical AI. In NVIDIA’s developer documentation, the lab is described as an ANAB-accredited ISO/IEC 17020 Inspection Body already working with over 43 companies, issuing inspection certificates that partners can take to third-party certifiers such as TÜV Rheinland, TÜV SÜD, SGS, exida, CERTX, or UL Solutions.

Laurie E. Locascio, President and CEO of ANSI, framed the accreditation as a credibility step for the field. “As AI-enabled robotics moves into industrial environments, the industry needs standardized, internationally recognized frameworks to assess safety across increasingly complex systems,” she said, according to NVIDIA. “ANAB’s accreditation of the NVIDIA Halos AI Systems Inspection Lab confirms the program has the competence and impartiality to evaluate robotic AI systems against recognized safety requirements, giving companies a rigorous and internationally recognized foundation for their path to certification.”

The technical specifics on the hardware sit in NVIDIA’s developer blog. IGX Thor is rated at up to 2,070 FP4 TFLOPs of AI performance with 14x Neoverse Arm CPU cores and 128 GB of memory at 273 GB/s bandwidth. NVIDIA says the chip carries a physically isolated safety island that is IEC 61508 SIL 3 capable and more than 22,000 safety mechanisms, while the Holoscan Sensor Bridge implements an end-to-end IEC 61508 SIL 2 safety protocol, per the same NVIDIA technical blog. The platform targets established functional-safety standards including IEC 61508, ISO 13849 and ISO/IEC TR 5469, according to Engineering.com.

Deepu Talla, vice president of robotics and edge AI at NVIDIA, positioned the launch as removing a barrier to deployment. “Physical AI is transforming how factories, warehouses and logistics operations work, and robotics teams need a unified safety architecture to scale autonomous systems into these environments,” he said, according to NVIDIA. “With NVIDIA Halos for Robotics, developers and system builders can harness NVIDIA’s proven autonomous vehicle safety foundation to develop safer robots faster and bring them into industrial operations alongside workers with greater confidence.”

The first named adopter is Agility Robotics, whose Digit humanoid is deployed at Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada. Engineering.com reports that Agility is the first to use NVIDIA Halos for Robotics to build safety into its humanoids, and The AI Insider similarly notes that Agility Robotics is the first company to publicly announce plans to use the platform. Peggy Johnson, CEO of Agility, said safety “has to be built into the robot and validated across the entire system” for humanoids to deliver value at scale, calling responsible automation “a nonnegotiable requirement,” according to NVIDIA.

The software is shipping initially in limited form. NVIDIA Halos Core for NVIDIA IGX is available in early access for registered developers in Linux and Linux plus QNX OS for Safety 8.0 configurations, and the open-source Outside-In Safety Blueprint is now available in early access on GitHub, according to NVIDIA.

Halos for Robotics extends the IGX Thor platform that reached general availability earlier this year, repurposing that industrial edge-AI compute module as the hardware foundation for the new safety stack.

What We Don’t Know

NVIDIA’s framing leans heavily on its own claims of being “first” — the first full-stack safety system and the first ANAB-accredited inspection program for physical AI. Those are descriptions provided by NVIDIA and ANSI; the materials reviewed here do not include independent benchmarking of how Halos performs against rival safety approaches, nor pricing for the commercial components beyond the early-access availability of Halos Core and the open-source blueprint.

The announcement also does not establish how many of the over 43 companies cited as working with the inspection lab have completed certification, or when broader commercial availability of Halos Core beyond early access will arrive. With Agility named as the first adopter, it remains to be seen how quickly other humanoid and mobile-robot makers integrate the stack, and whether third-party certifiers and regulators treat an NVIDIA-originated inspection certificate as a durable industry standard or a vendor-specific path.