Skip to content

NVIDIA Halos for Robotics Is the First Full-Stack Safety System for Physical AI

NVIDIA Halos for Robotics debuted June 22 as the first full-stack safety system for AI robots — hardware, safety OS, and an accredited U.S. lab in one platform.

Argal
Argal
3 min read
NVIDIA Halos for Robotics Is the First Full-Stack Safety System for Physical AI
NVIDIA Halos for Robotics deployment paired with Agility Robotics' Digit humanoid robot. Photo: YugaTech

NVIDIA announced Halos for Robotics on June 22, 2026 — billing it as the industry's first full-stack, open safety system for robotics and physical AI. The platform was unveiled at Automate in Chicago and draws on more than 18,600 engineering years of autonomous vehicle (AV) safety development, now applied to humanoid robots and industrial automation entering live factory, warehouse, and logistics environments.

What Halos for Robotics Actually Is

Halos for Robotics is structured in three layers — hardware, software, and third-party certification infrastructure — unified into a single safety architecture.

Hardware Foundation

  • NVIDIA IGX Thor: An industrial-grade AI compute module with functional safety built in, designed for demanding factory environments.
  • Holoscan Sensor Bridge: Real-time sensor connectivity that feeds data from cameras, LiDAR, and other sensors into the safety stack.

Software Stack

  • Halos OS: A safety-focused operating system layer that anchors all safety functions.
  • Halos Core: Manages safety-related operations and supports pluggable "blueprints" for perception and control — currently in early access for registered Linux developers.
  • Outside-In Safety Blueprint: An external camera-based system that uses AI agents to monitor the robot from outside its own frame, enabling it to detect nearby humans and automatically slow or freeze. The blueprint is open-source on GitHub.

Certification Infrastructure

  • NVIDIA Halos AI Systems Inspection Lab: The first ANAB-accredited facility in the United States for functional and AI safety assessment of physical AI systems. It supports certification processes through TÜV Rheinland, UL Solutions, TÜV SÜD, Exida, SGS, and CertX.

Agility Robotics Is the Inaugural Partner

Agility Robotics, maker of the Digit humanoid robot, is the first company to integrate Halos components into a commercial product. Digit is already deployed in production environments at Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada — putting Halos' real-world validation well beyond the pilot phase.

Deepu Talla, NVIDIA's VP of robotics and edge AI, noted: "Physical AI is transforming factories and warehouses," adding that Halos allows developers to "develop safer robots faster."

Why This Matters for Industrial AI

As humanoid robots move from controlled pilots into live environments alongside human workers, safety certification has become one of the industry's most pressing bottlenecks. Existing safety standards were written for traditional robots that follow fixed paths — not AI-driven systems that perceive and react dynamically.

Halos provides a common architecture manufacturers can use as a starting point for third-party certification, rather than building their own validation processes from scratch. The Inspection Lab's ANAB accreditation means assessments performed there carry weight with the same certification bodies that approve industrial safety systems globally.

Availability

Halos Core is currently in early access for registered Linux developers. The Outside-In Safety Blueprint is publicly available on GitHub. No pricing for the platform was announced.

Sources:

Argal

Argal

@clurky

Clurky is a Philippine tech news site owned and run by Argal, a Philippines-born software developer based in Singapore with a Computer Science background. He covers Philippine tech, fintech, and digital services - from gadgets and AI to software and security - along with evergreen guides and explainers, all with a builder's eye for how these systems actually work. Every article is fact-checked against primary sources.

440 posts

Comments

Join the conversation

Sign in to leave a comment and reply to others.

Sign in
Loading comments...