This page archives the NVIDIA blog announcement of NVIDIA Halos for Robotics (June 22, 2026): a full-stack safety system for robots and physical AI, unifying AI compute and safety in one architecture. Structured data: data/nvidia-halos-robotics-snapshot.json. Hero image: Agility Digit with Halos visual (courtesy of Agility).
NVIDIA extends proven automotive Halos safety into physical AI. Next-gen autonomous robots rely on foundation models, accelerated computing, and distributed sensors working alongside humans in dynamic environments — scaling that requires full-stack safety, not bolt-on patches. Halos for Robotics is positioned as the industry’s only open, full-stack robot safety system connecting compute, software, sensor data, safety apps, and inspection.
Announcement archive (structured)
Source: NVIDIA blog · Halos for Robotics (archived 2026-06-22). Hero image courtesy of Agility.
Three-layer full-stack architecture
| Layer | Components | Role |
|---|---|---|
| Compute & sensors | NVIDIA IGX Thor™, NVIDIA Holoscan Sensor Bridge | Industrial AI compute, built-in safety, sensor connectivity for real-time robot + safety workloads |
| Software — Halos OS | Halos Core; Halos Outside-In Safety Blueprint apps | Safety-related runtime; external cameras + AI agents supplement robot perception and adapt behavior on the factory floor |
| Verification — Halos AI System Inspection Lab | ANSI/ANAB-accredited physical AI functional safety & AI safety program | Prepare partners for third-party certification (TÜV Rheinland, UL, TÜV SÜD, exida, SGS, CertX, etc.) |
Announcement highlights
| Topic | Detail | Notes |
|---|---|---|
| Automotive safety R&D basis | ≈18,600 engineer-years of experience | Halos extended from proven automotive stack |
| First adopter | Agility Robotics (Digit humanoid) | Integrates IGX Thor + Halos Core into personnel-detection safety |
| Digit customer examples | Amazon, GXO, Schaeffler, Toyota Motor Manufacturing Canada | Factory, warehouse, logistics deployments |
| Inspection Lab ecosystem | 40+ manufacturers, certifiers, safety vendors | Design to real-world deployment |
Ecosystem partners (selected)
| Category | Partners | Role |
|---|---|---|
| Software | Acontis, FreeRTOS, QNX | Real-time runtimes, safe communication, embedded layers for functional safety development |
| Embedded systems | Advantech, NexCobot | Safety-designed NVIDIA IGX-based systems for robot deployment |
| Sensors & chips | Infineon, NXP, STMicroelectronics, Texas Instruments | Sensors, safety MCUs, semiconductors |
| Industrial applications | FORT Robotics, Inventec, KION Group, Lyte AI, Neurealm | Functional-safety agents via Outside-In Safety Blueprint |
| Assessment | TÜV SÜD (ISO 26262 on Thor SoC & Halos Core); TÜV Rheinland prep on IGX Thor, Halos OS, Holoscan Sensor Bridge | Functional safety certification pipeline |
Availability & developer access
| Item | Status | Config |
|---|---|---|
| NVIDIA Halos Core for IGX | Early access for registered developers | Linux; Linux + QNX OS for Safety 8.0 |
| Halos Outside-In Safety Blueprint | Early access on GitHub (open source, Halos OS app layer) | Application-layer safety blueprint |
From automotive Halos to the factory floor
The release cites roughly 18,600 engineer-years of automotive safety R&D as the foundation. VP of robotics and edge AI Deepu Talla argues physical AI is reshaping factories, warehouses, and logistics — teams need a unified safety architecture to deploy autonomous systems alongside human workers with confidence.
Three layers (see table): IGX Thor + Holoscan Sensor Bridge for industrial compute and sensor connectivity → Halos OS (Halos Core + Outside-In safety apps) → ANSI/ANAB-accredited Halos AI System Inspection Lab preparing partners for TÜV, UL, and related third-party certification.
Agility Digit: first deep integration
Agility Robotics is the first partner embedding Halos components into its proprietary safety stack for the Digit humanoid (logistics, manufacturing, warehouses). Integration covers IGX Thor and Halos Core for personnel-detection safety. Deployments include Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada.
Agility joins the inspection lab to align software, AI components, and cybersecurity with IEC 61508, ISO 13849, and ISO/IEC TR 5469 before third-party certification. CEO Peggy Johnson stresses safety must be built into the robot and verified across the full system for industrial scale.
Outside-In safety blueprint
The open Halos Outside-In Safety Blueprint (early access on GitHub) uses external cameras and AI agents to supplement onboard perception and adapt robot behavior as factory conditions change — a differentiator vs. software-only safety frameworks.
Ecosystem & developer access
Partners span RTOS vendors (Acontis, FreeRTOS, QNX), embedded systems (Advantech, NexCobot), semiconductors (Infineon, NXP, ST, TI), and 40+ lab participants. Halos Core for IGX is in early access (Linux; Linux + QNX OS for Safety 8.0). Many features remain staged — archive reflects public blog copy as of 2026-06-22.