Nvidia detailed its Halos safety architecture on Monday, outlining a full-stack platform designed to test and validate autonomous vehicles and industrial robots across hardware, software, and artificial intelligence models. The company pointed to market forecasts from ABI Research projecting an installed base of 49 million level 3 to level 5 autonomous vehicles by 2035, alongside Omdia estimates of roughly 60 million industrial robots deployed between 2026 and 2035.
Nvidia said scaling physical machines into environments shared with people requires continuous safety verification rather than a single pre-deployment check. The company structured its framework around four operational shifts: handling dynamic environments without fixed physical barriers, providing specific behavioral assurance under emerging standards like ISO/IEC TS 22440, maintaining safety across post-deployment software updates, and validating edge cases using simulation and synthetic data generation.
Autonomous vehicle architecture
Vehicle manufacturers using the automotive stack build on the Nvidia DRIVE AGX Thor computer and the Hyperion reference platform for level 4 automated driving. Operating capabilities run on Halos OS, which uses ASIL-D certified DriveOS as its base, alongside Halos Core and Middleware to isolate tasks and support deterministic communication. Nvidia also paired the system with Alpamayo, an open reasoning vision-language-action model built to navigate long-tail road scenarios.
Transportation companies have started integrating the vehicle components. Geely, Isuzu, Einride, and Nissan—working with software developer Wayve—are developing level 4-ready commercial and passenger vehicles using Hyperion and Halos OS. Ride-hailing and delivery operators Uber, Grab, and Lyft are also deploying Hyperion to scale their robotaxi programs.
Industrial robotics and sensing
Industrial robotics deployments rely on Nvidia IGX Thor, a specialized hardware module with an integrated Functional Safety Island designed for IEC 61508 and ISO 13849 standards. Halos Core for IGX handles operational safety functions, including fault monitoring, reporting, and sensor-to-actuator communication. The hardware connects to the Holoscan Sensor Bridge, which processes real-time sensor streams to detect invalid data and trigger programmed safety stops.
Robotics developers test behaviors in simulation through Nvidia Isaac Lab and Omniverse libraries before physical deployment. Nvidia also released the open-source Halos Outside-In Safety Blueprint, which pairs external facility cameras with vision AI agents to monitor hazards beyond a robot's onboard sensor line of sight. Agility is integrating IGX Thor and Halos Core into its Digit 5 humanoid robot, while KION Group is building safety agents for autonomous forklifts.
Testing and independent certification
Independent inspection bodies have conducted preliminary audits across the hardware and software layers. TÜV SÜD certified Nvidia's Automotive Product Lifecycle software process and DriveOS 6.0 to ISO 26262 ASIL D, alongside ISO/SAE 21434 automotive engineering standards, while also evaluating the Thor system-on-a-chip and Halos Core. TÜV Rheinland completed a UNECE safety assessment of DRIVE AV and is inspecting IGX Thor, Halos OS, and Holoscan Sensor Bridge for functional safety readiness.
Operating software partners include acontis and QNX, while Advantech and NexCOBOT produce safety-designed IGX computing units. Semiconductor suppliers Infineon, NXP, STMicroelectronics, and Texas Instruments provide sensor and safety-microcontroller hardware. The ANSI National Accreditation Board accredited the Nvidia Halos AI Systems Inspection Lab as an ISO/IEC 17020 inspection body, where partners including Bosch, Hesai, Lucid, onsemi, and Valeo inspect integrations prior to final commercial certification.
