Monday, September 21, 2026
Tech Beat

NVIDIA Halos Builds Full-Stack Safety for Physical AI at Scale

NVIDIA Halos unifies hardware, software, simulation and certification as autonomous vehicles and robots move toward deployment at unprecedented scale.

Listen to this briefingAudio briefing

Summary

On September 21, 2026, NVIDIA positioned Halos as the first and only full-stack physical AI safety system after a decade of AV work. ABI Research projects 49 million Level 3 to 5 autonomous vehicles installed by 2035; Omdia expects roughly 60 million industrial robots deployed from 2026 to 2035. Scaling demands continuous assurance across hardware, software, AI and operating environments, with context-aware safe states, design-time, runtime and validation guardrails, retesting after material changes, simulation, synthetic data and scenario reconstruction. ISO/IEC TS 22440 is beginning to address AI-specific risk.

For vehicles, DRIVE AGX Thor, Hyperion, Halos OS on ASIL-D certified DriveOS, Halos Core, Halos Middleware, Alpamayo reasoning vision language action models and the Halos Safety Evaluation Framework connect cloud-to-vehicle development and evidence. Robotics uses IGX Thor with a Functional Safety Island, Halos Core for IGX, Holoscan Sensor Bridge, Isaac Lab, Omniverse libraries and the open-source Outside-In Safety Blueprint, targeting IEC 61508 and ISO 13849. The Halos AI Systems Inspection Lab translates safety, cybersecurity and AI requirements into repeatable inspections.

Geely, Isuzu, Nissan with Wayve, and Einride are building Level 4-ready Hyperion vehicles; Uber, Grab and Lyft use Hyperion for robotaxis. Lab members span AUMOVIO, Bosch, Gatik, Hesai, Lucid, MIRA, onsemi, PlusAI, Sony, Valeo and Wayve. Robotics partners include acontis, QNX, Advantech, NexCOBOT, Infineon, NXP, STMicroelectronics and Texas Instruments; KION Group is developing autonomous-forklift safety agents, while Agility is adding IGX Thor and Halos Core to Digit 5. TÜV SÜD certified the Automotive Product Lifecycle process and DriveOS 6.0 to ISO 26262 ASIL D, plus automotive engineering to ISO/SAE 21434. TÜV Rheinland assessed DRIVE AV under UNECE and is inspecting IGX Thor, Halos OS and Holoscan Sensor Bridge after TÜV SÜD inspected Thor SoC and Halos Core for ISO 26262. ANAB accredited the lab under ISO/IEC 17020; independent bodies retain final certification.

Positives

  • 49 million Level 3 to 5 autonomous vehicles could be installed by 2035, expanding the market for certifiable safety platforms.
  • Roughly 60 million industrial robots could deploy from 2026 to 2035, accelerating demand for integrated functional safety.
  • NVIDIA Halos connects hardware, operating software, AI models, simulation and inspections across vehicle and robotics lifecycles.
  • TÜV SÜD certifications cover DriveOS 6.0, NVIDIA’s Automotive Product Lifecycle process and automotive cybersecurity engineering.
  • ANAB accredited the Halos AI Systems Inspection Lab as an ISO/IEC 17020 inspection body.
  • Geely, Isuzu, Nissan, Einride, Uber, Grab and Lyft are adopting Hyperion for autonomous vehicles or robotaxis.

Risks & concerns

  • Material software, model, task or operating-condition changes may require additional safety testing after deployment.
  • Dynamic roads, factories and warehouses cannot be made safe solely through static zones or physical barriers.
  • The volume and complexity of edge cases make real-world testing insufficient without simulation, synthetic data and scenario reconstruction.
  • ANAB accreditation does not replace final system certification, which remains the responsibility of independent third-party bodies.
  • AI behavior introduces risks beyond traditional functional safety, while ISO/IEC TS 22440 is only beginning to address them.
Primary sourceNVIDIA Bloghttps://blogs.nvidia.com/blog/physical-ai-halos-safety/
Read full article
Editorial note: Tech Beat summarizes and analyzes third-party reporting. The source link is the authoritative article. This page does not reproduce the full source text.

More From The Wire

Artificial Intelligence and RoboticsSep 16

Nvidia’s Les Karpas Identifies the Data Gap Blocking Robotics’ ChatGPT Moment

Artificial Intelligence and RoboticsSep 8

Danijar Hafner’s Dreamer AI Powers a Stealth Humanoid Robotics Startup

Artificial Intelligence and RoboticsAug 26

Anthropic Previews MHS to Standardize AI Control of Lab and Factory Hardware