How NVIDIA Halos Helps Make Physical AI Safer to Scale

Physical AI is moving rapidly from research into large-scale deployment. According to ABI Research, the installed base of Level 3–5 autonomous vehicles is projected to reach 49 million by 2035. Meanwhile, Omdia estimates that approximately 60 million industrial robots will be deployed between 2026 and 2035.

As autonomous machines enter roads, factories, warehouses and other environments shared with people, safety must scale alongside deployment.

Physical AI safety means proving that AI-powered machines can act safely when decisions are translated into physical actions. This applies to autonomous vehicles, humanoid robots, industrial robots and other autonomous systems. Safety requires protection and validation across hardware, software, AI, operating environments and the entire deployment lifecycle—not just a one-time check before launch.

Why Is Safety Essential for Scaling Physical AI?

After years of testing and benchmarking, autonomous vehicles continue to expand commercially. This progress required developers to demonstrate how automated systems address potential hardware and software failures, limitations in intended functionality and AI-specific risks.

Robotics is approaching a similar inflection point as autonomous machines move into factories, warehouses and other environments shared with people.

Across physical AI, manufacturers, regulators, insurers and workplace safety teams need evidence that hardware, software, AI behavior and operating environments can work together safely without human intervention.

Why Does Physical AI Need a New Safety Model?

Four major shifts are shaping new safety standards for autonomous vehicles and robotics.

• Dynamic environments require context-aware safety.
  Roads, factories and warehouses cannot be completely controlled through static zones or physical barriers. Autonomous systems must recognize changing conditions, adapt their behavior and reach a safe state when unexpected events occur.
• AI operations require dedicated safety guarantees.
  Testing should evaluate AI software alongside traditional functional safety. This includes design-time, run-time and validation guardrails. Emerging standards such as ISO/IEC TS 22440 are beginning to address AI-specific risks.
• Safety must continue throughout deployment.
  Autonomous vehicles and robots evolve through software and model updates, new tasks and changing operating conditions. Material changes may require additional safety testing.
• Large-scale validation requires simulation and synthetic data.
  The number and complexity of potential scenarios require real-world testing to be combined with simulation, synthetic data generation and scenario reconstruction.

These changes require safety to be operationalized across design, deployment and validation—from the underlying hardware to AI behavior and the operating environment.

What Is NVIDIA Halos?

Physical AI safety requires specialized engineering, data, processes and validation that few companies can replicate independently. NVIDIA’s safety foundation is built on more than a decade of autonomous-vehicle safety development, including expertise in functional safety, sensor fusion, AI assurance, vision AI, simulation and real-world verification.

NVIDIA Halos is a full-stack safety system for physical AI. It is designed to help developers engineer safety across every layer of design, validation and deployment. Although autonomous vehicles and robotics share certain safety principles, their platforms, standards and evidence requirements remain specific to each domain.

How NVIDIA Halos Supports Autonomous Vehicles

For autonomous-vehicle development, the NVIDIA Halos safety architecture includes the following components:

• Hardware:
  NVIDIA DRIVE AGX Thor delivers accelerated computing with a secure design. NVIDIA DRIVE Hyperion provides a full-stack vehicle platform and reference architecture for Level 4 autonomous vehicles.
• Operating system and middleware:
  Halos OS provides an integrated software foundation built on ASIL-D-certified DriveOS. Halos Core and Halos Middleware support system isolation, monitoring and deterministic communication.
• End-to-end AI model:
  NVIDIA Alpamayo provides an open reasoning vision-language-action model intended to bring explainability to long-tail scenarios.
• Simulation and validation:
  The NVIDIA Halos Safety Assessment Framework provides tools and guidelines for generating evidence to support autonomous-vehicle safety cases across different levels of automation.

By combining these elements, cloud-based AI development and simulation can be connected to in-vehicle deployment. This enables safety evidence to be traced throughout the vehicle lifecycle.

How NVIDIA Halos Supports Robotics

For robotics, the NVIDIA Halos architecture includes hardware, software, real-time sensing, simulation and outside-in safety capabilities.

• Industrial-grade hardware:
  NVIDIA IGX Thor combines accelerated computing and functional safety on one platform with a dedicated functional safety island. It is designed to support systems developed to standards such as IEC 61508 and ISO 13849.
• Safety software:
  Halos Core for IGX provides the communication and processing functions that connect sensors, actuators and other safety components. It also provides the software foundation for safety-related operational functions such as fault detection, monitoring and reporting.
• Real-time sensing:
  NVIDIA Holoscan Sensor Bridge connects sensor data to AI and safety-related processing. This enables systems to identify invalid information and take defined safety responses.
• Simulation and validation:
  NVIDIA Isaac Lab and the NVIDIA Omniverse Library enable developers to test robot behavior across relevant conditions and edge cases, complementing real-world validation.
• Outside-in safety:
  The open-source NVIDIA Halos Outside-In Safety Blueprint uses external cameras and vision AI agents to extend perception beyond onboard sensors. It supports facility-level surveillance and functional safety use cases.

How Is NVIDIA Halos Tested and Evaluated?

Across autonomous vehicles and robotics, the NVIDIA Halos AI System Testing Lab turns safety, cybersecurity and AI safety requirements into repeatable tests. The lab also helps prepare Halos integrations for final system-level certification by third-party agencies.

For autonomous vehicles:

• TÜV SÜD certified NVIDIA’s automotive product lifecycle software processes and DriveOS 6.0 to ISO 26262 ASIL D.
• TÜV SÜD certified NVIDIA’s automotive engineering processes to ISO/SAE 21434.
• TÜV Rheinland conducted an independent UNECE safety assessment of NVIDIA DRIVE AV.

For robotics:

• TÜV Rheinland is examining the functional safety certification readiness of NVIDIA IGX Thor, Halos OS and Holoscan Sensor Bridge.
• TÜV SÜD conducted an inspection of the Thor SoC and Halos Core for ISO 26262.

Across physical AI, ANAB certified the NVIDIA Halos AI Systems Inspection Lab as an ISO/IEC 17020 inspection laboratory. The lab examines targeted Halos integrations and helps companies prepare for final certification by an independent third party.

Who Is Building the NVIDIA Halos Safety Ecosystem?

NVIDIA Halos connects companies that build, integrate, evaluate and deploy physical AI solutions. The ecosystem includes product developers, software and embedded-systems providers, sensor and silicon companies, safety-solution developers and certification bodies.

Autonomous Vehicles

Geely, Isuzu, Nissan—powered by Wayve software—and Einride are building Level 4-capable vehicles on NVIDIA Hyperion supported by Halos OS.

Uber, Grab and Lyft, along with other mobility providers, are using Hyperion to expand the development and deployment of robotaxis.

Members of the NVIDIA Halos AI System Testing Lab include Aumovio, Bosch, Gatic, Hesai, Lucid, Mira, onsemi, PlusAI, Sony, Valeo and Wayve. These companies span autonomous-driving development, advanced driver assistance systems, sensors, silicon, system integration, validation and safety assurance.

Robotics

In robotics, Acontis and QNX provide embedded software required to perform safety functions predictably. Advantech and Neura Robotics build securely designed NVIDIA IGX systems. Infineon, NXP, STMicroelectronics and Texas Instruments contribute sensors, safety microcontrollers and other semiconductor technologies.

KION Group develops functional safety agents for self-driving forklifts. Agility integrates NVIDIA IGX Thor and Halos Core into its safety system for the Digit humanoid robot.

Why Safety-by-Design Matters for Physical AI

Companies scaling physical AI must do more than build capable systems. They must build systems that can be evaluated, certified, deployed and trusted in the real world.

Designing for functional safety from the beginning helps distinguish a prototype from a scalable solution. As autonomous vehicles and robots become more widespread, safety evidence must extend across hardware, software, AI models, simulation, operating environments and deployment updates.

Learn more about NVIDIA Halos for autonomous vehicles and robotics to explore a full-stack safety architecture for physical AI.