With 49 million level 3-5 autonomous vehicles projected on roads by 2035, and an estimated 60 million industrial robots deployed between 2026 and 2035, ensuring the safety of physical AI systems is no longer a future concern, it is a present imperative. NVIDIA is responding with Halos, a full-stack safety system designed to engineer safety “across every layer of design, validation and deployment.” The company asserts this approach extends beyond pre-deployment checks, demanding ongoing assurance of hardware, software, AI behavior, and the operating environment as these machines increasingly share space with people.
Projected Growth of Physical AI Demands Scalable Safety
This scale of machine deployment necessitates a shift from isolated pre-deployment safety checks to a continuous, lifecycle-integrated model, ensuring ongoing validation throughout operation and adaptation. NVIDIA’s Halos AI Systems Inspection Lab addresses this need by transforming safety, cybersecurity, and AI safety requirements into repeatable inspections, preparing integrations for certification by independent agencies.
Functional safety is now paramount for scaling physical AI solutions, demanding engineering expertise, specialized data, and rigorous processes that few organizations can replicate independently. NVIDIA draws on over a decade of experience in autonomous vehicle safety, building proficiency in areas like sensor fusion, AI behavior assurance, and real-world validation to establish a foundational safety framework.
This framework extends beyond hardware and software, encompassing the AI itself, the operational environment, and the entire deployment lifecycle, a point emphasized by the need for accreditation as an ISO/IEC 17020 inspection body, a status NVIDIA’s Halos lab achieved through the American National Standards Board, the company says. TÜV Rheinland’s independent assessment of NVIDIA DRIVE AV, alongside ongoing inspections of IGX Thor, Halos OS, and Holoscan Sensor Bridge for functional-safety certification readiness, demonstrates a commitment to third-party validation, according to the company.
The company’s approach connects developers, software providers, sensor manufacturers, and certification bodies within a unified ecosystem, facilitating a collaborative approach to safety assurance. “Designing functional safety from the start is what separates a prototype from a scalable solution,” NVIDIA asserts, highlighting the importance of proactive safety engineering rather than reactive problem-solving. This focus on scalable solutions is critical as companies strive not only to build capable systems, but also to ensure they can be reliably assessed, certified, and trusted in real-world applications.
Dynamic Environments & AI Behavior Drive New Safety Standards
The increasing complexity of real-world settings demands safety systems that move beyond pre-programmed parameters, as autonomous systems now require contextual awareness to navigate unpredictable conditions. Traditional safety measures relying on static zones or physical barriers are insufficient for roads, factories, and warehouses where environments are constantly changing, necessitating adaptive behavior and safe fallback mechanisms when unexpected events occur. This shift necessitates a move from simply detecting obstacles to understanding the broader context surrounding them, a challenge NVIDIA addresses with its Halos platform.
Assuring the behavior of the artificial intelligence itself is now a core component of physical AI safety, requiring testing methodologies that extend beyond traditional functional safety checks. Manufacturers are beginning to incorporate AI-specific risk assessments, evidenced by emerging standards like ISO/IEC TS 22440, which focus on the unique challenges posed by machine learning models.
Validation at scale is further complicated by the ongoing evolution of autonomous systems through software updates, new tasks, and changing operating conditions, meaning material changes may require additional safety testing. NVIDIA’s approach combines real-world testing with simulation and synthetic data generation to address the sheer number of potential scenarios an autonomous system might encounter.
NVIDIA has also secured ANAB accreditation for its Halos AI Systems Inspection Lab as an ISO/IEC 17020 inspection body, enabling it to inspect Halos integrations and prepare companies for final certification, the firm reports. NVIDIA’s hardware foundations for this safety framework include the DRIVE AGX Thor, providing safety-engineered accelerated compute, and the Hyperion platform, a full-stack vehicle reference architecture.
The principles guiding safety in autonomous vehicles are shared with robotics, though the specific platforms, standards, and evidence required differ between domains.
NVIDIA Halos: A Full-Stack System for Physical AI Safety
NVIDIA is collaborating with semiconductor manufacturers including Infineon, NXP, STMicroelectronics, and Texas Instruments to integrate safety-critical components into physical AI systems, extending beyond software to encompass the foundational hardware layer. KION Group is actively developing functional safety agents specifically for autonomous forklifts, demonstrating the application of NVIDIA’s Halos system to industrial robotics, while Agility Robotics is integrating NVIDIA IGX Thor and Halos Core into the safety architecture of its Digit 5 humanoid robot, the company states. These partnerships highlight a broadening ecosystem built around the Halos platform, connecting developers, integrators, and assessment bodies.
Independent verification of NVIDIA’s safety processes is being conducted by multiple agencies; TÜV SÜD certified the company’s Automotive Product Lifecycle software process and DriveOS 6.0 to ISO 26262 ASIL D, the highest automotive safety integrity level, and also assessed NVIDIA’s automotive engineering processes to ISO/SAE 21434 standards.
AV & Robotics Validation via NVIDIA’s Safety Ecosystem & Certifications
This accreditation signifies a shift toward ongoing assessment throughout the entire development and operational lifecycle, addressing the need for continuous safety validation as autonomous vehicles and robotics proliferate. Hardware components like the NVIDIA DRIVE AGX Thor and Hyperion platform provide a safety-engineered foundation for level 4 AVs, while the IGX system is being inspected by TÜV Rheinland for functional-safety certification readiness. Collaboration is central to NVIDIA’s approach, with a growing ecosystem of partners building and integrating Halos.
Companies like AUMOVIO, Bosch, and Hesai are members of the Halos AI Systems Inspection Lab, focusing on areas from autonomous driving development to sensor technology and safety assurance. Software and embedded-system providers, including acontis and QNX, contribute essential components for predictable safety function execution, while Advantech and NexCOBOT develop safety-designed NVIDIA IGX systems, by the company’s account. Uber, Grab, and Lyft are using Hyperion to accelerate robotaxi development, demonstrating the platform’s scalability for commercial deployment. These assessments, alongside the ANAB accreditation, reinforce confidence in the Halos system and its ability to meet stringent safety standards.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
