From the Lab to the Road — How NVIDIA Is Reshaping the Safety Foundation of Autonomous Driving

On a closed test track, a prototype intelligent vehicle moves forward slowly. A group of engineers stands nearby, not watching the steering wheel or sensors, but fixated on the real-time signals streaming across their monitors. This is another simulation test underway at NVIDIA's AI Systems Inspection Lab.

You may not have heard of this lab yet, but it's where NVIDIA and its partners validate the safety integration of their products with core elements of Halos. And Halos? It's not a product name — it's a safety framework. A bold attempt by NVIDIA to rebuild the safety foundation for intelligent assisted driving and create a system worthy of trust.

NVIDIA Halos is a full-stack safety framework for intelligent vehicles, integrating hardware, software, AI models, simulation environments, and validation services. At its core is the concept of physical AI—a safety architecture that spans the full process from cloud training to in-vehicle deployment.

Starting from infrastructure, NVIDIA first introduced the world's first AI Systems Inspection Lab. On that foundation, it gradually evolved Halos—a comprehensive safety system that stretches from cloud to edge to vehicle.

1.     AI Systems Inspection Lab: From Setting Standards to Earning Trust

In January 2025, NVIDIA unveiled the industry's first AI Systems Inspection Lab at CES. More than a symbolic milestone, it is certified by ANAB under ISO/IEC 17020, and integrates six key domains into a single framework: functional safety (ISO 26262), safety of intended functionality (ISO 21448), cybersecurity (ISO 21434), AI functional safety (ISO PAS 8800, ISO/IEC TR 5469), and relevant UN regulations (UN-R 79, 13-H, 152, 155, 157, and 171).

Built to support partners using NVIDIA DRIVE AGX Orin and Thor platforms, the lab helps verify whether AI-driven vehicle systems meet the industry's most rigorous safety requirements.

"Being a member of the AI Systems Inspection Lab means working at the forefront of automotive systems innovation and integrity," said Cristian Casorran Hontiyuelo, advanced driver-assistance system engineering and product manager at Ficosa.

OMNIVISION, onsemi, and Continental have also joined, validating sensors, central compute modules, and chip interfaces.

The methods developed and refined in the lab would go on to shape Halos.

2.  The Birth of Halos: From Philosophy to Systemic Safety Engine

In spring 2025, NVIDIA officially launched Halos at GTC. More than just a product, Halos embodies a "safety-first philosophy"—not adding safety as an afterthought, but building it in from the first line of code.

Halos integrates all of NVIDIA's core technologies: DRIVE AGX platforms (compute), DriveOS (operating system), DGX (AI model training), Omniverse and Cosmos (simulation), and the Inspection Lab (compliance and validation).

The Halos architecture spans the full lifecycle — from design and deployment to validation — and comprises four essential pillars：

Platform Safety: Halos features a safety-assessed system-on-a-chip (SoC) with hundreds of built-in safety mechanisms.

It also includes NVIDIA DriveOS software, a safety-certified operating system that extends from CPU to GPU; a safety-assessed base platform that delivers the foundational computer needed to enable safe systems for all types of applications; and DRIVE AGX Hyperion, a hardware platform that connects SoC, DriveOS and sensors in an electronic control unit architecture.

Algorithmic Safety: Halos includes libraries for safety data loading and accelerators, and application programming interfaces for safety data creation, curation and reconstruction to filter out, for example, undesirable behaviors and biases before training.

It also features rich training, simulation and validation environments harnessing the NVIDIA Omniverse Blueprint for AV simulation with NVIDIA Cosmos world foundation models to train, test and validate AVs. In addition, it boasts a diverse AV stack combining modular components with end-to-end AI models to ensure safety with cutting-edge AI models in the loop.

Ecosystem Safety: Halos includes safety datasets with diverse, unbiased data, as well as safe deployment workflows, comprising triaging workflows and automated safety evaluations, along with a data flywheel for continual safety improvements — demonstrating leadership in AV safety standardization and regulation.

AI Systems Inspection Lab: First of its kind to be accredited by ANAB, NVIDIA AI Systems Inspection Lab inspects and verifies the integration of partners' products with NVIDIA Halos' safety elements

Most importantly, Halos standardizes, modularizes, and formalizes these capabilities into a reusable toolkit—supporting full-lifecycle safety across data generation, training, simulation, deployment, and OTA updates.

"Halos is not just a technology stack—it's a philosophy that makes every phase of development traceable, testable, and explainable," said Riccardo Mariani, VP of Industry Safety, NVIDIA.

3. Process-Driven Safety: The Foundation of Halos

The roots of Halos trace back to NVIDIA's early partnerships with TÜV SÜD and TÜV Rheinland for DRIVE Orin and DRIVE AV platform certification. These collaborations helped formalize the "design–verify–deploy" model, now embedded in Halos.

The system is modular and developer-friendly: partners can build custom L3 solutions atop DRIVE OS or use Cosmos-based synthetic data to simulate edge cases — all within Halos' structured framework.

Use cases range from urban delivery to highway freight:

These examples show how Halos' physical AI security model supports scalability and domain adaptation.

4. Halos Beyond Cars: Expanding into Robotics

At GTC Paris in June 2025, NVIDIA announced the expansion of Halos beyond assisted driving — now including robotics under its AI safety framework.

In June 2025 at GTC Paris, NVIDIA extended Halos beyond vehicles into robotics.

Companies like Boston Dynamics, Advantech, Trimble, and Easyrain have joined the Lab ecosystem: Boston Dynamics validates emergency behavior and autonomous movement of its Atlas robot; Advantech uses Halos to test logistics robots under dynamic warehouse conditions.

This reflects Halos' broader mission: building trust in physical AI systems across industries, from streets to factories.

At the GTC Paris conference held on June 11, NVIDIA officially announced the expansion of its Halos framework from a focus solely on intelligent driving to encompass the broader robotics ecosystem, aiming to enhance safety across the full development lifecycle of AI-powered machines.

Specifically, the Halos AI Systems Inspection Lab—originally designed for intelligent driving—now also supports the functional safety assessment of robotics platforms.

With this broadened scope, several leading companies have joined the Halos Lab ecosystem, including Bosch, active safety developer Easyrain, positioning solutions provider Trimble, mobile robotics pioneer Boston Dynamics, IoT and embedded systems specialist Advantech, logistics operator ArcBest, autonomous agriculture innovator Bluewhite, safety radar manufacturer Inxpect, intelligent motion control solution provider NexCOBOT, and motion technology company Synapticon.

This marks a shift in Halos' role—from a symbol of automotive-grade safety to a foundational trust framework for all AI systems that engage in real-world physical decision-making.

5. China: A Regulatory Window of Opportunity

In late 2024, China's Ministry of Industry and Information Technology issued new regulations requiring type approval for L3+ driving systems. These policies prioritize explainability, system traceability, and AI cybersecurity—exactly the areas Halos addresses.

NVIDIA has aligned Halos with these evolving compliance needs. Without binding partners to any commercial stack, Halos provides modular tools—simulation, data governance, validation APIs—that help Chinese OEMs and suppliers navigate this new regulatory terrain.

Conclusion

As intelligent technologies accelerate the transformation of the automotive industry, bringing AI into vehicles takes more than just leaps in algorithm accuracy. The real challenge lies in building a standardized, sustainable safety framework that the entire industry can rely on.

That's precisely the thinking behind NVIDIA's Halos system. By connecting platform design, AI model training, simulation testing, and regulatory compliance, Halos weaves safety into every stage of vehicle development.

The road to smarter mobility may be long, but with Halos, the journey is faster—and far more secure.