NVIDIA has unveiled Halos, the first comprehensive safety system designed to help autonomous vehicles and robots prove they can operate safely in real-world conditions. As the autonomous vehicle industry accelerates toward mass deployment, with ABI Research projecting 49 million level 3-5 autonomous vehicles on roads by 2035, safety has become the critical bottleneck separating promising technology from trustworthy deployment.

Why Does Physical AI Safety Require a Completely New Approach?

Traditional safety testing for cars and industrial equipment relied on controlled environments and predictable scenarios. But autonomous systems operate in dynamic, unpredictable settings where roads, factories, and warehouses constantly change. A self-driving truck can't simply follow a pre-programmed route; it must perceive unexpected obstacles, adapt its behavior in real time, and reach a safe state if something goes wrong.

The challenge extends beyond hardware and software. Artificial intelligence itself introduces new risks that older safety standards never anticipated. A vehicle's AI model might behave unexpectedly in edge cases, rare situations that don't appear in training data. Regulators, insurers, and manufacturers now need evidence that hardware, software, AI behavior, and operating environments can work together safely without human intervention.

What Are the Four Shifts Defining New Safety Standards?

• Dynamic Environment Awareness: Roads, factories, and warehouses cannot be fully controlled through static zones or physical barriers. Autonomous systems must perceive changing conditions, adapt their behavior, and reach a safe state when something unexpected occurs.
• AI-Specific Assurance: Testing must assess AI software alongside traditional functional safety, using design-time, runtime, and validation-time guardrails. Emerging standards such as ISO/IEC TS 22440 are beginning to address these AI-specific risks.
• Ongoing Deployment Evolution: Autonomous vehicles and robots evolve through software and model updates, new tasks, and changing operating conditions. Material changes may require additional safety testing throughout the vehicle's lifecycle.
• Validation at Scale Through Simulation: The number and complexity of potential scenarios requires real-world testing to be combined with simulation, synthetic data generation, and scenario reconstruction.

Together, these shifts require safety to be operationalized across design, deployment, and validation, from underlying hardware to AI behavior and the operating environment.

How Does NVIDIA's Halos System Work Across Vehicles and Robots?

NVIDIA's Halos is built on more than a decade of autonomous vehicle safety development. The system connects specialized hardware, software, AI models, and validation tools into a unified framework that helps developers engineer safety across every layer of design, validation, and deployment.

For autonomous vehicles, Halos includes NVIDIA DRIVE AGX Thor, a safety-engineered computing platform, and NVIDIA Hyperion, a full-stack vehicle platform and reference architecture for level 4 autonomous driving. The software foundation runs on Halos OS, built on ASIL-D certified DriveOS, which ensures deterministic communication and system isolation. NVIDIA Alpamayo, an open reasoning vision language action model, brings explainability to long-tail scenarios where AI might encounter unexpected situations.

For industrial robotics, the approach adapts to different requirements. NVIDIA IGX Thor combines accelerated computing with functional safety on a single platform, designed to support systems developed for standards including IEC 61508 and ISO 13849. Halos Core for IGX provides the software foundation for safety-related operating functions, including fault detection, monitoring, and reporting.

Across both domains, the NVIDIA Halos AI Systems Inspection Lab turns safety, cybersecurity, and AI safety requirements into repeatable inspections. This helps prepare Halos integrations for final system-level certification by third-party agencies.

Which Companies Are Already Building on Halos?

NVIDIA's safety ecosystem is attracting major players across autonomous vehicles and robotics. In autonomous vehicles, Geely, Isuzu, Nissan (powered by Wayve software), and Einride are building level 4-ready vehicles on NVIDIA Hyperion, supported by Halos OS. Mobility providers including Uber, Grab, and Lyft are also using Hyperion to scale robotaxi development and deployment.

Einride, a Stockholm-based autonomous trucking company, recently announced a strategic collaboration with NVIDIA to accelerate its push into highway and suburban autonomous trucking. Einride currently operates hundreds of electric trucks for major shippers across the United States, Europe, and the Middle East, with autonomous trucks already operating in contracted customer deployments. The company expects to scale to 1,500 to 2,000 vehicles by 2028, with approximately 80 percent of that demand suitable for automation in the medium term.

"We have the customers, the operational experience and the technology. Building the next generation of the Einride Driver on the NVIDIA Hyperion platform will enable us to scale autonomous deployment across a freight network that's already serving customers today," stated Henrik Green, Chief Technology Officer at Einride.

Henrik Green, Chief Technology Officer at Einride

Einride will adapt the NVIDIA Hyperion platform for heavy-duty trucking, working directly with NVIDIA to extend the platform's compute, sensor, software, and safety architecture to the demands of heavy-duty freight. The company also intends to deploy NVIDIA Blackwell architecture at scale with an NVIDIA Exemplar Cloud partner to train, test, and refine its autonomous driving models. As part of its AI development workflow, Einride is using NVIDIA Cosmos to help improve the safety and reliability of its autonomous vehicles by searching and curating camera data to identify complex edge cases and augmenting real-world data with photorealistic synthetic scenarios.

How Are Regulators and Safety Bodies Validating These Systems?

Third-party certification is critical for building trust in autonomous systems. TÜV SÜD, an independent certification body, has certified NVIDIA's Automotive Product Lifecycle software process and DriveOS 6.0 to ISO 26262 ASIL D, the highest safety integrity level for automotive systems. TÜV SÜD also certified NVIDIA's automotive engineering processes to ISO/SAE 21434, a standard for cybersecurity in road vehicles.

TÜV Rheinland performed an independent UNECE safety assessment of NVIDIA DRIVE AV and is currently inspecting NVIDIA IGX Thor, Halos OS, and Holoscan Sensor Bridge for functional-safety certification readiness, building on TÜV SÜD's inspection of the Thor system-on-chip and Halos Core for ISO 26262.

The American National Standards Institute (ANSI) has accredited the NVIDIA Halos AI Systems Inspection Lab as an ISO/IEC 17020 inspection body. The lab inspects scoped Halos integrations and helps companies prepare for final certification by independent third-party bodies.

Beyond vehicle certification, infrastructure partnerships are emerging to support safe deployment. PrePass and Kodiak AI announced a collaboration to integrate enhanced inspections of Kodiak's autonomous trucks into established state roadside screening and enforcement systems. Using the PrePass platform, the companies created an automated bypass process that connects inspection information from Kodiak's autonomous trucks directly with enforcement personnel who can verify information and authorize bypass. Initial implementation began in Texas and Louisiana, with plans to expand nationwide.

"The next barrier to scaling driverless trucking isn't inside the truck. It's the infrastructure around it," noted Chas Wurster, Chief Technology Officer of PrePass.

Chas Wurster, Chief Technology Officer of PrePass

The collaboration supports Kodiak's operational and regulatory readiness as the company prepares to launch driverless commercial vehicle operations on public highways by the end of this year.

Steps to Understanding NVIDIA's Safety Approach for Autonomous Systems

• Recognize the Multi-Layer Challenge: Safety in autonomous systems requires validation across hardware, software, AI models, and operating environments, not just a single pre-deployment check.
• Understand AI-Specific Risks: Traditional functional safety standards don't address how AI models behave in unexpected scenarios, so new standards like ISO/IEC TS 22440 are being developed to fill this gap.
• Follow the Certification Path: Independent third-party bodies like TÜV SÜD and TÜV Rheinland validate safety systems against international standards, providing the evidence needed for regulatory approval and public trust.
• Track Real-World Deployment: Companies like Einride and Kodiak are moving from testing to actual customer operations, which requires both technological readiness and infrastructure partnerships with state agencies.

The autonomous vehicle and robotics industries are reaching an inflection point where safety assurance has become as important as technical capability. NVIDIA's Halos framework represents an attempt to standardize how companies prove their systems are safe, trustworthy, and ready for large-scale deployment. As 49 million autonomous vehicles are projected to enter roads by 2035, and 60 million industrial robots are expected to be deployed between 2026 and 2035, the ability to validate safety at scale will determine which companies succeed and which fall behind.