For Robotaxis, Safety Must Be Built In, Not Bolted On

Overview

NVIDIA introduced Halos OS, a unified safety foundation designed to build vehicle safety into the core of robotaxi systems rather than adding it later. The system runs on NVIDIA DRIVE Hyperion, meets the ISO 26262 ASIL D automotive safety standard, and is structured in layers that isolate safety-critical software from the rest of the driving stack. NVIDIA frames the move as a response to robotaxis shifting from test pilots to paid commercial rides in dozens of cities.

Key Takeaways

Halos OS is a production-ready safety foundation for AI-driven autonomous vehicles, built on the NVIDIA DRIVE Hyperion platform.
The core operating system meets ISO 26262 ASIL D, the strictest automotive functional safety level, and uses a hypervisor to wall off safety-critical functions.
NVIDIA organizes the system into layers: a certified core, a middleware SDK, safety applications, and cloud infrastructure for training and validation.
The Halos Safety Evaluation Framework gives developers tools to build a documented safety case backed by research and patents.
Named adopters and partners include BYD, Geely, Isuzu, Nissan, Uber, Foxconn, and VinFast for Level 4 vehicles and robotaxi programs.
NVIDIA argues safety must be engineered into the architecture from the start because regulators require proof of reliable, fault-tolerant behavior.

Stats & Key Facts

#330+ research papers underpin the Halos Safety Evaluation Framework, according to NVIDIA.
#1,000 patents were developed within Halos OS and feed into its safety evaluation work.
#Dozens of cities now host commercial robotaxi services, the scale NVIDIA cites as the reason for built-in safety.
#ISO 26262 ASIL D is the highest of the four automotive safety integrity levels the core OS conforms to.
#4 core layers make up the architecture: Core, SDK, Applications, and Infrastructure.
#Multiple global automakers, including BYD, Geely, Isuzu, and Nissan, have adopted DRIVE Hyperion for Level 4 vehicles.

For Robotaxis, Safety Must Be Built In, Not Bolted On

What Halos OS Is and the DRIVE Hyperion Platform It Runs On

NVIDIA positions Halos OS as a single safety foundation for self-driving cars rather than a loose collection of features.

Halos OS is a full-stack safety system built on NVIDIA DRIVE Hyperion, the company's hardware-and-software platform for Level 4 autonomous vehicles. The goal is to give robotaxi makers one production-ready base for safety instead of stitching together separate parts.

DRIVE Hyperion supplies the redundant computing, sensors, and functional safety mechanisms that the standards require. NVIDIA describes the combination as covering the full path from cloud to car, including validation across large numbers of driving scenarios.

Why NVIDIA Says Safety Must Be Built In, Not Bolted On

The headline phrase reflects a design philosophy about where safety belongs in the system.

›Robotaxis have moved from prototype milestones to paying customers in dozens of cities, raising the stakes for reliability.
›Regulators want proof that a vehicle behaves predictably, isolates faults, and stays within its designed operating limits.
›Bolting safety on as an afterthought leaves gaps that an integrated architecture is meant to close.
›Building safety from the operating system upward lets every layer above inherit the same guarantees.

The Halos Core Operating System and ISO 26262 ASIL D Certification

The bottom layer is a certified operating system that anchors the rest of the stack.

Halos Core is built on NVIDIA's next-generation DriveOS and conforms to ISO 26262 ASIL D, the most demanding level in the main automotive functional safety standard. That rating signals the system is engineered for the highest tolerance against dangerous failures.

The core uses a hypervisor, software that partitions the computer so safety-critical functions run in isolation from everything else. It also provides safety-certified support for NVIDIA's CUDA and TensorRT software, so AI workloads run on a foundation that carries the same safety pedigree.

The SDK and Applications Layers That Sit Above the Core

Two middle layers connect the certified core to the actual driving software.

›The Halos SDK is middleware with sensor and vehicle abstraction layers, so a driving stack does not have to be rewritten for every hardware change.
›It adds deterministic scheduling and zero-copy communication between processes, which keeps timing predictable and reduces wasted data copying.
›Halos Applications add rule-based safety guardrails around the AI, including world-model perception.
›The active safety stack covers automatic emergency braking, lane departure warning, blind spot monitoring, and collision warning.

The Halos Safety Evaluation Framework and the Inspection Lab

NVIDIA pairs the on-vehicle stack with cloud tools and an independent certification path.

The Halos Safety Evaluation Framework gives developers tools and guidelines to assemble a credible safety case, from driver-assistance features up to full Level 4 robotaxis. NVIDIA says the framework draws on more than 330 research papers and 1,000 patents, and it measures key performance indicators that supply statistical evidence for a safety argument.

Supporting research includes Sim2Val, a method developed with collaborators at Harvard and Stanford that blends real-world and simulated test results to cut the cost of physical road miles. Separately, the NVIDIA Halos AI Systems Inspection Lab, accredited by ANAB, offers independent inspection and certification of robotaxi fleets, AV stacks, and sensors.

Automakers and Mobility Partners Adopting the Platform

NVIDIA names a growing list of companies building on DRIVE Hyperion and Halos.

›Global automakers including BYD, Geely, Isuzu, and Nissan have adopted DRIVE Hyperion for Level 4 vehicles.
›Uber and Autobrains are tied to a robotaxi effort using agentic AI.
›Foxconn is expanding robotaxi fleet deployment, and VinFast is pursuing Level 4 vehicles for Southeast Asia.
›The expansion spans markets in Asia, Europe, and the Middle East as commercial services scale up.

Frequently Asked Questions

What is NVIDIA Halos OS?

Halos OS is a unified, production-ready safety system for AI-driven autonomous vehicles. It runs on NVIDIA DRIVE Hyperion and is organized into a certified core operating system, a middleware SDK, safety applications, and cloud infrastructure for training and validation.

What does ISO 26262 ASIL D mean for a robotaxi?

ISO 26262 is the main automotive functional safety standard, and ASIL D is its strictest level. A system rated at ASIL D is engineered to the highest tolerance against failures that might cause harm, which matters when no human driver is in the seat.

What does built in, not bolted on mean here?

It means safety is designed into the system architecture from the operating system upward rather than added later as separate features. NVIDIA argues this approach better satisfies regulators who require proof of reliable, fault-isolated behavior.

Which companies are using NVIDIA's robotaxi platform?

NVIDIA names automakers including BYD, Geely, Isuzu, and Nissan adopting DRIVE Hyperion for Level 4 vehicles, along with mobility partners such as Uber, Foxconn, and VinFast tied to robotaxi programs.

How does NVIDIA prove a robotaxi is safe?

Through the Halos Safety Evaluation Framework, which provides tools to build a documented safety case backed by research and key performance indicators. An ANAB-accredited inspection lab also offers independent certification of fleets, AV stacks, and sensors.

NVIDIA is betting that the next phase of the robotaxi business depends on safety that is engineered into the core of the vehicle's software and certified to recognized standards. Halos OS packages that approach into one foundation aimed at automakers and mobility operators scaling driverless rides across more cities.