Nvidia used its GTC Taipei keynote to name four new robotaxi partners building on its DRIVE Hyperion platform and to position a new safety-certified operating system, Halos OS, as the regulatory backbone for autonomous fleets. Uber and Autobrains are launching a robotaxi service in Munich on Hyperion. Foxconn is expanding its Nvidia collaboration to deploy fleets in Taiwan. VinFast is bringing Level 4 vehicles to Southeast Asia, and Saudi Arabia's HUMAIN is putting Hyperion-powered robotaxis on the road in the Middle East.
The four deals land as robotaxi services now operate in dozens of cities, a shift Nvidia frames as the industry crossing from prototype milestones to commercial operations. The pitch to automakers and operators is that scaling a fleet across jurisdictions is no longer a perception problem alone — it is a certification problem.
That is the gap Halos OS is designed to fill. Halos is Nvidia's full-stack safety system, and the OS sits on DRIVE Hyperion as the production software foundation. It bundles a certified kernel, standardized sensor and vehicle interfaces, AI safety guardrails, and a cloud-side validation framework into one stack.
Key facts
- 01Uber and Autobrains will launch a robotaxi service in Munich on Nvidia's DRIVE Hyperion platform.
- 02Foxconn, VinFast and Saudi Arabia's HUMAIN all signed deals to deploy DRIVE Hyperion robotaxi fleets across Taiwan, Southeast Asia and the Middle East.
- 03Halos Core, the OS foundation, is certified to ISO 26262 ASIL D, the highest automotive functional-safety standard.
- 04The Halos Safety Evaluation Framework draws on more than 330 research papers and 1,000 patents developed inside Nvidia Halos.
- 05Robotaxi services now operate in dozens of cities, moving from prototype to commercial deployment.
The argument Nvidia is making is regulatory, not just technical. Accurate perception and sound decision-making are well-studied problems, and progress is real. But fleets cannot operate at scale without auditable evidence that the entire system fails safely.
“Regulators require something more: proof that the overall system behaves reliably, isolates faults before they escalate and never operates outside the boundaries it was designed for.”— Riccardo Mariani, Nvidia VP of Industry Safety
Halos Core, the foundation layer, is the next generation of Nvidia DriveOS and is certified to ISO 26262 ASIL D — the highest automotive functional-safety classification. It includes a hypervisor that isolates safety-critical functions from non-critical ones so a failure in one domain cannot reach the vehicle controls. Safety-certified support for Nvidia CUDA and TensorRT is built in, alongside the TensorRT Edge-LLM open-source framework for on-vehicle large language model inference.
On top of the kernel sits the Halos SDK, which standardizes the messy middle of a robotaxi: cameras, radar, lidar and other sensors that each stream data in their own format at their own rate. A sensor abstraction layer decouples the autonomous driving stack from individual drivers, so swapping a lidar no longer cascades through application code. A vehicle abstraction layer does the same for the chassis. The SDK also ships a deterministic scheduler, zero-copy inter-process communication and a scenario data recorder.
The Halos Applications layer is where Nvidia tries to reconcile end-to-end AI driving with rule-based regulatory expectations. It pairs the top-rated Nvidia DRIVE active safety stack — automatic emergency braking, lane departure warning, blind spot monitoring, collision warning — with the Nvidia Alpamayo family of open models for autonomous vehicle development. Alpamayo runs chain-of-thought reasoning to continuously evaluate the road and adapt to changing conditions, while the deterministic guardrails sit underneath as the floor a regulator can audit.
“AI models can match human driving behavior, but regulators require more than performance.”— Riccardo Mariani, Nvidia VP of Industry Safety
The validation piece is Halos Infra and the recently released Halos Safety Evaluation Framework. Nvidia says the framework draws on more than 330 research papers and 1,000 patents developed inside Halos, and provides the tooling to build a credible safety case spanning L2 driver assistance through L4 robotaxis. Infra runs on Nvidia's three-computer architecture: DGX systems for training in the data center, Omniverse on OVX for simulation and synthetic data, and AGX in the vehicle for real-time sensor processing.
The competitive read here is that Nvidia is trying to do for robotaxi software what it already did for AI compute — make its stack the default substrate so every new entrant builds on Hyperion instead of rolling their own. Uber, Foxconn, VinFast and HUMAIN all picking Hyperion in the same week is a tell. Each is entering a different regulatory regime, and a pre-certified OS shortens the path through each one.
The open question is whether the Halos certification carries weight with regulators outside the markets where ISO 26262 is the operative standard. Saudi Arabia, Taiwan and Southeast Asia each have their own approval processes, and a single ASIL D claim on the kernel does not automatically translate into an operating permit. The Munich launch with Uber will be the most-watched test — German type-approval is among the most demanding in the world.
If Halos OS becomes the default robotaxi stack, Nvidia captures the most defensible layer of the autonomy market: the certified base on which every fleet operator's differentiation runs. The chip business already gives Nvidia the AGX in-vehicle computer and the DGX training cluster. Owning the safety-certified OS in between turns three product lines into one platform, and makes it materially harder for a Foxconn or a VinFast to swap suppliers later. That is the real prize in this announcement — not the four deals, but the architecture they all just agreed to share.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
