On June 10, 2026 (local time), NVIDIA published a detailed look at Halos OS, its safety foundation for robotaxis, on the company's official blog[1]. The core design philosophy is that safety must be built into the operating system from the ground up, not bolted on afterward. At GTC Taipei, NVIDIA also announced new collaborations with Uber, Foxconn, VinFast and others to bring robotaxis built on its platform to cities around the world[1].

Robotaxi Programs Spinning Up Worldwide

The app says your ride has arrived, and when you get in, no one is sitting in the driver's seat. In the dozens of cities now hosting robotaxi services, this is already everyday reality[1]. The industry has moved past prototype milestones into commercial operations, and several collaborations announced at GTC Taipei illustrate how quickly deployments are expanding[1].

The key announcements are as follows. Uber and Autobrains are launching a robotaxi program in Munich, Germany, on the NVIDIA DRIVE Hyperion platform. Foxconn is expanding its collaboration with NVIDIA to deploy robotaxi fleets, combining its services with DRIVE Hyperion for rapid integration and scaling in Taiwan. Vietnam's VinFast is working with Autobrains to bring level 4 vehicles built on DRIVE Hyperion to the Southeast Asian market. In addition, Saudi Arabia's HUMAIN is working to bring DRIVE Hyperion-powered robotaxis to the Middle East, extending the platform's footprint to a global scale[1].

Safety Cannot Be an Afterthought: Four Challenges

Discussions of level 4 autonomy tend to focus on what the vehicle can perceive and decide. But NVIDIA points out that regulators and certification bodies demand something more: proof that the overall system behaves reliably, isolates faults before they escalate, and never operates outside the boundaries it was designed for[1].

NVIDIA argues that robotaxi safety requires solving four challenges simultaneously: a safety-certifiable operating system, safe and standardized hardware and software interfaces, AI that operates within verifiable guardrails, and validation at scale before vehicles touch public roads[1]. Halos OS, offered as a component of the NVIDIA Halos full-stack safety system, is positioned as a unified, production-ready safety foundation built on DRIVE Hyperion to address these challenges[1].

Halos OS consists of three main layers. At the foundation is Halos Core, the next generation of NVIDIA DriveOS, compliant with ISO 26262 ASIL D, the automotive functional safety standard. A hypervisor — a specialized software layer for virtualization — isolates safety-critical functions so that failures cannot reach vehicle controls. It includes safety-certified support for CUDA and TensorRT, along with TensorRT Edge-LLM, an open source framework for high-performance large language model inference in the vehicle[1].

Above that sits the Halos SDK, a standardized middleware layer for handling cameras, radar, lidar and other sensors that each stream data in different formats and at different rates. Its sensor abstraction layer keeps additions or swaps of sensors from rippling through application code, and it provides the runtime building blocks that safety-critical software demands: a deterministic scheduler, zero-copy inter-process communication, and a comprehensive system error-handling framework[1].

The top layer, Halos Applications, provides safety guardrails for AI. It is built from deterministic, rule-based functions designed to behave within defined bounds, and includes the NVIDIA DRIVE active safety stack featuring automatic emergency braking, lane departure warning, blind spot monitoring and more. The layer is also designed to be combined with end-to-end AI models where explainability is essential, including NVIDIA Alpamayo, a family of open models that uses chain-of-thought reasoning to continuously evaluate the road[1].

Halos Infra: Cloud-Side Development and Validation

Beyond the vehicle itself, NVIDIA also provides cloud-side infrastructure. Halos Infra enables autonomous vehicle training, simulation and validation at scale, and serves as the foundation for the recently released Halos Safety Evaluation Framework (SEF). SEF provides the tools and guidelines needed to build a credible safety case, from level 2 driver assistance to level 4 robotaxis, and draws on more than 330 research papers and 1,000 patents developed within NVIDIA Halos OS[1].

Halos Infra runs on NVIDIA's three-computer autonomous driving solution: NVIDIA DGX systems for training the AI stack in the data center, NVIDIA Omniverse on NVIDIA OVX systems for simulation and synthetic data generation, and the NVIDIA AGX in-vehicle computer for real-time sensor processing and safety. Halos OS thus spans the full development lifecycle, from training and simulation to inference in the vehicle itself[1].

Summary

Halos OS is a safety foundation for robotaxis that answers four challenges — a certified OS foundation, standardized interfaces, AI guardrails and validation at scale — with a single stack[1]. Adoption is spreading among companies such as Uber, Foxconn, VinFast and HUMAIN, with deployments taking shape from Munich to Taiwan, Southeast Asia and the Middle East. Robotaxis are still rare on Japanese roads, but foundations like this, which build safety in from the ground up, are likely to determine how fast they spread.

Source: https://blogs.nvidia.com/blog/halos-os-robotaxi-safety/