Robotaxi Safety: Built-in from Design, Not Added On

What Happened?

NVIDIA has unveiled HALOS OS, a safety-certifiable operating system specifically designed for Level 4 robotaxis. This system integrates into the DRIVE Hyperion platform, already used by companies like Uber, Autobrains, Foxconn, VinFast, and HUMAIN in commercial deployments in Munich, Taiwan, Southeast Asia, and Saudi Arabia. The key to HALOS OS is that safety is not added later but is present from the design of both software and hardware. According to NVIDIA, HALOS OS is designed to meet the strictest functional safety standards, such as ISO 26262 ASIL-D and the cybersecurity standard ISO 21434, making it the first operating system for autonomous vehicles to seek safety certification from its core.

The announcement was made during GTC Taipei, where NVIDIA also revealed new collaborations: Uber and Autobrains will launch a robotaxi program in Munich based on DRIVE Hyperion, using Autobrains' agentic AI for scalable operations. Foxconn will expand its collaboration with NVIDIA to deploy robotaxi fleets in Taiwan, combining its services with DRIVE Hyperion for rapid integration. VinFast will work with Autobrains to bring Level 4 vehicles built on DRIVE Hyperion to the Southeast Asian market. HUMAIN will bring DRIVE Hyperion-powered robotaxis to Saudi Arabia, expanding the platform's global footprint into the Middle East. These deployments are not prototypes but commercial operations already underway in cities like Munich, Taipei, and Riyadh.

Why Is This Important?

The robotaxi industry has moved from prototypes to commercial operations, but safety remains the biggest hurdle for regulation and public acceptance. Until now, many systems focused on perception and decision-making, but regulators demand proof that the entire system is reliable, isolates faults, and never operates outside its limits. HALOS OS addresses these four challenges: a certifiable operating system, secure hardware, fault management, and clear operational boundaries. Historically, safety approaches in autonomous vehicles have been reactive, such as the Uber incident in 2018 where a vehicle failed to detect a pedestrian due to failures in risk management. HALOS OS introduces a proactive approach: safety is mathematically modeled from the design, using formal verification techniques and partition isolation to ensure that failures in one module do not propagate to the entire system.

Additionally, HALOS OS integrates with NVIDIA's hardware, such as the Drive Thor SoC and the sensors of the Hyperion platform, which includes 12 cameras, 9 radars, 12 ultrasonics, and 3 lidars. This vertical integration allows the operating system to directly manage safety-critical peripherals, reducing latency and eliminating intermediate software layers that could introduce vulnerabilities. NVIDIA claims that HALOS OS can detect and isolate faults in less than 10 milliseconds, a crucial time for emergency maneuvers.

Market Implications

• Standardization: NVIDIA aims for DRIVE Hyperion and HALOS OS to become the global reference platform, accelerating regulatory homologation. Companies like Uber and Foxconn have already adopted the platform, suggesting NVIDIA could replicate the Android model in the robotaxi world: an open but certifiable ecosystem that reduces market fragmentation. It is expected that by 2027, more than 50 cities will have commercial robotaxi operations, and NVIDIA aspires for at least half to use its platform.
• Collaborations: Uber and Autobrains in Munich, Foxconn in Taiwan, VinFast in Southeast Asia, and HUMAIN in Saudi Arabia demonstrate that the platform is scalable and adaptable to different markets. Each collaboration brings specific advantages: Autobrains contributes agentic AI to handle complex scenarios; Foxconn offers mass manufacturing capability; VinFast provides local market knowledge in Southeast Asia; and HUMAIN facilitates entry into the Middle East, a region with high demand for autonomous mobility due to its smart city investments.
• Competition: Waymo and Cruise also have their own approaches, but NVIDIA's bet on an open and certifiable system could make a difference. Waymo uses its own software and hardware stack, while Cruise relies on GM hardware and proprietary software. However, both have faced regulatory issues: Cruise had to suspend operations in San Francisco after an accident in 2023. By offering a pre-certified platform, NVIDIA reduces the regulatory burden for its partners, potentially accelerating deployments. Additionally, companies like Tesla, which bet on pure vision, could be pressured to adopt a more holistic approach if HALOS OS proves effective.

What Readers Should Know

Safety in robotaxis is not just about AI algorithms; it requires a holistic approach spanning from the operating system to hardware. HALOS OS represents a paradigm shift: safety as a design requirement, not a patch. For users, this means greater confidence that autonomous vehicles meet rigorous standards, such as the ability to operate safely even if a sensor or software module fails. For companies, adopting these platforms can reduce development costs by up to 40% according to NVIDIA estimates, by avoiding the need to design certifiable safety systems from scratch. Additionally, the platform enables secure over-the-air (OTA) updates, facilitating continuous improvement without needing to recall entire fleets.

However, challenges remain: certification of HALOS OS is still in progress, and NVIDIA has not disclosed specific timelines for approval by bodies like NHTSA or UNECE. Moreover, reliance on a single hardware and software vendor could pose vendor lock-in risks. Finally, public acceptance remains a critical factor: according to a 2025 AAA survey, only 30% of Americans would trust a robotaxi. HALOS OS, by focusing on transparency and certification, could help improve that perception, but cultural change will be gradual.

“Safety is not an add-on; it is the foundation upon which trust in autonomous mobility is built.” — Source: NVIDIA Blog