Safety in Robotaxis: More than an Add-on
A car pulls up to the curb. The app notifies: "Your vehicle has arrived." No one in the driver's seat. For those living in cities with robotaxi services, this scene is already common. The robotaxi industry has evolved from prototypes to commercial operations, with an expanding ecosystem accelerating the pace of deployment. New partnerships announced at NVIDIA GTC Taipei show robotaxi programs emerging globally. Uber and Autobrains are launching a program in Munich, using the NVIDIA DRIVE Hyperion platform. Foxconn expands its collaboration with NVIDIA to deploy fleets in Taiwan. VinFast and Autobrains are bringing level 4 vehicles to Southeast Asia. Meanwhile, HUMAIN is bringing robotaxis to Saudi Arabia, expanding the platform's presence in the Middle East.
Safety: The Foundation of the Industry
With the growth of the robotaxi industry, safety is crucial. Regulators, certification bodies, and developers are paying attention to what is needed for safe large-scale deployment. Discussions about level 4 autonomy usually focus on what the vehicle can perceive and decide. And it makes sense. Accurate perception, sound decisions, and dealing with the unexpected are complex challenges being addressed. But perception and decision are not everything. Regulators demand more: proof that the system behaves reliably, isolates faults before they escalate, and never operates outside its designed limits.
Robotaxi safety requires solving four challenges simultaneously: a certifiable operating system, standardized and secure hardware and software interfaces, AI that operates within verifiable limits, and validation at scale before vehicles hit the streets. To help address these challenges, the newly launched Halos Operating System (OS) offers a unified, production-ready safety foundation for AI-powered vehicles, built on NVIDIA DRIVE Hyperion.
Halos: The Safety Foundation for Robotaxis
At the heart of the NVIDIA Halos OS is the Halos Core, the next generation of NVIDIA DriveOS, certified for automotive safety standards. It is audited, documented, and proven to behave predictably under fault conditions, with a hypervisor that isolates safety-critical functions so failures don't reach vehicle controls. The Halos Core is compliant with ISO 26262 ASIL D, includes certified support for NVIDIA CUDA and TensorRT, and offers the open-source TensorRT Edge-LLM framework for high-performance language model inference.
The Halos SDK removes the burden of integrating cameras, radars, and other sensors. Its sensor abstraction layer decouples the autonomous driving stack from individual drivers. This allows adding or swapping sensors without impacting the application code.
Additionally, the Halos SDK provides the necessary building blocks for safety-critical software. It includes a deterministic scheduler for predictable timing and zero-copy inter-process communication. The SDK also features a comprehensive error handling framework. A robust scenario data recorder is also part of the package. These elements are fundamental to ensuring the reliability and low latency of automotive applications.
Guardrails and Safety Evaluation
AI models can match human driving behavior, but regulators demand more than performance. The Halos Applications layer provides safety guardrails for AI through deterministic, rule-based functions. These functions are designed to operate within defined limits. It includes world model perception and the NVIDIA DRIVE active safety stack, featuring automatic emergency braking, lane departure warning, blind spot monitoring, and more.
Halos Infra is the cloud development infrastructure that enables the training, simulation, and validation of autonomous vehicles at scale. It is the foundation for the recently launched Halos Safety Evaluation Framework (SEF), which offers tools and guidelines to build a credible safety case, from level 2 driver assistance to level 4 robotaxis.
Ultimately, safety in robotaxis must be built-in from the start, not just added as an accessory. And with the advancement of technology, the promise of safe and reliable autonomous transportation is closer to becoming a reality.
Comments (0)
Comments are moderated and if they violate our Terms and Conditions of use, the comment will be deleted. Persistence in violation will result in a ban of your account.