Physical AI is accelerating the development of autonomous vehicles (AVs), offering up new robotics and autonomy possibilities. Safe drivers, passengers, and pedestrians require suitable frameworks and technology.
For this reason, NVIDIA today unveiled NVIDIA Halos, a holistic safety system that combines its state-of-the-art AI research in AV safety with its range of automotive hardware and software safety solutions.
With an emphasis on AI-based, end-to-end AV stacks, NVIDIA Halos combines chips, software, tools, and services to assist ensure the safe development of AVs from the cloud to the automobile.
“With the launch of Halos, we’re empowering partners and developers to choose the state-of-the-art technology elements they need to build their own unique offerings, driving forward a shared mission to create safe and reliable autonomous vehicles.” “Halos can potentially speed up standardization and regulatory compliance while enhancing current safety practices.”
At the Heart of NVIDIA Halos
Three distinct but complimentary layers make up the comprehensive safety system known as Halos.
It covers platform, algorithmic, and ecological safety at the technological level. It comprises design-time, deployment-time, and validation-time constraints at the development stage. Additionally, it uses three powerful computers the NVIDIA DGX for AI training, the NVIDIA Omniverse and Cosmos running on the NVIDIA OVX for simulation, and the NVIDIA DRIVE AGX for deployment to span the computing level from AI training to deployment.
According to Marco Pavone, lead AV researcher at NVIDIA, “Halos’ holistic approach to safety is particularly critical in a setting where companies want to harness the power of generative AI for end-to-end development of increasingly powerful AV systems that eliminate the need for conventional compositional design and verification.
AI Systems Inspection Lab
The NVIDIA AI Systems Inspection Lab is a gateway to NVIDIA Halos that enables developers and automakers to confirm that their products safely integrate with NVIDIA technologies.
The AI Systems Inspection Lab is the first global program to receive accreditation from the ANSI National Accreditation Board for an inspection plan that combines cybersecurity, functional safety, AI safety, and regulations into a single safety framework. This accreditation was disclosed during the CES trade show earlier this year.
Ficosa, OMNIVISION, Onsemi, and Continental are the first members of the AI Systems Inspection Lab.
Key Elements of NVIDIA Halos
Platform safety, algorithmic safety, and ecosystem safety are the three main pillars around which Halos is based.
Platform Safety
Halos has hundreds of integrated safety features in its safety-assessed system-on-a-chip (SoC).
Additionally, it comes with DRIVE AGX Hyperion, a hardware platform that links SoC, DriveOS, and sensors in an electronic control unit architecture; a safety-assessed base platform that provides the fundamental computer required to enable safe systems for all kinds of applications; and NVIDIA DriveOS software, a safety-certified operating system that spans from CPU to GPU.
Algorithmic Safety
Halos comprises application programming interfaces for safety data production, curation, and reconstruction to remove, for instance, biases and undesired behaviors prior to training, as well as libraries for safety data loading and accelerators.
In order to train, test, and verify AVs, it also has comprehensive training, simulation, and validation environments that use the NVIDIA Omniverse Blueprint for AV simulation with NVIDIA Cosmos world foundation models. It also has a varied AV stack that combines end-to-end AI models with modular components to guarantee safety while utilizing state-of-the-art AI models.
Ecosystem Safety
Halos demonstrates leadership in AV safety standardization and regulation by providing safety datasets with a variety of objective data, safe deployment procedures that include automated safety evaluations and triaging workflows, and a data flywheel for ongoing safety improvements.
Safety Track Record
NVIDIA’s extensive safety-focused technology research, development, deployment, partnerships, and collaborations are all combined in Halos. These include:
- More than 15,000 engineering years have been devoted to automobile safety.
- Contributions to international standards committees totaling over 10,000 hours
- more than 1,000 AV-safety patents submitted
- 240+ research papers on AV safety have been published.
- More than thirty cybersecurity and safety certifications
Additionally, it aligns with previous noteworthy safety certifications and evaluations of NVIDIA automotive solutions, such as:
- The operating system NVIDIA DriveOS 6.0 complies with ISO 26262 criteria for automotive safety integrity level (ASIL D).
- NVIDIA received ISO/SAE 21434 Cybersecurity Process certification from TÜV SÜD for its software engineering, platform, and automotive SoC processes.
- For the United Nations Economic Commission for Europe, TÜV Rheinland conducted an independent safety evaluation of NVIDIA DRIVE AV in relation to the safety standards for intricate electronic systems.
