NVIDIA Launches Halos Safety Framework for Scalable Physical AI Deployment

NVIDIA has unveiled the Halos OS, a unified software foundation built on ASIL-D certified DriveOS, aimed at ensuring safety in physical AI applications. This framework is part of a broader strategy to support the projected 49 million level 3-5 autonomous vehicles expected to be installed by 2035 and the deployment of approximately 60 million industrial robots between 2026 and 2035. The Halos Safety Evaluation Framework provides tools for generating evidence of safety, while Halos Core focuses on system isolation, monitoring, and communication.

Key components include the NVIDIA DRIVE AGX Thor, a safety-engineered compute platform for autonomous vehicles, and the NVIDIA IGX Thor, an industrial-grade module for robotics that emphasizes functional safety. The Halos Outside-In Safety Blueprint is an open-source framework that utilizes external sensors to enhance safety measures.

NVIDIA's commitment to safety is further underscored by certifications such as ISO 26262 ASIL D for its automotive software processes and ISO/SAE 21434 for its engineering processes. The company asserts that scaling physical AI requires specialized engineering, data, and validation that few can replicate independently.

NVIDIA's Halos AI Systems Inspection Lab includes members like Bosch, Sony, and Lucid, focusing on safety, cybersecurity, and AI safety inspections. This initiative follows NVIDIA's previous discussions on open versus closed AI models at TechCrunch Disrupt 2026 and its establishment of the AI Data Center Power Coalition with Google and Emerald AI. The introduction of Halos OS and associated safety frameworks marks a significant step in addressing the complexities of deploying physical AI at scale.