NVIDIA says safety must be built into every layer of physical AI, from the underlying hardware to the AI models and the operating environment. The company argues that traditional safety checks are no longer enough for systems that must adapt to changing conditions in roads, factories and warehouses.

The Halos framework spans hardware, software and simulation. For autonomous vehicles, it includes safety-engineered compute platforms, a unified operating system foundation and tools for generating evidence to support safety cases. For robotics, it offers industrial-grade modules, real-time sensing bridges and open-source blueprints that extend awareness beyond onboard sensors.

NVIDIA claims Halos is the first full-stack safety system for physical AI, drawing on more than a decade of experience in autonomous vehicle safety. The company says the principles are shared across AVs and robotics, while the platforms and standards remain specific to each domain.