NVIDIA used a September 21 explainer to argue that the safety of AI-driven vehicles and robots must be assessed across their computing hardware, software and operating conditions. The company describes Halos as the framework connecting that work. Its account sets out a development approach, not evidence that every machine using its products is safe.
The company first announced Halos for autonomous vehicles on March 18, 2025. That announcement brought automotive hardware, software and AI research under a common safety framework, covering model training, simulation and deployment in a vehicle.
The latest explanation extends the discussion across vehicles and robotics. NVIDIA says changing environments, software updates and new tasks can require further validation. It presents simulation as a way to examine difficult scenarios alongside testing in the real world.
In its original announcement, NVIDIA also described an AI Systems Inspection Lab through which developers could assess how their products integrate with NVIDIA technology. The 2026 explainer says scoped inspections help companies prepare for final system certification by independent bodies. Assessing an integration is therefore distinct from certifying a complete vehicle or robot.
For readers following physical AI, that distinction matters: a safety framework describes how developers can organize evidence and testing. It does not establish the outcome for every deployment. NVIDIA is both the supplier and the source of these descriptions; this article does not independently certify its products.