AI-rewritten: This is a summary of an article from Ars Technica, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Nvidia has launched a full-stack safety system called Halos to reduce risks for autonomous vehicles and physical AI technologies like humanoid robots. The company introduced Halos for self-driving cars in 2025, followed by Halos for Robotics in June 2026. This expansion aims to enable safe deployments of mobile robots in warehouses, factory walkers, and surgical devices. Amit Goel, head of robotics ecosystem and edge computing at Nvidia, stated that while AI models and hardware are becoming capable, safety is the next bottleneck requiring this new system.

The Halos system combines specific hardware and software tools. It includes the IGX Thor computing module, which features an independent processor for safety workloads running alongside functional systems on the same silicon. The Halos operating system monitors every hardware block and software library to spot failures quickly while isolating critical tasks. Additionally, the Nvidia Holoscan Sensor Bridge connects sensor data with processing to identify corrupted information, often embedded in microcontrollers or Field Programmable Gate Arrays.

Adapting this system for robotics required redefining safety standards because different robots face unique environments. Goel explained that while driving safety definitions are consistent across companies, a vacuum cleaner poses different risks than a forklift handling heavy loads. The system allows developers to define custom safety functions without breaking the underlying infrastructure. Agility Robotics was the first to use Halos in its Digit 5 humanoid robot, allowing it to operate safely near people without needing isolated workstations or physical barriers.

Boston Dynamics is another early partner using Halos for its Spot robot dogs, wheeled Stretch robots, and Atlas humanoids, with plans to deploy Atlas in Hyundai factories by 2028. Other companies like KION Group and LG are also incorporating the technology. Goel noted that earlier robots were confined to single stations, but new generalized robots need to perform multiple tasks in diverse settings like hospitals and factories. He emphasized that improving safety is essential for these robots to operate effectively in production environments where unstructured elements can appear unexpectedly.