Nvidia expands its Halos safety system from robotaxis to robots

In short: Nvidia is extending its Halos safety system from self-driving cars to a wider range of robots, and several robotics companies are already using it.

What happened

Nvidia has spent billions on what it calls “physical AI,” meaning AI used in real-world machines like self-driving cars and robots. The company says safety is becoming a key bottleneck, because these machines increasingly work close to people.

Nvidia launched Halos in 2025 for autonomous vehicles, like robotaxis. In June 2026, it announced Nvidia Halos for Robotics, which aims to help companies build safer warehouse robots, humanoid robots in factories, and even surgical robots.

Halos is described as a “full-stack” safety system, which means it covers multiple layers at once. It includes special computing hardware with a separate safety processor (like having an independent co-pilot watching for danger), software that constantly checks for failures, tools to spot corrupted sensor data, and virtual testing through simulations (practice runs in a digital world). Nvidia also runs an inspection lab program so partners can get feedback on safety issues during development.

Why it matters

Robots are moving out of fenced-off areas and into places where people walk and work. A safety system like Halos could make it easier for companies to deploy robots without building new barriers and sensor setups each time, as long as it works as promised.

Nvidia says companies including Agility Robotics, Boston Dynamics, KION Group, and LG are incorporating Halos in different ways. The big open question is whether safety tools like this can keep up as robots take on more tasks in more unpredictable settings, like busy warehouses and hospitals.

Source: Arstechnica