Nvidia introduced the Halos safety suite in 2025 as a combination of hardware and software designed to place guardrails around self-driving cars. In June 2026 the company announced Halos for Robotics, extending the same safety architecture to autonomous mobile robots used in warehouses, factory floors and even surgical settings. The CEO said the physical-AI segment is already generating close to $10 billion of yearly revenue.
Beyond the platform, Nvidia has taken equity stakes in several robotics start-ups and forged collaborations that embed its technology. Investments include the humanoid-robot firms Agility Robotics and Figure AI, while a joint effort with the Chinese company Unitree supplies an open reference design for research-grade humanoids. These moves aim to seed the ecosystem that will rely on the Halos safety stack.
The hardware foundation of the robotics offering centers on the IGX Thor module, a compute board that houses a dedicated processor for safety-critical tasks. According to Nvidia engineer Goel, the functional and safety systems must run on the same silicon to ensure reliable physical AI operation. This integration is intended to keep safety logic isolated yet tightly coupled with the main workload.
On the software side, the Halos operating system continuously audits every hardware block and software library, enabling rapid detection of anomalies. It also segregates safety-sensitive computations to prevent interference from non-critical processes. Complementing this, the Holoscan Sensor Bridge routes sensor streams into the safety pipeline and can be embedded directly in microcontrollers or field-programmable gate arrays for low-level validation.
Nvidia also bundles virtual testing tools and an inspection lab service with the Halos suite. The simulation environment lets developers evaluate robot behavior under a wide range of fault conditions before hardware is built, while the lab program provides partner companies with quick feedback on any safety artifacts discovered during development. These resources aim to accelerate certification and deployment timelines.
Early adopters illustrate the breadth of applications. Agility Robotics integrated Halos into its Digit 5 humanoid, consolidating safety sensors inside the machine and removing the need for external perimeter devices. Boston Dynamics joined the Halos accreditation program for its Spot, Stretch and Atlas robots, with plans for Atlas to appear in Hyundai’s automotive factories by 2028. The German KION Group and South Korean LG are also deploying the stack in autonomous forklifts and a new humanoid built on Nvidia’s Isaac GROOT model.