NVIDIA Unveils Halos for Robotics, a Full-Stack Safety System for Physical AI

NVIDIA has announced Halos for Robotics, touted as the industry's first full-stack safety system designed for physical AI and autonomous robots, marking a major step toward the safe integration of machines that operate alongside humans in industrial environments. The system was unveiled at the Automate 2026 conference in Chicago on June 22, 2026, and leverages over 18,600 engineering years of NVIDIA's autonomous vehicle safety expertise.

Halos for Robotics provides a unified safety architecture that spans four key layers: platform safety (hardware), safety operating system, algorithm safety, and ecosystem safety. At the hardware level, NVIDIA's IGX Thor computing platform features a dedicated "safety island" physically isolated from the main AI compute to ensure emergency functions can operate even if the primary system fails. The Holoscan Sensor Bridge aggregates data from diverse sensors with low-latency synchronization, achieving SIL 2 safety assurance. The Halos OS software stack, which can run on a Linux-only or Linux-plus-QNX hybrid architecture, isolates safety-critical tasks from AI workloads to prevent software faults from affecting safety functions.

A standout feature is the Outside-In Safety Blueprint, which uses external cameras and independent AI agents to monitor robot behavior from a third-person perspective. This allows robots to operate at higher efficiency when the environment is safe and immediately intervene when humans enter danger zones. NVIDIA has open-sourced the core safety framework, inviting the broader robotics community to adopt and contribute to the standard.

Agility Robotics is the first company to integrate Halos into its Digit humanoid robots, which are already deployed in factories and warehouses for customers including Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada. Beyond Agility, NVIDIA has expanded its Halos ecosystem to include more than 43 partners, such as Boston Dynamics, lidar manufacturer Hesai Technology, and safety robotics company FORT Robotics, indicating broad industry traction.

To support certification, NVIDIA has established the Halos AI Systems Inspection Lab, which received ANAB accreditation under ISO/IEC 17020—a first in the physical AI field. This lab helps partners prepare for third-party certification by leading bodies such as TÜV Rheinland, UL Solutions, and TÜV SÜD. The system is designed to streamline the previously fragmented safety certification process for sensors, controllers, and AI systems into a single, coherent framework.

NVIDIA's vice president of robotics and edge AI, Deepu Talla, emphasized the need for a unified safety architecture to scale autonomous systems. Agility Robotics CEO Peggy Johnson noted that safety is a prerequisite for humanoid robots to enter industrial workflows. With Halos, NVIDIA now completes its full-stack robotics portfolio—covering simulation (Isaac Sim), foundation models (GR00T), world models (Cosmos), and edge inference (Jetson Thor)—effectively defining a standard for how robots should be built and certified.