NVIDIA Unveils Halos for Robotics, the Industry's First Full-Stack Physical AI Safety System
NVIDIA has officially unveiled NVIDIA Halos for Robotics, the industry’s first full-stack safety system that brings AI computing and safety capabilities together under a single, standardized architecture. Designed for the development, validation, and industrial deployment of robots and physical AI, Halos marks a pivotal step toward deploying autonomous machines at scale in human-centric environments.
Humanoid robotics and physical AI company Agility Robotics will be the first to adopt NVIDIA Halos for Robotics, integrating comprehensive safety capabilities into its industrial humanoid robots. Agility’s machines are slated for deployment in factories, warehouses, and logistics settings — environments that demand robust, certifiable safety infrastructure.
Next-generation autonomous robots must navigate ever-changing environments alongside humans, relying on AI foundation models, accelerated computing, and distributed sensor networks to perform their tasks. Moving from pilot programs to large-scale deployment requires more than isolated safety features; it demands a unified architecture spanning hardware, software, sensors, applications, and certification processes. NVIDIA Halos delivers exactly that.
A Three-Pillar Architecture
The Halos system is built around three core components:
Hardware: IGX Thor and Holoscan Sensor Bridge — The hardware foundation provides industrial-grade AI compute for real-time robotic and safety workloads, built-in safety mechanisms, and seamless sensor connectivity.
Software: Halos OS — The software layer consists of Halos Core, which handles safety-critical runtime functions, and the External Perception Safety Blueprint, an open-source framework that leverages external cameras and AI agents to complement the robot’s onboard perception. The blueprint dynamically adjusts robot behavior in response to changing conditions on the factory floor.
Validation: Halos AI System Validation Lab — This is the world’s first program accredited by the ANSI National Accreditation Board (ANAB) to cover both physical AI functional safety and AI safety. The lab helps partners prepare for third-party certification from leading bodies including TÜV Rheinland, UL Solutions, TÜV SÜD, exida, SGS, and CertX.
Early Access Now Available
NVIDIA Halos Core for IGX is already available to registered developers with early access, supporting both Linux and Linux + QNX OS for Safety 8.0 configurations. The open-source External Perception Safety Blueprint, part of the Halos OS application layer, is also available for early access on GitHub.
With Halos for Robotics, NVIDIA is laying the groundwork for a future where autonomous robots can be deployed safely and reliably at industrial scale — not just in controlled labs, but everywhere people work.