NVIDIA Halos is an umbrella functional safety platform for physical AI originally announced at CES 2025 for autonomous vehicles and extended on June 22, 2026 to robotics. The platform transfers NVIDIA's autonomous vehicle heritage to robotics โ over 18,600 engineering years of safety work โ and covers a full hardware, software and process stack.
The Halos stack comprises five components: (1) Halos OS โ a safety-rated operating system in Linux and Linux+QNX configurations, certified to IEC 61508 SIL 3, (2) Halos Core SDK in early access for robotics application developers, (3) a hardware Functional Safety Island (FSI) integrated with the IGX Thor module, (4) the Outside-In Safety Blueprint released as open source and (5) the NVIDIA Halos AI Systems Inspection Lab accredited by ANAB under ISO/IEC 17020.
At launch, the Halos ecosystem spans over 40 companies โ including Infineon, Texas Instruments, TรV Rheinland and UL Solutions โ providing hardware components, software and certification services compatible with the platform. The first reference deployment in the humanoid robot segment is the fifth-generation Digit humanoid from Agility Robotics, on a path to fenceless cooperative safety operation at Amazon, GXO, Schaeffler and Toyota facilities by late 2026.
Physical AI platforms are software products that connect physical robots and industrial production processes with artificial intelligence models. Unlike a pure SDK or middleware, a Physical AI Platform ships as an end-to-end stack: perception, task planning, process parameter optimization, multi-robot coordination and operator UI. They target use cases such as surface treatment (grinding, painting, polishing), inspection, assembly and packing โ where business value comes from combining hardware, AI and domain knowledge.
An SDK (Software Development Kit) is a curated set of libraries, interfaces, tools, sample code, and documentation intended for building applications and integrating with a specific hardware device, platform, or service. In robotics, an SDK typically exposes device control, telemetry, sensor access, configuration, and execution functions, significantly reducing the time-to-first-integration for developers targeting a specific robot or platform.
A Runtime is the environment or execution layer used to run code, load libraries, manage dependencies, and operate applications or services โ either in real time or during normal system operation. In robotics this includes real-time operating system (RTOS) runtimes, ROS 2 executor runtimes, containerised execution environments (Docker, podman), and embedded C++ runtimes on microcontrollers.
The functional safety role describes software responsible for ensuring that a robot operating alongside humans does not cause harm even in the event of hardware failure, software error or unforeseen environmental event. The safety component implements: deterministic execution of critical tasks (hard timing determinism), fault containment, continuous system integrity monitoring, transitions to safe failure states, and compliance with functional safety standards โ IEC 61508 (SIL), ISO 13849 (Performance Level), ISO/IEC TR 5469 (AI in safety systems). It typically runs on an isolated hardware subsystem (e.g. an FSI island) independent of the main compute stack.
The real-time control role describes software responsible for closed control loops executed at a guaranteed, deterministic frequency โ typically 1 to 10 kHz for current/torque loops in actuators. Unlike soft real-time, in a hard real-time regime missing a deadline is treated as a system error. Implemented on deterministic kernels (Linux PREEMPT_RT, QNX Neutrino RTOS) or dedicated microcontrollers. Critical for biped balance stability, force control, impedance control and functional safety.
The benchmarking and evaluation role describes software responsible for standardized measurement of robot and AI model capabilities. The benchmark component includes: a set of defined tasks (manipulation, locomotion, perception, instruction), automated success and execution quality metrics, test scenarios covering generalization dimensions (lighting, background, camera noise, control delay), reproducible processes for running multiple trials with result aggregation, model leaderboards (e.g. ฯ0.5, GR00T, ฯ0). Modern systems use VLM (Vision-Language Models) for auto-evaluation of complex qualitative criteria inaccessible to simple numerical metrics.
The NVIDIA Halos family comprises a full functional safety stack for physical AI: the Halos OS operating system, Halos Core SDK, the hardware FSI on IGX Thor and an accredited inspection laboratory.
Agility Robotics โ fifth-generation Digit humanoid (cooperative safety deployment targeted for late 2026, eventually fenceless operation at Amazon, GXO, Schaeffler, Toyota facilities).
The Halos ecosystem spans over 40 partner companies (Infineon, Texas Instruments, TรV Rheinland, UL Solutions). Early access for registered developers; AI Systems Inspection Lab accredited by ANAB (ISO/IEC 17020).
QNX Neutrino โ the BlackBerry QNX microkernel RTOS with a POSIX API. Dominant in automotive ADAS / infotainment (235M+ vehicles) and safety-critical robotics.
The full Halos stack requires the hardware FSI available only on the NVIDIA IGX Thor module. Individual stack components (e.g. the Outside-In Safety Blueprint) can run more broadly.
License family: Proprietary โ Commercial
License family: Permissive
