Name: NVIDIA Halos (for Robotics)
Brand: NVIDIA

Description

NVIDIA Halos for Robotics is the first comprehensive functional safety platform for physical AI, announced on June 22, 2026. Halos transfers NVIDIA's autonomous vehicle heritage to robotics: over 18,600 engineering years of safety work and more than 7 million lines of certifiable code. The first deployment is the fifth-generation Digit humanoid from Agility Robotics, targeted for late 2026.

Three-layer architecture

Hardware layer — NVIDIA IGX Thor module with isolated Functional Safety Island (FSI) and Holoscan Sensor Bridge (HSB) connecting sensors to the compute platform with low latency over 10 GbE.

Software layer — Halos OS, a safety-rated operating system available in Linux and Linux+QNX configurations. Halos Core SDK available in early access for registered developers.

Certification layer — NVIDIA Halos AI Systems Inspection Lab accredited by ANAB (ISO/IEC 17020) as an inspection body. Humanoid manufacturers verify in the lab that their software, AI components and cybersecurity protections meet IEC 61508, ISO 13849 and ISO/IEC TR 5469 standards.

Outside-In Safety Blueprint

The open-source blueprint available on GitHub inverts the classic Inside-Out logic: a traditional robot perceives the world only through its own sensors and slows down upon detecting an obstacle. Outside-In redirects data from external facility cameras into a four-stage pipeline: SIPP (Sensor Input Processing Pipeline) converts video streams to tracking events, SAIM (Safety AI Monitor) detects environmental degradation and anomalies, SEI (Safety Event Integrator) validates events, SDM (Safety Decision Maker) makes decisions. In a warehouse scenario, when the external system confirms no humans are present, the SDM can temporarily lift the robot's speed constraints.

Partner ecosystem

Over 40 companies participate in the Halos ecosystem — from semiconductor suppliers (Infineon, Texas Instruments) to certification bodies (TÜV Rheinland, UL Solutions). TÜV Rheinland is inspecting the IGX Thor module and Halos OS for functional safety certification readiness — the results will be the first real-world test of the system's credibility.

First customer: Agility Robotics

Agility Robotics has so far deployed Digit exclusively behind physical barriers (workcells). The fifth-generation Digit targeted for Q4 2026 will achieve cooperative safety — operating alongside humans without fencing. The key is the IGX Thor module running Halos Core executing certified software. According to CEO Peggy Johnson, this opens the path to deployments at Amazon, GXO, Schaeffler and Toyota facilities.

MLOps Lifecycle

7/11 supported

Model Registry

Versioning — model artifact versioning

Approval workflows — approval workflow before production

Immutable artifacts — immutability of stored versions

Lineage tracking — tracking data and model relationships

4 / 4 supported · none unsupported

Monitoring

Data drift detection — input data drift detection

Concept drift detection — concept drift detection

Hallucination monitoring — LLM hallucination monitoring

Bias evaluation tools — bias evaluation tooling

2 / 4 supported · 2 unsupported hidden

Human-in-the-Loop

Labeling services — data labeling tools

RLHF workflows — reinforcement learning from human feedback

Manual override — manual override of model decisions

1 / 3 supported · 2 unsupported hidden

Applications

3

Architecture & Mechanisms

3

Security

Developer Ecosystem

SDK Languages

C+C / C++PyPython

API Type

RESTgRPC

Community & resources

Templates library

Quickstarts

API Reference

Tutorials

Pricing & Business Model

See full pricing

Pricing models

Tiered subscription

Resource quotas

Per project

Per user

Cost alerting

SLA & Support

Enterprise 24/7Standard

Robotics & Humanoids Extension

Robotics-Ready

Communication protocols

2

ROS 2 DDS

Robotics middlewareOpen Robotics / ROS 2 ecosystem

docs.ros.org

Data Distribution Service (DDS)

Real-time communicationObject Management Group (OMG)

omg.org

Robotics standards

URDF Support
OpenUSD Interoperability
Sim-to-Real Pipelines

Edge Orchestration

OTA updates (over-the-air)
Real-time kernel support

Description

Three-layer architecture

Outside-In Safety Blueprint

Partner ecosystem

First customer: Agility Robotics

MLOps LifecycleiMLOps LifecycleFull model lifecycle: registry, feature store, prompt management, monitoring and human-in-the-loop.

Model Registry

Monitoring

Human-in-the-Loop

ApplicationsiAI ApplicationsDomains and use cases this platform is best suited for — from RAG and fine-tuning to scientific research.

Architecture & MechanismsiArchitecture & MechanismsArchitectural foundations and modern AI processing methods that are natively supported or used by this platform.

SecurityiEnterprise SecurityCertifications, access controls and data-protection features essential for corporate deployments and cloud privacy compliance.

Developer EcosystemiDeveloper EcosystemDeveloper resources: available SDKs, supported programming languages, and infrastructure features and model-deployment methods.

Pricing & Business ModeliPricing & Business ModelBilling models (usage-based, provisioned throughput), resource limits and SLA parameters (uptime, support tiers).

Robotics & Humanoids ExtensioniRobotics & Humanoids ExtensionSimulation engines (Isaac Sim, Gazebo, MuJoCo), communication protocols (ROS2, MQTT, Zenoh), robotics standards (URDF, OpenUSD) and edge orchestration.

SourcesiDocumentation VaultCentralized hub of links to official sources, technical guides, repositories and release notes.