> This page is for version 1.4.0 (Latest) (default).
> For other versions, use one of these documentation indexes:
> - 1.4.0 (Latest) (default): https://docs.nvidia.com/halos/outside-in-safety/1.4.0/llms.txt

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/halos/outside-in-safety/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/halos/outside-in-safety/_mcp/server.

# NVIDIA Halos Outside-In Safety Blueprint

> Outside-in safety agents that extend robot perception with infrastructure cameras, AI perception, and safety logic.

NVIDIA Halos Outside-In Safety Blueprint extends robot perception beyond onboard sensors by using external infrastructure cameras, AI perception, and safety logic. Running on NVIDIA IGX, it helps robots operate alongside workers at higher efficiency while adapting to occlusions and complex facility layouts.

Onboard sensors have a limited field of view when walls, payloads, workers, or vehicles block them. An outside-in safety agent expands that view with sensors and cameras placed throughout the facility, so the system can detect safety events, communicate mitigations, and keep a verifiable log.

#### [Start here](/halos/outside-in-safety/getting-started)

Bring up a first working system: prerequisites, AI perception, the Safety Core, SIL, HIL, agentic deploy, or NVIDIA Brev.

#### [Skip ahead: Safety Core only](/halos/outside-in-safety/getting-started/safety-core)

Perception is already publishing events. Install and run the Safety Core on that feed.

#### [Skip ahead: closed loop in simulation](/halos/outside-in-safety/getting-started/sil)

Run perception, the Safety Core, and NVIDIA Isaac Sim on a single host with no cameras.

#### [Skip ahead: deploy on NVIDIA Brev](/halos/outside-in-safety/getting-started/brev-launchable)

Provision a cloud GPU instance and bring up the SIL loop from a notebook.

## What you can do

* Monitor safety parameters of an industrial environment in real time
* Detect potential safety events proactively
* Communicate safety events to corresponding mitigation mechanisms
* Maintain continuous safety verification
* Provide reliable logging and auditing

The reference use case is Automated Trailer Loading: fixed cameras and AI perception monitor workers and autonomous forklifts at a warehouse dock, and the Safety Core decides when it is safe for the forklift to enter the trailer.

## Key features

1. **Modular architecture** — Integrate the framework into the target environment and adapt it to a variety of sensor processing pipelines.
2. **Safety-centric design** — Designed adhering to safety standards.
3. **Real-time capabilities** — Low-latency communication and priority-based handling for critical safety events.
4. **Scalable implementation** — From small embedded systems to large distributed systems, with incremental feature rollout.

## Documentation

#### [Getting Started](/halos/outside-in-safety/getting-started)

First working system, from the AI perception pipeline to the Safety Core.

#### [Deployment Guide](/halos/outside-in-safety/deployment)

Run the Safety Core with Docker, from prerequisites to troubleshooting.

#### [Integration Guide](/halos/outside-in-safety/integration)

Architecture, running without Docker, and building decision makers and event sources.

#### [Safety Core Components](/halos/outside-in-safety/integration/components)

Each component's role, configuration, failure behaviour and architecture.

#### [Closed-Loop Testing](/halos/outside-in-safety/testing)

SIL, HIL, and the Regression Testing Reporter.

#### [Reference](/halos/outside-in-safety/reference)

Launcher options, configuration keys, runtime profiles, and terminology.

The C and C++ headers that integrators build against, and the generated C API reference, are described in [API Documentation](/halos/outside-in-safety/api-documentation).