Is this page helpful?

Deployment Overview#

Halos Outside-In Safety deploys by profile on top of a VSS Blueprint perception backend. Choose the profile that matches your target, then follow its guide.

Profile	Description	Guide
`base`	Safety Core on an existing perception feed; the MUTE / UNMUTE decision is rendered as the VST `halo_safety` overlay. Runs on x86, or on IGX Thor (CCPLEX or FSI). No simulation.	Base Profile
`sil`	Full single-host closed loop: NVIDIA Isaac Sim drives a forklift and the safety decision is fed back over ROS.	SIL Quick Start
`hil` 🚧	Hardware-in-the-loop: the Safety Core runs on an NVIDIA Thor device. Under development.	N/A

Before you start#

These steps apply to every profile; each profile’s guide then adds its profile-specific configuration.

You need an NGC API key with Early Access to nvidia/halos-outside-in, and Docker 28.3.3 or later but below 29.5.0 with Compose. Docker Engine 29.5.0 and later breaks NGC image pulls, so pin to 28.3.3 if needed. Hardware requirements depend on the profile: base follows the VSS Warehouse hardware requirements, and sil adds an Isaac Sim GPU (see Prerequisites).

Log in to NGC and clone the repository:

docker login nvcr.io -u '$oauthtoken' -p "$NGC_CLI_API_KEY"
git clone https://github.com/NVIDIA/halos-outside-in-safety.git
cd halos-outside-in-safety

Deploy the VSS Warehouse 3.2 (2D Vision AI profile) perception backend first, with Kafka enabled, via the vss-deploy-profile skill or the public VSS Warehouse Blueprint. Each profile’s guide notes any profile-specific VSS settings.

Deploy with the agent (recommended)#

The hoisa-deploy-profile skill brings up the VSS Blueprint perception backend and a chosen profile from a single agent prompt. See Quick Start with Halos Deployment Skill.

For detailed Safety Core configuration, runtime parameters, and error handling, see the Safety Core Guide.