Skip to content

Supported Platforms

AirStack targets one compute platform per Docker Compose profile: each service in robot/docker/docker-compose.yaml extends the shared robot_base and pins the base image and platform toggles for its target. This page is the honest status matrix — what CI actually tests, what has been used in the field, and what merely exists as a profile. Deployment topologies (which autonomy stack runs where) are a separate axis, covered in Autonomy Modes.

Platform Matrix

Platform Compose profile → service(s) Base image (from compose build.args) Status
x86-64 desktop/laptop (development + simulation) desktop → robot-desktop (+ gcs) nvidia/cuda:13.0.2-base-ubuntu24.04 CI-tested (sim): system-tests.yml runs build, liveliness, sensor, and flight marks against this profile on ephemeral x86 GPU runners; test_build_docker.py builds the robot-desktop, gcs, isaac-sim, and ms-airsim images
x86-64 desktop, split-topology debugging desktop_split → robot-desktop-onboard + robot-offboard (+ gcs) Same image as robot-desktop Profile exists; shares the CI-tested desktop image, but the split topology itself is not CI-exercised
x86-64 ground station (field, offboard half) offboard → robot-offboard + gcs-real Same image as robot-desktop; GCS from osrf/ros:jazzy-desktop-full Profile exists for field use paired with l4t_lite/voxl vehicles; not CI-exercised
NVIDIA Jetson Orin AGX/NX (full stack onboard) l4t → robot-l4t (+ robot-l4t-stack-base, zed-l4t) dustynv/ros:jazzy-ros-base-r36.4.0-cu128-24.04 via the intermediate Dockerfile.l4t-stack-base image Supported, field-used: the installation guide documents tested install and run on Jetson Orin AGX/NX with Ubuntu 22.04 (L4T / JetPack); not covered by CI
NVIDIA Jetson Orin, lite (global planning offloaded) l4t_lite → robot-l4t-onboard Same image as robot-l4t Same support level as l4t; defaults to the lite_default stack
ModalAI VOXL 2 voxl (alias voxl_onboard) → robot-voxl-onboard ubuntu:24.04 (aarch64, REAL_ROBOT=true, no CUDA) Profile exists; docs in progress (per About); defaults to the lite_default stack (compute-constrained); do NOT assume test coverage

Companion services on the same profiles: zed-l4t (profile l4t, base dustynv/ros:jazzy-desktop-r36.4.0-cu128-24.04) runs only the ZED stereo camera driver next to robot-l4t; simple-robot (profile simple) and robot-test (profile test) are desktop-image variants for the lightweight simulator and colcon-test runs, not separate platforms.

OS / JetPack Requirements

Only requirements actually stated by a doc or Dockerfile in this repo:

Platform Stated requirement Source
Development machine Ubuntu 22.04 or 24.04, NVIDIA GPU (RTX 3070 minimum, RTX 4080+ recommended), 16GB+ RAM, ~100GB free storage About — FAQ
Jetson Orin AGX/NX Ubuntu 22.04 (L4T / JetPack); container stack pinned to L4T r36.4.0 Installation guide; r36.4.0 image tags in robot/docker/docker-compose.yaml

What CI Does and Does Not Cover

GPU simulation jobs (system-tests.yml) run on [self-hosted, airstack-ephemeral] x86 runners — every CI-verified result is the desktop x86 simulation path only. No CI job builds or runs the l4t, l4t_lite, or voxl images; their status above comes from the installation docs and profile definitions, not automated testing.

See Also