This repository contains Dockerfiles for base and application images for Jetson Nano and Jetson AGX Orin devices. Below is a quick summary, with full category pages for details and per-version instructions.
- Base images: docs/base.md
- Minimal and CUDA images: docs/minimal.md
- ROS Humble images: docs/ros.md
- PyTorch images: docs/pytorch.md
- ROS + PyTorch images: docs/ros-pytorch.md
Jump into a page to see supported tags, sizes, how to use, and links to detailed build/test guides. “jetson-” images are custom images; “l4t-” images are NVIDIA official.
Run the following command on a AMD64 computer to setup buildx to build arm64 docker containers.
docker buildx create --use --driver-opt network=host --name MultiPlatform --platform linux/arm64Run the following command on a Jetson device to setup buildx to build arm64 docker containers.
docker buildx create --use --driver=docker-container --name=container --buildkitd-flags '--debug' --bootstrap- base/ — base Dockerfiles and assets
- minimal/ — minimal and CUDA Dockerfiles
- ros/ — ROS 2 Humble Dockerfiles and entrypoint
- pytorch/ — PyTorch Dockerfiles
- ros-pytorch/ — combined ROS + PyTorch Dockerfiles
Use the orchestrator workflows to build images in sequence on native ARM runners, without juggling dependencies manually:
- r32.7.1: Actions → “Orchestrate r32.7.1 pipeline”
- r35.2.1: Actions → “Orchestrate r35.2.1 pipeline”
- r36.3.0: Actions → “Orchestrate r36.3.0 pipeline”
What they do
- Dispatch each workflow (Base, Minimal, ROS, PyTorch, ROS+PyTorch) in the right order and wait for completion between steps.
- Child workflows also define automatic triggers, but builds triggered by workflow completion are skipped to avoid duplicates. Orchestrators take priority.
Tips
- You can still run any individual workflow via its “Run workflow” button or on push/PR.
- All CI runs use GitHub’s ubuntu-24.04-arm runners (no QEMU emulation).