Jetson Nano Image

tl;dr; Build sd-card flashable images for Jetson Nano dev kits using Docker and Github Actions.

Problem

After buying a hobby robot and companion Jetson Nano, I wanted to update the base image for my robot and iterate on my robot software. I struggled to find a simple set of instructions and scripts that would let me create my own images and flash them to an SD card from my Mac laptop.

Nvidia privides a set of docs, scripts, and guides from their linux-for-tegra environment but I found these cumbersome to understand and modify. Some of the scripts required a linux environment to even run.

I wanted something simpler. I wanted to be able to iterate on the base image quickly, and when ready, use Github Actions to automatically build a sd-card flash ready image.

Solution

Using Docker and buildx this repository is setup to create arm64 docker images. These images can then be turned into sd-card flashable .img files using the create-image.sh script. This script use nvidia l4t scripts to configure the rootfs with the correct boot files.

Finally, all of this is automatically run with Github actions. After pushing a change to the repo, actions run and produce artifacts with .img files that I can flash to an sd-card.

Here's a screenshot of the artifact ready to download.

Once downloaded I can flash the image to the sd card.

Now I'm ready to boot the nano.

Customizing

You can make your own images by forking this repo and modifying the Dockerfile. Your fork will automatically run the forked Github Actions and you'll end up with ready-to-flash images from your changes.

Credentials

The default credentials:

username: jetson
password: jetson

Local Development

One advantage of using Docker to setup the root file system is the ability to iterate locally and test your changes.

Here are a few commands you can use to work locally and make sure everything installs before you push your changes to CI.

Build the rootfs image

docker buildx build --platform linux/arm64 -t jetson-nano-image .

Run the image (without any init system)

docker run -it --rm --user 1000:1000 jetson-nano-image /bin/bash

Run the built image and invoke systemd init to see what runs on startup

docker run -it --rm --cap-add SYS_ADMIN -v /sys/fs/cgroup/:/sys/fs/cgroup:ro jetson-nano-image /sbin/init

Make a flashable image

If you are on linux, you can turn the Docker image into a flashable image

# Export the rootfs image to a folder on your file-system
# Nvidia l4t tools turn this folder into a .img file you can flash
docker export $(docker create --name nano-rootfs --platform linux/arm64 jetson-nano-image) -o rootfs.tar

mkdir -p /tmp/jetson-builder/rootfs
sudo tar --same-owner -xf rootfs.tar -C /tmp/jetson-builder/rootfs

# Create a jetson.img from the `rootfs` you can flash to an SD card
sudo -E ./create-image.sh

Supported boards:

• Jetson nano
• Jetson nano 2GB

References

This work builds upon the learnings from this great post by pythops:

• https://pythops.com/post/create-your-own-image-for-jetson-nano-board.html
• https://github.com/pythops/jetson-nano-image

Additional links

• https://developer.nvidia.com/embedded/linux-tegra
• https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%20Linux%20Driver%20Package%20Development%20Guide/updating_jetson_and_host.html
• https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%20Linux%20Driver%20Package%20Development%20Guide/flashing.html#wwpID0E0CM0HA

License

MIT