Replicate Earthrover Robot Software on your Raspberry Pi

This article turns a Raspberry Pi into a robot laboratory. One installer sets up a web server, the camera stack, OpenCV, the TensorFlow Lite runtime and Coral USB Accelerator support, then fetches the Earthrover source code and the machine learning models and puts them where the web server can reach them. When it finishes, you open a page in a browser and drive the robot.

Updated for the current Raspberry Pi OS. The earlier version of this article installed a script written for Raspberry Pi OS Buster. That script no longer works on current images — the camera stack, the GPIO tools and the TensorFlow runtime have all been replaced since it was written. Everything below has been tested on Raspberry Pi OS 13 (Trixie) on a Raspberry Pi 3A+ and a Raspberry Pi 4 Model B, installed from a fresh card.

The code is on GitHub as robotics-level-4. That repository is the authoritative reference for what the installer does and how the software is laid out; this article is the walkthrough.

Step 1: Prepare a micro SD card with Raspberry Pi OS

Use Raspberry Pi Imager to write Raspberry Pi OS (64-bit) to the card. Either current release works — OS 12 (Bookworm) or OS 13 (Trixie). The installer checks the version and stops on anything older, because the packages it needs do not exist there.

Before Imager writes the card, open its customisation settings and fill in your Wi-Fi network, a hostname, and enable SSH. This matters more than it used to: the old trick of dropping a wpa_supplicant.conf file onto the boot partition stopped working when Raspberry Pi OS moved to NetworkManager, and Imager is now the way to set up a board you intend to run without a monitor and keyboard.

Step 2: Download the installer

Boot the Pi, connect it to your network, and open a terminal. Fetch the script:

curl -fsSL https://raw.githubusercontent.com/jiteshsaini/robotics-level-4/main/earthrover/setup_level4.sh -o setup_level4.sh

Downloading the file first, rather than piping it straight into a shell, means you can read it before you run it. It is worth a minute of your time — it is your Pi, and this script installs a lot on it.

Step 3: Run the installer

bash setup_level4.sh

Not with sudo. The script calls sudo itself where it needs to, and it will stop and tell you so if you try. Run it as the ordinary user you logged in as.

Allow an hour or more for a first run. The script brings the whole OS up to date before it installs the camera and vision packages, so that they are not built against an older kernel. Nothing reboots by itself; if a restart is needed, the script says so at the end.

You do not need to enable the camera separately. The installer writes camera_auto_detect=1 to the boot configuration for you — the old raspi-config step from previous versions of this article is gone.

Step 4: Open the control panel

When the script finishes it prints the address to open. From any laptop or phone on the same network, in a browser:

http://<your-pi-ip>/earthrover/

One page drives everything. The five machine learning projects run along the top, the live camera feed fills the middle, and the direction pad and speed sit on the right, with the distance reading, the lights and the speaker around the edges. Reload the page and the controls come back as you left them — they report what the robot is actually doing, rather than what was last clicked.

The green badge beside AI ROBOTICS is worth pointing out. It says which backend the models will use: Coral USB Accelerator when one is plugged in, CPU when it is not. The accelerator is optional — the models run either way, and it is detected at runtime, so there is nothing to configure and no separate build to install. With one plugged in the models run roughly ten times faster. The badge asks the same code the robot does, so the panel and the rover cannot disagree about it.

What the installer does

In order:

• Brings the OS fully up to date
• Installs Apache, PHP, the GPIO library, the camera stack, and the speech and audio tools
• Installs the TensorFlow Lite runtime (ai-edge-litert) and OpenCV
• Fetches the code and the machine learning models into /var/www/html
• Enables the camera interface
• Generates a self-signed certificate and switches on https, which the phone-sensor pages require
• Installs Coral USB Accelerator support — harmless if you do not have one
• Gives the web server the group memberships and file ownership it needs

An existing installation is moved aside rather than overwritten, so running the script again on a Pi that already has it is safe.

Two of those deserve a note, because they are the things that changed since the original article. The GPIO library is rpi-lgpio rather than the older RPi.GPIO, which does not work on a Raspberry Pi 5. And raspi-gpio, the command line tool a lot of older tutorials still reach for, was removed from Raspberry Pi OS — pinctrl replaces it.

If your board has 512 MB of memory

On a Raspberry Pi 3A+, the desktop is the difference between a detector that runs and one that lives in swap — about 0.1 frames per second against 10. If that is your board, turn the desktop off:

sudo systemctl set-default multi-user.target
sudo systemctl disable lightdm
sudo systemctl disable --now wayvnc
sudo reboot

The installer points this out when it finds such a board, but does not do it for you, because it also closes VNC. On a Pi 4 or 5 there is no need.

When something does not work

The script can check its own installation without changing anything, and without asking for a password:

bash setup_level4.sh --verify

That is the first thing to reach for. Two other options exist: --fix-perms re-applies web-root ownership without reinstalling, and --help prints the notes at the top of the script.

If a machine learning feature misbehaves once the panel is up, the worker logs are in the logs/ folder inside the installed code.

Starting smaller

This installs the complete robot. If you would rather build up to it, the earlier levels are separate repositories that each add one layer: level 1 for motors and a web control panel, level 2 adds the camera, level 3 adds the phone sensors and voice control, and level 4 — this one — adds machine learning on the robot itself.

All you need is a Raspberry Pi and a camera to get started. Other hardware can be added subsequently.

While building this robot, i realised that it is not enough to just share the source code. The important part is the environment in which this source code runs. Setting up the required environment can be a challenging task if you are just starting. So, i created this bash script, which sequentially installs everything that is required to run the software of Earthrover on your Raspberry Pi. Follow the above steps as described and you can replicate my work on your device.

All the best.