Rock Paper Scissors Game with Arduino UNO Q

A real-time Rock-Paper-Scissors game running on the Arduino UNO Q using an Edge Impulse object detection model.

The camera detects your hand gesture (rock, paper, or scissors) via machine learning inference, while the Arduino picks a random move. Are you going to win the Arduino UNO Q?

Deployment

Prerequisites

• Arduino UNO Q with Arduino App Lab
• USB camera connected to the board
• Edge Impulse machine learning model trained to detect rock, paper, and scissors as you can find in this public project here. Clone it and re-train it to improve the accuracy with your light and background.

Step 1: Transfer the app

Clone this repository to your local machine.

Copy the entire Rock Paper Scissors folder to the Arduino UNO Q board:

scp -r Rock-Paper-Scissors-Arduino-UNO-Q/ arduino@<device-ip>:/home/arduino/ArduinoApps/RPS-game

or use the Arduino App Lab Create new App button in the My Apps section and import the application.

Step 2: Deploy the model

Get into the Rock Paper Scissor app into the Arduino App Lab.

Click in the Brick Video Object Detection and then click Train new AI model in the bottom.

Log In into your Arduino account and the Edge Impulse account and then train your own Rock Paper Scissors model or clone this public project and re-train it.

Go to deploy the model as Arduino UNO Q or as Linux aarch64.

Then the deployed models will appear in the brick of the Arduino App Lab when you will go to the AI models tab. Select the Rock paper scissors model.

And check that it's being added in the app.yaml file of the app.

Step 3: Start the app

Launch the Arduino App Lab in your local machine and get into your Arduino UNO Q.

Go to My Apps and you may see the Rock Paper Scissors application there. Click on it and then click Run.

Alternatively, via SSH you can start the application using the Arduino App Lab CLI.

arduino-app-cli app start user:rock-paper-scissors-game

Once successfully started, navigate to http://<device-ip>:5001 in your browser and start playing!

Good luck!

Game flow

1. Show your hand gesture (rock, paper, or scissors) to the camera.
2. The detection panel on the left shows what the model sees in real-time after running inference on a local object detection Edge Impulse model.
3. Click Play Round — your gesture is locked in at that moment.
4. The Arduino reveals its random move and the winner is shown

Configuration

All settings are in python/main.py at the top:

Setting	Default	Description
CONFIDENCE_THRESHOLD	0.6	Minimum confidence to accept a detection
PORT	5001	Flask web server port (also set FLASK_PORT env var)
COUNTDOWN_SECS	3	Countdown duration before evaluating
RESULT_HOLD_SECS	3	How long the result stays on screen

Improving the model

In case that you want to create your own object detection model using Edge Impulse.

Collect data, label it and train the neural network. Test it in Edge Impulse and when you will feel confident, deploy it as an Arduino UNO Q model or Linux aarch64.

Then follow the same instructions that you performed to add it to the app's brick.

Become an Edge Impulse expert

Want to learn more about how Edge Impulse ork? Try one of the Edge Impulse courses.

Troubleshooting

"No gesture detected" every round:

• Check that the brick is initialized: look for [BRICK] VideoObjectDetection initialized in logs
• Check that App.run() is active: look for [MODE] App runner: yes in logs
• Look for [BRICK-RAW] lines — if absent, the brick callback isn't firing
• Ensure your model labels match rock, paper, scissors (lowercase)

"App runner: no" in logs:

• The App class couldn't be imported. Make sure you're running via arduino-app-cli app start, not python3 main.py directly

Model not found:

• Verify the .eim file exists at the path in app.yaml
• Ensure the file is executable: chmod +x /home/arduino/.arduino-bricks/ei-models/rcp-model.eim

Feel free to reach out to us on the Edge Impulse forum or the Edge Impulse Discord server if you need help.

Disclaimer

This project is intended for educational and experimental purposes only. It is not hardened for production use. Do not deploy in any safety-critical environments without proper security, testing, and validation.