Skip to main content

Object Detection Tutorial

Create a custom YOLO-Pro model for your FIRST robot in 10 steps.

Prerequisites​

  • ✅ Access to FIRST Robotics Edge Impulse
  • ✅ 50-100+ images of game pieces from your robot's camera
  • ✅ Camera mounted in competition position

Step 1: Create Your Project​

  1. Go to firstrobotics.edgeimpulse.com
  2. Create an account on the login page or sign in if you already have an account
  3. Click Create new project
  4. Name it (e.g., "2025-Game-Piece-Detection")
  5. Click Create project

Step 2: Create your AI Model​

Follow steps 2-4 from the Edge Impulse Object Detection Docs with these important modifications:

During Step 3 - Designing an Impulse:

  • Use your camera mounted on your robot
  • Set width and height to 640x640 (required for PhotonVision)

During Step 4 - Configuring the Transfer Learning Model:

  • Change AI Model to YOLO-Pro
  • Change Training Processor to GPU
  • Your configuration should look like:

model_config

  • Change the model size to tradeoff performance with accuracy
  • Nano is our recommendation for most teams

Step 3: Download Model​

  1. Go to the Deployment tab
  2. Select C++ Library
  3. Select Tensorflow
  4. Select Quantized Int8
  5. Click Build

Your configuration should look like this:

model_config

Step 4: Deploy to Robot​

note

PhotonVision to update how they want the deployment to work. We'll update this section when we know more.

Step 5: Iterate & Improve​

Machine learning is iterative. Your first model is just the beginning!

Next Steps​

🎉 Congratulations! You've created your first custom object detection model.

Continue learning:

  1. Read Best Practices for advanced tips
  2. Talk with others on Chief Delphi!
Remember

The best models come from continuous iteration. Each competition provides valuable data for improvement. Good luck! 🤖