Mabni
AI LITERACYmicro:bit AI Maker
Trained on this computer

Teach it. Test it. Make it move.

Your photos become a model. Your model gives a robot its next move.

1

Teach with examples

Give every group different examples. Keep your test photos separate.

2

Train & try

Loading the local image model…
A new practice cardAn unseen practice card
Train your model, then try a new photo.

Photos, training and predictions stay here. The model does not use the picture’s filename.

3

Bring it to life

Waiting for your model

The simulator works without a board. Forward spins the wheels; Turn steers; Stop holds still.

USB is optional. Chrome or Edge on the main PC.

Download AI Bridge HEX →Edit the bridge in MakeCode →
Use real motors or a servo

An adult must check the external motor driver, separate battery supply and common ground. GPIO pins are signals, not motor power.

Wiring & motor blocks

Add an example

Start the camera in the test panel first. Nothing is sent to a server.

Your own offline AI workshop

  1. Collect: use the practice cards, upload your own photos, or capture camera frames. Teach at least two examples in every group. Different backgrounds and viewpoints make a better lesson.
  2. Train: a local MobileNet image model extracts 256 features. Your labels train a nearest-neighbour classifier. Test on new pictures, rather than memorised training examples.
  3. Improve: compare the closest examples. A high vote is not a guarantee. “Not sure” stops the robot. Change examples, train again, then test.
  4. Connect: download Mabni AI Bridge HEX, copy it to MICROBIT, then connect with USB serial. Image inference runs on the computer; the micro:bit receives short commands. Internet is never required.

Practice cards are original Mabni artwork. This is an independent Teachable Machine style learning activity, not Google Teachable Machine or the online micro:bit CreateAI service. It does not import their models.

Camera and USB serial require a secure browser context. Use http://localhost:8000 on the main PC, or a properly secured classroom server. Students using a plain LAN link can train from local images and use the robot simulator.

Models and uploaded pictures are saved in this browser’s local database, separately for each Mabni profile. Use Save model for a portable copy. Training changes stop outputs until you train again.

Local model provenance · Hardware setup