Skip to main content

02microbit_CreateAI

Overview​

micro:bit CreateAI is a free, web-based tool that makes it easy for students to explore AI through movement and machine learning (ML). You can use micro:bit CreateAI to train an ML model and then run it on your BBC micro:bit V2.

Machine learning (ML) is a type of AI where computers can learn from and make decisions based on data. It's widely used to power many kinds of technology, including smart fitness trackers.

image-20250815105929398

Setup​

  • A computer with Chrome or Edge browser
  • A micro:bit V2
  • USB data cable
  • Batteries and battery pack

STEP​

Connect your micro:bit​

To collect data to train your ML model, you need to connect a micro:bit to your computer. We call this micro:bit the data collection micro:bit. Because you'll be collecting movement data, you will connect using a wireless connection. There are two ways to do this, Bluetooth and radio link.

Open micro:bit CreateAI (https://createai.microbit.org/) and click the "Get started".

image-20250815110035517

In the new interface, click "New Session". If you have ongoing session or saved session, you can open them using the two options above.

image-20250815110110527

If this is a new session, the interface will display "No data samples". Click "Connect" at the bottom of the screen.

image-20250815110209117

Follow the prompts to establish a wired connection between your micro:bit and computer.

image-20250815110341492

image-20250815110424134

image-20250815110503822

After completing the wired connection, insert the battery pack (pre-assembled), turn on the power, then disconnect from the computer.

image-20250815110539652

image-20250815110603307

image-20250815110638943

Record data samples​

The data collection micro:bit uses its accelerometer to measure changes in movement in three dimensions: x, y and z. It sends the data wirelessly to micro:bit CreateAI.

image-20250815110804790

micro:bit CreateAI shows this data in a live graph that responds in real time to the movements of your micro:bit.

image-20250815110905225

Step 1: Mount your micro:bit

Mount the micro:bit onto the 3D-printed wand

Step 2: Design your action

Wave the wand and design simple action, ensuring distinct differences between each motion for reliable recognition.

Step 3: Name your action

Name your action and select icons.

image-20250815110953713

Step 4: Record data samples

Press the 'Record' button to record data samples of your movement. Each sample lasts 1 second.

image-20250815111036493

Step 5: Record more data samples

You need at least 3 data samples for each action to train the model, but ML models usually work better with more data.

image-20250815111058999

Step 6: Check your data quality

Check your data quality. Use a process of data cleaning and remove any data you think might be bad (data outliers) and record more samples if you need to.

image-20250815111119080

Step 7: Repeat for more actions

Use the 'Add action' button to create more actions. Collect data for each one. You need at least 2 actions to train the ML model. When you have enough actions and data, you are ready to train your ML model.

image-20250815111150897

Tips:

  • It is best to record a static-state sample—specifically, a sample of holding the wand without waving it.
  • Keep the micro:bit in a fixed orientation.
  • For samples of the same action, record as many as possible.
  • Review samples of each action, observe the curve of every sample. If any sample differs significantly from others, it should be treated as erroneous data and re-recording is recommended.

Train and test your ML model​

When you have enough actions and data, the 'Train model' button will become active. Press the 'Train model' button and you're ready to test your model.

image-20250815111246458image-20250815111311098

After entering the testing Model interface, focus primarily on the following three sections:

Certainty

The certainty shows how certain or confident the ML model is that you are doing that movement. The higher the percentage, the more confident it is.

Recognition point

The confidence threshold for CreateAI to decide you are doing a particular action. You can adjust the recognition point to see how it changes the way the model recognises your movements.

Estimated action

The estimated action is what the model recognises as the current movement you are making.

image-20250815111404836

image-20250815111446386

Example​

In this example, 5 gestures were selected for demonstration: Up, Down, Left, Right, and Lightning.

image-20250815111513444