Today you will build a pose model in Teachable Machine that can recognise when you tilt left, tilt right, or stay straight.

You will create three classes, add your own body-position examples, train and test the model, improve it when it gets things wrong, and export a link so you can use the model later.

Before you start capturing, think: what might make it hard for the model to tell a left tilt from a right tilt?

There is no code here that says what leaning left looks like. You give the model examples of each position and it finds the pattern itself.

A pose model watches where your body is rather than the colours in the picture, so what matters is that your examples really are different from each other. Record left, right and straight properly and the model works; record three near-identical positions and it cannot tell them apart, no matter how many you add.

First we need to open the Google's Teachable Machine website to create our model.

Click on the Get Started button.

Click on the Pose Project button to create a new pose model project.

This will bring you to the screen where we can create our classes for our pose model.

In an AI model, a class is a category that the model can recognize. For example, in our space game, we will have three classes: tilt left, tilt right, and no tilt. The AI will learn to recognize images of each class and be able to tell them apart.

There will be two classes already added to your model called 'Class 1' and 'Class 2'.

Add a third class by clicking on the 'Add a class' button and then rename your classes to:

• Tilt Left
• Tilt Right
• No Tilt