Discover our
resources

This activity explores how a color sensor works with the LEGO Spike robot, and howartificial intelligence can learn to recognize and react to specific colors, via a process called supervised learning.

Students will build a simple tri-object, train an AI model, and observe how it can operate autonomously.

🧩 Activity objectives

• Understand how a color sensor works.
• Train an AI model to react to specific colors.
• Experiment with generalization: what happens if a color has not been seen during learning?
• Discover the limits and strengths of machine learning in a real-life context.

🛠️ Materials required

• 1 LEGO Spike Prime set per group of 2 to 3 students.
• Computer with AlphAI software installed.
• Objects in different colors (ideally LEGO bricks and non-LEGO objects).

🚀 Activity steps

1. Configuring the color sensor
   • In AlphAI, go to the "Sensors" tab , select "Color sensor", then choose "LEGO bricks - 9 colors".
2. Testing color recognition
   • Pass different bricks under the sensor to see which are correctly recognized.
   • Discuss the limits: some close nuances are sometimes misinterpreted.
3. Define actions associated with colors
   • In the "Actions" tab , select simple actions (sounds, emoticons, movements).
   • Assign one action for each color detected.
4. Training the model
   • Activate "Learning" mode .
   • Show the sensor a color, then click on the action to associate.
   • Repeat for each color. The training data counter increases with each example.
5. Test stand-alone mode
   • Click on the "Stand-alone" button.
   • The robot then automatically executes the action corresponding to the detected color, based on what it has learned.
6. Testing generalization
   • Try a color not used for training (e.g. orange if only red has been trained). The robot then chooses the closest action.

🧹 Data cleansing

Training data can be reviewed:

• Delete mislabeled data.
• Observe the impact of biased or too little data.

This highlights the importance of a clean, balanced dataset in any AI project.

🌈 Advanced mode: RGB detection

To go one step further, activate the "RGB " mode in the sensor tab:

• The sensor no longer identifies fixed colors, but reads Red, Green and Blue values (between 0 and 255).
• This makes it possible to detect non-LEGO objects in a wider range of colors.

Students can then train new behaviors based on the RGB reading, making sorting more realistic for everyday objects.

🧠 Summary

This activity highlights :

• The foundations ofsupervised learning.
• The importance of training data in the performance of an AI model.
• Practical applications of AI in simple but realistic scenarios (waste sorting, visual recognition, etc.).

‍