🤖 machine learning

Teaching Machine Learning Fundamentals with LEGO Robotics

This paper introduces "Machine Learning with Bricks," an open-source, web-based platform and two-day course that uses programming-free LEGO robotics activities to effectively teach core machine learning algorithms to students aged 12–17, resulting in statistically significant improvements in their technical understanding, motivation, and engagement.

Original authors: Viacheslav Sydora, Guner Dilsad Er, Michael Muehlebach

Published 2026-04-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Viacheslav Sydora, Guner Dilsad Er, Michael Muehlebach

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to explain how a computer learns to recognize a cat in a photo. Usually, you'd have to dive into complex math and lines of code, which feels like trying to explain how to bake a cake by discussing the molecular structure of flour. It's intimidating and often boring for kids.

This paper introduces a new way to teach these tricky concepts called "Machine Learning with Bricks." Think of it as turning the "black box" of AI into a transparent, interactive playground using LEGO robots.

Here is the breakdown of their idea, explained simply:

1. The Big Idea: No Code, Just Play

The researchers built a website that connects to LEGO robots (specifically the LEGO SPIKE Prime kits, which many schools already have). The best part? You don't need to write a single line of code.

Instead of typing commands, students use a web interface to:

Collect Data: Like a scientist gathering clues.
Train Models: Teaching the robot what to do.
Watch it Learn: Seeing the robot figure things out in real-time.

It's like teaching a dog tricks. You don't need to know how the dog's brain works; you just show it a treat (reward) when it does the right thing, and over time, it learns the pattern.

2. The Three "Games" (Experiments)

The course uses three different LEGO robots to teach three different types of machine learning. Here are the analogies:

The Fruit Detective (K-Nearest Neighbors):
- The Setup: A robot arm with a color sensor and a ruler.
- The Lesson: Imagine you have a basket of apples and bananas. You measure their color and length. Now, you find a mystery fruit. How do you know what it is? You look at the fruits closest to it in your basket. If it looks most like the apples nearby, it's an apple!
- The Takeaway: This teaches pattern recognition. The robot learns by comparing new things to old examples.
The Pitcher (Linear Regression):
- The Setup: A robot arm that launches table tennis balls.
- The Lesson: If you spin the motor slowly, the ball goes a short distance. If you spin it fast, it goes far. The students launch balls at different speeds and measure how far they go. They draw a line through the dots on a graph.
- The Takeaway: This teaches prediction. Once the robot understands the "line" connecting speed to distance, you can tell it, "I want the ball to go 2 meters," and it calculates exactly how fast to spin the motor.
The Crawler (Q-Learning / Reinforcement Learning):
- The Setup: A wheeled robot with an arm that moves forward.
- The Lesson: The robot doesn't know how to move at first. It tries random moves. If it moves forward, it gets a "point" (reward). If it hits a wall, it gets nothing. It keeps trying, remembering which moves gave it points.
- The Takeaway: This teaches trial and error. The robot learns a strategy by exploring the world and figuring out what actions lead to the best results, just like a video game character learning a level.

3. Did It Work? (The Results)

The researchers tested this with 14 students aged 12 to 17 over two days. The results were like finding a treasure chest:

Confidence Boost: Before the workshop, the kids felt like AI was magic or scary. Afterward, they felt like they understood the "magic tricks." They could explain the concepts to friends.
New Vocabulary: Before, they used words like "robots" and "future." After, they started using technical words like "agents," "rewards," and "linear regression." They stopped seeing AI as a monster and started seeing it as a tool they could understand.
Fun Factor: The kids loved it. They rated the hands-on activities and the visual tools very highly. It wasn't a boring lecture; it was a game.
Future Dreams: More students said they wanted to pursue careers in AI or robotics after the workshop.

4. Why This Matters

Usually, schools wait until university to teach how AI actually works. In the meantime, kids treat AI like a magic box. This project proves that if you give kids tangible tools (LEGO) and visual feedback (graphs and charts), they can grasp complex math and logic without needing to be professional programmers.

In a nutshell: This paper shows that you don't need to be a wizard to understand magic. If you give kids the right toys and let them play, they can learn how the "magic" of Artificial Intelligence actually works.

1. Problem Statement

The paper addresses a critical gap in K-12 Artificial Intelligence (AI) education. While AI literacy is becoming essential, current curricula often treat AI systems as "black boxes," focusing on ethics or high-level applications rather than underlying algorithms.

Complexity Barrier: Core machine learning (ML) algorithms are typically reserved for university levels due to their mathematical complexity.
Cognitive Load: Existing educational tools often rely on block-based programming, which shifts student focus toward syntax and implementation details rather than conceptual understanding.
Accessibility: Many tangible robotics solutions require custom hardware not widely available in schools, limiting scalability.

The authors aim to create a programming-free, accessible platform that demystifies ML algorithms for students aged 12–17 using widely available LEGO robotics kits.

2. Methodology

The authors developed "Machine Learning with Bricks," an open-source, web-based platform, and evaluated it through a two-day workshop curriculum.

A. The Platform Design

The platform integrates interactive visualizations with LEGO Education SPIKE Prime kits. It eliminates coding requirements by using a graphical web interface that communicates with the robot via Bluetooth. The system focuses on three core algorithms, selected for their mathematical accessibility (coordinate systems, linear equations, tabular updates):

K-Nearest Neighbors (KNN):
- Robot: Fruit Detector (uses color and length sensors).
- Activity: Students collect data on apples and bananas, visualize the 2D feature space, and observe how the algorithm classifies unknown fruits based on proximity to neighbors. They also explore decision boundaries and data quality (poisoning).
Linear Regression:
- Robot: Pitcher (launches a table tennis ball).
- Activity: Students vary motor speed to control flight distance. They collect data points to visualize the linear relationship between speed and distance, manually adjust line parameters to minimize loss, and observe the computer's best-fit solution.
Q-Learning (Reinforcement Learning):
- Robot: Crawler (wheeled platform with an arm).
- Activity: The robot learns to move toward a target. Students interact with a Q-table, adjusting the exploration rate and discount factor to understand the exploration-exploitation trade-off and the importance of future rewards.

B. Experimental Setup

Participants: 14 students (ages 12–17) recruited from a local makerspace.
Structure: A two-day workshop (approx. 6–7 hours total) involving pair work with shared LEGO kits.
Pedagogy: The course follows a constructionist framework, alternating between concise theory and hands-on application. It uses an iterative theory-practice cycle and no-code interfaces.
Evaluation: Pre- and post-course surveys were administered to measure:
- Self-reported knowledge of ML concepts (5-point Likert scale).
- Changes in AI-related terminology (open-ended word association).
- Platform usability and course engagement.
- Motivation for future AI engagement.

3. Key Contributions

The paper makes three primary contributions:

C1. The Platform: An open-source, web-based tool (ml-with-bricks) that combines interactive visualizations, video tutorials, and LEGO robotics to teach KNN, linear regression, and Q-learning without programming.
C2. The Curriculum: A structured two-day workshop featuring three tangible activities that scaffold learning from data collection to model training and inference.
C3. Empirical Evidence: Statistical validation showing that this approach significantly improves student understanding and shifts their vocabulary toward technical ML terms.

4. Results

Data from 14 participants yielded the following findings:

Learning Outcomes:
- Statistically Significant Gains: Students showed significant improvements ( $p < 0.01$ ) in understanding ML paradigms (L2, L3) and specific algorithms like KNN, linear regression, and reinforcement learning (L5–L8).
- General Concepts: Improvements were observed in general ML concepts and data importance, though these did not reach statistical significance, likely due to pre-existing familiarity.
Terminology Shift:
- Pre-course responses used abstract terms (e.g., "future," "danger," "intelligent").
- Post-course responses incorporated specific technical vocabulary (e.g., "agent," "rewards," "linear regression," "supervised learning"), indicating a shift from abstract perception to technical understanding.
Usability and Engagement:
- The platform received high usability ratings (Mean = 4.00/5).
- Presentations and hands-on activities were rated highest (Mean = 4.54/5), confirming high engagement.
Motivation:
- Interest in pursuing a career in AI/robotics increased from 5 to 8 participants.
- Participants expressed high motivation to continue learning ( $M=3.92$ ) and perceived the skills as useful for future careers ( $M=3.69$ ).

5. Significance and Limitations

Significance:
The study demonstrates that tangible, visualization-based approaches can effectively teach complex algorithmic concepts to young learners without the barrier of programming syntax. By making the "black box" transparent through real-time data visualization and physical interaction, the platform fosters a mechanistic understanding of ML. This suggests a viable pathway for democratizing AI education in K-12 settings using standard, affordable hardware.

Limitations:

Sample Size: The study involved only 14 participants from a single location, limiting generalizability.
Self-Reported Data: Reliance on self-assessed knowledge rather than objective performance tasks.
Short-Term Scope: The evaluation was immediate; long-term retention of concepts was not measured.
Lack of Control Group: Without a control group, it is difficult to isolate the specific impact of the platform compared to other teaching methods.

Future Work:
The authors plan to expand the curriculum to include neural networks and decision trees, integrate deeper ethics training, and conduct large-scale longitudinal studies within the "Robotics in Germany" (RIG) education track.