pith. sign in

arxiv: 2601.19376 · v2 · submitted 2026-01-27 · 💻 cs.RO · cs.AI· cs.CY· cs.HC· cs.LG

Teaching Machine Learning Fundamentals with LEGO Robotics

Viacheslav Sydora , Guner Dilsad Er , Michael Muehlebach This is my paper

Pith reviewed 2026-05-16 10:59 UTC · model grok-4.3

classification 💻 cs.RO cs.AIcs.CYcs.HCcs.LG
keywords machine learning educationLEGO roboticsKNNlinear regressionQ-learningtangible interfacesAI for studentseducational robotics
0
0 comments X

The pith

A web-based platform pairs LEGO robots with visualizations so students aged 12-17 can learn KNN, linear regression, and Q-learning by collecting data and training models without writing code.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Machine Learning with Bricks, an open-source web platform that combines physical LEGO robot activities with interactive visualizations to teach three machine learning algorithms to middle- and high-school students. Learners gather sensor data from the robots, train models for classification with KNN, prediction with linear regression, and decision-making with Q-learning, then observe the results in real time through the browser interface. Pre- and post-surveys with 14 participants report statistically significant increases in self-described understanding, a shift toward more technical AI vocabulary, high usability ratings, and greater interest in further learning. The work positions tangible, programming-free robotics as a concrete entry point that preserves technical depth while lowering barriers for young learners.

Core claim

The platform enables students to learn the fundamentals of KNN, linear regression, and Q-learning by physically collecting data with LEGO robots and interacting with training visualizations in a web browser, producing measurable gains in self-reported algorithm comprehension and motivation after a two-day course.

What carries the argument

The web interface that links real-time data collection from LEGO robots to side-by-side visualizations for training and testing KNN classification, linear regression, and Q-learning agents.

If this is right

  • Students acquire hands-on experience mapping real sensor data to model parameters without first learning a programming language.
  • The two-day format can be replicated in schools or after-school programs using only standard LEGO kits and a browser.
  • Open release of the platform and video tutorials supports adoption by educators who lack ML expertise.
  • Increased motivation reported by participants suggests the approach may encourage continued study of AI topics.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Physical robot feedback may strengthen intuition about model errors more effectively than screen-only simulations for this age group.
  • The same data-collection loop could be extended to additional algorithms such as decision trees or neural networks by adding new visualization modules.
  • Longer-term studies could check whether early exposure through robots improves later performance in formal programming-based ML courses.

Load-bearing premise

Self-reported survey answers from the fourteen students accurately reflect genuine understanding of the algorithms and that any observed changes result from the platform rather than novelty, social pressure, or other unmeasured factors.

What would settle it

A controlled experiment that gives the same students an objective test of algorithm mechanics before and after the course and compares gains against a matched group taught the same material through lectures or simulations only.

Figures

Figures reproduced from arXiv: 2601.19376 by Guner Dilsad Er, Michael Muehlebach, Viacheslav Sydora.

Figure 1
Figure 1. Figure 1: Landing page of the Machine Learning with Bricks platform showing the available experiments. Each experiment page provides step-by-step instructions on conducting the experiment, including guidance for robot assembly, connecting the LEGO hub to the platform, per￾forming data collection, model training, and inference. The instructional materials are complemented by video tutori￾als2 , which visually guide l… view at source ↗
Figure 2
Figure 2. Figure 2: Fruit detector consisting of a hub, a color sensor, and a caliper mechanism equipped with a distance sensor. Web interface. The web interface visualizes collected data on a 2D plot with color on the x-axis and length on the y￾axis as shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Pitcher consisting of a hub, a distance sensor, and motors that, through a gear set, accelerate a pitching arm to launch a table tennis ball toward a target. Web interface. The interface is shown in [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: User interface of the Fruit Detector experiment in inference mode. The 2D plot visualizes collected samples and the decision boundary. The interface also includes a data table for editing or deleting samples, and controls for switching between training and inference modes and adjusting the number of neighbors participating in the vote. Educational scaffolding. Participants first brainstorm pos￾sible featur… view at source ↗
Figure 7
Figure 7. Figure 7: User interface of the Crawler experiment displaying the Q-table (left), the reward diagram illustrating distance crawled for each transition (center), and the control panel (right), which includes options to pause, resume, or reset training, adjust the exploration rate, and toggle the discount factor to account for future rewards. the introduction of the Q-table, after which training of the robot begins. I… view at source ↗
Figure 6
Figure 6. Figure 6: Crawler consisting of a hub and a distance sensor mounted on a wheeled base, actuated by a two-limb arm that enables forward movement. Web interface. The interface is displayed in [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Shift in participants’ interest in pursuing a career in [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
read the original abstract

This paper presents the web-based platform Machine Learning with Bricks and an accompanying two-day course designed to teach machine learning concepts to students aged 12 to 17 through programming-free robotics activities. Machine Learning with Bricks is an open source platform and combines interactive visualizations with LEGO robotics to teach three core algorithms: KNN, linear regression, and Q-learning. Students learn by collecting data, training models, and interacting with robots via a web-based interface. Pre- and post-surveys with 14 students indicate statistically significant improvements in self-reported understanding of machine learning algorithms, changes in AI-related terminology toward more technical language, high platform usability, and increased motivation for continued learning. This work suggests that tangible, visualization-based approaches can make machine learning concepts accessible and engaging for young learners while maintaining technical depth. The platform is freely available at https://learning-and-dynamics.github.io/ml-with-bricks/, with video tutorials guiding students through the experiments at https://youtube.com/playlist?list=PLx1grFu4zAcwfKKJZ1Ux4LwRqaePCOA2J.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper presents the open-source 'Machine Learning with Bricks' web platform that combines interactive visualizations with LEGO robotics to teach KNN, linear regression, and Q-learning to students aged 12-17 without programming. It describes a two-day course structure and evaluates effectiveness via pre- and post-surveys of 14 students, reporting statistically significant gains in self-reported understanding, shifts to more technical AI terminology, high usability scores, and increased motivation for further learning.

Significance. If the platform produces genuine learning gains, the work would demonstrate a promising tangible-visual approach for making core machine-learning algorithms accessible to young learners while preserving technical content. The open-source release and accompanying video tutorials strengthen potential for adoption and replication in educational settings.

major comments (2)
  1. [Results] Results section: The pre-post survey design with n=14 reports statistically significant self-reported gains but includes no control group, so changes cannot be attributed to the platform rather than novelty effects, repeated testing, or social desirability.
  2. [Evaluation] Evaluation section: All outcome measures are subjective self-reports of understanding and motivation; no objective performance metrics (e.g., students correctly training models on held-out data or debugging robot policies) are provided to validate that perceived gains reflect actual algorithmic competence.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'statistically significant improvements' should specify the exact tests, effect sizes, and p-values for transparency.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and have made partial revisions to the text to strengthen the discussion of limitations and future directions.

read point-by-point responses

  1. Referee: [Results] Results section: The pre-post survey design with n=14 reports statistically significant self-reported gains but includes no control group, so changes cannot be attributed to the platform rather than novelty effects, repeated testing, or social desirability.

    Authors: We agree that the lack of a control group prevents strong causal attribution of the observed gains. Our study was an initial pilot evaluation of platform feasibility in a real classroom setting with a small cohort, where a controlled design was not practical. We have added a dedicated Limitations subsection that explicitly discusses potential confounds including novelty effects, repeated testing, and social desirability bias, and we outline plans for future controlled experiments with larger samples. revision: partial

  2. Referee: [Evaluation] Evaluation section: All outcome measures are subjective self-reports of understanding and motivation; no objective performance metrics (e.g., students correctly training models on held-out data or debugging robot policies) are provided to validate that perceived gains reflect actual algorithmic competence.

    Authors: We acknowledge that objective performance metrics would provide stronger validation of actual learning. The current evaluation prioritizes self-reported understanding and motivation as key indicators for an accessible, non-programming platform aimed at young learners. We have revised the Discussion and Future Work sections to propose concrete objective measures for follow-up studies, such as task-based assessments of model training and policy debugging, while noting that the combination of significant self-report gains and shifts toward technical terminology offers preliminary support for the approach. revision: partial

Circularity Check

0 steps flagged

No circularity: claims rest on independent external surveys

full rationale

The paper presents an educational platform and evaluates its effectiveness solely through pre- and post-survey responses from 14 students. These survey measures (self-reported understanding, terminology shifts, usability, motivation) are collected externally and do not depend on any fitted parameters, internal definitions, or equations within the platform. No derivation chain, ansatz, or self-citation reduces the reported outcomes to the inputs by construction. The evaluation design is self-contained against external benchmarks and contains no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper introduces an educational platform and evaluation but relies on standard algorithms and conventional survey-based assessment methods without introducing new mathematical constructs, fitted parameters, or postulated entities.

axioms (1)
  • domain assumption Pre- and post-surveys reliably measure changes in students' understanding of machine learning concepts
    The central claim of improved understanding rests directly on the validity and sensitivity of these self-report instruments.

pith-pipeline@v0.9.0 · 5504 in / 1399 out tokens · 67193 ms · 2026-05-16T10:59:41.883740+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Three Interviews About K-12 AI Education in America, Europe, and Singapore,

    F. Heintz, “Three Interviews About K-12 AI Education in America, Europe, and Singapore,”KI - K ¨unstliche Intelligenz, vol. 35, no. 2, pp. 233–237, 2021. [Online]. Available: https://doi.org/10.1007/s13218-021-00730-w

    work page doi:10.1007/s13218-021-00730-w 2021

  2. [2]

    AI literacy in K-12: a systematic literature review,

    L. Casal-Otero, A. Catala, C. Fern ´andez-Morante, M. Taboada, B. Cebreiro, and S. Barro, “AI literacy in K-12: a systematic literature review,”International Journal of STEM Education, vol. 10, no. 29, pp. 1–17, 2023. [Online]. Available: https://doi.org/10.1186/ s40594-023-00418-7

    work page 2023

  3. [3]

    A systematic review of teaching and learning machine learning in K-12 education,

    I. T. Sanusi, S. S. Oyelere, H. Vartiainen, J. Suhonen, and M. Tukiainen, “A systematic review of teaching and learning machine learning in K-12 education,” Education and Information Technologies, vol. 28, no. 5, pp. 5967–5997, 2023. [Online]. Available: https://doi.org/10.1007/s10639-022-11416-7

    work page doi:10.1007/s10639-022-11416-7 2023

  4. [4]

    Principles of Explanatory Debugging to Personalize Interactive Machine Learning,

    T. Kulesza, M. Burnett, W.-K. Wong, and S. Stumpf, “Principles of Explanatory Debugging to Personalize Interactive Machine Learning,”International Conference on Intelligent User Interfaces, p. 126–137, 2015. [Online]. Available: https://doi.org/10.1145/2678025. 2701399

    work page doi:10.1145/2678025 2015

  5. [5]

    AI Programming by Children using Snap! Block Programming in a Developing Country,

    K. Kahn, R. Megasari, E. Piantari, and E. Junaeti, “AI Programming by Children using Snap! Block Programming in a Developing Country,”European Conference on Technology Enhanced Learning, vol. 2193, pp. 1–14, 2018. [Online]. Available: https: //ceur-ws.org/V ol-2193/paper1.pdf

    work page 2018

  6. [6]

    Teaching Tech to Talk: K-12 Conversational Artificial Intelligence Literacy Curriculum and Development Tools,

    J. Van Brummelen, T. Heng, and V . Tabunshchyk, “Teaching Tech to Talk: K-12 Conversational Artificial Intelligence Literacy Curriculum and Development Tools,”AAAI Conference on Artificial Intelligence, vol. 35, no. 17, pp. 15 655–15 663, 2021. [Online]. Available: https://doi.org/10.1609/aaai.v35i17.17844

    work page doi:10.1609/aaai.v35i17.17844 2021

  7. [7]

    Why Are We Not Teaching Machine Learning at High School? A Proposal,

    I. Evangelista, G. Blesio, and E. Benatti, “Why Are We Not Teaching Machine Learning at High School? A Proposal,”World Engineering Education Forum - Global Engineering Deans Council, pp. 1– 6, 2018. [Online]. Available: https://doi.org/10.1109/ WEEF-GEDC.2018.8629750

    work page arXiv 2018

  8. [8]

    Beyond Black- Boxing: Building Intuitions of Complex Machine Learning Ideas Through Interactives and Levels of Abstraction,

    B. Broll, S. Grover, and D. Babb, “Beyond Black- Boxing: Building Intuitions of Complex Machine Learning Ideas Through Interactives and Levels of Abstraction,”ACM Conference on International Com- puting Education Research, p. 21–23, 2022. [Online]. Available: https://doi.org/10.1145/3501709.3544273

    work page doi:10.1145/3501709.3544273 2022

  9. [9]

    Text Classifi- cation for AI Education,

    T. Reddy, R. Williams, and C. Breazeal, “Text Classifi- cation for AI Education,”ACM Technical Symposium on Computer Science Education, p. 1381, 2021. [Online]. Available: https://doi.org/10.1145/3408877.3439689

    work page doi:10.1145/3408877.3439689 2021

  10. [10]

    SmileyCluster: supporting accessible machine learning in K-12 scientific discovery,

    X. Wan, X. Zhou, Z. Ye, C. K. Mortensen, and Z. Bai, “SmileyCluster: supporting accessible machine learning in K-12 scientific discovery,”Interaction Design and Children Conference, p. 23–35, 2020. [Online]. Available: https://doi.org/10.1145/3392063.3394440

    work page doi:10.1145/3392063.3394440 2020

  11. [11]

    Introducing Reinforcement Learning to K-12 Students with Robots and Augmented Reality,

    Z. Zhang, K. Lavigne, W. Church, J. Sinapov, and C. Rogers, “Introducing Reinforcement Learning to K-12 Students with Robots and Augmented Reality,”Robotics in Education, pp. 351–365, 2023. [Online]. Available: https://doi.org/10.1007/978-3-031-38454-7 29

    work page doi:10.1007/978-3-031-38454-7 2023

  12. [12]

    Calypso for Cozmo: Robotic AI for Everyone (Abstract Only),

    D. S. Touretzky and C. Gardner-McCune, “Calypso for Cozmo: Robotic AI for Everyone (Abstract Only),” ACM Technical Symposium on Computer Science Education, p. 1110, 2018. [Online]. Available: https: //doi.org/10.1145/3159450.3162200

    work page doi:10.1145/3159450.3162200 2018

  13. [13]

    PopBots: Designing an Artificial Intelligence Curriculum for Early Childhood Education,

    R. Williams, H. W. Park, L. Oh, and C. Breazeal, “PopBots: Designing an Artificial Intelligence Curriculum for Early Childhood Education,”AAAI Conference on Artificial Intelligence, vol. 33, no. 1, pp. 9729–9736, 2019. [Online]. Available: https://doi.org/10.1609/aaai.v33i01.33019729

    work page doi:10.1609/aaai.v33i01.33019729 2019

  14. [14]

    Zhorai: Designing a Conversational Agent for Children to Explore Machine Learning Concepts,

    P. Lin, J. Van Brummelen, G. Lukin, R. Williams, and C. Breazeal, “Zhorai: Designing a Conversational Agent for Children to Explore Machine Learning Concepts,” AAAI Conference on Artificial Intelligence, vol. 34, no. 9, pp. 13 381–13 388, 2020. [Online]. Available: https://doi.org/10.1609/aaai.v34i09.7061

    work page doi:10.1609/aaai.v34i09.7061 2020

  15. [15]

    “AlphAI”: Teaching AI Algorithms to K12 by Training Learning Robots and Visualizing How It Works,

    M. Absalon and T. Deneux, ““AlphAI”: Teaching AI Algorithms to K12 by Training Learning Robots and Visualizing How It Works,”AAAI Conference on Artificial Intelligence, vol. 39, no. 28, pp. 29 096–29 103,

    work page

  16. [17]

    Intro- ducing children to machine learning concepts through hands-on experience,

    T. Hitron, I. Wald, H. Erel, and O. Zuckerman, “Intro- ducing children to machine learning concepts through hands-on experience,”ACM Conference on Interaction Design and Children, p. 563–568, 2018. [Online]. Available: https://doi.org/10.1145/3202185.3210776

    work page doi:10.1145/3202185.3210776 2018

  17. [18]

    Artificial intelligence and computer science in education: From kindergarten to university,

    M. Kandlhofer, G. Steinbauer, S. Hirschmugl-Gaisch, and P. Huber, “Artificial intelligence and computer science in education: From kindergarten to university,” IEEE Frontiers in Education Conference, pp. 1–9,

    work page

  18. [19]

    Available: https://doi.org/10.1109/FIE

    [Online]. Available: https://doi.org/10.1109/FIE. 2016.7757570

    work page doi:10.1109/fie 2016

  19. [20]

    Envisioning AI for K-12: What Should Every Child Know about AI?

    D. Touretzky, C. Gardner-McCune, F. Martin, and D. Seehorn, “Envisioning AI for K-12: What Should Every Child Know about AI?”AAAI Conference on Artificial Intelligence, vol. 33, no. 1, pp. 9795–9799,

    work page

  20. [21]

    Louis, G

    [Online]. Available: https://doi.org/10.1609/aaai. v33i01.33019795

    work page doi:10.1609/aaai

  21. [22]

    What is AI Literacy? Competencies and Design Considerations,

    D. Long and B. Magerko, “What is AI Literacy? Competencies and Design Considerations,”CHI Conference on Human Factors in Computing Systems, p. 1–16, 2020. [Online]. Available: https://doi.org/10.1145/3313831.3376727

    work page doi:10.1145/3313831.3376727 2020

  22. [23]

    Conceptualizing AI literacy: An exploratory review,

    D. T. K. Ng, J. K. L. Leung, S. K. W. Chu, and M. S. Qiao, “Conceptualizing AI literacy: An exploratory review,”Computers and Education: Artificial Intelligence, vol. 2, p. 100041, 2021. [Online]. Available: https://doi.org/10.1016/j.caeai.2021.100041

    work page doi:10.1016/j.caeai.2021.100041 2021

  23. [24]

    PhET: Simulations That Enhance Learning,

    C. E. Wieman, W. K. Adams, and K. K. Perkins, “PhET: Simulations That Enhance Learning,”Science, vol. 322, no. 5902, pp. 682–683, 2008. [Online]. Available: https://doi.org/10.1126/science.1161948

    work page doi:10.1126/science.1161948 2008

  24. [25]

    Situating Constructionism,

    S. Papert, “Situating Constructionism,” inConstruction- ism: Research Reports and Essays. Ablex, 1991, pp. 1–11

    work page 1991

  25. [26]

    USA: Basic Books, Inc., 1980

    ——,Mindstorms: Children, Computers, and Powerful Ideas. USA: Basic Books, Inc., 1980

    work page 1980

  26. [27]

    Youth making machine learning models for gesture-controlled interactive media,

    A. Zimmermann-Niefield, S. Polson, C. Moreno, and R. B. Shapiro, “Youth making machine learning models for gesture-controlled interactive media,”Interaction De- sign and Children Conference, p. 63–74, 2020. [Online]. Available: https://doi.org/10.1145/3392063.3394438

    work page doi:10.1145/3392063.3394438 2020

  27. [28]

    Can Children Understand Machine Learning Concepts?: The Effect of Uncovering Black Boxes,

    T. Hitron, Y . Orlev, I. Wald, A. Shamir, H. Erel, and O. Zuckerman, “Can Children Understand Machine Learning Concepts?: The Effect of Uncovering Black Boxes,”CHI Conference on Human Factors in Computing Systems, no. 415, pp. 1–11, 2019. [Online]. Available: https://doi.org/10.1145/3290605.3300645

    work page doi:10.1145/3290605.3300645 2019

  28. [29]

    Youth Learning Machine Learning through Building Models of Athletic Moves,

    A. Zimmermann-Niefield, M. Turner, B. Murphy, S. K. Kane, and R. B. Shapiro, “Youth Learning Machine Learning through Building Models of Athletic Moves,” ACM International Conference on Interaction Design and Children, p. 121–132, 2019. [Online]. Available: https://doi.org/10.1145/3311927.3323139

    work page doi:10.1145/3311927.3323139 2019

  29. [30]

    The Machine Learning Machine: A Tangible User Interface for Teaching Machine Learning,

    M. H. Kaspersen, K.-E. K. Bilstrup, and M. G. Petersen, “The Machine Learning Machine: A Tangible User Interface for Teaching Machine Learning,”International Conference on Tangible, Embedded, and Embodied Interaction, no. 19, pp. 1–12, 2021. [Online]. Available: https://doi.org/10.1145/3430524.3440638

    work page doi:10.1145/3430524.3440638 2021

  30. [31]

    Teacher Perspectives on How To Train Your Robot: A Middle School AI and Ethics Curriculum,

    R. Williams, S. P. Kaputsos, and C. Breazeal, “Teacher Perspectives on How To Train Your Robot: A Middle School AI and Ethics Curriculum,”AAAI Conference on Artificial Intelligence, vol. 35, no. 17, pp. 15 678–15 686, 2021. [Online]. Available: https: //doi.org/10.1609/aaai.v35i17.17847

    work page doi:10.1609/aaai.v35i17.17847 2021

  31. [32]

    IRobot: Teaching the Basics of Artificial Intelligence in High Schools,

    H. Burgsteiner, M. Kandlhofer, and G. Steinbauer, “IRobot: Teaching the Basics of Artificial Intelligence in High Schools,”AAAI Conference on Artificial Intelligence, vol. 30, no. 1, pp. 4126–4127, 2016. [Online]. Available: https://doi.org/10.1609/aaai.v30i1. 9864

    work page doi:10.1609/aaai.v30i1 2016

  32. [33]

    Teaching AI using LEGO Mindstorms,

    S. Parsons and E. Sklar, “Teaching AI using LEGO Mindstorms,”AAAI Spring Symposium,

    work page

  33. [34]

    Available: https://aaai.org/papers/ 0003-ss04-01-003-teaching-ai-using-lego-mindstorms/

    [Online]. Available: https://aaai.org/papers/ 0003-ss04-01-003-teaching-ai-using-lego-mindstorms/

    work page

  34. [35]

    The ICAP Framework: Linking Cognitive Engagement to Active Learning Outcomes,

    M. T. H. Chi and R. Wylie, “The ICAP Framework: Linking Cognitive Engagement to Active Learning Outcomes,”Educational Psychologist, vol. 49, no. 4, pp. 219–243, 2014. [Online]. Available: https://doi.org/ 10.1080/00461520.2014.965823

    work page doi:10.1080/00461520.2014.965823 2014

  35. [36]

    Likert,The Method of Constructing an Attitude Scale

    R. Likert,The Method of Constructing an Attitude Scale. Wiley, 1967. [Online]. Available: https://www.sfu.ca/ ∼palys/ Likert-1933-TheMethodOfConstructingAnAttitudeScale. pdf

    work page 1967