Mighty Thymio for University-Level Educational Robotics

Jérôme Guzzi, Alessandro Giusti, Gianni A. Di Caro, Luca Maria GambardellaAAAI 2018 · DOI

Abstract

The "Mighty Thymio" project introduces a low-cost, modular upgrade for the popular K-12 Thymio robot, transforming it into a sophisticated platform for university-level robotics and AI education. By integrating an Odroid Linux board, a camera, and full ROS support, the authors provide a portable, $130 platform capable of complex tasks like SLAM and deep learning-based obstacle avoidance.

Introduction: Bridging the "Missing Middle"

In robotics education, there is often a "missing middle" between simple K-12 toys and expensive, bulky research platforms. The original Thymio robot is a staple in primary and secondary schools because it is sturdy, affordable, and Lego-compatible. However, it lacks the onboard processing power, vision capabilities, and industry-standard software support (like ROS) required for university-level curriculum.

While platforms like the Turtlebot 3 exist for higher education, they are often too large and expensive for every student to own or take home. The Mighty Thymio project bridges this gap by creating an incremental, tool-free hardware extension that turns the humble Thymio into a sophisticated platform suitable for graduate-level research and education.

System Architecture and Hardware

The authors developed a hardware and software "upgrade kit" designed for easy assembly by students. The hardware upgrade utilizes a Lego-based frame added to the Thymio base to house an Odroid C1 (a quad-core ARM Linux board with 1GB RAM), a 720p webcam, a Wi-Fi dongle, and a dual-battery system.

Labeled hardware components of the Mighty Thymio upgrade kit, showing the Odroid C1, camera, and power systems.

The power setup includes a main power pack and a small UPS, allowing students to hot-swap batteries without rebooting the Linux system. This modularity ensures that students spend their time on robotics and AI logic rather than mechanical engineering or soldering.

Software Integration and Simulation

The system runs a custom Linux image pre-configured with the Robot Operating System (ROS). A specialized "Aseba-ROS bridge" allows ROS nodes to communicate with the Thymio’s native sensors and motors. To facilitate development, the authors provided URDF models for visualization in rviz and integrated the platform with the V-REP simulator.

The Mighty Thymio digital twin in a simulation environment, used for prototyping SLAM and navigation algorithms.

This "digital twin" approach allows students to prototype algorithms in a virtual environment and deploy them to the physical robot with minimal code changes, bridging the gap between Aseba (a simple, event-based language) and ROS (the complex middleware used in professional robotics).

Educational Case Studies

The platform has been successfully deployed in two distinct academic settings:

  1. Graduate Robotics: Students implemented landmark-based SLAM (Simultaneous Localization and Mapping) using AR markers and path planning using RRT (Rapidly-exploring Random Trees).
  2. Bachelor Machine Learning: Students trained a Convolutional Neural Network (CNN) for obstacle avoidance. By programming the robot to explore randomly, students collected a self-supervised dataset where the robot's simple IR sensors provided ground-truth labels for the complex camera input. Notably, the robot could navigate even when its physical proximity sensors were covered, relying entirely on its learned visual perception.

Conclusion and Future Work

The Mighty Thymio demonstrates that existing low-cost hardware can be "upcycled" into a high-performance platform. Its small form factor makes it "home-work friendly," allowing students to continue research outside the lab. While the Odroid C1 is powerful, running heavy real-time computer vision and SLAM simultaneously can push its limits. Future work involves refining the software bridge for lower latency and expanding the platform's use to more diverse graduate courses.

Summary updated

BibTeX
@inproceedings{guzzi2018mighty,
  author = {Jérôme Guzzi and Alessandro Giusti and Gianni A. Di Caro and Luca Maria Gambardella},
  title = {{Mighty Thymio for University-Level Educational Robotics}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.1609/aaai.v32i1.11402},
}
Figure