Foundations of Autonomous Vehicles: A Curriculum Model for Developing Competencies in Artificial Intelligence and the Internet of Things for Grades 7-10

Elham Buxton, Elahe Javadi, Matthew HagamanAAAI 2024 · DOI

Abstract

This paper presents a one-week curriculum designed for students in grades 7–10, bridging the gap between Artificial Intelligence (AI) and the Internet of Things (IoT) through the lens of autonomous vehicles. Utilizing low-cost Elegoo robot cars and Google Teachable Machine, the curriculum facilitates experiential learning across the "Five Big Ideas in AI": perception, representation, learning, natural interaction, and societal impact. Pilot studies indicate that this hands-on approach effectively demystifies AI "black boxes," helping students understand the complexities of real-world data distribution shifts and hardware integration.

Introduction

While AI literacy is increasingly mandated in K-12 education, many existing programs remain purely software-based or treat AI and IoT as isolated subjects. This separation prevents students from understanding how AI models interact with physical environments. To address this, the researchers developed a curriculum based on Kolb’s Experiential Learning Cycle, moving from concrete experimentation with hardware to abstract conceptualization of machine learning principles.

Methodology

The curriculum is structured into three modules:

  1. IoT Module: Students assemble Elegoo Smart Cars and interface with ultrasonic and infrared sensors using microcontrollers (Arduino/ESP32).
  2. AI Module: Students explore digital representation and use Google Teachable Machine to train computer vision models.
  3. Integration: Students bridge AI and IoT using Python and Socket programming, where a PC processes the car's camera feed and sends real-time movement commands back to the vehicle.
The curriculum maps the 'Five Big Ideas in AI' to practical robotics activities.

Data Collection and Distribution Shift

A core component of the curriculum is student-generated data. Students capture hand gestures and traffic signs using both high-quality webcams and the robot's lower-resolution ESP32 camera. This comparison provides an "aha" moment regarding distribution shift: a model that works perfectly on a laptop may fail on the robot due to differences in perspective and image quality.

Students collect custom datasets, such as traffic signs, to train their autonomous vehicle models.

Key Insights

  • Hardware as a Reality Check: Physical consequences, such as a robot crashing, make abstract concepts like model bias and inference latency tangible.
  • The IoT-AI Symbiosis: AI perception is fundamentally an IoT challenge; without sensors and data protocols, AI remains disconnected from the physical world.
  • Accessibility: The use of affordable kits (~$70) makes advanced technical education accessible to underserved communities.

Results and Discussion

The curriculum was piloted with 29 students, showing 90% engagement. Final projects demonstrated significant innovation, such as "anti-social robots" that avoid humans or vehicles equipped with visual-trigger mechanisms. Students successfully applied societal impact concepts, like the "Trolley Problem," to their own builds.

Limitations and Future Work

Current limitations include Wi-Fi latency in the control loop. Future iterations aim to deploy models directly to ESP32 edge devices for real-time inference. Additionally, the authors plan to add more rigorous mathematical modules for older students and adapt the 20-hour instruction time for standard classroom schedules.

Summary updated

BibTeX
@inproceedings{buxton2024foundations,
  author = {Elham Buxton and Elahe Javadi and Matthew Hagaman},
  title = {{Foundations of Autonomous Vehicles: A Curriculum Model for Developing Competencies in Artificial Intelligence and the Internet of Things for Grades 7-10}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2024},
  doi = {10.1609/aaai.v38i21.30375},
}
Figure