2026 ASEE Annual Conference & Exposition

The accelerating integration of Artificial Intelligence (AI) and Edge AI into modern engineering applications highlights the urgent need to expose undergraduate students to practical, hands-on experiences in these domains. This paper presents the development of a structured laboratory framework that bridges theoretical AI principles with embedded, real-time implementation through accessible Edge AI experiments. The work builds on an instructional case study using the Arduino Nano 33 BLE Sense and Edge Impulse, an open-source machine-learning platform designed for low-power devices. Within this framework, students collect and label image data using an onboard camera, design and train lightweight convolutional models such as MobileNetV2 (0.35) with FOMO (Faster Objects, More Objects) architecture, and deploy them on-device to perform object classification and counting.
The proposed experiments emphasize the complete machine-learning life cycle—from data acquisition and preprocessing to model training, evaluation, and edge deployment—without the need for GPUs or cloud resources. By employing real-time inference and visualization via Firebase and Streamlit, students witness how embedded AI systems sense, decide, and communicate autonomously. This approach not only reinforces core machine-learning and embedded-systems concepts but also cultivates problem-solving, iteration, and experimental design skills essential to modern engineering education.
Preliminary implementation within the undergraduate embedded-systems context demonstrates that the Edge AI experiments enhance student engagement and conceptual understanding of AI pipelines, model efficiency, and hardware–software co-design. The paper concludes by outlining best practices for integrating AI-and-Edge-AI laboratories into electrical and computer engineering curricula, illustrating how accessible hardware and cloud-free platforms can facilitate AI education and prepare students for the next generation of intelligent, data-driven embedded systems.