This work introduces an open-source Edge AI pipeline that integrates Arduino’s ecosystem with Zant, a Zig-based SDK designed for efficient neural network (NN) deployment on microcontrollers. By leveraging hardware-agnostic optimizations, this solution offers scalable, vendor-independent Edge AI implementations, reducing development complexity and enhancing accessibility.
While AI deployment on edge devices remains complex due to resource limitations and fragmented ecosystems, this work addresses these challenges through an open-source, end-to-end pipeline for Edge AI inference. By combining Arduino’s development tools with Zant’s lightweight inference engine, the approach democratizes Edge AI development while ensuring efficiency in memory usage, computational performance, and power consumption. This pipeline fosters standardization and interoperability, eliminating vendor lock-in and promoting widespread adoption across diverse hardware platforms.
The pipeline was validated through the implementation of a device for identifying obstructive sleep apnea (OSA). The prototype utilizes an ear-mounted wearable form factor, integrating an high-sensitivity pulse oximeter and heart-rate sensor for physiological signal acquisition. Real-time data processing is performed onboard using the Arduino Nicla Sense Me, an Arduino board with embedded sensors.