Agile AI and firmware management in IoT: DevOps for low-power microcontroller-based platforms

In this paper, we introduce a Development and Operations (DevOps)-based IoT system tailored for the dynamic management of firmware and AI models across distributed IoT environments. The system offers scalability, resource efficiency, and cost-effectiveness for updating the AI model and firmware on the IoT devices without a need for human intervention. To accomplish this, we have developed a continuous integration and continuous deployment (CI/CD) pipelines that operate across multiple platforms, leveraging the capabilities of Kubernetes and GitLab-Runner. Moreover, the system is specifically designed for Microcontroller-based low-power devices capable of running tiny AI models. The new deployment is sent to the IoT devices despite their location to start interacting with the surrounding environment and perform predictions regarding its application. Through empirical experiments, we demonstrate the system effectiveness with promising results in terms of scalability, resource utilization, and deployment efficiency.