Advancing Tiny Machine Learning Operations: Robust Model Updates in the Internet of Intelligent Vehicles

Abstract

The Internet of Intelligent Vehicles is becoming increasingly important, and embedded machine learning is gaining popularity due to new development paradigms. However, the demand for machine learning model updates on embedded systems has become relevant in multiple scenarios. This article proposes a methodology for tiny machine learning operations within the context of the Internet of Intelligent Vehicles, utilizing affordable microcontrollers based on the ESP32 platform. The solution presented in the article consists of two ESP32 devices: one functioning as a radio station (RS) and the other as the microcontroller of an onboard diagnostic (OBD-II) scanner. The RS hosts the updated model and transmits it to the OBD-II scanner using the Espressif Systems Peer-to-Peer Over Wi-Fi communication protocol over 802.11 Wi-Fi. Experimental results demonstrate significant improvement in model performance postupdate, but the article also identifies critical challenges to model robustness because of the use of the interpreter method on microcontrollers.