Controle de veículo através de palavras-chave faladas utilizando TinyML

Abstract

Technological evolution has enabled the development of devices with increasingly greater processing power and storage capacity while reducing their physical size and energy consumption. Among these devices are microcontrollers, components capable of executing software (referred to in this context as firmware) and controlling hardware. Alongside the evolution of electronic devices, machine learning, a subfield of AI that is responsible for developing devices capable of learning and improving from experience, has also significantly emerged. This work aims to explore and interconnect these two major areas of computer science. This subfield is known as TinyML. In this project, an edge device was developed, capable of classifying keywords using a neural network in order to control a vehicle. The neural network model is executed on a microcontroller, which detects the keyword and sends the corresponding command to a second microcontroller responsible for controlling the vehicle’s peripherals. The execution of the AI model on a microcontroller is made possible through the Edge Impulse platform, where it is possible to manage the dataset, create, train, and test the model, and conclude by exporting the model in a format compatible with the microcontroller. The results obtained showed that, despite the hardware’s limitations, it was possible to achieve a good level of accuracy, and the model could also be extended to classify more keywords without interfering with the vehicle’s operation.