Direção autônoma com aprendizado por reforço e TinyML

Abstract

Machine learning plays an important role in the development of autonomous vehicles. This work aims to tackle the autonomous driving problem using only images in a low-cost mini-vehicle prototype. For this, reinforcement learning algorithms were employed to train the artificial intelligence model in a simulated virtual environment and posteriorly TinyML techniques were used to make it compatible to be executed by microcontrollers. This resulted in a model capable of successfully drive the vehicle in a road and a comparison between the conventional cloud based and TinyML approaches showed a significant reduction in the system latency.