Estudo da aplicabilidade de algoritmos de aprendizado por reforço em um pêndulo invertido

Abstract

This work addresses an exploratory research on the applicability of reinforcement learning algorithms (RLA) in an inverted pendulum system, a classic problem of an unstable open loop system. The research carried out, addressed the study and application of some specific algorithms such as Q-Learning and Deep-Q-Networks. The work presents a theoretical research of the algorithms until the software implementation of the same ones. In a second step, a physical prototype of the pendulum system was built, in which stochastic filters were used in order to enhance the sensor reading quality, leaving open a possibility of continuity of the development for future works.