Desenvolvimento de ambiente para teste e treinamento de agentes inteligentes por aprendizado por reforço

Abstract

This work proposes the development of a software environment dedicated to testing and training intelligent agents that utilize reinforcement learning algorithms. Inspired by the Gym and Gymnasium libraries maintained by OpenAI, the project aims to create an alternative and intuitive tool for developers and students to explore and implement reinforcement learning algorithms. The main focus is on reinforcement learning, where the agent interacts with the environment through actions and rewards, adjusting its behavior to maximize efficiency over time. The proposed environment seeks to offer a flexible and user-friendly platform, enabling experimentation with different algorithms and the evaluation of their performance in simulated tasks. The work also includes the implementation of a functional prototype, tested with the Deep Q-Network (DQN) algorithm, aiming to validate the effectiveness of the developed environment. The developed tool represents an advancement in knowledge in the field by facilitating learning and research in artificial intelligence, particularly in the context of reinforcement learning.