Aprendizado por reforço profundo para navegação de um veículo guiado automaticamente

Abstract

Automated Guided Vehicles (AGVs) are related to the problem of autonomous navigation in internal logistics, which is a very relevant area in the current industry. With the use of mobile robots, which are controlled without direct human interaction, it is possible to reduce costs, increase efficiency and facilitate the maintenance of internal logistics environments. Among the various solutions, there is the use of reinforcement learning applied to robot control. This learning technique has the characteristic of having an exchange of information between environment and agent, with the agent seeking goals and the environment returning rewards for each of its actions. Such characteristics allow learning to be conducted without prior knowledge of the environment, which is useful in the robotics field, due to its adaptability to dynamic and diverse environments. The present work aims to evaluate the Deep Q-learning (DQN) algorithm for use in autonomous navigation of a mobile robot in a simulation environment and to analyze the results obtained.