Aprendizado por reforço profundo paralelo aplicado a navegação de drones em três dimensões

Abstract

The work presented in this text consists of a study on deep reinforcement learning applied to mapless navigation of drones in three dimensions. For this, two methods were used, D4PG and DSAC, both with parallelization and distribution capabilities. In addition, a prioritized memory was employed in each of the methods. The drone used in the study was the Hydrone, a hybrid quadrotor drone that operates only in the air. The methods were trained in environments of varying complexity, from obstacle-free environments to environments with multiple obstacles in three dimensions. The results showed a good learning capacity of the methods, which were able to achieve the vast majority of the proposed objectives. Additionally, the generalization of the methods was tested by applying them in unseen environments, which showed that the methods had good generalization capacity. In summary, this work presented a study on deep reinforcement learning applied to mapless navigation of drones in three dimensions, with promising results and potential applications in various contexts related to robotics and autonomous air navigation.