Optimización de trayectorias y tiempos para navegación autónoma de robots dentro de un proceso industrial aplicando Industria 4.0

Abstract

The present work is based on the design of a control algorithm for the optimization of trajectories and their travel time, implementing the model in a simulated environment for the autonomous navigation of robots, focusing for its development on industry-based tools. 4.0 and the application of neural networks to evaluate the actions executed by the controller in such a way that the error in the path of the trajectory is reduced, a reinforcement learning method is also added to the system that allows the model to know when an executed action was correct or incorrect, this is because its objective is to maximize its reward level, due to this the system will learn by exploring the environment to avoid obstacles and reach the objective, thus allowing the path to be followed to be optimal, The controller is based on the RMSprop optimizer algorithm, which allows it to give greater importance to the current paths than to the earlier, allowing learning to grow gradually, since over time the robot in its first training scenarios collides due to the fact that the amount of information is null or almost null, which is considered as an insufficient data source , however, as the training progresses, the robot, trying to increase its reward level, reaches the goal more frequently, giving greater importance to the routes where it began to learn than those where it collided. The communication of the system occurs through nodes controlled by a ROS master, this allows the exchange of information through messages published on topics, which gives rise to an adequate reading of the LIDAR sensor in charge of determining objects around the robot and a correct sending. of data by the DQN network to control the actions.