Resumen

Introducción In recent years, smart factories have transformed their processes into digital ones thanks to the integration of technologies that allow better communication and handling of information in production lines, giving rise to processes that handle a high amount of data, allowing the application of systems Based on machine learning and data control methodologies, nowadays predictive systems are highly important because they can statistically predict future critical conditions on production lines[1].Recently, in most globalized countries where industries are adapting their processes to industry 4.0, the use of statistical methods for the analysis of data models is considered, developing advanced regression and classification systems, it is considered that autonomous predictive models can be risky as they can lead to positive or negative changes within the manufacturing process [2], the use of systems based on machine learning are presented as a novel method that allows the analysis and design of algorithms that consider historical data to generate optimized models of control systems based on statistical regression, whose The purpose of this project is to use industry 4.0 bases for the development of a multiplatform and open source system that allows the control of a robot based on a DQN network by reinforcement learning, so that this device through training simulated be able to reconfigure its action parameters in real time to follow a new and optimal route avoiding mobile and fixed obstacles, with the aim of reaching a route to the goal in a reduced way and in less time.

Objetivos Design a control system based on neural networks that implements an algorithm based on the optimization of trajectories and that reduces the error due to collision or delays for the continuous operation of the mobile robot.

Método This research is based on a DQN (Deep Q Learning) network model to control the trajectory of a simulated mobile robot in Gazebo, the information exchanged between the environment and the agent will be through a ROS system and automatic learning will be by reward which will depend on the position and rotation of the robot with respect to the goal and if it completes the trajectory or collides.

Principales resultados Control system developed in open source that allows the control of trajectories in real time that implements an algorithm based on the optimization of trajectories and that reduces the error due to collision or delays for the continuous operation of the mobile robot.

Conclusiones The algorithm allows the training of the robot within the network to be simulated, which makes experimentation possible by sampling other optimization and error reduction parameters, which leads to determining the most suitable one for the reduction of trajectories, reducing the time of mobile robot operation.