Aplicación del algoritmo de aprendizaje por refuerzo Q-Learning para la generación de trayectorias óptimas en plataformas robóticas

Abstract

Technological innovation is fundamental for business development and efficiency in Industry 4.0, which demands staying updated with the latest technologies. The rapid growth in the use of artificial intelligence (AI) offers high benefits and low operational costs, creating more stable and automated work environments. However, one of the greatest challenges in applying AI is the complexity of trajectory planning for mobile robots, as their behavior varies according to the scenario and algorithms used, making it difficult to compare learning and performance between different methods. This research, as part of the project titled "Use of Deep Learning Techniques for Trajectory Planning of Mobile Robots within an Industrial Process," developed a trajectory planning algorithm using Q-learning and a multi-agent system that collaborate in decision-making. This algorithm employs odometry and laser sensor signals to manage states and rewards. Tests were conducted in a ROS simulation environment and replicated in a real-world scenario with the KUKA Youbot robot to implement the algorithm's actions. The simulated environment recreates a space with obstacles, where a master agent evaluates the decisions made by the odometry agent and the laser sensor agent to autonomously make a final decision. This contributes to the comparison of AI algorithms in terms of efficiency and effectiveness within the mentioned project, laying a foundation for future improvements in trajectory planning on mobile platforms.