Sistema de control basado en Inteligencia Artificial aplicado al seguimiento de trayectoria de un brazo robótico industrial para corte de materiales

Abstract

Abstract: This project addresses the design and implementation of a trajectory tracking system for material cutting, based on artificial intelligence, using Convolutional Neural Networks (CNN) developed with Pytorch and TensorFlow libraries. Additionally, a reinforcement learning module is integrated within the HiwinReacher environment of IsaacGym/Sim. The hardware infrastructure comprises the Epson ProSix C12 robot, its RC 700A controller, a 1080p HD webcam, and a Lenovo Legion 7 computer. The software architecture includes Python for developing CNN algorithms with Pytorch and TensorFlow. The IsaacGym/Sim reinforcement learning simulation environment utilizes Python and Pytorch, focusing on reward and penalty strategies for the Hiwin RA620 robot. Ten evaluation tests were conducted for the algorithms with Pytorch and TensorFlow, both approaches demonstrate robustness across various trajectories and encounter difficulties with abrupt changes in direction. The results highlight the excellent performance of the CNNs in the implementation of the tracking system, with each library revealing its strengths and specific weaknesses.