Simulación de Controladores de Seguimiento de Trayectoria Basados en Inteligencia Artificial para un Robot Humanoide NAO : Simulación de Controladores de Seguimiento de Trayectoria Basado en Redes Neuronales para el Seguimiento de la Trayectoria de un Robot Humanoide.

Abstract

This work presents a machine learning technique based on soft computing for the trajectory tracking control of a humanoid robot using neural networks. It is proposed the development of a kinematic controller based on neural networks designed from the direct kinematics of a mobile robot to obtain better accuracy and performance in trajectory tracking. The neural controller starts from an initial tuning performed by the designer, while in the application phase using the backpropagation algorithm, the values of the neural network weights will be updated which is equivalent to the controller gains. For the learning stage the algorithm constantly evaluates the closed-loop behavior of the system. The results obtained will be analyzed and compared with a conventional kinematic controller, demonstrating qualitatively and quantitatively that the proposed approach presents improvements in terms of performance under different trajectories.