Desarrollo de un algoritmo de aprendizaje computacional para el control de movimiento del robot Emerge

Abstract

This thesis presents a motion control algorithm for the Emerge robot based on computational learning, for this purpose a motion control algorithm was developed implementing deep learning using the TensorFlow - Keras library in the training of neural networks for chain, L-type and block or quadpod morphologies; the second motion control applied for the comparison of the developed technique were the motion tables following the technique of (Morantes et al., 2020) applying it for the three morphologies previously mentioned. To elaborate the local controller of the robot, a Raspberry PI 3 was implemented together with an Arduino Uno, two ultrasonic sensors and two voltage regulators, one for the power supply of the Raspberry and the other for the power supply of the sensors and the Arduino. With this work, a new technique for motion control in modular robots is incorporated by applying artificial intelligence and opening the field of study in this control technique for these robots in the morphologies chain, ELE and block type, as well as improving the displacement times and the offset angle with respect to the origin in the Y axis in the three morphologies with respect to the comparison technique; the application of multilayer neural networks in the motion control of modular robots is also validated.