Algoritmo de operación para robot asistencial autónomo enfocado a alimentación

The present work outlines the implementation of an algorithm for the control of a robot assistant, which focuses on assisted feeding. The algorithm applied to the robot has 3 fundamental pillars for its operation: detection of existence or not of food, decision making in the face of situations of obstruction of its trajectory with the user’s hand and execution of the feeding task reaching the point of contact. For this, artificial intelligence techniques based on deep learning are used, as well as the use of a RGB-D camera in charge of capturing information from the environment, so that it can be processed to perform assistance. For the detection of the states of the mouth, to know if the user is chewing or waiting for food, a neuronal network with long short-term memory was used, obtaining a 99.3% accuracy in its validation tests. The recognition of the existence or not of food in the dish, was made with a convolutional neuronal network, which reached a performance of 98.7%. Regarding the detection of obstacles, the user's hand is defined as the obstacle, which is recognized and localized by means of a convolutional neural network based on regions, achieving a 77.4% mean precision in the interception over union of the location of the estimated boxes against the original (or labeled) ones. With the functionalities implemented, a graphic user interface is created in which all the algorithms are coupled within a single system, to generate the task of assisting a user in their feeding. The emulation of this task is done through the interaction of the real environment, where the user is, and a simulated environment, where the robot will perform the movements, passing the real three-dimensional information of the situations presented to the simulated environment. With this, the system's performance tests are carried out, demonstrating a high performance of each of the functions within 3 variations of the real environment, altering the lighting, ranging from a low quality of lighting to a bright one.