Análise comparativa de aplicações de sistemas de inferência nebulosa na navegação de robôs móveis em ambientes desconhecidos

• Ano de defesa: 2023
• Autor(a) principal: Leandro Daros Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA; Programa de Pós-Graduação em Engenharia Elétrica; UFMG
• Palavras-chave em Português: Campo potencial artificial nebuloso; Múltiplo sistema de inferência nebulosa; Sistema de inferência neuro-nebuloso adaptativo; Evitar obstáculos; Navegação de robôs móveis; Engenharia elétrica; Robôs; Robótica; Inferência (Lógica); Algoritmos - Simulação por computador; Caminhada
• Link de acesso: http://hdl.handle.net/1843/74155

The navigation task of a mobile robot on a collision-free path is a central problem in mobile robotics, and when it comes to navigating in unknown environments, this task can become challenging both in terms of ensuring its completion and in terms of implementing the solution, so that several studies on the subject have been carried out in the past few decades. This work seeks to investigate the application of fuzzy inference systems (FIS) in solving the navigation task of a non-holonomic Two-Wheeled Mobile Robot (TWMR) in unknown environments, dividing this task into two behaviors: I) go to goal; and II) obstacle avoidance. A comparative analysis of three mobile robot navigation methods that uses FIS was carried out: I) Fuzzy Artificial Potential Field (FAPF) in which a fuzzy inference system is used to weigh the value of the attractive force and repulsive force of the artificial potential field method; II) Multiple Fuzzy Inference System (MFIS) in which two FIS controllers are used for each behavior, seeking to directly control the robot’s left and right wheel velocities; and III) Adaptive Neuro-Fuzzy Inference System (ANFIS) in which a hybrid algorithm is used to tune and update the premise and consequent parameters of a FIS based on the gradient descent method and the least mean squares method, seeking to determine the robot angular velocity in the obstacle avoidance behavior. To validate and compare the methods, six scenarios were used, in which the difficulty level of the task increases from one scenario to another: I) an environment with a narrow gap; II) an environment with a narrow path; III) an environment with an obstacle generating a local minima problem; IV) a cluttered environment; V) a dense cluttered environment; and VI) an environment with an obstacle generating a local minima problem followed by a narrow gap and a narrow path. The simulation results were presented using MATLAB software integrated with CoppeliaSim and the classical Artificial Potential Field (APF) method was used as a basis for comparison. Statistical analyses were performed using Jamovi software. The FAPF method presented a good performance related to the traveled distance, however, it presented a low performance related to the task execution time. The MFIS method presented good performance related to task execution time, however, in certain cases it presented low performance in relation to the traveled distance. The ANFIS method presented good performance both in terms of traveled distance and task execution time.