Improving mobile robot navigation through odometry optimization using particle swarm optimization at kinematics model

• Ano de defesa: 2023
• Autor(a) principal: SANTOS, Lucas Henrique Cavalcanti
• Orientador(a): BARROS, Edna Natividade da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Pernambuco
• Programa de Pós-Graduação: Programa de Pos Graduacao em Ciencia da Computacao
• País: Brasil
• Palavras-chave em Português: Engenharia da computação; Odometria; Navegação autônoma; Cinemática; Robôs móveis; PSO
• Link de acesso: https://repositorio.ufpe.br/handle/123456789/51666

Autonomous navigation is crucial for mobile robots to move and interact with their surroundings. This requires the integration of intelligence, perception, and control in the robots. The first step in modelling the movement of robots is to create a kinematic model that explains how actuators influence their movement. The wheel velocity and the kinematic model are used to calculate the robot’s velocity and then the path traveled by integrating velocity over time, known as odometry. Odometry is the foundation of robotics navigation, but due to systematic errors in the kinematic model, it may have translation and rotation errors that accumulate over time. This study introduces a method to improve odometry accuracy using Particle Swarm Optimization (PSO). The method employs wheel velocity data and an inertial sensor to optimize the robot’s kinematic model. The technique involves experiments with the robot to record its velocity and position and to simulate the traveled path using the kinematic model. The simulation is evaluated using root-mean-square error compared to the ground-truth positions. The PSO method optimizes the kinematic parameters by minimizing the error between the simulation and the ground-truth positions. The proposed optimization technic improved odometry by 75%, from a mean squared error of 0.37 to 0.09. The result showed that the final position of a 6-meter path had an error of less than 5 cm, while previous methods achieved a minimum error of 10 cm. The optimization allows robots to navigate with greater autonomy without external information or additional sensors and is also efficient for low-power embedded computers.