Automatic State Estimation of an Over-Sensored Robotic Manipulator

• Autor(a) principal: João Pedro Ribeiro Moreira
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/133037
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

There is an increasing demand of robotic manipulators for performing more complex and versatile tasks. In order to fulfill this need, expeditious calibration and estimation techniques are required as a first step for the correct usage of the manipulator. Only after these problems are solved, can it be used for higher level tasks such as generic tool placement and object manipulation. There are currently several techniques for automatic calibration of a wide variety of sensors, as well as several filters to fuse their data into useful information. This dissertation aims at finding a subset of these algorithms that could be used in a generic manipulator and should allow for its prompt use. The techniques used were chosen with the purpose of being modular and therefore usable in a wide variety of manipulators. They also assume a minimal amount of requirements necessary for their use, making them suitable for an unequipped user. A previously developed manipulator is used to realistically test the performance of the implemented methods. It is equipped with incremental encoders, inertial measurement units and load cells. A calibration methodology for the inertial sensors is described and the calibrated measurements are used together with the encoders' to determine the pose of the manipulator. Two models for the representation of the pose of the manipulator are described and used in the state estimation problem. One defines the state vector as the dynamics of the angles of each joint. The other uses the orientation of each link in an inertial frame independently. The state of the first model is estimated with the Unscented Kalman Filter and the second one with the Multiplicative Extended Kalman Filter. The results of implementation are tested and some performance metrics are obtained using both the algorithms' output and an external system.