A self-localization system based on topometric and topographic map of outdoor semi-structured environments

• Ano de defesa: 2025
• Autor(a) principal: Horita, Luiz Alberto Hiroshi
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Altitude; Ambiente Externo; Imagens; Images; Localização; Localization; Navegação Topométrica; Outdoor; Topometric Navigation
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-22052025-102507/

There are already plenty of works advancing researches in mobile robotics, mainly for urban scenarios, with great solutions for the autonomous concept in all of its main tasks of localization, mapping, and navigation. However, advancing technologies applied to rural environments is also important, to improve security, sustainability, and productivity, since in the next decades the demand for basic necessities will be much higher due to the worlds population growth. Some companies are already developing and testing some autonomous tractors in the field, although all of them are guided mainly by GNSS. The researches applied in rural mobile robots are, most of them, focused on precise localization in the field, so it can navigate in line and does not run over the plantation, however, localization between an operation base and the plantation does not need a sub-metric precision since it can navigate safely within the road. Thus, a discrete localization approach based on a topological map must be enough in this context. For the localization task, a robot needs a previous knowledge of the environment and a perception system, which are given by maps and sensors. The GNSS-based localization methods, largely used, can be inconvenient when the sensor is submitted to signal blockage or reflection conditions, due to undesired weather condition or large obstacles. To overcome the GNSS-denied cases, some researchers proposed using other local-sensors, such as LiDAR, cameras, IMU, wheel encoders, among which, LiDAR is usually mechanically complex and very expensive, although robust to different environmental conditions, such as light variation. Recurrent methods for localization involve using road geometry identification through curbs, multi-session appearance-based maps for place recognition, processing and sharing multi-session maps on clouds. However, in rural environments, the roads are not well defined by curbs and with regular width, the appearance can change a lot due to the own natural changes and the changes of plantation cultures, and the telecommunication infrastructure for internet connection is not reliable. Then, this project proposes a hybrid topometric-topographic approach for self-localization, in which a camera will be used for road geometry identification as landmarks for pose correction, encoders and IMU will be used for odometry, and barometric altimeter will be used for altitude offset estimation to constraint and optimize the pose estimation. With this proposal, a more affordable solution is expected, using long-term, invariant and easy to map features from the environment, reducing dramatically the time spent for map construction, compared to the state-of-the-art proposals reviewed.