Hybrid mapping for static and non-static indoor environments

• Autores: Clara Gómez Blázquez
• Directores de la Tesis: Ramón I. Barber Castaño (dir. tes.)
• Lectura: En la Universidad Carlos III de Madrid ( España ) en 2021
• Idioma: español
• Tribunal Calificador de la Tesis: Carlos Balaguer Bernaldo de Quirós (presid.), Javier González Jiménez (secret.), Nancy Marie Amato (voc.)
• Programa de doctorado: Programa de Doctorado en Ingeniería Eléctrica, Electrónica y Automática por la Universidad Carlos III de Madrid
• Enlaces
  • Tesis en acceso abierto en: e-Archivo
• Resumen

  • Indoor environments populated by humans, such as houses, offices or universities, involve a great complexity due to the diversity of geometries and situations that they may present. Apart form the size of the environment, they can contain multiple rooms distributed into floors and corridors, repetitive structures and loops, and they can get as complicated as one can imagine. In addition, the structure and situations that the environment present may vary over time as objects could be moved, doors can be frequently opened or closed and places can be used for different purposes. Mobile robots need to solve these challenging situations in order to successfully operate in the environment. The main tools that a mobile robot has for dealing with these situations relate to navigation and perception and comprise mapping, localization, path planning and map adaptation. In this thesis, we try to address some of the open problems in robot navigation in non-static indoor environments. We focus on house-like environments as the work is framed into the HEROITEA research project that aims attention at helping elderly people with their everyday-life activities at their homes. This thesis contributes to HEROITEA with a complete robotic mapping system and map adaptation that grants safe navigation and understanding of the environment. Moreover, we provide localization and path planning strategies within the resulting map to further operate in the environment.

    The first problem tackled in this thesis is robot mapping in static indoor environments. We propose a hybrid mapping method that structures the information gathered from the environment into several maps. The hybrid map contains diverse knowledge of the environment such as its structure, the navigable and blocked paths, and semantic knowledge, such as the objects or scene in the environment. All this information is separated into different components of the hybrid map that are interconnected so the system can, at any time, benefit from the information contained in every component. In addition to the conceptual conception of the hybrid map, we have also developed building procedures and an exploration algorithm to autonomous build the hybrid map.

    However, indoor environments populated by humans are far from being static as the environment may change over time. For this reason, the second problem tackled in this thesis is the adaptation of the map to non-static environments. We propose an object-based probabilistic map adaptation that calculates the likelihood of moving or remaining in its place for the different objects in the environment.

    Finally, a map is just a description of the environment whose importance is mostly related to how the map is used. In addition, map representations are more valuable as long as they offer a wider range of applications. Therefore, the third problem that we approach in this thesis is exploiting the intrinsic characteristics of the hybrid map in order to enhance the performance of localization and path planning methods. The particular objectives of these approaches are precision for robot localization and efficiency for path planning in terms of execution time and traveled distance.

    We evaluate our proposed methods in a diversity of simulated and real-world indoor environments. In this extensive evaluation, we show that hybrid maps can be efficiently built and maintained over time and they open up for new possibilities for localization and path planning. In this thesis, we show an increase in localization precision and robustness and an improvement in path planning performance.

    In sum, this thesis makes several contributions in the context of robot navigation in indoor environments, and especially in hybrid mapping. Hybrid maps offer higher efficiency during map building and other applications such as localization and path planning. In addition, we highlight the necessity of dealing with the dynamics of indoor environments and the benefits of combining topological, semantic and metric information to the autonomy of a mobile robot.