Modelado de entornos con técnicas basadas en redes de petri borrosas para la exploración y planificación de robots autónomos

Abstract

El objetivo de esta tesis es el desarrollo de una arquitectura de control de robots autónomos en la que se combinan tareas de navegación, aprendizaje del modelo del entorno mediante exploración y planificación de rutas. La navegación se plantea desde un punto de vista reactivo que dota al sistema con características de robustez y flexibilidad. La planificación de rutas incluye elementos reflexivos sobre los que se apoya la toma de decisiones en la elaboración del plan. La integración de ambos planteamientos supone necesariamente tomar como referencia una aproximación híbrida para nuestra arquitectura de control. Los mecanismos a los que hacemos referencia han sido probados en un robot móvil nomad-200. Los entornos para las pruebas son los laboratorios y pasillos de la facultad de informática de la universidad politécnica de Madrid. The aim of this thesis is the development of a control architecture for autonomous robots that combines tasks for navigating, world modelling through exploration and route planning. The navigation is established from the reactive or behaviour-based point of view that endows to the system with robustness and flexibility. The route planning includes deliberative elements for making decisions in the accomplishment of the plan. The integration of reactive and deliberative elements involves taking the framework of a hybrid approximation for the control architecture. The control architecture is based on a cyclic executive for guaranteeing the real time activities of the robot. The processes constitute the elements of the lowest abstraction level in the architecture. The grouping of processes forms the elements of the second abstraction level: the behaviours. A behaviour is identified as an operation mode of the control system that must achieve a short-term goal. The behaviours can be inhibited, can momentarily stay out of the system. This inhibition mechanism originates the different strategies that can be used for navigating. The strategies are included in the next abstraction level of the control architecture and their actions can be observed at medium-term. The behaviour selection for resolving a same problem with different results defines the elements of the upper abstraction level: the activities. An activity causes an operation mode of the system that can be observed at long-term. We distinguish three kind of activities. In the environment directed activity the robot wanders in the environment without a long-term goal, only the robot must avoid collisions with the obstacles. In the exploration directed activity the robot explores and constructs an environment model. The robot decisions in this activity are determined by the need of improvement the knowledge about the world. In the suggestion directed activity the robot uses the environment model for planning routes to a place. Each activity requires the definition of specific mechanisms. Specifically, we have to resolve the problems of the collision-free navigation, the detection and recognition of reference places in the environment, the definition of schemes that determine the actions during the exploration stage, and the efficient route planning to specific places in the environment. The control architecture and the above mentioned mechanisms have been tested on a NOMAD-200 mobile robot platform. The office-like environments for testing are the laboratories and corridors of the Computer Science Faculty at the Technical University of Madrid.