| Abstract |
A primary goal in robotics research is to provide means for mobile platforms to perform autonomously within their environment. Depending on the task at hand, autonomous performance can be defined as the execution by the robot, without human interaction, of certain navigational tasks, such as finding its position or planning its way towards a specific target. In mobile robotics literature, commonly addressed navigation tasks include the localization, mapping, path-planning and obstacle avoidance tasks. Solving any of these tasks is a hard problem by itself. The reason stems from the inherent complexity associated with both the robot and its environment, each of them being an extremely complex dynamical system with many degrees of freedom. Adoption of appropriate, application-depended constrains, as well as careful modelling of both the robot's and the environment's dynamics, may contribute in reducing the complexity and making the problem tractable. In line with the above, we propose in this thesis a probabilistic framework for indoor (structured environment) mobile robot navigation, based on switching state-space models (SSSMs). SSSMs are hybrid models, combining two of the most successful classes of models already used for mobile robot navigation, i.e. discrete state models (hidden Markov models) and continuous state models (Kalman filter based models). In discrete models the robot's state-space is discretized into a finite number of states. The discretization permits the arithmetic computation of robot dynamics, thus eliminating the need for assumptions about any of the involved distributions. On the other hand, in continuous state models, robot's state-space is assumed to be continuous. States, as well as all involved distributions are represented and computed analytically as gaussian distributions (usually via the Kalman filter update equations). Comparative results among these two major classes of approaches indicate the superiority of the Kalman filter approaches with respect to computational efficiency, scalability, and accuracy. On the other hand, discrete state approaches have been proved to be more robust in the presence of noise and/or unreliable odometry information. Based on SSSMs, a complete framework for mobile robot navigation is presented in this thesis. For localization and map building, two novel algorithms have been developed respectively, that combine advantages from both discrete and continuous approaches while relaxing, at the same time, inherent limitations in each of them. Both algorithms utilize native properties of SSSMs in order to let discrete Markovian dynamics handle the topological aspects of the localization and mapping problems, while letting Kalman filters handle the metric aspects. More specifically, for localization, discrete Hidden Markov Model (HMM) update equations are used in order to update the probabilities assigned to a fixed, small number of discrete states, while Kalman filter based trackers, operating within each discrete state, are used in order to provide accurate metric information. For map building, an EM-mapping algorithm is proposed that utilizes the global localization properties of the above mentioned localizer in order to correctly identify already mapped areas and, hence, ensure topological correctness of produced maps. For global path planning, a dynamic programming algorithm, namely the value iteration algorithm, has been adapted to our framework, while, for short term motion planning and collision avoidance, an algorithm based on the vector field histogram method has been employed. To facilitate the implementation of the proposed navigation framework, a powerful set of high-level features is utilized, namely line segments and corner points. These features are extracted directly from laser range data. For detecting possibly hazardous obstacles, invisible by the laser range finder, visual depth information is also used by the motion planning and obstacle avoidance module. Visual range data are extracted though stereovision, wherever range data are detected to be inconsistent with the visual information. Experimental results have shown the applicability and effectiveness of the proposed framework for indoor navigation tasks where the robustness and the global localization capabilities of the discrete approaches and the efficiency and accuracy of the Kalman filter based approaches are required at the same time.
|
Collections
