In principle, general approaches to optimal nonlinear filtering can be described in a unified way from the recursive Bayesian approach. The central idea to this recur- sive Bayesian estimation is to determine the probability density function of the state vector of the nonlinear systems conditioned on the available measurements. However, the optimal exact solution to this Bayesian filtering problem is intractable since it requires an infinite dimensional process. For practical nonlinear filtering applications approximate solutions are required. Recently efficient and accurate approximate non- linear filters as alternatives to the extended Kalman filter are proposed for recursive nonlinear estimation of the states and parameters of dynamical systems. First, as sampling-based nonlinear filters, the sigma point filters, the unscented Kalman fil- ter and the divided difference filter are investigated. Secondly, a direct numerical nonlinear filter is introduced where the state conditional probability density is calcu- lated by applying fast numerical solvers to the Fokker-Planck equation in continuous- discrete system models. As simulation-based nonlinear filters, a universally effective algorithm, called the sequential Monte Carlo filter, that recursively utilizes a set of weighted samples to approximate the distributions of the state variables or param- eters, is investigated for dealing with nonlinear and non-Gaussian systems. Recentparticle filtering algorithms, which are developed independently in various engineer- ing fields, are investigated in a unified way. Furthermore, a new type of particle filter is proposed by integrating the divided difference filter with a particle filtering framework, leading to the divided difference particle filter. Sub-optimality of the ap- proximate nonlinear filters due to unknown system uncertainties can be compensated by using an adaptive filtering method that estimates both the state and system error statistics. For accurate identification of the time-varying parameters of dynamic sys- tems, new adaptive nonlinear filters that integrate the presented nonlinear filtering algorithms with noise estimation algorithms are derived. For qualitative and quantitative performance analysis among the proposed non- linear filters, systematic methods for measuring the nonlinearities, biasness, and op- timality of the proposed nonlinear filters are introduced. The proposed nonlinear optimal and sub-optimal filtering algorithms with applications to spacecraft orbit es- timation and autonomous navigation are investigated. Simulation results indicate that the advantages of the proposed nonlinear filters make these attractive alterna- tives to the extended Kalman filter.

Acoustic source (speaker) tracking in the room environment plays an important role in many speech and audio applications such as multimedia, hearing aids and hands-free speech communication and teleconferencing systems; the position information can be fed into a higher processing stage for high-quality speech acquisition, enhancement of a specific speech signal in the presence of other competing talkers, or keeping a camera focused on the speaker in a video-conferencing scenario. Most of existing systems focus on the single source tracking problem, which assumes one and only one source is active all the time, and the state to be estimated is simply the source position. However, in practical scenarios, multiple speakers may be simultaneously active, and the tracking algorithm should be able to localise each individual source and estimate the number of sources. This thesis contains three contributions towards solutions to multiple acoustic source tracking in a moderate noisy and reverberant environment. The first contribution of this thesis is proposing a time-delay of arrival (TDOA) estimation approach for multiple sources. Although the phase transform (PHAT) weighted generalised cross-correlation (GCC) method has been employed to extract the TDOAs of multiple sources, it is primarily used for a single source scenario and its performance for multiple TDOA estimation has not been comprehensively studied. The proposed approach combines the degenerate unmixing estimation technique (DUET) and GCC method. Since the speech mixtures are assumed window-disjoint orthogonal (WDO) in the time-frequency domain, the spectrograms can be separated by employing DUET, and the GCC method can then be applied to the spectrogram of each individual source. The probabilities of detection and false alarm are also proposed to evaluate the TDOA estimation performance under a series of experimental parameters. Next, considering multiple acoustic sources may appear nonconcurrently, an extended Kalman particle filtering (EKPF) is developed for a special multiple acoustic source tracking problem, namely “nonconcurrent multiple acoustic tracking (NMAT)”. The extended Kalman filter (EKF) is used to approximate the optimum weights, and the subsequent particle filtering (PF) naturally takes the previous position estimates as well as the current TDOA measurements into account. The proposed approach is thus able to lock on the sharp change of the source position quickly, and avoid the tracking-lag in the general sequential importance resampling (SIR) PF. Finally, these investigations are extended into an approach to track the multiple unknown and time-varying number of acoustic sources. The DUET-GCC method is used to obtain the TDOA measurements for multiple sources and a random finite set (RFS) based Rao-blackwellised PF is employed and modified to track the sources. Each particle has a RFS form encapsulating the states of all sources and is capable of addressing source dynamics: source survival, new source appearance and source deactivation. A data association variable is defined to depict the source dynamic and its relation to the measurements. The Rao-blackwellisation step is used to decompose the state: the source positions are marginalised by using an EKF, and only the data association variable needs to be handled by a PF. The performances of all the proposed approaches are extensively studied under different noisy and reverberant environments, and are favorably comparable with the existing tracking techniques.

We introduce Simultaneous Localization and Tracking (SLAT), the problem of tracking a target in a sensor network while simultaneously localizing and calibrating the nodes of the network. Our proposed solution, LaSLAT, is a Bayesian filter providing on-line probabilistic estimates of sensor locations and target tracks. It does not require globally accessible beacon signals or accurate ranging between the nodes. When applied to a network of 27 sensor nodes, our algorithm can localize the nodes to within one or two centimeters.

Ces travaux se focalisent sur une problématique fondamentale de la robotique autonome: l'estimation d'état. En effet, la plupart des approches actuelles permettant à un robot autonome de réaliser une tâche requièrent tout d'abord l'extraction d'une information d'état à partir de mesures capteurs bruitées. Ce vecteur d'état contient un ensemble de variables caractérisant le système à un instant t, comme la position du robot, sa vitesse, etc. En robotique comme dans de nombreux autres domaines, le filtrage bayésien est devenu la solution la plus populaire pour estimer l'état d'un système de façon robuste et à haute fréquence. Le succès du filtrage bayésien réside dans sa relative simplicité, que ce soit dans la mise en oeuvre des équations récursives de filtrage, ou encore dans la représentation simplifiée et intuitive du système au travers du modèle de Markov caché d'ordre 1. Généralement, un filtre bayésien repose sur une description minimaliste de l'état du système. Cette représentation simplifiée permet de conserver un temps d'exécution réduit, mais est également la conséquence de notre compréhension partielle du fonctionnement du système physique. Tous les aspects inconnus ou non modélisés du système sont ensuite représentés de façon globale par l'adjonction de composantes de bruit. Si ces composantes de bruit constituent une représentation simple et unifiée des aspects non modélisés du système, il reste néanmoins difficile de trouver des paramètres de bruit qui sont pertinents dans tous les contextes. En effet, à l'opposé de ce principe de modélisation, la problématique de navigation autonome pose le problème de la multiplicité d'environnements différents pour lesquels il est nécessaire de s'adapter intelligemment. Cette problématique nous amène donc à réviser la modélisation des aspects inconnus du systèmes sous forme de bruits stationnaires, et requiert l'introduction d'une information de contexte au sein du processus de filtrage. Dans ce cadre, ces travaux se focalisent spécifiquement sur l'amélioration du modèle état-observation sous-jacent au filtre bayésien afin de le doter de capacités d'adaptation vis-à-vis des perturbations contextuelles modifiant les performances capteurs. L'objectif principal est donc ici de trouver l'équilibre entre complexité du modèle et modélisation précise des phénomènes physiques représentés au travers d'une information de contexte. Nous établissons cet équilibre en modifiant le modèle état-observation afin de compenser les hypothèses simplistes de bruit stationnaire tout en continuant de bénéficier du faible temps de calcul requis par les équations récursives. Dans un premier temps, nous définissons une information de contexte basée sur un ensemble de mesures capteurs brutes, sans chercher à identifier précisément la typologie réelle de contextes de navigation. Toujours au sein du formalisme bayésien, nous exploitons des méthodes d'apprentissage statistique pour identifier une distribution d'observation non stationnaire et dépendante du contexte. cette distribution repose sur l'introduction de deux nouvelles composantes: un modèle destiné à prédire le bruit d'observation pour chaque capteur, et un modèle permettant de sélectionner un sous-ensemble de mesures à chaque itération du filtre. Nos investigations concernent également l'impact des méthodes d'apprentissage: dans le contexte historique du filtrage bayésien, le modèle état-observation est traditionnellement appris de manière générative, c'est à dire de manière à expliquer au mieux les paires état-observation contenues dans les données d'apprentissage. Cette méthode est ici remise en cause puisque, bien que fondamentalement génératif, le modèle état-observation est uniquement exploité au travers des équations de filtrage, et ses capacités génératives ne sont donc jamais utilisées[...] / Prevalent approaches for endowing robots with autonomous navigation capabilities require the estimation of a system state representation based on sensor noisy information. This system state usually depicts a set of dynamic variables such as the position, velocity and orientation required for the robot to achieve a task. In robotics, and in many other contexts, research efforts on state estimation converged towards the popular Bayes filter. The primary reason for the success of Bayes filtering is its simplicity, from the mathematical tools required by the recursive filtering equations, to the light and intuitive system representation provided by the underlying Hidden Markov Model. Recursive filtering also provides the most common and reliable method for real-time state estimation thanks to its computational efficiency. To keep low computational complexity, but also because real physical systems are not perfectly understood, and hence never faithfully represented by a model, Bayes filters usually rely on a minimum system state representation. Any unmodeled or unknown aspect of the system is then encompassed within additional noise terms. On the other hand, autonomous navigation requires robustness and adaptation capabilities regarding changing environments. This creates the need for introducing contextual awareness within the filtering process. In this thesis, we specifically focus on enhancing state estimation models for dealing with context-dependent sensor performance alterations. The issue is then to establish a practical balance between computational complexity and realistic modelling of the system through the introduction of contextual information. We investigate on achieving this balance by extending the classical Bayes filter in order to compensate for the optimistic assumptions made by modeling the system through time-homogeneous distributions, while still benefiting from the recursive filtering computational efficiency. Based on raw data provided by a set of sensors and any relevant information, we start by introducing a new context variable, while never trying to characterize a concrete context typology. Within the Bayesian framework, machine learning techniques are then used in order to automatically define a context-dependent time-heterogeneous observation distribution by introducing two additional models: a model providing observation noise predictions and a model providing observation selection rules.The investigation also concerns the impact of the training method we choose. In the context of Bayesian filtering, the model we exploit is usually trained in the generative manner. Thus, optimal parameters are those that allow the model to explain at best the data observed in the training set. On the other hand, discriminative training can implicitly help in compensating for mismodeled aspects of the system, by optimizing the model parameters with respect to the ultimate system performance, the estimate accuracy. Going deeper in the discussion, we also analyse how the training method changes the meaning of the model, and how we can properly exploit this property. Throughout the manuscript, results obtained with simulated and representative real data are presented and analysed.

京都大学 / 新制・課程博士 / 博士(工学) / 甲第24613号 / 工博第5119号 / 新制||工||1979(附属図書館) / 京都大学大学院工学研究科原子核工学専攻 / (主査)教授 村上 定義, 教授 横峯 健彦, 教授 宮寺 隆之 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

Um sistema artificial pode usar raciocínio espacial qualitativo para inferir informações sobre seu ambiente tridimensional a partir de imagens bidimensionais. Inferências realizadas com base em raciocínio espacial qualitativo devem ser capazes de lidar com incertezas. Neste trabalho investigamos a utilização de técnicas probabilísticas para tornar o raciocínio espacial qualitativo mais robusto a incertezas e aplicável a agentes móveis em ambientes reais. Investigamos uma formalização de raciocínio espacial com lógica de descrição probabilística em um subdomínio de tráfego. Desenvolvemos também um método que combina raciocínio espacial qualitativo com um filtro Bayesiano para desenvolver dois sistemas que foram aplicados na auto localização de um robô móvel. Executamos dois experimentos de auto localização; um utilizando a teoria de relações qualitativas percebíveis sobre sombra com filtro Bayesiano; e outro utilizando o cálculo de oclusão de regiões e o cálculo de direção com filtro Bayesiano. Ambos os sistemas obtiveram resultados positivos onde somente o raciocínio espacial qualitativo não foi capaz de inferir a localização do robô. Os experimentos com dados reais mostraram robustez aos ruídos e à informação parcial. / An artificial system can use qualitative spatial reasoning to obtain information about its tridimensional environment, from bi-dimensional images. Inferences produced by qualitative spatial reasoning must be able to deal with uncertainty. This work investigates the use of probabilistic techniques to make qualitative spatial reasoning more robust against uncertainty, and better applicable to mobile agents in real environments. The work investigates a formalization of spatial reasoning using probabilistic description logics in a traffic domain. Additionally, a method is presented that combines qualitative spatial reasoning with a Bayesian filter, to develop two systems that are applied to self-localization of mobile robots. Two experiments are described; one using the theory of perceptual qualitative relations about shadows; the other using occlusion calculus and direction calculus. Both systems are combined with a Bayesian filter producing positive results in situations where qualitative spatial reasoning alone cannot infer robot location. Experiments with real data show robustness to noise and partial information.

Localization and tracking objects or people in real time in indoor environments have gained great importance. In the literature and market, many different location estimation and tracking solutions using received signal strength indication (RSSI) are proposed. But there is a lack of information on the comparison of these techniques revealing their weak and strong behaviors over each other. There is a need for the answer to the question / &ldquo / which localization/tracking method is more suitable to my system needs?&rdquo / . So, one purpose of this thesis is to seek the answer to this question. Hence, we investigated the behaviors of commonly proposed localization methods, mainly nearest neighbors based methods, grid based Bayesian filtering and particle filtering methods by both simulation and experimental work on the same test bed. The other purpose of this thesis is to propose an improved method that is simple to install, cost effective and moderately accurate to use for real life applications. Our proposed method uses an improved type of sampling importance resampling (SIR) filter incorporating automatic calibration of propagation model parameters of logv distance path loss model and RSSI measurement noise by using reference tags. The proposed method also uses an RSSI smoothing algorithm exploiting the RSSI readings from the reference tags. We used an active RFID system composed of 3 readers, 1 target tag and 4 reference tags in a home environment of two rooms with a total area of 36 m&sup2 / . The proposed method yielded 1.25 m estimation RMS error for tracking a mobile target.

Um sistema artificial pode usar raciocínio espacial qualitativo para inferir informações sobre seu ambiente tridimensional a partir de imagens bidimensionais. Inferências realizadas com base em raciocínio espacial qualitativo devem ser capazes de lidar com incertezas. Neste trabalho investigamos a utilização de técnicas probabilísticas para tornar o raciocínio espacial qualitativo mais robusto a incertezas e aplicável a agentes móveis em ambientes reais. Investigamos uma formalização de raciocínio espacial com lógica de descrição probabilística em um subdomínio de tráfego. Desenvolvemos também um método que combina raciocínio espacial qualitativo com um filtro Bayesiano para desenvolver dois sistemas que foram aplicados na auto localização de um robô móvel. Executamos dois experimentos de auto localização; um utilizando a teoria de relações qualitativas percebíveis sobre sombra com filtro Bayesiano; e outro utilizando o cálculo de oclusão de regiões e o cálculo de direção com filtro Bayesiano. Ambos os sistemas obtiveram resultados positivos onde somente o raciocínio espacial qualitativo não foi capaz de inferir a localização do robô. Os experimentos com dados reais mostraram robustez aos ruídos e à informação parcial. / An artificial system can use qualitative spatial reasoning to obtain information about its tridimensional environment, from bi-dimensional images. Inferences produced by qualitative spatial reasoning must be able to deal with uncertainty. This work investigates the use of probabilistic techniques to make qualitative spatial reasoning more robust against uncertainty, and better applicable to mobile agents in real environments. The work investigates a formalization of spatial reasoning using probabilistic description logics in a traffic domain. Additionally, a method is presented that combines qualitative spatial reasoning with a Bayesian filter, to develop two systems that are applied to self-localization of mobile robots. Two experiments are described; one using the theory of perceptual qualitative relations about shadows; the other using occlusion calculus and direction calculus. Both systems are combined with a Bayesian filter producing positive results in situations where qualitative spatial reasoning alone cannot infer robot location. Experiments with real data show robustness to noise and partial information.

Target tracking is an important part of video surveillance from a UAV. Tracking a target in an urban environment can be difficult because of the number of occlusions present in the environment. If multiple UAVs are used to track a target and the target behavior is learned autonomously by the UAV then the task may become easier. This thesis explores the hypothesis that an existing cooperative control algorithm can be enhanced by a language modeling algorithm to improve over time the target tracking performance of one or more ground targets in a dense urban environment. Observations of target behavior are reported to the Sequence Memoizer which uses the observations to create a belief model of future target positions. This belief model is combined with a kinematic belief model and then used in a cooperative auction algorithm for UAV path planning. The results for tracking a single target using the combined belief model outperform other belief models and improve over the duration of the mission. Results from tracking multiple targets indicate that algorithmic enhancements may be needed to find equivalent success. Future target tracking algorithms should involve machine learning to enhance tracking performance.

[pt] SLAM (Mapeamento e Localização Simultânea) é uma das áreas mais pesquisadas na Robótica móvel. Trata-se do problema, num robô móvel, de construir um mapa sem conhecimento prévio do ambiente e ao mesmo tempo manter a sua localização nele. Embora a tecnologia ofereça sensores cada vez mais precisos, pequenos erros na medição são acumulados comprometendo a precisão na localização, sendo estes evidentes quando o robô retorna a uma posição inicial depois de percorrer um longo caminho. Assim, para melhoria do desempenho do SLAM é necessário representar a sua formulação usando teoria das probabilidades. O SLAM com Filtro Extendido de Kalman (EKF-SLAM) é uma solução básica, e apesar de suas limitações é a técnica mais popular. O Fast SLAM, por outro lado, resolve algumas limitações do EKF-SLAM usando uma instância do filtro de partículas conhecida como Rao-Blackwellized. Outra solução bem sucedida é o DP-SLAM, o qual usa uma representação do mapa em forma de grade de ocupação, com um algoritmo hierárquico que constrói mapas 2D bastante precisos. Todos estes algoritmos usam informação de dois tipos de sensores: odômetros e sensores de distância. O Laser Range Finder (LRF) é um medidor laser de distância por varredura, e pela sua precisão é bastante usado na correção do erro em odômetros. Este trabalho apresenta uma detalhada implementação destas três soluções para o SLAM, focalizado em ambientes fechados e estruturados. Apresenta-se a construção de mapas 2D e 3D em terrenos planos tais como em aplicações típicas de ambientes fechados. A representação dos mapas 2D é feita na forma de grade de ocupação. Por outro lado, a representação dos mapas 3D é feita na forma de nuvem de pontos ao invés de grade, para reduzir o custo computacional. É considerado um robô móvel equipado com apenas um LRF, sem nenhuma informação de odometria. O alinhamento entre varreduras laser é otimizado fazendo o uso de Algoritmos Genéticos. Assim, podem-se construir mapas e ao mesmo tempo localizar o robô sem necessidade de odômetros ou outros sensores. Um simulador em Matlab é implementado para a geração de varreduras virtuais de um LRF em um ambiente 3D (virtual). A metodologia proposta é validada com os dados simulados, assim como com dados experimentais obtidos da literatura, demonstrando a possibilidade de construção de mapas 3D com apenas um sensor LRF. / [en] Simultaneous Localization and Mapping (SLAM) is one of the most widely researched areas of Robotics. It addresses the mobile robot problem of generating a map without prior knowledge of the environment, while keeping track of its position. Although technology offers increasingly accurate position sensors, even small measurement errors can accumulate and compromise the localization accuracy. This becomes evident when programming a robot to return to its original position after traveling a long distance, based only on its sensor readings. Thus, to improve SLAM s performance it is necessary to represent its formulation using probability theory. The Extended Kalman Filter SLAM (EKF-SLAM) is a basic solution and, despite its shortcomings, it is by far the most popular technique. Fast SLAM, on the other hand, solves some limitations of the EKFSLAM using an instance of the Rao-Blackwellized particle filter. Another successful solution is to use the DP-SLAM approach, which uses a grid representation and a hierarchical algorithm to build accurate 2D maps. All SLAM solutions require two types of sensor information: odometry and range measurement. Laser Range Finders (LRF) are popular range measurement sensors and, because of their accuracy, are well suited for odometry error correction. Furthermore, the odometer may even be eliminated from the system if multiple consecutive LRF scans are matched. This works presents a detailed implementation of these three SLAM solutions, focused on structured indoor environments. The implementation is able to map 2D environments, as well as 3D environments with planar terrain, such as in a typical indoor application. The 2D application is able to automatically generate a stochastic grid map. On the other hand, the 3D problem uses a point cloud representation of the map, instead of a 3D grid, to reduce the SLAM computational effort. The considered mobile robot only uses a single LRF, without any odometry information. A Genetic Algorithm is presented to optimize the matching of LRF scans taken at different instants. Such matching is able not only to map the environment but also localize the robot, without the need for odometers or other sensors. A simulation program is implemented in Matlab to generate virtual LRF readings of a mobile robot in a 3D environment. Both simulated readings and experimental data from the literature are independently used to validate the proposed methodology, automatically generating 3D maps using just a single LRF.