For a long time stereo-cameras have been deployed in visual Simultaneous Localization And Mapping (SLAM) systems to gain 3D information. Even though stereo-cameras show good performance, the main disadvantage is the complex and expensive hardware setup it requires, which limits the use of the system. A simpler and cheaper alternative are monocular cameras, however monocular images lack the important depth information. Recent works have shown that having access to depth maps in monocular SLAM system is beneficial since they can be used to improve the 3D reconstruction. This work proposes a deep neural network that predicts dense high-resolution depth maps from monocular RGB images by casting the problem as a supervised regression task. The network architecture follows an encoder-decoder structure in which multi-scale information is captured and skip-connections are used to recover details. The network is trained and evaluated on the KITTI dataset achieving results comparable to state-of-the-art methods. With further development, this network shows good potential to be incorporated in a monocular SLAM system to improve the 3D reconstruction.

Esta pesquisa tem o intuito de abordar um movimento literário e cultural que se expressa pela performance de poesias autorais, chamado de batalha de poesia ou slam de poesia com influência no Poetry Slam e Spoken Words, e que recentemente cresce no Brasil, particularmente na cidade de São Paulo. O objetivo desta pesquisa visa apreender como este movimento relativamente novo no Brasil, se insere no conceito de Literatura Marginal e da ideia de Cultura de Periferia, tanto pelas questões internas ao discursos, das poesias faladas e publicadas, como por questões externas ligadas a ações sociais e conjuntas. Foi utilizada a metodologia de etnografia em campo com visitas ao Slam da Guilhermina, ZAP!Slam e Slam das Minas SP; levantamento das publicações de livros lançados nos slams e mapeamento cartográfico a partir dos locais de realização dos mesmos na cidade e região metropolitana de São Paulo. Como resultado de pesquisa foi possível definir os temas mais recorrentes entre os poemas que concorreram às batalhas e o ano e região de maior concentração dos slams em São Paulo. Podemos concluir com esta pesquisa que todos os livros são do gênero de poemas e foram publicados de forma independente ou a partir de editoras que se definem como independentes; que a maioria dos coletivos de slam nasceram entre 2016 e 2017 com maior concentração na Zona Leste de São Paulo / This research aims to address a literary and cultural movement that is expressed by the performance of poetry authored, called the \"battle of poetry\" or poetry slam with influence in Poetry Slam and Spoken Words, and which has recently grown in Brazil, particularly in Brazil. Sao Paulo City. The objective of this research is to understand how this relatively new movement in Brazil is inserted in the concept of Marginal Literature and the idea of Culture of Periphery, as much for the internal questions to the speeches, of the spoken and published poetry, as for external questions related to social actions and joint. The ethnography methodology was used in the field with visits to Guilhermina Slam, ZAP! Slam and Slam from SP mines; survey of the publications of books published in slam\'s and cartographic mapping from the places of accomplishment of the same ones in the city and metropolitan region of São Paulo. As a result of research it was possible to define the most recurrent themes among the poems that competed for the battles and the year and region of greater concentration of slam\'s in São Paulo. We can conclude from this research that all books are of the genre of poems and have been published independently or from publishers who define themselves as independent; that most of the slam collectives were born between 2016 and 2017 with greater concentration in the East Zone of São Paulo

Les travaux de cette thèse adressent la problématique de la coopération aéro-terrestre pour la cartographie de l’espace navigable. La nécessité d’une carte pour la navigation et la planification de chemins pour les robots terrestres n’est plus à prouver. L’utilisation d’une coopération aéro-terrestre pour créer une carte navigable à destination du robot terrestre a plusieurs intérêts. Premièrement, le drone peut cartographier rapidement une zone grâce à son champ de vision étendu et ses capacités de déplacement. Deuxièmement, la fusion des cartes créées par ces deux agents permet de tirer le meilleur profit des deux points de vue : la cohérence de la vue aérienne globale et la précision de la vue terrestre locale. Pour répondre à cette problématique, nous proposons une méthode qui s’appuie sur la création de cartes hybrides et leur fusion. Les cartes sont construites en utilisant le squelette de l’espace navigable terrestre comme support d’un graphe contenant également des informations métriques locales de l’environnement. La mise en correspondance des cartes aérienne et terrestre s’effectue à l’aide d’un appariement point à point déterminé grâce à une mesure de dissimilarité appropriée. Cette dernière est définie pour répondre aux critères d’invariance et de discriminance dans ce contexte. La mise en correspondance est ensuite utilisée pour fusionner les cartes entre elles. Les cartes fusionnées peuvent être utilisées par le robot au sol pour effectuer sa mission. Elles permettent également de propager des informations telles que des coordonnées GPS à des robots et dans des lieux où ce dispositif n’est pas disponible. Des expérimentations en environnements virtuels et réels sont réalisées pour valider cette approche et en tracer les perspectives. / This work aims to study the problem of air-ground robotic cooperation for collaborative traversability mapping. The need for a map for navigation and path planning for terrestrial robots is no longer to be proven. The use of air-ground cooperation to create a navigable map for the ground robots has several interests. First, the drone can quickly map an area through its large field of vision and traveling capabilities. Second, the fusion of maps based on these two agents makes it possible to draw the best benefits from both points of views: the coherence of the global aerial view and the accuracy of the local ground view. To answer this problem, we propose a method that relies on the construction of a unified model of hybrid maps and their fusion.The maps are built using the skeleton of the traversability space as a support for graphs also containing local metric and potentialy semantic information of the environment. The maching of aerial and ground maps is done using a point to point correlation based on an appropriate dissimilarity measure. This measure is defined to meet invariance and discriminance criteria. The matching is then used to merge the maps into an augmented traversability map. The merged maps can be used by the ground robot to perform its mission. They also make it possible to deploy information such as GPS coordinates to robots in GPS denied environments. Experiments in virtual and real world environments have been carried out to validate this approach and map out future perspetives.

La fusion entre un ensemble de capteurs et de bases de données dont les erreurs sont indépendantes est aujourd’hui la solution la plus fiable et donc la plus répandue de l’état de l’art au problème de la localisation. Les véhicules semi-autonomes et autonomes actuels, ainsi que les applications de réalité augmentée visant les contextes industriels exploitent des graphes de capteurs et de bases de données de tailles considérables, dont la conception, la calibration et la synchronisation n’est, en plus d’être onéreuse, pas triviale. Il est donc important afin de pouvoir démocratiser ces technologies, d’explorer la possibilité de l’exploitation de capteurs et bases de données bas-coûts et aisément accessibles. Cependant, ces sources d’information sont naturellement plus incertaines, et plusieurs obstacles subsistent à leur utilisation efficace en pratique. De plus, les succès récents mais fulgurants des réseaux profonds dans des tâches variées laissent penser que ces méthodes peuvent représenter une alternative peu coûteuse et efficace à certains modules des systèmes de SLAM actuels. Dans cette thèse, nous nous penchons sur la localisation à grande échelle d’un véhicule dans un repère géoréférencé à partir d’un système bas-coût. Celui-ci repose sur la fusion entre le flux vidéo d’une caméra monoculaire, des modèles 3d non-texturés mais géoréférencés de bâtiments,des modèles d’élévation de terrain et des données en provenance soit d’un GPS bas-coût soit de l’odométrie du véhicule. Nos travaux sont consacrés à la résolution de deux problèmes. Le premier survient lors de la fusion par terme barrière entre le VSLAM et l’information de positionnement fournie par un GPS bas-coût. Cette méthode de fusion est à notre connaissance la plus robuste face aux incertitudes du GPS, mais est plus exigeante en matière de ressources que la fusion via des fonctions de coût linéaires. Nous proposons une optimisation algorithmique de cette méthode reposant sur la définition d’un terme barrière particulier. Le deuxième problème est le problème d’associations entre les primitives représentant la géométrie de la scène(e.g. points 3d) et les modèles 3d des bâtiments. Les travaux précédents se basent sur des critères géométriques simples et sont donc très sensibles aux occultations en milieu urbain. Nous exploitons des réseaux convolutionnels profonds afin d’identifier et d’associer les éléments de la carte correspondants aux façades des bâtiments aux modèles 3d. Bien que nos contributions soient en grande partie indépendantes du système de SLAM sous-jacent, nos expériences sont basées sur l’ajustement de faisceaux contraint basé images-clefs. Les solutions que nous proposons sont évaluées sur des séquences de synthèse ainsi que sur des séquence urbaines réelles sur des distances de plusieurs kilomètres. Ces expériences démontrent des gains importants en performance pour la fusion VSLAM/GPS, et une amélioration considérable de la robustesse aux occultations dans la définition des contraintes. / The fusion between sensors and databases whose errors are independant is the most re-liable and therefore most widespread solution to the localization problem. Current autonomousand semi-autonomous vehicles, as well as augmented reality applications targeting industrialcontexts exploit large sensor and database graphs that are difficult and expensive to synchro-nize and calibrate. Thus, the democratization of these technologies requires the exploration ofthe possiblity of exploiting low-cost and easily accessible sensors and databases. These infor-mation sources are naturally tainted by higher uncertainty levels, and many obstacles to theireffective and efficient practical usage persist. Moreover, the recent but dazzling successes ofdeep neural networks in various tasks seem to indicate that they could be a viable and low-costalternative to some components of current SLAM systems.In this thesis, we focused on large-scale localization of a vehicle in a georeferenced co-ordinate frame from a low-cost system, which is based on the fusion between a monocularvideo stream, 3d non-textured but georeferenced building models, terrain elevation models anddata either from a low-cost GPS or from vehicle odometry. Our work targets the resolutionof two problems. The first one is related to the fusion via barrier term optimization of VS-LAM and positioning measurements provided by a low-cost GPS. This method is, to the bestof our knowledge, the most robust against GPS uncertainties, but it is more demanding in termsof computational resources. We propose an algorithmic optimization of that approach basedon the definition of a novel barrier term. The second problem is the data association problembetween the primitives that represent the geometry of the scene (e.g. 3d points) and the 3d buil-ding models. Previous works in that area use simple geometric criteria and are therefore verysensitive to occlusions in urban environments. We exploit deep convolutional neural networksin order to identify and associate elements from the map that correspond to 3d building mo-del façades. Although our contributions are for the most part independant from the underlyingSLAM system, we based our experiments on constrained key-frame based bundle adjustment.The solutions that we propose are evaluated on synthetic sequences as well as on real urbandatasets. These experiments show important performance gains for VSLAM/GPS fusion, andconsiderable improvements in the robustness of building constraints to occlusions.

Après la démocratisation des robots industriels, la tendance actuelle est au développement de robots sociaux dont la fonction principale est l'interaction avec ses utilisateurs. Le déploiement de telles plate-formes dans des boutiques, des musées ou des gares relance différentes problématiques dont celle de la localisation pour les robots mobiles. Cette thèse traite ainsi de la localisation du robot Pepper en milieu intérieur non-contraint. Présent dans de nombreuses boutiques au Japon, Pepper est utilisé par des personnes non-expertes et doit donc être le plus autonome possible. Cependant, les solutions de localisation autonome de la littérature souffrent des limitations de la plate-forme. Les travaux de cette thèse s'articulent autour de deux grands axes. D'abord, le problème de la relocalisation dans un environnement visuellement redondant est étudié. La solution proposée consiste à combiner la vision et le Wi-Fi dans une approche probabiliste basée sur l'apparence. Ensuite, la question de la création d'une carte métrique cohérente est approfondie. Pour compenser les nombreuses pertes de suivi d'amers visuels causées par une fréquence d'acquisition basse, des contraintes odométriques sont ajoutées à une optimisation par ajustement de faisceaux. Ces solutions ont été testées et validées sur plusieurs robots Pepper à partir de données collectées dans différents environnements intérieurs sur plus de 7 km. / After the democratization of industrial robots, the current trend is the development of social robots that create strong interactions with their users. The deployment of such platforms in shops, museums or train stations raises various issues including the autonomous localization of mobile robots. This thesis focuses on the localization of Pepper robots in a non-constrained indoor environment. Pepper robots are daily used in many shops in Japan and must be as autonomous as possible. However, localization solutions in the literature suffer from the limitations of the platform. This thesis is split into two main themes. First, the problem of relocalization in a visually redundant environment is studied. The proposed solution combines vision and Wi-Fi in a probabilistic approach based on the appearance. Then, the question of a consistent metrical mapping is examined. In order to compensate the numerous losses of tracking caused by the low acquisition frequency, odometric constraints are added to a bundle adjustment optimization. These solutions have been tested and validated on several Pepper robots, from data collected in different indoor environments over more than 7 km.

Localization and mapping are vital capabilities for a mobile robot. These two capabilities strongly depend on each other and simultaneously executing both of these operations is called SLAM (Simultaneous Localization and Mapping). SLAM problem requires the environment to be represented with an abstract mapping model. It is possible to construct a map from point cloud of environment via scanner sensor systems. On the other hand, extracting higher level of features from point clouds and using these extracted features as an input for mapping system is also a possible solution for SLAM. In this work, a 4D feature based EKF SLAM system is constructed and open form of equations of algorithm are presented. The algorithm is able to use center of mass and direction of features as input parameters and executes EKF SLAM via these parameters. Performance of 4D feature based EKF SLAM was examined and compared with 3D EKF SLAM via monte-carlo simulations. By this way / it is believed that, contribution of adding a direction vector to 3D features is investigated and illustrated via graphs of monte-carlo simulations. At the second part of the work, a scanner sensor system with IR distance finder is designed and constructed. An algorithm was presented to extract planar features from data collected by sensor system. A noise model was proposed for output features of sensor and 4D EKF SLAM algorithm was executed via extracted features of scanner system. By this way, performance of 4D EKF SLAM algorithm is tested with real sensor data and output results are compared with 3D features. So in this work, contribution of using 4D features instead of 3D ones was examined via comparing performance of 3D and 4D algorithms with simulation results and real sensor data.

Localization and mapping are vital capabilities for a mobile robot. These two capabilities strongly depend on each other and simultaneously executing both of these operations is called SLAM (Simultaneous Localization and Mapping). SLAM problem requires the environment to be represented with an abstract mapping model. It is possible to construct a map from point cloud of environment via scanner sensor systems. On the other hand, extracting higher level of features from point clouds and using these extracted features as an input for mapping system is also a possible solution for SLAM. In this work, a 4D feature based EKF SLAM system is constructed and open form of equations of algorithm are presented. The algorithm is able to use center of mass and direction of features as input parameters and executes EKF SLAM via these parameters. Performance of 4D feature based EKF SLAM was examined and compared with 3D EKF SLAM via monte-carlo simulations. By this way / it is believed that, contribution of adding a direction vector to 3D features is investigated and illustrated via graphs of monte-carlo simulations. At the second part of the work, a scanner sensor system with IR distance finder is designed and constructed. An algorithm was presented to extract planar features from data collected by sensor system. A noise model was proposed for output features of sensor and 4D EKF SLAM algorithm was executed via extracted features of scanner system. By this way, performance of 4D EKF SLAM algorithm is tested with real sensor data and output results are compared with 3D features. So in this work, contribution of using 4D features instead of 3D ones was examined via comparing performance of 3D and 4D algorithms with simulation results and real sensor data.

The major aim in search and rescue using mobile robots is to reach trapped survivors and to support rescue operations through the disaster environments. Our motivation is based on the fact that a search and rescue (SAR) robot can navigate within and penetrate a disaster area only if the area in question possesses connected voids Traversability or penetrability of a disaster area is a primary factor that guides the navigation of a search and rescue (SAR) robot, since it is highly desirable that the robot, without hitting a dead end or getting stuck, keeps its mobility for its primary task of reconnaissance and mapping when searching the highly unstructured environment We propose two novel guided prioritized exploration system: 1) percolation guided methodology where a percolator estimates the existence of connected voids in the upcoming yet unexplored region ahead of the robot so as to increase the efficiency of reconnaissance operation by the superior ability of the percolation guidance in speedy coverage of the area / 2) the hybrid exploration methodology that makes the percolation guided exploration collaborate with entropy based SLAM under a switching control dependent on either priority given to position accuracy or to map accuracy This second methodology has proven to combine the superiority of both methods so that the active SLAM becomes speedy, with high coverage rate of the area as well as accurate in localization.

Dans le contexte de la navigation autonome en environnement urbain, une localisation précise du véhicule est importante pour une navigation sure et fiable. La faible précision des capteurs bas coût existants tels que le système GPS, nécessite l'utilisation d'autres capteurs eux aussi à faible coût. Les caméras mesurent une information photométrique riche et précise sur l'environnement, mais nécessitent l'utilisation d'algorithmes de traitement avancés pour obtenir une information sur la géométrie et sur la position de la caméra dans l'environnement. Cette problématique est connue sous le terme de Cartographie et Localisation Simultanées (SLAM visuel). En général, les techniques de SLAM sont incrémentales et dérivent sur de longues trajectoires. Pour simplifier l'étape de localisation, il est proposé de découpler la partie cartographie et la partie localisation en deux phases: la carte est construite hors-ligne lors d'une phase d'apprentissage, et la localisation est effectuée efficacement en ligne à partir de la carte 3D de l'environnement. Contrairement aux approches classiques, qui utilisent un modèle 3D global approximatif, une nouvelle représentation égo-centrée dense est proposée. Cette représentation est composée d'un graphe d'images sphériques augmentées par l'information dense de profondeur (RGB+D), et permet de cartographier de larges environnements. Lors de la localisation en ligne, ce type de modèle apporte toute l'information nécessaire pour une localisation précise dans le voisinage du graphe, et permet de recaler en temps-réel l'image perçue par une caméra embarquée sur un véhicule, avec les images du graphe, en utilisant une technique d'alignement d'images directe. La méthode de localisation proposée, est précise, robuste aux aberrations et prend en compte les changements d'illumination entre le modèle de la base de données et les images perçues par la caméra. Finalement, la précision et la robustesse de la localisation permettent à un véhicule autonome, équipé d'une caméra, de naviguer de façon sure en environnement urbain. / In an autonomous navigation context, a precise localisation of the vehicule is important to ensure a reliable navigation. Low cost sensors such as GPS systems are inacurrate and inefficicent in urban areas, and therefore the employ of such sensors alone is not well suited for autonomous navigation. On the other hand, camera sensors provide a dense photometric measure that can be processed to obtain both localisation and mapping information. In the robotics community, this problem is well known as Simultaneous Localisation and Mapping (SLAM) and it has been studied for the last thirty years. In general, SLAM algorithms are incremental and prone to drift, thus such methods may not be efficient in large scale environments for real-time localisation. Clearly, an a-priori 3D model simplifies the localisation and navigation tasks since it allows to decouple the structure and motion estimation problems. Indeed, the map can be previously computed during a learning phase, whilst the localisation can be handled in real-time using a single camera and the pre-computed model. Classic global 3D model representations are usually inacurrate and photometrically inconsistent. Alternatively, it is proposed to use an ego-centric model that represents, as close as possible, real sensor measurements. This representation is composed of a graph of locally accurate spherical panoramas augmented with dense depth information. These augmented panoramas allow to generate varying viewpoints through novel view synthesis. To localise a camera navigating locally inside the graph, we use the panoramas together with a direct registration technique. The proposed localisation method is accurate, robust to outliers and can handle large illumination changes. Finally, autonomous navigation in urban environments is performed using the learnt model, with only a single camera to compute localisation.

Autonomous features in vehicles is already a big part of the automobile area and now many companies are looking for ways to make vehicles fully autonomous. Autonomous vehicles need to get information about the surrounding environment. The information is extracted from exteroceptive sensors and today vehicles often use laser scanners for this purpose. Laser scanners are very expensive and fragile, it is therefore interesting to investigate if cheaper radar sensors could be used. One big challenge when it comes to autonomous vehicles is to be able to use the exteroceptive sensors and extract a position of the vehicle and at the same time get a map of the environment. The area of Simultaneous Localization and Mapping (SLAM) is a well explored area when using laser scanners but is not that well explored when using radars. It has been investigated if it is possible to use radar sensors on a truck to create a map of the area where the truck drives. The truck has been equipped with ego-motion sensors and radars and the data from them has been fused together to get a position of the truck and to get a map of the surrounding environment, i.e. a SLAM algorithm has been implemented. The map is represented by an Occupancy Grid Map (OGM) which should only consist of static objects. The OGM is updated probabilistically by using a binary Bayes filter. To localize the truck with help of motion sensors an Extended Kalman Filter (EKF) is used together with a map and a scan match method. All these methods are put together to create a SLAM algorithm. A range rate filter method is used to filter out noise and non-static measurements from the radar. The results of this thesis show that it is possible to use radar sensors to create a map of a truck's surroundings. The quality of the map is considered to be good and details such as space between parked trucks, signs and light posts can be distinguished. It has also been proven that methods with low performance on their own can together with other methods work very well in the SLAM algorithm. Overall the SLAM algorithm works well but when driving in unexplored areas with a low number of objects problems with positioning might occur. A real time system has also been implemented and the map can be seen at the same time as the truck is manoeuvred.