Auswahl der wissenschaftlichen Literatur zum Thema „Indoor robotics“

Our dissertation concerns robotic navigation in dynamic indoor environments using image-based visual homing. Image-based visual homing infers the direction to a goal location S from the navigator’s current location C using the similarity between panoramic images IS and IC captured at those locations. There are several ways to compute this similarity. One of the contributions of our dissertation is to identify a robust image similarity measure – mutual image information – to use in dynamic indoor environments. We crafted novel methods to speed the computation of mutual image information with both parallel and serial processors and demonstrated that these time-savers had little negative effect on homing success. Image-based visual homing requires a homing agent tomove so as to optimise themutual image information signal. As the mutual information signal is corrupted by sensor noise we turned to the stochastic optimisation literature for appropriate optimisation algorithms. We tested a number of these algorithms in both simulated and real dynamic laboratory environments and found that gradient descent (with gradients computed by one-sided differences) works best.

Les humains sont en mesure d’interpréter l’environnement qui les entourent avec peu d’effort grâce à système visuel très performant. Par analogie, un système de vision capable de recueillir les mêmes informations sur l’environnement est hautement souhaitable en robotique autonome pour effectuer des tâches complexes et ainsi interagir avec les humains.À cet égard, nous nous sommes particulièrement intéressés aux environnements intérieurs, dans lesquels les humains passent presque toute leur vie. Dans ce travail, pour faire une analyse efficace et rapide des scènes, nous avons opté pour l’utilisation de caméras non conventionnelles : l’imagerie 360° et les capteurs 3D. Ces systèmes ont la particularité d’acquérir en une seule prise de vue soit la totalité de l’environnement qui entoure le robot (caméras 360°) soit l’information 3D.C’est ainsi que cette thèse aborde les problèmes de description hiérarchique d’une scène d’intérieur avec des capteurs non conventionnels allant de l’estimation de la disposition des pièces ; de la détection et la localisation des objets à la modélisation de la forme des objets 3D.Ces différents points font l’objet de contribution dans ce travail. Dans un premier temps, nous nous sommes intéressés à l'estimation de la disposition 3D de la pièce à partir d'une seule image à 360°. Pour ce faire, nous exploitons l'hypothèse de Manhattan World et les techniques d'apprentissage profond pour proposer des modèles qui gèrent les parties occultées de la pièce sur l'image. A vu de la particularité des images considérées, nous avons développé de nouveaux filtres de convolution d’image tenant compte des fortes distorsions des images équirectangulaires.Par la suite, et dans l’objectif de permettre au robot de faire une analyse contextuelle de hauts niveaux de la scène qui l’entoure, nous nous sommes intéressés au problème de la localisation et de la segmentation des objets. C’est ainsi que nous avons une nouvelle fois exploité les images 360° en tenant compte de la disposition des objets 2D dans l’image dans le but de les décrire par leur modèle 3D en adéquation avec la disposition de la pièce préalablement estimée.La dernière contribution de ce travail tire parti des capteurs 3D pour étudier la forme des objets. Dans ce cadre, nous utilisons une modélisation explicite de la non-rigidité de objets et caractérisons leurs symétries afin de détecter, par un apprentissage profond non supervisé, ces points d’intérêt 3D.Toutes ces contributions nous ont permis de faire progresser l’état de l’art sur les problèmes posés et ont toutes fait l’objet d’évaluation sur des bases de données de référence dans notre communauté<br>Humans understand environments effortlessly, under a wide variety of conditions, by the virtue of visual perception. Computer vision for similar visual understanding is highly desirable, so that machines can perform complex tasks by interacting with the real world, to assist or entertain humans. In this regard, we are particularly interested in indoor environments, where humans spend nearly all their lifetime.This thesis specifically addresses the problems that arise during the quest of the hierarchical visual understanding of indoor scenes.On the side of sensing the wide 3D world, we propose to use non-conventional cameras, namely 360º imaging and 3D sensors. On the side of understanding, we aim at three key aspects: room layout estimation; object detection, localization and segmentation; and object category shape modeling, for which novel and efficient solutions are provided.The focus of this thesis is on the following underlying challenges. First, the estimation of the 3D room layout from a single 360º image is investigated, which is used for the highest level of scene modelling and understanding. We exploit the assumption of Manhattan World and deep learning techniques to propose models that handle invisible parts of the room on the image, generalizing to more complex layouts. At the same time, new methods to work with 360º images are proposed, highlighting a special convolution that compensates the equirectangular image distortions.Second, considering the importance of context for scene understanding, we study the problem of object localization and segmentation, adapting the problem to leverage 360º images. We also exploit layout-objects interaction to lift detected 2D objects into the 3D room model.The final line of work of this thesis focuses on 3D object shape analysis. We use an explicit modelling of non-rigidity and a high-level notion of object symmetry to learn, in an unsupervised manner, 3D keypoints that are order-wise correspondent as well as geometrically and semantically consistent across objects in a category.Our models advance state-of-the-art on the aforementioned tasks, when each evaluated on respective reference benchmarks

Indoor tracking and navigation is a fundamental need for pervasive and context-aware applications. However, no practical and reliable indoor positioning solution is available at present. The major challenge of a practical solution lies in the fact that only the existing devices and infrastructure can be utilized to achieve high positioning accuracy. This thesis presents a robust indoor positioning system with the lifelong learning ability. The typical features of the proposed solution is low-cost, accurate, robust, and scalable. This system only takes the floor plan and the existing devices, e.g. phones, pads, etc. and infrastructure such as WiFi/BLE access points for the sake of practicality. This system has four closely correlated components including, non-line-of-sight identification and mitigation (NIMIT), robust pedestrian dead reckoning (R-PDR), lightweight map matching (MapCraft), and lifelong learning. NIMIT projects the received signal strength (RSS) from WiFi/BLE to locations. The R-PDR component converts the data from inertial measurement unit (IMU) sensors ubiquitous in mobile devices and wearables to the trajectories of the user. Then MapCraft fuses trajectories estimated from the R-PDR and the coarse location information from NIMIT with the floor plan and provides accurate location estimations. The lifelong learning component then learns the various parameters used in all other three components in an unsupervised manner, which continuously improves the the positioning accuracy of the system. Extensive real world experiments in multiple sites show how the proposed system outperforms state-of-the art approaches, demonstrating excellent sub-meter positioning accuracy and accurate reconstruction of tortuous trajectories with zero training effort. As proof of its robustness, we also demonstrate how it is able to accurately track the position regardless of the users, devices, attachments, and environments. We believe that such an accurate and robust approach will enable always-on background localization, enabling a new era of location-aware applications to be developed.

Exploration is of interest for autonomous mapping and rescue applications using unmanned vehicles. The objective is to, without any prior information, explore all initially unmapped space. We present a system that can perform fast and efficient exploration of large scale arbitrary 3D environments. We combine frontier exploration planning (FEP) as a global planning strategy, together with receding horizon planning (RH-NBVP) for local planning. This leads to plans that incorporate information gain along the way, but do not get stuck in already explored regions. Furthermore, we make the potential information gain estimation more efficient, through sparse ray-tracing, and caching of already estimated gains. The worked carried out in this thesis has been published as a paper in Robotand Automation letters and presented at the International Conference on Robotics and Automation in Montreal 2019.