Doctor of Philosophy (PhD) / This thesis addresses the problem of building statistical models for multi-sensor perception in unstructured outdoor environments. The perception problem is divided into three distinct tasks: recognition, representation and association. Recognition is cast as a statistical classification problem where inputs are images or a combination of images and ranging information. Given the complexity and variability of natural environments, this thesis investigates the use of Bayesian statistics and supervised dimensionality reduction to incorporate prior information and fuse sensory data. A compact probabilistic representation of natural objects is essential for many problems in field robotics. This thesis presents techniques for combining non-linear dimensionality reduction with parametric learning through Expectation Maximisation to build general representations of natural features. Once created these models need to be rapidly processed to account for incoming information. To this end, techniques for efficient probabilistic inference are proposed. The robustness of localisation and mapping algorithms is directly related to reliable data association. Conventional algorithms employ only geometric information which can become inconsistent for large trajectories. A new data association algorithm incorporating visual and geometric information is proposed to improve the reliability of this task. The method uses a compact probabilistic representation of objects to fuse visual and geometric information for the association decision. The main contributions of this thesis are: 1) a stochastic representation of objects through non-linear dimensionality reduction; 2) a landmark recognition system using a visual and ranging sensors; 3) a data association algorithm combining appearance and position properties; 4) a real-time algorithm for detection and segmentation of natural objects from few training images and 5) a real-time place recognition system combining dimensionality reduction and Bayesian learning. The theoretical contributions of this thesis are demonstrated with a series of experiments in unstructured environments. In particular, the combination of recognition, representation and association algorithms is applied to the Simultaneous Localisation and Mapping problem (SLAM) to close large loops in outdoor trajectories, proving the benefits of the proposed methodology.

Doctor of Philosophy (PhD) / This thesis addresses the problem of building statistical models for multi-sensor perception in unstructured outdoor environments. The perception problem is divided into three distinct tasks: recognition, representation and association. Recognition is cast as a statistical classification problem where inputs are images or a combination of images and ranging information. Given the complexity and variability of natural environments, this thesis investigates the use of Bayesian statistics and supervised dimensionality reduction to incorporate prior information and fuse sensory data. A compact probabilistic representation of natural objects is essential for many problems in field robotics. This thesis presents techniques for combining non-linear dimensionality reduction with parametric learning through Expectation Maximisation to build general representations of natural features. Once created these models need to be rapidly processed to account for incoming information. To this end, techniques for efficient probabilistic inference are proposed. The robustness of localisation and mapping algorithms is directly related to reliable data association. Conventional algorithms employ only geometric information which can become inconsistent for large trajectories. A new data association algorithm incorporating visual and geometric information is proposed to improve the reliability of this task. The method uses a compact probabilistic representation of objects to fuse visual and geometric information for the association decision. The main contributions of this thesis are: 1) a stochastic representation of objects through non-linear dimensionality reduction; 2) a landmark recognition system using a visual and ranging sensors; 3) a data association algorithm combining appearance and position properties; 4) a real-time algorithm for detection and segmentation of natural objects from few training images and 5) a real-time place recognition system combining dimensionality reduction and Bayesian learning. The theoretical contributions of this thesis are demonstrated with a series of experiments in unstructured environments. In particular, the combination of recognition, representation and association algorithms is applied to the Simultaneous Localisation and Mapping problem (SLAM) to close large loops in outdoor trajectories, proving the benefits of the proposed methodology.

Cette thèse est centralisée sur cette question : comment est-ce que le système visuel peut coder efficacement la position des objets en mouvement, en dépit des diverses sources d'incertitude ? Cette étude déploie une hypothèse sur la connaissance a priori de la cohérence temporelle du mouvement (Burgi et al 2000; Yuille and Grzywacz 1989). Nous avons ici étendu le cadre de modélisation précédemment proposé pour expliquer le problème de l'ouverture (Perrinet and Masson, 2012). C'est un cadre d'estimation de mouvement Bayésien mis en oeuvre par un filtrage particulaire, que l'on appelle la prévision basé sur le mouvement (MBP). Sur cette base, nous avons introduit une théorie du codage de position basée sur le mouvement, et étudié comment les mécanismes neuronaux codant la position instantanée de l'objet en mouvement pourraient être affectés par le signal de mouvement le long d'une trajectoire. Les résultats de cette thèse suggèrent que le codage de la position basé sur le mouvement peut constituer un calcul neuronal générique parmi toutes les étapes du système visuel. Cela peut en partie compenser les effets cumulatifs des délais neuronaux dans le codage de la position. En outre, il peut expliquer des changements de position basés sur le mouvement, comme par example, l'Effect de Saut de Flash. Comme un cas particulier, nous avons introduit le modèle de MBP diagonal et avons reproduit la réponse anticipée de populations de neurones dans l'aire cortical V1. Nos résultats indiquent qu'un codage en position efficace et robuste peut être fortement dépendant de l'intégration le long de la trajectoire. / Coding the position of moving objects is an essential ability of the visual system in fulfilling precise and robust tracking tasks. This thesis is focalized upon this question: How does the visual system efficiently encode the position of moving objects, despite various sources of uncertainty? This study deploys the hypothesis that the visual systems uses prior knowledge on the temporal coherency of motion (Burgi et al 2000; Yuille and Grzywacz 1989). We implemented this prior by extending the modeling framework previously proposed to explain the aperture problem (Perrinet and Masson, 2012), so-called motion-based prediction (MBP). This model is a Bayesian motion estimation framework implemented by particle filtering. Based on that, we have introduced a theory on motion-based position coding, to investigate how neural mechanisms encoding the instantaneous position of moving objects might be affected by motion. Results of this thesis suggest that motion-based position coding might be a generic neural computation among all stages of the visual system. This mechanism might partially compensate the accumulative and restrictive effects of neural delays in position coding. Also it may account for motion-based position shifts as the flash lag effect. As a specific case, results of diagonal MBP model reproduced the anticipatory response of neural populations in the primary visual cortex of macaque monkey. Our results imply that an efficient and robust position coding might be highly dependent on trajectory integration and that it constitutes a key neural signature to study the more general problem of predictive coding in sensory areas.

Dans cette thèse, nous proposons une approche progressive (forward) pour la représentation probabiliste d'Equations aux Dérivées Partielles (EDP) nonlinéaires et nonconservatives, permettant ainsi de développer un algorithme particulaire afin d'en estimer numériquement les solutions. Les Equations Différentielles Stochastiques Nonlinéaires de type McKean (NLSDE) étudiées dans la littérature constituent une formulation microscopique d'un phénomène modélisé macroscopiquement par une EDP conservative. Une solution d'une telle NLSDE est la donnée d'un couple $(Y,u)$ où $Y$ est une solution d' équation différentielle stochastique (EDS) dont les coefficients dépendent de $u$ et de $t$ telle que $u(t,cdot)$ est la densité de $Y_t$. La principale contribution de cette thèse est de considérer des EDP nonconservatives, c'est-à- dire des EDP conservatives perturbées par un terme nonlinéaire de la forme $Lambda(u,nabla u)u$. Ceci implique qu'un couple $(Y,u)$ sera solution de la représentation probabiliste associée si $Y$ est un encore un processus stochastique et la relation entre $Y$ et la fonction $u$ sera alors plus complexe. Etant donnée la loi de $Y$, l'existence et l'unicité de $u$ sont démontrées par un argument de type point fixe via une formulation originale de type Feynmann-Kac. / This thesis performs forward probabilistic representations of nonlinear and nonconservative Partial Differential Equations (PDEs), which allowto numerically estimate the corresponding solutions via an interacting particle system algorithm, mixing Monte-Carlo methods and non-parametric density estimates.In the literature, McKean typeNonlinear Stochastic Differential Equations (NLSDEs) constitute the microscopic modelof a class of PDEs which are conservative. The solution of a NLSDEis generally a couple $(Y,u)$ where $Y$ is a stochastic process solving a stochastic differential equation whose coefficients depend on $u$ and at each time $t$, $u(t,cdot)$ is the law density of the random variable $Y_t$.The main idea of this thesis is to consider this time a non-conservative PDE which is the result of a conservative PDE perturbed by a term of the type $Lambda(u, nabla u) u$. In this case, the solution of the corresponding NLSDE is again a couple $(Y,u)$, where again $Y$ is a stochastic processbut where the link between the function $u$ and $Y$ is more complicated and once fixed the law of $Y$, $u$ is determined by a fixed pointargument via an innovating Feynmann-Kac type formula.