Une tâche fondamentale en robotique consiste à naviguer entre deux endroits. En particulier, la navigation dans le monde réel nécessite une planification à long terme à l'aide d'images RVB (RGB) en haute dimension, ce qui constitue un défi considérable pour les approches d'apprentissage de bout-en-bout. Les méthodes semi-paramétriques actuelles parviennent plutôt à atteindre des objectifs éloignés en combinant des modèles paramétriques avec une mémoire topologique de l'environnement, souvent représentée sous forme d'un graphe ayant pour nœuds des images précédemment vues. Cependant, l'utilisation de ces graphes implique généralement l'ajustement d'heuristiques d'élagage afin d'éviter les arêtes superflues, limiter la mémoire requise et permettre des recherches raisonnablement rapides dans le graphe. Dans cet ouvrage, nous montrons comment les approches de bout-en-bout basées sur l'apprentissage auto-supervisé peuvent exceller dans des tâches de navigation à long terme. Nous présentons initialement Duckie-Former (DF), une approche de bout-en-bout pour la navigation visuelle dans des environnements routiers. En utilisant un Vision Transformer (ViT) pré-entraîné avec une méthode auto-supervisée, nous nous inspirons des champs de potentiels afin de dériver une stratégie de navigation utilisant en entrée un masque de segmentation d'image de faible résolution. DF est évalué dans des tâches de navigation de suivi de voie et d'évitement d'obstacles. Nous présentons ensuite notre deuxième approche intitulée One-4-All (O4A). O4A utilise l'apprentissage auto-supervisé et l'apprentissage de variétés afin de créer un pipeline de navigation de bout-en-bout sans graphe permettant de spécifier l'objectif à l'aide d'une image. La navigation est réalisée en minimisant de manière vorace une fonction de potentiel définie de manière continue dans l'espace latent O4A. Les deux systèmes sont entraînés sans interagir avec le simulateur ou le robot sur des séquences d'exploration de données RVB et de contrôles non experts. Ils ne nécessitent aucune mesure de profondeur ou de pose. L'évaluation est effectuée dans des environnements simulés et réels en utilisant un robot à entraînement différentiel. / A fundamental task in robotics is to navigate between two locations. Particularly, real-world navigation can require long-horizon planning using high-dimensional RGB images, which poses a substantial challenge for end-to-end learning-based approaches. Current semi-parametric methods instead achieve long-horizon navigation by combining learned modules with a topological memory of the environment, often represented as a graph over previously collected images. However, using these graphs in practice typically involves tuning various pruning heuristics to prevent spurious edges, limit runtime memory usage, and allow reasonably fast graph queries. In this work, we show how end-to-end approaches trained through Self-Supervised Learning (SSL) can excel in long-horizon navigation tasks. We initially present Duckie-Former (DF), an end-to-end approach for visual servoing in road-like environments. Using a Vision Transformer (ViT) pretrained with a self-supervised method, we derive a potential-fields-like navigation strategy based on a coarse image segmentation model. DF is assessed in the navigation tasks of lane-following and obstacle avoidance. Subsequently, we introduce our second approach called One-4-All (O4A). O4A leverages SSL and manifold learning to create a graph-free, end-to-end navigation pipeline whose goal is specified as an image. Navigation is achieved by greedily minimizing a potential function defined continuously over the O4A latent space. O4A is evaluated in complex indoor environments. Both systems are trained offline on non-expert exploration sequences of RGB data and controls, and do not require any depth or pose measurements. Assessment is performed in simulated and real-world environments using a differential-drive robot.

The Northern Norrbotten ore district hosts a multitude of Sweden’s mineral deposits including world class deposits such as the Malmberget and Kirunavaara Iron oxide apatite deposits, the Aitik Iron oxide copper gold deposit, and a multitude of smaller deposits. Northern Norrbotten has been shaped by tectonothermal events related to the evolution of the Fennoscandian Shield and is a geologically complex environment. Without extensive rock outcropping and with most drilling localized to known deposits the regional to local scale of mineralization is not fully understood. To better understand the evolution and extent of the mineralization’s cross-disciplinary geosciences must be applied, where geophysical methods allow for interpretations of the deep and non-outcropping subsurface. Common earth modelling is a term describing a joint model derived from all available geoscientific data in an area, where geophysical models provide the framework.This study describes the geophysical modeling of two IOA deposits in Norrbotten, the Malmberget deposit in Gällivare and the Per-Geijer deposit in Kiruna. To better put these two deposits into a semi-regional setting magnetotelluric (MT) measurements have been conducted together with LKAB. LTU and LKAB have measured more than 200 MT stations in the two areas from 2016-2023. These measurements have then been robustly processed into magnetic transfer functions (impedances) for the broadband MT frequency spectrum (1000Hz,1000s). Then, all processed data judged to be of sufficient quality have been used for 3D inversion modelling using the ModEM code. The resulting conductivity/resistivity models reveals the local conductivity structure of the area, believed to be closely tied to the mineralization due to the conductive properties of the iron bearing minerals. Both areas yielded believable models which pinpointed known mineralization’s at surface as conductive anomalies and their connections to deeper regional anomalies.During modelling a robust iteratively re-weighted least square (IRLS) scheme has been implemented in the inversion algorithms. This scheme allows for objective re-weighting of data errors based on the ability for a given model discretization to predict individual datums. This, to better identify measurements which have been contaminated by local electromagnetic noise due to anthropogenic sources (mainly the power grid and railway). Due to the mathematical properties of the scheme, it allows for models which minimizes the L1 data error-norm instead of usual L2 minimization. This has yielded models whit sharper contrasts in resistivity and successfully emphasizes data believed to be reliable. Results indicate that the scheme was implemented successfully and the tradeoffs in data-fit are deemed acceptable.In addition, in the Kiruna study potential field data (magnetic total field and gravimetry) have been 3D modelled for the same area. These data sets have been inversion modelled in 3D using the MR3D-code developed at LTU with partners. Resulting 3D models have then been interpreted collectively both traditionally and with the use of machine learning methods. To guide interpretations more than 100 rock samples have been collected in the area and their petrophysical properties (density, magnetic susceptibility, electrical resistivity) have been measured at LTU. These petrophysical properties have been used to guide the machine learning methods for the 3D models by first using K-mean clustering on normalized petrophysical data and then using the resulting centroid vectors as input for a Gaussian mixture model of the similarly normalized 3D models. Resulting clusters show potential in being able to pick up sharp geological boundaries but expectedly is unable to fully capture geological structures one to one.

Robotics has been the subject of academic study from as early as 1948. For much of this time, study has focused on very specific applications in very well controlled environments. For example, the first commercial robots (1961) were introduced in order to improve the efficiency of production lines. The tasks undertaken by these robots were simple, and all that was required of a control algorithm was speed, repetitiveness and reliability in these environments. Now however, robots are being used to move around autonomously in increasingly unpredictable environments, and the need for robotic control algorithms that can successfully react to such conditions is ever increasing. In addition to this there is an ever-increasing array of robots available, the control algorithms for which are often incompatible. This can result in extensive redesign and large sections of code being re-written for use on different architectures. The thesis presented here is that a new generic approach can be created that provides robust high quality smooth paths and time-optimal path tracking to substantially increase applicability and efficiency of autonomous motion plans. The control system developed to support this thesis is capable of producing high quality smooth paths, and following these paths to a high level of accuracy in a robust and near time-optimal manner. The system can control a variety of robots in environments that contain 2D obstacles of various shapes and sizes. The system is also resilient to sensor error, spatial drift, and wheel-slip. In achieving the above, this system provides previously unavailable functionality by generically creating and tracking high quality paths so that only minor and clear adjustments are required between different robots and also be being capable of operating in environments that contain high levels of perturbation. The system is comprised of five separate novel component algorithms in order to cater for five different motion challenges facing modern robots. Each algorithm provides guaranteed functionality that has previously been unavailable in respect to its challenges. The challenges are: high quality smooth movement to reach n-dimensional goals in regions without obstacles, the navigation of 2D obstacles with guaranteed completeness, high quality smooth movement for ground robots carrying out 2D obstacle navigation, near time-optimal path tracking, and finally, effective wheel-slip detection and compensation. In meeting these challenges the algorithms have tackled adherence to non-holonomic constraints, applicability to a wide range of robots and tasks, fast real-time creation of paths and controls, sensor error compensation, and compensation for perturbation. This thesis presents each of the above algorithms individually. It is shown that existing methods are unable to produce the results provided by this thesis, before detailing the operation of each algorithm. The methodology employed is varied in accordance with each of the five core challenges. However, a common element of methodology throughout the thesis is that of gradient descent within a new type of potential field, which is dynamic and capable of the simultaneous creation of high-quality paths and the controls required to execute them. By relating global to local considerations through subgoals, this methodology (combined with other elements) is shown to be fully capable of achieving the aims of the thesis. It is concluded that the produced system represents a novel and significant contribution as there is no other system (to the author’s knowledge) that provides all of the functionality given. For each component algorithm there are many control systems that provide one or more of its features, but none that are capable of all of the features. Applications for this work are wide ranging as it is comprised of five component algorithms each applicable in their own right. For example, high quality smooth paths may be created and followed in any dimensionality of space if time optimality and obstacle avoidance are not required. Broadly speaking, and in summary, applications are to ground-based robotics in the areas of smooth path planning, time optimal travel, and compensation for unpredictable perturbation.

Commercial computer gaming is a large growing industry that already has its major contributions in the entertainment industry of the world. One of the most important among different types of computer games are Real Time Strategy (RTS) based games. RTS games are considered being the major research subject for Artificial Intelligence (AI). But still the performance of AI in these games is poor by human standards due to some fundamental AI problems those require more research to be better solved for the RTS games. There also exist some AI algorithms those can help us solve these AI problems. Anytime- Algorithms (AA) are algorithms those can optimize their memory and time resources and are considered best for the RTS games. We believe that by making AI algorithms anytime we can optimize their behavior to better solve the AI problems. Although many anytime algorithms are available to solve various kinds of AI problems, but according to our research no such study is been done to compare the performances of different anytime algorithms for an AI problem in RTS games. This study will take care of that by building our own research platform specifically design for comparing performances of our selected anytime algorithms for an AI problem. / Address: NaN Mob. +46 - 737 - 40 19 17

Commercial computer gaming is a large growing industry, that already has its major contributions in the entertainment industry of the world. One of the most important among different types of computer games are Real Time Strategy (RTS) based games. RTS games are considered being the major research subject for Artificial Intelligence (AI). But still the performance of AI in these games is poor by human standards because of some broad sets of problems. Some of these problems have been solved with the advent of an open real time research platform, named as ORTS. However there still exist some fundamental AI problems that require more research to be better solved for the RTS games. There also exist some AI algorithms that can help us solve these AI problems. Anytime- Algorithms (AA) are algorithms those can optimize their memory and time resources and are considered best for the RTS games. We believe that by making AI algorithms anytime we can optimize their behavior to better solve the AI problems for the RTS games. Although many anytime algorithms are available to solve various kinds of AI problems, but according to our research no such study is been done to compare the performances of different anytime algorithms for each AI problem in RTS games. This study will take care of that by building our own research platform specifically design for comparing performances of selected anytime algorithms for an AI problem