Cette thèse présente l'implémentation d'une stratégie visuelle bio-inspirée sur un aéroglisseur miniature totalement actionné, qui lui permet de naviguer dans le plan horizontal d'un tunnel inconnu. L'élaboration de ce pilote automatique, nommé LORA, fait suite aux études éthologiques menées sur l'abeille depuis ces dernières décennies et nous ont amené à énoncer le principe de la régulation du flux optique pour le contrôle du vol de croisière. Ce pilote automatique est un double régulateur de flux optique latéral constitué de deux boucles visuo-motrices interdépendantes contrôlant conjointement la vitesse d'avance et la position du robot par rapport aux obstacles sans avoir à mesurer ou estimer aucun de ces paramètres. La clé de voûte de ce système de guidage est une troisième boucle destinée à maintenir le cap grâce à un micro-gyromètre et un micro-compas magnétique permettant au robot d'effectuer des mouvements de translation qui génèrent sur son œil composé artificiel du flux optique de translation, seul dépendant du ratio vitesse/distance aux obstacles. Cet œil estime le flux optique grâce à ses deux ou quatre Détecteurs élémentaires de mouvement (total de 4 ou 8 pixels). L'aéroglisseur est alors capable de franchir sans collision, à la manière d'une abeille, divers tunnels : droit, fuselé ou présentant une pente, un virage, une absence de texture sur un mur ou même une zone non-stationnaire. Cette stratégie visuelle bio-inspirée fournit non seulement une solution de navigation élégante à destination de robots totalement actionnés mais elle permet aussi d'expliquer comment une abeille de 100mg peut naviguer sans l'aide de SONAR, RADAR, LIDAR, ou GPS. / In this work, we present for the first time a bio-inspired motion vision-based navigation strategy embedded on a miniature fully-actuated hovercraft allowing it to navigate safely on the horizontal plane of an unknown corridor. The design of this autopilot, called LORA, follows the ethological findings made on honeybees these last decades, which led us to elaborate the principle of the optic flow regulation which might be used by insects to control their flight. The bee-inspired LORA autopilot is a dual optic flow regulator which consists in two intertwined visuomotor feedback loops which control jointly the forward speed of the robot and its clearance to the obstacles. The keystone of this bio-inspired guidance system is a heading-lock system enabling the robot to move in translations and therefore experience a purely translational optic flow which depends only on the ratio speed/clearance to obstacles thanks to a micro-gyrometer and a micro-magnetic compass. The estimation of optic flow is made by a minimalist compound eye, made of two or four Elementary Motion Detectors (only 4 or 8 pixels). The hovercraft is therefore able to cross without crashing a straight or a tapered corridor, presenting a frontal sloping terrain, a bend, a textureless wall, or even a non-stationary section by automatically adapting both its forward speed and its clearance to the walls imitating the honeybee. This bio-inspired visual strategy not only provides an elegant navigation solution in an unknown environment aimed to equip fully-actuated miniature vehicles but also to explain how a 100mg honeybee can navigate with few computational ressources, i.e., without any SONAR, RADAR, LIDAR or GPS.

Deep learning has been rapidly growing in recent years obtaining excellent results for many computer vision applications, such as image classification and object detection. One aspect for the increased popularity of deep learning is that it mitigates the need for hand-crafted features. This thesis work investigates deep learning as a methodology to solve the problem of autonomous collision avoidance for a small robotic car. To accomplish this, transfer learning is used with the VGG16 deep network pre-trained on ImageNet dataset. A dataset has been collected and then used to fine-tune and validate the network offline. The deep network has been used with the robotic car in a real-time manner. The robotic car sends images to an external computer, which is used for running the network. The predictions from the network is sent back to the robotic car which takes actions based on those predictions. The results show that deep learning has great potential in solving the collision avoidance problem.

This thesis is based on the work carried out in the field of safety systems for Autonomous Guided Vehicles(AGV). With autonomous vehicles being more prominent today, safe traversing of these is a major concern. The same is true for AGVs working in industry environment like forklift trucks etc. Our work applies to industrial robots. The method described here is developed by closely following an algorithm developed for safe traversing of a robot using a warning field. The report describes the literature review with work related to the safe traversing, path planning and collision avoidance in robots. The next part is dedicated to describing the methodology of implementation of the Adaptive Warning Field Method and the Dynamic Window Approach. The evaluation of the Adaptive Warning Method with the previous developed Warning Field Methods is done and test cases are designed to test the working of the designed method. Vrep simulation environment and Industrial data is used to run a simulation of the robot using the method developed in this work. We find that the method performs better compared to the previous methods in the designed scenarios. Lastly we conclude the report with the future work that can be carried out to improve and extend the algorithm.

Avec cette génération d'appareils, la réalité virtuelle (RV) s'est réellement installée dans les salons des utilisateurs finaux. Ces appareils disposent de 6 degrés de liberté de suivi, ce qui leur permet de se déplacer naturellement dans les mondes virtuels. Cependant, pour une locomotion naturelle dans le virtuel, il faut un espace libre correspondant dans l'environnement réel. L'espace disponible est souvent limité. Les objets de la vie quotidienne peuvent rapidement devenir des obstacles pour les utilisateurs de RV s'ils ne sont pas éliminés. Les systèmes actuellement disponibles n'offrent qu'une aide rudimentaire pour résoudre ce problème. Il n'y a pas de détection d'objets potentiellement dangereux. Cette thèse montre comment les obstacles peuvent être détectés automatiquement avec des caméras d'imagerie à distance et comment les utilisateurs peuvent être avertis efficacement de leur présence dans l'environnement virtuel. 4 métaphores visuelles ont été évaluées à l'aide d'une étude des utilisateurs. / With this generation of devices, virtual reality (VR) has actually made it into the living rooms of end-users. These devices feature 6 degrees of freedom tracking, allowing them to move naturally in virtual worlds. However, for a natural locomotion in the virtual, one needs a corresponding free space in the real environment. The available space is often limited. Objects of daily life can quickly become obstacles for VR users if they are not cleared away. The currently available systems offer only rudimentary assistance for this problem. There is no detection of potentially dangerous objects. This thesis shows how obstacles can be detected automatically with range imaging cameras and how users can be effectively warned about them in the virtual environment. 4 visual metaphors were evaluated with the help of a user study.

The main goal of this work is to gain knowledge of how and to what extent state-of-the-artsimulation tools can be used in a conceptual development phase for vehicle dynamics control atVolvo Car Corporation (VCC).The first part of the thesis deals with an evaluation of vehicle dynamics simulation tools and theiruses. The three simulation tools selected for the study, namely Mechanical Simulation CarSim 8.2.1,IPG CarMaker 4.0.5, and VI-Grade CarRealTime V14, are briefly described and discussed. In order toevaluate and compare these tools with respect to application for vehicle dynamics control, a criterialist is developed covering aspects such as tool requirements and intended usage. Based on thecriteria list and certain identified drawbacks, a ranking of the tools is made possible. Furthermore,the process of developing vehicle models for the different tools is discussed in detail, along with theprocedure of validating the vehicle models.In the second part, the concept of Collision Avoidance Driver Assistance (CADA) function isintroduced and possible approaches for developing CADA functions are discussed in brief. It isimportant to note that the CADA functions in this work are based on cornering the vehicle i.e.maneuvering around the threat, rather than solely reducing vehicle speed. A number ofimplementations of the functions are developed in Simulink. A frequency analysis of a simplifiedlinear vehicle model is performed to investigate the influence of steering, differential braking, andtheir combination on the resultant lateral displacement of the vehicle during an evasive maneuver.The developed CADA functions are then simulated using the vehicle simulation tools. Two specificmetrics - Lateral Displacement gain and DeltaX - are formulated to evaluate the effectiveness of theCADA functions. Based on these metrics, the assistance obtained due to the functions for a specificevasive maneuver is compared.From the evaluation process of the three tools, two were considered suitable for the purpose ofsimulating collision avoidance functions. The evaluation of the CADA functions demonstrates thatcombined assistive steering with differential braking provides considerable assistance in order toavoid collisions. The simulation results also present interesting trends which provide a usefuldirection regarding the conditions for intervention by such collision avoidance functions during anevasive maneuver. The use of simulation tools makes it possible to observe these trends and utilizethem in the development process of the functions.

Cette thèse porte sur des techniques de commande à base d’optimisation dans le cadre des systèmes dynamiques Multi-Agents sous contraintes, plus particulièrement liées à l’évitement des collisions. Dans un contexte ensembliste, l’évitement des collisions au sein de la formation se traduit par des conditions de non intersection des régions de sécurité caractéristiques à chaque agent/obstacle. Grace à sa capacité à gérer les contraintes, la commande prédictive a été choisie parmi les méthodes de synthèse fondées sur des techniques d’optimisation. Tout d’abord, une structure de type leader-suiveur est considérée comme une architecture décentralisée élémentaire. La zone de fonctionnement de chaque suiveur est décidée par le leader et puis une loi de commande locale est calculée afin de garantir que les suiveurs restent à l’intérieur de la zone autorisée, permettant d’éviter les collisions. Ensuite, un déploiement des agents fondé sur l’approche de commande prédictive décentralisée, utilisant des partitions dynamiques de Voronoi, est proposé, permettant de ramener chaque agent vers l’intérieur de sa cellule Voronoi. Une des contributions a été de considérer le centre de Chebyshev comme cible à l’intérieur de chaque cellule. D’autres solutions proposent l’utilisation du centre de masse ou du centre obtenu par l’interpolation des sommets. Finalement, des méthodes ensemblistes sont utilisées pour construire un niveau supplémentaire de détection de défauts dans le cadre du système Multi-Agents. Cela permet l’exclusion des agents défectueux ainsi que l’intégration des agents extérieurs certifiés sans défauts dans la formation en utilisant des techniques de commande prédictive centralisée. / This thesis presents optimizationbased control techniques for dynamical Multi-Agent systems (MAS) subject to collision avoidance constraints. From the set-theoretic point of view, collision avoidance objective can be translated into non-overlapping conditions for the safety regions characterizing each agent/obstacle while maintaining the convergence towards a specified formation. Among the successful optimizationbased control methods, Model Predictive Control (MPC) is used for constraints handling. First, a leader-follower structure is considered as a basic decentralized architecture. The followers functioning zone assignment is decided by the leader and then the local linear feedback control is computed such that the follower operates strictly inside its authorized zone, offering anti-collision guarantees. Second, a dynamic Voronoi partition based deployment of the agents using an inner target driver is developed. The main novelty is to consider the Chebyshev center as the inner target for each agent, leading to an optimization-based decentralized predictive control design. In the same topic, other inner targets are considered such as the center of mass or vertex interpolated center. Third, set-theoretic tools are used to design a centralized FDI layer for dynamical MAS, leading to the exclusion of a faulty agent from the MAS formation and the integration of an external healthy/recovered agent in the current formation. The set-based FDI allows detecting and isolating these faulty agents to protect the current formation using centralized predictive control techniques.

Unmanned aerial vehicles are becoming more popular in today’s society, which results in the rise of laws intended to maintain safety. To abide by these, while allowing the technology to expand, functioning path-planning algorithms are required.This also includes having methods for detecting and managing obstacles. This project aims to improve an existing path-planning algorithm that is based on A* and implemented in Python.The solution consisted of using functions for finding polytopeintersection,as well as optimizing the collision avoidance and the search algorithm. In addition to that, realistic constraints were implemented on the generated trajectory in order to reflect real-life limitations. The results demonstrated that the paths were always feasible, with respect to input and position constraints. The program’s computation time was also reduced up to 89% of the original run-time. There is, however, still room for improvement since the original code generated a shorter path for the three scenarios it was created for. On the other hand,the improved algorithm could handle a new scenario, which the original code failed to do. / Obemannade flygfarkoster blir alltmer vanliga i dagens samhälle, vilket resulterar i uppkomsten av nya lagar ämnade åt att upprätthålla säkerhet. För att förhålla sig till dessa, samtidigt som teknologin tillåts expandera, krävs fungerande vägplaneringsalgoritmer. Där ingår det även att ha metoder för att upptäcka och hantera hinder. Detta projekt syftar till att förbättra en befintlig vägplaneringsalgoritm som är baserad på A* och implenterad i Python. Lösningsmetoden bestod av att använda inbyggda Python-funktioner ämnade åt att finna skärningar mellan polytoper, samt optimera kollisionshantering och sökalgoritmen. Dessutom infördes realistiska krav på den framställda vägen i syfte om att reflektera verlighetens begränsningar. Resultatet visade att vägarna alltid var genomförbara, med avseende på inmatningsoch positionsrelaterade villkor. Programmets beräkningstid hade även reducerats upptill 89% av den ursprungliga körtiden. Det finns dock utrymme för förbättringar då den ursprungliga koden generar en kortare väg för de tre scenarion den tillverkades för. Däremot kinde den förbättrade algoritmen hantera ett nytt scenario, en ursprungliga koden misslyckades med. / Kandidatexjobb i elektroteknik 2022, KTH, Stockholm

Multi-Agent coordination and control has been studied for a long time, but has recently gained more interest because of technology improvements allowing smaller, more versatile robots and other types of agents. To facilitate multi-agent experiments between heterogeneous agents, including robots and UAVs, we have created a test-bed with both simulation and hardware capabilities. This thesis discusses the creation of this unique, versatile test-bed for multi-agent experiments, also a unique graph creation algorithm, and some experimental results obtained using the test-bed.

Unmanned Aerial Vehicles (UAVs) can navigate in indoor environments and through environments that are hazardous or hard to reach for humans. This makes them suitable for use in search and rescue missions and by emergency response and law enforcement to increase situational awareness. However, even for an experienced UAV tele-operator controlling the UAV in these situations without colliding into obstacles is a demanding and difficult task. This thesis presents a human-UAV interface along with a collision avoidance method, both optimized for a human tele-operator. The objective is to simplify the task of navigating a UAV in indoor environments. Evaluation of the system is done by testing it against a number of use cases and a user study. The results of this thesis is a collision avoidance method that is successful in protecting the UAV from obstacles while at the same time acknowledges the operator’s intentions. / Obemannad luftfarkoster (UAV:er) kan navigera i inomhusmiljöer och genom miljöer som är farliga eller svåra att nå för människor. Detta gör dem lämpliga för användning i sök- och räddningsuppdrag och av akutmottagning och rättsväsende genom ökad situationsmedvetenhet. Dock är det även för en erfaren UAV-teleoperatör krävande och svårt att kontrollera en UAV i dessa situationer utan att kollidera med hinder. Denna avhandling presenterar ett människa-UAV-gränssnitt tillsammans med en kollisionsundvikande metod, båda optimerade för en mänsklig teleoperatör. Målet är att förenkla uppgiften att navigera en UAV i inomhusmiljöer. Utvärdering av systemet görs genom att testa det mot ett antal användningsfall och en användarstudie. Resultatet av denna avhandling är en kollisionsundvikande metod som lyckas skydda UAV från hinder och samtidigt tar hänsyn till operatörens avsikter.

No description available.