Over the past decade there has been a growing demand for interactive multimedia applications deployed over public IP networks. To achieve acceptable Quality of Ser- vice (QoS) without significantly modifying the existing infrastructure, the end-to-end applications need to optimize their behavior and adapt according to network char- acteristics. Most existing application optimization techniques are based on reactive strategies, i.e. reacting to occurrences of congestion. We propose the use of predic- tive control to address the problem in an anticipatory manner. This research deals with developing models to predict end-to-end single flow characteristics of Wide Area Networks (WANs). A novel signal, in the form of single flow packet accumulation, is proposed for feedback purposes. This thesis presents a variety of effective predictors for the above signal using Auto-Regressive (AR) models, Radial Basis Functions (RBF) and Sparse Basis Functions (SBF). The study consists of three sections. We first develop time- series models to predict the accumulation signal. Since encoder bit-rate is the most logical and generic control input, a statistical analysis is conducted to analyze the effect of input bit-rate on end-to-end delay and the accumulation signal. Finally, models are developed using this bit-rate as an input to predict the resulting accu- mulation signal. The predictors are evaluated based on Noise-to-Signal Ratio (NSR) along with their accuracy with increasing accumulation levels. In time-series models, RBF gave the best NSR closely followed by AR models. Analysis based on accu- racy with increasing accumulation levels showed AR to be better in some cases. The study on effect of bit-rate revealed that bit-rate may not be a good control input on all paths. Models such as Auto-Regressive with Exogenous input (ARX) and RBF were used to develop models to predict the accumulation signal using bit-rate as a modeling input. ARX and RBF models were found to give comparable accuracy, with RBF being slightly better.

The problem of estimating an adequate level of detail for an object for a specific view is one of the important problems in computer 3d-graphics and is especially important in real-time applications. The well-known continuous level-of-detail technique, Real-time Optimally Adapting Meshes (ROAM), has been employed with success for almost 10 years but has at present, due to rapid development of graphics hardware, been found to be inadequate. Compared to many other level-of-detail techniques it cannot benefit from the higher triangle throughput available on graphics cards of today. This thesis will describe the implementation of the new version of ROAM (informally known as ROAM 2.0) for the purpose of massive planetary data visualization. It will show how the problems of the old technique can be bridged to be able to adapt to newer graphics card while still benefiting from the advantages of ROAM. The resulting implementation that is presented here is specialized on spherical objects and handles both texture and geometry data of arbitrary large sizes in an efficient way.

Despite the importance of hair for characters in games, the process for creating realistic looking real-time hair is a highly difficult task that can only be performed by a few artists. As the industry is growing so does expectations of games graphics and the need for higher fidelity hairstyles. This paper sets out to explain current modeling techniques for hairstyling, and presents an alternate way for styling haircard models using curve deformers. Our Blende radd-on, based on this method is then previewed by industry professionals to evaluate their perception of the method’s effectiveness based on “control”, “time” and “preference”. Our conclusion is that per wisp curve based controls are easily learnt, relatively quick to use and gives adequate control over each haircard. The method therefore shows promise instream lining artists’ production pipelines. This test serves as a pilot for further exploration of curve based tools for production of real-time hair.

Machine learning systems have been demonstrated to be highly effective in various fields, such as in vision tasks for autonomous driving. However, the deployment of these systems poses a significant challenge in terms of ensuring their reliability and safety in diverse and dynamic environments. Online Unsupervised Domain Adaptation (UDA) aims to address the issue of continuous domain changes that may occur during deployment, such as sudden weather changes. Although these methods possess a remarkable ability to adapt to unseen domains, they are hindered by the high computational cost associated with constant adaptation, making them unsuitable for real-world applications that demand real-time performance. In this work, we focus on the challenging task of semantic segmentation. We present a framework for real-time domain adaptation that utilizes novel strategies to enable online adaptation at a rate of over 29 FPS on a single GPU. We propose a clever partial backpropagation in conjunction with a lightweight domain-shift detector that identifies the need for adaptation, adapting appropriately domain-specific hyperparameters to enhance performance. To validate our proposed framework, we conduct experiments in various storm scenarios using different rain intensities and evaluate our results in different domain shifts, such as fog visibility, and using the SHIFT dataset. Our results demonstrate that our framework achieves an optimal trade-off between accuracy and speed, surpassing state-of-the-art results, while the introduced strategies enable it to run more than six times faster at a minimal performance loss. / Maskininlärningssystem har visat sig vara mycket effektiva inom olika områden, till exempel i datorseende uppgifter för autonom körning. Spridning av dessa system utgör dock en betydande utmaning när det gäller att säkerställa deras tillförlitlighet och säkerhet i olika och dynamiska miljöer. Online Unsupervised Domain Adaptation (UDA) syftar till att behandla problemet med kontinuerliga domänändringar som kan inträffas under systemets användning, till exempel plötsliga väderförändringar. Även om dessa metoder har en anmärkningsvärd förmåga att anpassa sig till okända domäner, hindras de av den höga beräkningskostnaden som är förknippad med ständig nöndvändighet för anpassning, vilket gör dem olämpliga för verkliga tillämpningar som kräver realtidsprestanda. I detta avhandling fokuserar vi på utmanande uppgiften semantisk segmentering. Vi presenterar ett system för domänanpassning i realtid som använder nya strategier för att möjliggöra onlineanpassning med en hastighet av över 29 FPS på en enda GPU. Vi föreslår en smart partiell backpropagation i kombination med en lätt domänförskjutningsdetektor som identifierar nãr anpassning egentligen behövs, vilket kan konfigureras av domänspecifika hyperparametrar på lämpligt sätt för att förbättra prestandan. För att validera vårt föreslagna system genomför vi experiment i olika stormscenarier med olika regnintensiteter och utvärderar våra resultat i olika domänförskjutningar, såsom dimmasynlighet, och med hjälp av SHIFT-datauppsättningen. Våra resultat visar att vårt system uppnår en optimal avvägning mellan noggrannhet och hastighet, och överträffar toppmoderna resultat, medan de introducerade strategierna gör det möjligt att köra mer än sex gånger snabbare med minimal prestandaförlust.

The Quality of Service (QoS) routing approach is gaining an increasing interest in the Internet community due to the new emerging Internet applications such as real-time multimedia applications. These applications require better levels of quality of services than those supported by best effort networks. Therefore providing such services is crucial to many real time and multimedia applications which have strict quality of service requirements regarding bandwidth and timeliness of delivery. QoS routing is a major component in any QoS architecture and thus has been studied extensively in the literature. Scalability is considered one of the major issues in designing efficient QoS routing algorithms due to the high cost of QoS routing both in terms of computational effort and communication overhead. Localized quality of service routing is a promising approach to overcome the scalability problem of the conventional quality of service routing approach. The localized quality of service approach eliminates the communication overhead because it does not need the global network state information. The main aim of this thesis is to contribute towards the localised routing area by proposing and developing some new models and algorithms. Toward this goal we make the following major contributions. First, a scalable and efficient QoS routing algorithm based on a localised approach to QoS routing has been developed and evaluated. Second, we have developed a path selection technique that can be used with existing localized QoS routing algorithms to enhance their scalability and performance. Third, a scalable and efficient hierarchical QoS routing algorithm based on a localised approach to QoS routing has been developed and evaluated.

In the past decade, automobile manufacturers have gone through the initial adoption phase of electric mobility. The increasing momentum behind electric vehicles (EV) suggests that electrified storage systems will play an important role in electric mobility going forward. Lithium ion batteries have become one of the most common solutions for energy storage due to their light weight, high specific energy, low self-discharge rate, and non-memory effect. To fully benefit from a lithium-ion energy storage system and avoid its physical limitations, an accurate battery management system (BMS) is required. One of the key issues for successful BMS implementation is the battery model. A robust, accurate, and high fidelity battery model is required to mimic the battery dynamic behavior in a harsh environment. This dissertation introduces a robust and accurate model-based approach for lithium-ion battery management system. Many strategies for modeling the electrochemical processes in the battery have been proposed in the literature. The proposed models are often highly complex, requiring long computational time, large memory allocations, and real-time control. Thus, model-order reduction and minimization of the CPU run-time while maintaining the model accuracy are critical requirements for real-time implementation of lithium-ion electrochemical battery models. In this dissertation, different modeling techniques are developed. The proposed models reduce the model complexity while maintaining the accuracy. The thermal management of the lithium ion batteries is another important consideration for a successful BMS. Operating the battery pack outside the recommended operating conditions could result in unsafe operating conditions with undesirable consequences. In order to keep the battery within its safe operating range, the temperature of the cell core must be monitored and controlled. The dissertation implements a real-time electrochemical, thermal model for large prismatic cells used in electric vehicles' energy storage systems. The presented model accurately predicts the battery's core temperature and terminal voltage. / Thesis / Doctor of Philosophy (PhD)

Dans le domaine des systèmes temps-réel embarqués critiques, les méthodes de conception et de spécification et leurs outils associés doivent permettre le développement de systèmes au comportement temporel déterministe et, par conséquent, reproductible afin de garantir leur sûreté de fonctionnement. Pour atteindre cet objectif, on s’intéresse aux méthodologies de développement basées sur le paradigme Time-Triggered (TT). Dans ce contexte, nombre de propriétés et, en particulier, les contraintes temps-réel de-bout-en-bout, se voient satisfaites par construction. Toutefois, garantir la sûreté de fonctionnement de tels systèmes reste un défi. En général, les outils de développement existants n’assurent pas par construction le respect de l’intégralité des spécifications, celles-ci doivent, en général, être vérifiées à posteriori. Avec la complexité croissante des applications embarquées, celle de leur validation à posteriori devient, au mieux, un facteur majeur dans les coûts de développement et, au pire, tout simplement impossible. Il faut, donc, définir une méthode qui, tout en permettant le développement des systèmes corrects par constructions, structure et simplifie le processus de spécification. Les méthodologies de conception de haut niveau à base de composants, qui permettent la conception et la vérification des systèmes temps-réels critiques, présentent une solution ultime pour la structuration et la simplification du processus de spécification de tels systèmes.L’objectif de cette thèse est d'associer la méthodologie BIP (Behaviour-Interaction-Priority) qui est une approche de conception basée sur composants avec la plateforme d'exécution PharOS, qui est un système d'exploitation temps-réel déterministe orienté sûreté de fonctionnement. Le flot de conception proposé dans cette thèse est une approche transformationnelle qui permet de conserver les propriétés fonctionnelles des modèles originaux de BIP. Il est composé essentiellement de deux étapes. La première étape, paramétrée par un mapping de tâche défini par l'utilisateur, permet de transformer un modèle BIP en un modèle plus restreint qui représente une description haut niveau des implémentations basées sur des primitives de communication TT. La deuxième étape permet la génération du code pour la plateforme PharOS à partir de ce modèle restreint.Un ensemble d'outils a été développé dans cette thèse afin d'automatiser la plupart des étapes du flot de conception proposé. Ceci a permis de tester cette approche sur deux cas d'étude industriels ; un simulateur de vol et un relais de protection moyenne tension. Dans les deux applications, on vise à comparer les fonctionnalités du modèle BIP avec celles du modèle intermédiaire et du code généré. On fait varier les stratégies de mapping de tâche dans la première application, afin de tester leur impact sur le code généré. Dans la deuxième application, on étudie l'impact de la transformation sur le code généré en comparant quelques aspects de performance du code générer avec ceux d'une version de l'application qui a été développée manuellement. / In hard real-time embedded systems, design and specification methods and their associated tools must allow development of temporally deterministic systems to ensure their safety. To achieve this goal, we are specifically interested in methodologies based on the Time-Triggered (TT) paradigm. This paradigm allows preserving by construction number of properties, in particular, end-to-end real-time constraints. However, ensuring correctness and safety of such systems remains a challenging task. Existing development tools do not guarantee by construction specification respect. Thus, a-posteriori verification of the application is generally a must. With the increasing complexity of embedded applications, their a-posteriori validation becomes, at best, a major factor in the development costs and, at worst, simply impossible. It is necessary, therefore, to define a method that allows the development of correct-by-construction systems while simplifying the specification process.High-level component-based design frameworks that allow design and verification of hard real-time systems are very good candidates for structuring the specification process as well as verifying the high-level model.The goal of this thesis is to couple a high-level component-based design approach based on the BIP (Behaviour-Interaction-Priority) framework with a safety-oriented real-time execution platform implementing the TT approach (the PharOS Real-Time Operating System). To this end, we propose an automatic transformation process from BIPmodels into applications for the target platform (i.e. PharOS).The process consists in a two-step semantics-preserving transformation. The first step transforms a BIP model coupled to a user-defined task mapping into a restricted one, which lends itself well to an implementation based on TT communication primitives. The second step transforms the resulting model into the TT implementation provided by the PharOS RTOS.We provide a tool-flow that automates most of the steps of the proposed approach and illustrate its use on an industrial case study for a flight Simulator application and a medium voltage protection relay application. In both applications, we compare functionalities of both original, intermediate and final model in order to confirm the correctness of the transformation. For the first application, we study the impact of the task mapping on the generated implementation. And for the second application, we study the impact of the transformation on some performance aspects compared to a manually written version.

[pt] Nesta tese, propomos novas arquiteturas para redes neurais profundas utlizando métodos de atenção e álgebra multilinear para aumentar seu desempenho. Também exploramos convoluções em grafos e suas particularidades. Nos concentramos aqui em problemas relacionados à estimativa de pose em tempo real. A estimativa de pose é um problema desafiador em visão computacional com muitas aplicações reais em áreas como realidade aumentada, realidade virtual, animação por computador e reconstrução de cenas 3D. Normalmente, o problema a ser abordado envolve estimar a pose humana 2D ou 3D, ou seja, as partes do corpo de pessoas em imagens ou vídeos, bem como seu posicionamento e estrutura. Diveros trabalhos buscam atingir alta precisão usando arquiteturas baseadas em redes neurais de convolução convencionais; no entanto, erros causados por oclusão e motion blur não são incomuns, e ainda esses modelos são computacionalmente pesados para aplicações em tempo real. Exploramos diferentes arquiteturas para melhorar o tempo de processamento destas redes e, como resultado, propomos dois novos modelos de rede neural para estimativa de pose 2D e 3D. Também apresentamos uma nova arquitetura para redes de atenção em grafos chamada de atenção em grafos semânticos. / [en] This thesis proposes new architectures for deep neural networks with attention enhancement and multilinear algebra methods to increase their performance. We also explore graph convolutions and their particularities. We focus here on the problems related to real-time pose estimation. Pose estimation is a challenging problem in computer vision with many real applications in areas including augmented reality, virtual reality, computer animation, and 3D scene reconstruction. Usually, the problem to be addressed involves estimating the 2D and 3D human pose, i.e., the anatomical keypoints or body parts of persons in images or videos. Several papers propose approaches to achieve high accuracy using architectures based on conventional convolution neural networks; however, mistakes caused by occlusion and motion blur are not uncommon, and those models are computationally very intensive for real-time applications. We explore different architectures to improve processing time, and, as a result, we propose two novel neural network models for 2D and 3D pose estimation. We also introduce a new architecture for Graph attention networks called Semantic Graph Attention.

Virtual Environments have transformed over the years the way software is built and distributed. The recent growth of services such as Amazon EC2 or Google Cloud is representative of this trend and encourages developers to build software intended for virtual environments like virtual machines or containers. Despite all the benefits that virtualization brings (isolation, security, energy efficiency, stability, portability, etc.), the extra layer of software between the virtual environment and the hardware, called the hypervisor, increases the complexity of a system and the interpretation of its metrics. In this paper, we explore the situation of software performing latency measurements from a virtual environment. This is an example of a use-case where latency from the hypervisor could lead to measurable noise on the virtual environment if the hypervisor makes our environment wait for resources for milliseconds. To solve this problem, we propose an algorithm that will filter out this noise from computed metrics from the virtual environment. This algorithm was developed studying correlation between those metrics and hypervisor-induced latency spikes. We also try to be hypervisor agnostic which means that this work stays relevant whether a virtual environment is deployed locally or on a Cloud Service with different (and constantly evolving) hypervisor technologies. This research gives an overview of hypervisor technologies and how latency can appear when executing processes on virtual environments. As we will see, computing the metric and running the algorithm make network quality measurements from virtual environments more reliable and can explain unexpected latencies. / Virtuella miljöer(virtual environments) har under åren förändrat hur mjukvara(software) byggs och distribueras. Den senaste tidens tillväxt av tjänster som Amazon EC2 eller Google Cloud är representativ för denna trend och uppmuntrar utvecklare att bygga programvara avsedd för virtuella miljöer som virtuella maskiner eller behållare. Trots alla fördelar som virtualisering ger (isolering, säkerhet, energieffektivitet, stabilitet, portabilitet, etc.), ökar det extra lagret av mjukvara mellan den virtuella miljön och hårdvaran, kallad hypervisor, komplexiteten hos ett system och tolkning av dessa måtvärden. I denna artikel utforskar vi situationen för programvara som utför latensmätningar från en virtuell miljö. Detta är ett exempel på ett användningsfall där latens från hypervisorn kan leda till mätbart brus i den virtuella miljön om hypervisorn får vår miljö att vänta på resurser i millisekunder. För att lösa detta problem föreslår vi en algoritm som kommer att filtrera bort detta brus från beräknade mätvärden från den virtuella miljön. Denna algoritm utvecklades för att studera korrelationen mellan dessa mätvärden och hypervisor-inducerade latensspikar. Vi försöker också vara hypervisoragnostiska vilket innebär att detta arbete förblir relevant oavsett om en virtuell miljö distribueras lokalt eller på en molntjänst med olika (och ständigt utvecklande) hypervisorteknologier. Denna forskning ger en översikt över hypervisorteknologier och hur latens kan uppstå när processer körs i virtuella miljöer. Som vi kommer att se gör beräkning av måtten och körning av algoritmen mätningar av nätverkskvalitet från virtuella miljöer mer tillförlitliga och kan förklara oväntade latenser. / Les environments virtuels transforment depuis des années la manière de développer et distribuer du logiciel. La récente croissance de services comme Amazon EC2 ou Google Cloud reflète bien cette tendance et encourage les développeurs à construire du logiciel déployable sur des environnements virtuels comme des machines virtuels ou des conteneurs. Malgré tous les bénéfices que la virtualisation apporte (isolation, sécurité, efficacité énergétique, stabilité, portabilité, etc.), la couche logiciel supplémentaire entre l’environnement virtuel et le hardware, appelée hyperviseur, augmente la complexité d’un système et l’interprétation de ces métriques. Dans ce mémoire de projet de fin d’études, nous explorons la situation où un logiciel effectue des tests de latence depuis un envirronnement virtuel. Cette situation est un exemple d’un cas d’utilisation où la latence introduite par un hyperviseur peut mener à un bruit mesurable si l’hyperviseur fait attendre notre environnement dans l’ordre de grandeur de la milliseconde. Pour résoudre ce problème, nous proposons un algorithme qui filtre ce bruit à partir de métriques calculées directement depuis l’environnement virtuel. Cet algorithme est dévelopé en étudiant la corrélation entre nos métriques et une latence dite "hypervisor-induced". Cette approche permet donc une grande flexibilité dans la technologie sous-jacente de l’hôte puisque celui-ci peut utiliser des hyperviseurs différents ou même faire partie d’un service Cloud sans que notre solution en soit impactée. Ce mémoire donne aussi un aperçu de la technologie derrière un hyperviseur et comment de la latence peut s’introduire dans l’exécution d’un processus dans un environnement virtuel.

Le problème de l'efficacité énergétique dans le domaine des transports a comme principal défi savoir comment utiliser la source d'énergie pour que l'efficacité énergétique puisse être maximisée, c'est-à-dire comment le véhicule doit être conduit de telle sorte que la quantité minimale d’énergie est utilisée. Ce problème est le principal problème considéré dans cette thèse. Le véhicule est un prototype impliqué dans la course européenne Shell Eco-Marathon. La dynamique du véhicule est d'abord obtenu par l'identification expérimentale des paramètres. Une stratégie en boucle ouverte de conduite optimale en termes de consommation électrique est calculée. Plusieurs approches ont été étudiées pour le suivi de la référence optimale (stratégie de conduite optimale). Ces approches doivent prendre en compte les ressources limitées en taille mémoire et capacité de calcul. Une commande prédictive (MPC) basée sur la dynamique linéarisée est tout d'abord synthétisée. Le problème de poursuite nécessite une MPC avec contraintes variant dans le temps. La stabilité et la convergence de la commande prédictive sont prouvées à l'aide du formalisme des ensembles invariants. En troisième partie, à partie du modèle LPV, une adaptation de techniques standards basées sur des fonctions de Lyapunov quadratiques et à paramètres variants avec calculs hors-ligne est proposée. Elle est implémentée sur un banc de test. Enfin, une technique adaptative robuste avec identification en ligne de la dynamique est proposée et implémentée dans le véhicule. Cette technique a été testée et validée en course. Les résultats expérimentaux obtenus montrent de bonnes performances de la stratégie de conduite / In the field of transportation, the research on energy efficiency has been carried out for few decades by the automotive industry, where one of the main objectives is to reduce the energetic consumption. This particular problem can be rephrased as how the vehicle must be driven so that the minimum quantity of energy is used. This is the optimal driving strategy. In this project, a suitable model of the Vir'volt electric vehicle involved in the European Shell Eco-Marathon is obtained. The unknown parameters involved in the vehicle dynamics are estimated using Parameter identification from experimental data. The identified dynamics is used to derive an optimal driving strategy that is intended to be tracked on-line during the driving task. The tracking task is subject to time-varying polytopic constraint on the input and/or the state. A MPC-based tracking strategy that uses an homothetic transformation as a suitable time-varying invariant set is used. The time-varying invariant set guarantees the asymptotic stability of the control law. The problem of the MPC tracking for Linear Parametric Varying (LPV) systems is introduced. A new explicit MPC strategy for LPV systems is developed. This strategy uses a Parameter dependent Lyapunov Function (PDLF) to involve explicitly the time-varying parameter in the control law and so it reduces conservatism. A benchmark is used to test the performances of the optimal driving strategy and the explicit MPC tracking strategy. Finally, a robust adaptive technique with on-line identification of the dynamics is has been proposed and tested in the race showing good performances of the adaptive driving strategy