No description available.

Les systèmes distribués ont réussi à étendre la technologie de l’information à un public plus large, en termes d’emplacement et de nombre. Cependant, ces systèmes géo-répliqués doivent être évolutifs afin de répondre aux demandes toujours croissantes. De plus, le système doit pouvoir traiter les messages dans un ordre équivalent à celui de leur création afin d’éviter des effets indésirables. L’exécution suivant des ordres partiels fournit un ordonnancement d’événements que tous les nœuds suivront, ce qui permet donc le traitement des messages dans un ordre adéquat. Un système qui applique un ordre partiel simplifie le développement des applications distribuées et s’assure que l’utilisateur final n’observera pas des comportements défiant la causalité. Dans cette thèse, nous présentons des modèles statistiques pour différentes contraintes d’ordre partiel, en utilisant différentes distributions de modèles de latence. Étant donné un modèle de latence, qui donne le temps qu’il faut pour qu’un message passe d’un nœud à un autre, notre modèle s’appuie sur lui pour donner le temps supplémentaire qu’il faut pour appliquer un ordre partiel spécifique. Nous avons proposé les modèles suivants. Tout d’abord, dans une communication entre un et plusieurs nœuds, la probabilité que le message soit délivré dans tous les nœuds avant un temps donné. Deuxièmement, après la réception d’un message, la probabilité que tous les autres nœuds aient exécuté ce message avant temps donné. Troisièmement, dans une communication de un à plusieurs nœuds, la probabilité que le message soit arrivé à au moins un sous-ensemble d’entre eux avant un temps donné. Quatrièmement, l’ordre FIFO ou causal qui détermine si un message est prêt à être livré, dans un nœud ou plusieurs. Tout cela favorise la compréhension du comportement des systèmes distribués en présence d’ordres partiels. En outre, en utilisant cette connaissance, nous avons construit un algorithme qui utilise ces modèles de comportement du réseau pour établir un système de livraison causal fiable. Afin de valider nos modèles, nous avons développé un outil de simulation qui permet d’exécuter des scénarios adaptés à nos besoins. Nous pouvons définir les différents paramètres du modèle de latence, le nombre de clients et les charges de travail des clients. Cette simulation nous permet de comparer les valeurs générées de façon aléatoire pour chaque configuration spécifique avec les résultats prévus de notre modèle. Une des applications qui peuvent tirer profit de notre modèle, est un algorithme de livraison causale fiable. Il utilise l’information causale pour détecter les éléments manquants et réduit le besoin d’acquittement de message en contactant d’autres répliques seulement lorsque le message est supposé manquant. Cette information est fournie par notre modèle, qui définit les temporisateurs d’attente en fonction des statistiques du réseau et de la consommation des ressources. Enfin, cette application a été testée dans le même simulateur que les modèles, avec des résultats prometteurs, puis évaluée dans une expérience réelle utilisant Amazon EC2 comme plate-forme / Distributed systems have managed to extend technology to a broader audience, in both terms of location and numbers. However these geo-replicated systems need to be scalable in order to meet the ever growing demands. Moreover, the system has to be able to process messages in an equivalent order that they were created to avoid unwanted side effects. Partial order enforcement provides an ordering of events that all nodes will follow therefore processing the messages in an adequate order. A system that enforces a partial order simplifies the challenge of developing distributed applications, and ensures that the end-user will not observe causality defying behaviors. In this thesis we present models for different partial order enforcements, using different latency model distributions. While a latency model, which yields the time it takes for a message to go from one node to another, our model builds on it to give the additional time that it takes to enforce a given partial order. We have proposed the following models. First, in a one to many nodes communication, the probability for the message to be delivered in all the nodes before a given time. Second, in a one to many nodes communication from the receivers, the probability that all the other nodes have delivered the message after a given time of him receiving it. Third, in a one to many nodes communication, the probability that the message has arrived to at least a subset of them before a given time. Fourth, applying either FIFO or Causal ordering determining if a message is ready for being delivered, in one node or many. All of this furthers the understanding of how distributed systems with partial orders behave. Furthermore using this knowledge we have built an algorithm that uses the insight of network behavior to provide a reliable causal delivery system. In order to validate our models, we developed a simulation tool that allows to run scenarios tailored to our needs. We can define the different parameters of the latency model, the number of clients, and the clients workloads. This simulation allows us to compare the randomly generated values for each specific configuration with the predicted outcome from our model. One of the applications that can take advantage of our model, is a reliable causal delivery algorithm. It uses causal information to detect missing elements and removes the need of message acknowledgment by contacting other replicas only when the message is assumed missing. This information is provided by our model, that defines waiting timers according to the network statistics and resource consumption. Finally this application has been both tested in the same simulator as the models, with promising results, and then evaluated in a real-life experiment using Amazon EC2 for the platform

No description available.

As software systems are growing in complexity and size, reliability becomes a major concern. A large degree of industrial and academic efforts for increasing software reliability are directed towards design, testing and validation—activities performed before the software is deployed. While such activities are fundamental for achieving high levels of confidence in software systems, bugs still occur after deployment resulting in costly software failures. This dissertation presents assertion-based repair, a novel approach for error recovery from insidious bugs that occur after the system is deployed. It describes the design and implementation of a repair framework for Java programs and evaluates the efficiency and effectiveness of the approach on repairing data structure errors in both software libraries and open-source stand-alone applications. Our approach introduces a new form of assertions, assertAndRepair, for developers to use when checking the consistency of the data structures manipulated by their programs with respect to a set of desired structural and data properties. The developer provides the properties in a Java boolean method, repOk, which returns a truth value based on whether a given data structure satisfies these properties. Upon an assertion violation due to a faulty structure, instead of terminating the execution, the structure is repaired, i.e., its fields are mutated such that the resulting structure satisfies the desired properties, and the program proceeds with its execution. To aid developers in detecting the causes of the fault, repair-logs are generated which provide useful information about the performed mutations. The repair process is performed using a novel algorithm that uses a systematic search based on symbolic execution to determine valuations for the structures’ fields that result in a valid structure. Our experiments on repairing both library data structures, as well as, stand-alone applications demonstrate the utility and efficiency of the approach in repairing large structures, enabling programs to recover from crippling errors and proceed with their executions. Assertion-based repair presents a novel post-deployment mechanism that integrates with existing and newly developed software, providing them with the defensive ability to recover from unexpected runtime errors. Programmers already understand the advantages of using assertions and are comfortable with writing them. Providing new analyses and powerful extensions for them presents an attractive direction towards building more reliable software. / text

Software systems are now pervasive throughout our world. The reliability of these systems is an urgent necessity. A large degree of research effort on increasing software reliability is dedicated to requirements, architecture, design, implementation and testing---activities that are performed before system deployment. While such approaches have become substantially more advanced, software remains buggy and failures remain expensive. We take a radically different approach to reliability from previous approaches, namely contract-driven data structure repair for runtime error recovery, where erroneous executions of deployed software are corrected on-the-fly using rich behavioral contracts. Our key insight is to transform the software contract---which gives a high level description of the expected behavior---to an efficient implementation which repairs the erroneous data structures in the program state upon an error. To improve efficiency, scalability, and effectiveness of repair, in addition to rich behavioral contracts, we leverage the current erroneous state, dynamic behavior of the program, as well as repair history and abstraction. A core technical problem our approach to repair addresses is construction of structurally complex data that satisfy desired properties. We present a novel structure generation technique based on dynamic programming---a classic optimization approach---to utilize the recursive nature of the structures. We use our technique for constraint-based testing. It provides better scalability than previous work. We applied it to test widely-used web browsers and found some known and unknown bugs. Our use of dynamic programming in structure generation opens a new future direction to tackle the scalability problem of data structure repair. This research advances our ability to develop correct programs. For programs that already have contracts, error recovery using our approach can come at a low cost. The same contracts can be used for systematically testing code before deployment using existing as well as our new techniques. Thus, we enable a novel unification of software verification and error recovery. / text

xi, 51 p. : ill. (some col.) / Conducting field studies for human centric research often demands a significant amount of time and effort. Virtual Environments (VE) can be a potential alternative to reduce such requirements and help scale the field studies. However, we may experience a performance difference between (1) a virtual trial, and (2) a field trial of the same study. To learn under what circumstances a VE can successfully replace a field study and when it fails, this thesis describes a route-following experiment that compares the participants' performance between a simple VE and a field setup. The experiment results unveil that there is a significant difference in performance between a physical and a virtual setup for more challenging navigational tasks, whereas no significant difference is observed for simpler tasks. This finding encourages us to replace a less challenging field study with a simple VE, and explore the possibilities for a complex one. / Committee in charge: Dr. Stephen Fickas, Chairperson; Dr. Christopher Wilson, Member

A Machine Learning approach for error recovery in an industrial robot for the plastic mold industry isproposed in this master thesis project. The goal was to improve the present error recovery method byproviding a learning algorithm to the system instead of using the traditional algorithm-based control.The chosen method was the Support Vector Machine (SVM) due to the robustness and the goodgeneralization performance in real-world applications. Furthermore, SVM generates good classifierseven with a minimal number of training examples. In production, there will be no need for a humanoperator to train the SVM with hundreds or thousands of training examples to achieve goodgeneralization. The advantage with SVM is that good accuracy can be achieved with only a couple oftraining examples if the training examples are well designed.Firstly, the algorithm proposed was evaluated experimentally. The experiments consisted of correcthandling of classification performance on training examples, which was a hand-coded data set createdwith defined in- and output signals. Secondly, the results from the experiments were tested in asimulated environment. By using only a few training examples the SVM reached perfect performance.In conclusion, SVM is a good tool for classification and a suitable method for error recovery on theindustrial robot for the plastic mold industry. / En maskininlärningsstrategi för felhantering på industrirobotar inom plastformindustrin presenteras idetta examensarbete. Målet var att förbättra den nuvarande felhanteringen genom att applicera eninlärningsalgoritm istället för det traditionella förprogrammerade systemet till roboten. Den valdametoden är Support Vector Machine (SVM), då SVM är en robust metod som ger bra prestanda iverkliga tillämpningar. SVM genererar bra klassificerare även med ett minimalt antal träningsexempel.Fördelen med SVM är att god precision kan uppnås med bara ett par träningsexempel förutsatt attträningsexemplen är väldesignade. Detta betyder att operatörerna i produktionen inte behöver tränahundratals eller tusentals träningsexempel med SVM för att uppnå en god generalisering.I projektet utvärderasdes SVM metoden experimentellt varefter den testades i ett simuleringsprogram.Resultatet visade att SVM metoden gav en perfekt precision med hjälp av endast ett fåtal träningsdata.En slutsats från denna studie är att SVM är en bra metod för klassificering och lämplig för felhanteringpå industrirobotar inom plastindustrin.

No description available.

The ability to learn is essential for robots to function and perform services within a dynamic human environment. Robot programming by demonstration facilitates learning through a human teacher without the need to develop new code for each task that the robot performs. In order for learning to be generalizable, the robot needs to be able to grasp the underlying structure of the task being learned. This requires appropriate knowledge abstraction and representation. The goal of this thesis is to develop a learning by imitation system that abstracts knowledge of human demonstrations of a task and represents the abstracted knowledge in a hierarchical framework. The learning by imitation system is capable of performing both action and object recognition based on video stream data at the lower level of the hierarchy, while the sequence of actions and object states observed is reconstructed at the higher level of the hierarchy in order to form a coherent representation of the task. Furthermore, error recovery capabilities are included in the learning by imitation system to improve robustness to unexpected situations during task execution. The first part of the thesis focuses on motion learning to allow the robot to both recognize the actions for task representation at the higher level of the hierarchy and to perform the actions to imitate the task. In order to efficiently learn actions, the actions are segmented into meaningful atomic units called motion primitives. These motion primitives are then modeled using dynamic movement primitives (DMPs), a dynamical system model that can robustly generate motion trajectories to arbitrary goal positions while maintaining the overall shape of the demonstrated motion trajectory. The DMPs also contain weight parameters that are reflective of the shape of the motion trajectory. These weight parameters are clustered using affinity propagation (AP), an efficient exemplar clustering algorithm, in order to determine groups of similar motion primitives and thus, performing motion recognition. The approach of DMPs combined with APs was experimentally verified on two separate motion data sets for its ability to recognize and generate motion primitives. The second part of the thesis outlines how the task representation is created and used for imitating observed tasks. This includes object and object state recognition using simple computer vision techniques as well as the automatic construction of a Petri net (PN) model to describe an observed task. Tasks are composed of a sequence of actions that have specific pre-conditions, i.e. object states required before the action can be performed, and post-conditions, i.e. object states that result from the action. The PNs inherently encode pre-conditions and post-conditions of a particular event, i.e. action, and can model tasks as a coherent sequence of actions and object states. In addition, PNs are very flexible in modeling a variety of tasks including tasks that involve both sequential and parallel components. The automatic PN creation process has been tested on both a sequential two block stacking task and a three block stacking task involving both sequential and parallel components. The PN provides a meaningful representation of the observed tasks that can be used by a robot to imitate the tasks. Lastly, error recovery capabilities are added to the learning by imitation system in order to allow the robot to readjust the sequence of actions needed during task execution. The error recovery component is able to deal with two types of errors: unexpected, but known situations and unexpected, unknown situations. In the case of unexpected, but known situations, the learning system is able to search through the PN to identify the known situation and the actions needed to complete the task. This ability is useful not only for error recovery from known situations, but also for human robot collaboration, where the human unexpectedly helps to complete part of the task. In the case of situations that are both unexpected and unknown, the robot will prompt the human demonstrator to teach how to recover from the error to a known state. By observing the error recovery procedure and automatically extending the PN with the error recovery information, the situation encountered becomes part of the known situations and the robot is able to autonomously recover from the error in the future. This error recovery approach was tested successfully on errors encountered during the three block stacking task.

The ability to learn is essential for robots to function and perform services within a dynamic human environment. Robot programming by demonstration facilitates learning through a human teacher without the need to develop new code for each task that the robot performs. In order for learning to be generalizable, the robot needs to be able to grasp the underlying structure of the task being learned. This requires appropriate knowledge abstraction and representation. The goal of this thesis is to develop a learning by imitation system that abstracts knowledge of human demonstrations of a task and represents the abstracted knowledge in a hierarchical framework. The learning by imitation system is capable of performing both action and object recognition based on video stream data at the lower level of the hierarchy, while the sequence of actions and object states observed is reconstructed at the higher level of the hierarchy in order to form a coherent representation of the task. Furthermore, error recovery capabilities are included in the learning by imitation system to improve robustness to unexpected situations during task execution. The first part of the thesis focuses on motion learning to allow the robot to both recognize the actions for task representation at the higher level of the hierarchy and to perform the actions to imitate the task. In order to efficiently learn actions, the actions are segmented into meaningful atomic units called motion primitives. These motion primitives are then modeled using dynamic movement primitives (DMPs), a dynamical system model that can robustly generate motion trajectories to arbitrary goal positions while maintaining the overall shape of the demonstrated motion trajectory. The DMPs also contain weight parameters that are reflective of the shape of the motion trajectory. These weight parameters are clustered using affinity propagation (AP), an efficient exemplar clustering algorithm, in order to determine groups of similar motion primitives and thus, performing motion recognition. The approach of DMPs combined with APs was experimentally verified on two separate motion data sets for its ability to recognize and generate motion primitives. The second part of the thesis outlines how the task representation is created and used for imitating observed tasks. This includes object and object state recognition using simple computer vision techniques as well as the automatic construction of a Petri net (PN) model to describe an observed task. Tasks are composed of a sequence of actions that have specific pre-conditions, i.e. object states required before the action can be performed, and post-conditions, i.e. object states that result from the action. The PNs inherently encode pre-conditions and post-conditions of a particular event, i.e. action, and can model tasks as a coherent sequence of actions and object states. In addition, PNs are very flexible in modeling a variety of tasks including tasks that involve both sequential and parallel components. The automatic PN creation process has been tested on both a sequential two block stacking task and a three block stacking task involving both sequential and parallel components. The PN provides a meaningful representation of the observed tasks that can be used by a robot to imitate the tasks. Lastly, error recovery capabilities are added to the learning by imitation system in order to allow the robot to readjust the sequence of actions needed during task execution. The error recovery component is able to deal with two types of errors: unexpected, but known situations and unexpected, unknown situations. In the case of unexpected, but known situations, the learning system is able to search through the PN to identify the known situation and the actions needed to complete the task. This ability is useful not only for error recovery from known situations, but also for human robot collaboration, where the human unexpectedly helps to complete part of the task. In the case of situations that are both unexpected and unknown, the robot will prompt the human demonstrator to teach how to recover from the error to a known state. By observing the error recovery procedure and automatically extending the PN with the error recovery information, the situation encountered becomes part of the known situations and the robot is able to autonomously recover from the error in the future. This error recovery approach was tested successfully on errors encountered during the three block stacking task.