Phd Thesis Detailed

The complete list of all the PhD thesis (and HdRs) defended in our team, accessible from the IRIT's publications database, is here or you can explore this database from this form.

Back to the bare list of PhDs.

 

Ongoing PhD Theses


 

  • RECOVAC : conditions de REprise de CONtrôle par auto-observation des situations au sein d'un Véhicule Autonome Connecté, PhD in collaboration with the CLLE laboratory, prepared by Kristell Aguilar, under the supervision, for SMAC, of Marie-Pierre Gleizes and, for the CLLE, Loïc Caroux. Start: 10/2018.

Abstract:  This thesis project concerns the design of connected autonomous vehicles and in particular level-3 autonomous vehicles, i.e. vehicles that do not permanently require the driver to control the vehicle. The main objective of this thesis is to develop a computer system whose function is to help the driver regain control in certain situations and vice versa: to allow the vehicle to regain control over the human being. The objective of this thesis is to design a self-adaptive multi-agent system that learns by self-observation of the execution contexts for which the vehicle will no longer be able to control itself and vice versa, for which the driver can return the control. This system will be based on the analysis of the conditions for regaining control defined by specialists in cognitive ergonomics.

Related papers


 

  • Design of a vehicle fleet by self-adaptive multi-agent systems, PhD prepared by Guilhem Marcillaud, under the supervision of Marie-Pierre Gleizes and Valérie Camps. Start: 10/2018.

Abstract:  The theme addressed in this thesis concerns the autonomous and connected vehicle. It has to prove itself from the point of view of technological reliability, to which most projects focusing on technological developments answer. It is essential, after making a vehicle more reliable, to study how several connected autonomous vehicles will be able to interact in order to maximize the safety of the collective. The Society of American Automotive Engineers (SAE International) has defined 5 levels of autonomy: level 1, in which the driver performs all maneuvers, at level 5, in which the driver is completely autonomous and can do without driver and / or passenger. This thesis concerns both level 4 autonomous vehicles in which the vehicle drives and the supervised driver (without having to do so at all times) and can regain control of driving, as well as level 5 vehicles ( total autonomy of the vehicle).
In this thesis project, it is a question of studying how each vehicle communicates with its neighboring vehicles and how it behaves / reacts in response to the information provided by its neighborhood. A lock consists in particular in determining the information that is relevant to communicate among all the data recovered from the numerous sensors scattered in the vehicle and effectors.

Related papers



  • Meshing and Planning Large Area Acquisitions Support by Cooperative Multi-Agent Systems and Learning Techniques, PhD in collaboration with IRT Saint Exupéry and ISAE-Supaero, prepared by Timothée Jammot, under the supervision, for SMAC, of Pierre Glize and Elsy Kaddoum, and Serge Rainjonneau (IRT St Exupéry) and Emmanuel Rachelson (ISAE-Supaero). Start: 11/2017.

Abstract:   During the course of their missions Earth observation satellites can perform acquisitions of large zones on the surface of the Earth, a process called "large area acquisition". These tasks may require satellite fleets to acquire partial images ("meshes") of the zone over the span of several months to fully capture the large area. Images that are acquired this way must respect quality criteria which are dependent of parameters fluctuating during the mission, such as the cloud cover of the area of interest.
The problem to solve considers the scheduling of acquisitions of partial images by satellites on long planning periods (from a few weeks up to several months). Parallel to optimization algorithms, machine learning techniques have recently shown good results for various classification problems transposable to choices being performed in the frame of planning algorithms. Adaptive multi-agents systems have also demonstrated their ability to adapt and provide very satisfying solutions to combinatorial optimization problems in evolving environments. The thesis project combines both approaches in the context of satellite mission planning for large area coverage to take into account the uncertainties present in the long term planning process.

Related papers  


  • Dynamic Learning of the Environment for Eco-citizen Behaviours, joint PhD with the University of Palermo and ICAR of CNR, prepared by Davide Guastella, under the supervision, for SMAC, of Marie-Pierre Gleizes and Valérie Camps, for the University of Palermo, Cesare Valenti, and, for ICAR, Massimo Cossentino. Start: 10/2017.

Abstract:   The reality of a smart campus or more generally of a smart city passes through a regular observation of the environment for a better knowledge of human activities and the conditions in which these activities are taken. The direct way to obtain this knowledge is to provide the environment with a sufficient number of heterogeneous connected sensors. This knowledge of the environment is used to act with automatic devices to improve the well-being of users. Moreover deploying a large number of ad-hoc sensors in order to monitor the environmental parameters of the campus can be expensive. The costs are mainly related to the installation, the maintenance and the infrastructures of sensors in existing buildings. For these reasons, we aim to reduce the number of sensors to be installed and reduce the related costs. We consider that in the University of Toulouse III Paul Sabatier the number of people per day on the campus is around 20 000 among which 80% own a smartphone leading to about 500 000 daily data. Thus, in a smart campus context the use of smartphones is a good solution for avoiding the installation of ad-hoc sensors.
The objective of my PhD thesis is to employ a cooperative multi-agent approach to learn the state of environmental variables using partial and intermittent information coming from smartphones of persons situated in the campus of the university Toulouse III Paul Sabatier. The use of different heterogeneous devices will help the system to determine the most exactly possible the values of the environmental information (temperature, luminosity, humidity, noise...) and deduce the state of surrounding devices (lights, shutters, doors...) in every place.

Related papers  


  • Modèles multi-agents à dires d'acteurs, PhD in collaboration with GEODE Lab, prepared by Doryan Kaced, under the supervision, for SMAC, of Benoit Gaudou, and, for GEODE, of Mehdi Saqalli. Start: 10/2017.

Abstract:   The questions of the impact of oil pollution on the life and health of Ecuadorian farmersor the implementation of new water management policies at the scale of a watershed lead to consider these questions according to a systemic approach considering these systems as socio-environmental systems. The agent-based modeling approach is particularly well suited to model these systems, but requires to model the different aspects of these socio-environmental systems: bio-physical or social dynamics, or even individual decisions-making process. This thesis focuses on this last dimension by proposing an innovative and interdisciplinary methodology of dynamic formalization of the interactions between qualitative and quantitative aspects of individual behaviors. This modeling will be based on real field data, quantitative (territory, practical) but also qualitative data from PBRM, semi-structured surveys. This will be achieved in particular by the definition of a dedicated agent architecture to manage these these representational biases.

Related papers  


  • Lifelong Learning by Endogenous Feedback, Application to a Robotic System, PhD prepared by Bruno Dato, under the supervision of Marie-Pierre Gleizes and Frédéric Migeon. Start: 10/2017.

Abstract:  This thesis aims to study and design a robotic system capable of learning continuously through endogenous feedback. The core of this thesis is to design the learning and decision-making abilities of this robotic system based on adaptive multi-agents systems. This work follows a thesis of the team whose object was the learning by demonstration for a robot.
In this thesis, the robotic system will have to learn not by demonstration, that is to say by imitating what the tutor showed it, but depending on the interactions it will have with its environment. It will thus have to be able to modify the goals it pursues. We name this systems agnostic systems. Theses systems do not have an intrinsic knowledge of the task they have to solve. It will therefore be necessary to provide the system with the capacity to self-observe and to forge a representation of the consequences of its activity on its environment. These abilities will enable it to modify its action on the environment and thus modify its task. The acquisition of emerging processes by the AMAS theory will make possible to carry out collective behaviors functionally adequate in real time.

Related papers


 

  • Design of an Intelligent Engine for Opportunistic Software Composition in Ambient and Mobile Environments, PhD prepared by Walid Younes, under the supervision of Jean-Paul Arcangeli, Sylvie Trouilhet and Françoise Adreit. Start: 10/2017.

Abstract:   Ambient and mobile systems consist of networked devices and software components surrounding human users and providing services. From the services present in the environment, other services can be composed opportunistically and automatically by an intelligent system and pushed to the user.
The latter must not only to be aware of existing services but also be kept in the loop in order to both control actively the services and influence the automated decision.
In this thesis, we analyze the requirements for such an ambient intelligence placing the user in the loop: service presentation, acceptation, edition... strengthening of the composition system through learning and feedback. We propose an approach aimed at answering the identified requirements. The originality of our approach consists in the use of a model-driven engineering paradigm.

Related papers   Poster


  • User-Centric Modeling for an Opportunistic Composition System, in collaboration with SM@RT team, PhD prepared by Maroun Koussaifi, under the supervision, for SMAC, of Jean-Paul Arcangeli and Sylvie Trouilhet, and, for SM@RT, of Jean-Michel Bruel. Start: 10/2017.

Abstract:   Ambient and mobile systems consist of networked devices and software components surrounding human users and providing services. From the services present in the environment, other services can be composed opportunistically and automatically by an intelligent system and pushed to the user.
The latter must not only to be aware of existing services but also be kept in the loop in order to both control actively the services and influence the automated decision.
In this thesis, we analyze the requirements for such an ambient intelligence placing the user in the loop: service presentation, acceptation, edition... strengthening of the composition system through learning and feedback. We propose an approach aimed at answering the identified requirements. The originality of our approach consists in the use of a model-driven engineering paradigm.

Related papers   Poster


  • Dynamiques des parties non denses dans des réseaux complexes, in collaboration with XSyS, by Mehdi Djellabi, under the supervision, for XSyS, of Bertrand Jouve and, for SMAC, Frédéric Amblard. Start: 2017.
    Abstract:

Abstract:   Le terme de "réseau" désigne un système d’interactions entre entités (personnes, animaux, neurones, gènes, entreprises...) qui admet une représentation mathématique abstraite sous forme de graphe : les entités correspondent aux sommets du graphe et les interactions ou les relations aux arêtes. De nombreuses recherches ont consisté à préciser les propriétés partagées par les grands réseaux que l’on rencontre en pratique afin élaborer et par la suite d’améliorer les modèles génériques. Ces recherches de régularités et de lois d’évolutions, caractéristiques de la physique et des sciences de la nature, ont été largement portées par les chercheurs de ces domaines.
Il est cependant maintenant largement convenu que ces propriétés topologiques partagées ne sont pas suffisantes pour expliquer la diversité et la complexité des architectures des réseaux réels rencontrés. La dynamique et la structure multi-échelles de ces réseaux sont des caractéristiques au fondement même de ces architectures. Une approche possible pour capturer la structure multi-échelles peut consister à identifier les petits mondes qui correspondent souvent à des sous-parties denses en connexion (recouvrantes ou non) du réseau (appelées aussi « communautés ») et à étudier le réseau mésoscopique d’assemblage de ces communautés et sa dynamique. Voire d’étudier ensuite séparément chacune de ces communautés.
L’objet de cette thèse est de compléter ces outils de modélisation réseau en prenant le contre-pied de cette approche par communautés et en considérant que les parties non-denses d’un réseau sont aussi des éléments structurants de son organisation. Cette posture rejoint en sociologie des réseaux celle de l’importance des trous structuraux de Burt qui jusqu’à aujourd’hui n’est pas implémentée dans des modèles formels de réseaux sociaux. L’objectif sera de montrer que des études hybrides basées à la fois sur des parties denses et non dense permettent des avancées qualitative et quantitative dans la description des réseaux complexes et de leurs dynamiques, et in fine de proposer des modèles génératifs plus réalistes basés sur ces deux dimensions.

Related papers



  • Modelling and Simulation of Social Processes in Rural Territories for the Evaluation of Public Local Development Policies, in collaboration with the IDETCOM, PhD prepared by Marcos Aurelio Santos da Silva, under the supervision, for SMAC, of Christophe Sibertin-Blanc and, for the IDETCOM, of Pascal Roggero. Start: 04/2016.

Abstract:   bserving the strategic dimension of the territory at the regional level represents an important step for the development of public policies, for the allocation of resources and for the development of collective spaces. In rural territories, we have several social processes that establish the actors' social game, the institutional relations of power and the relations of forces that constraint decision-making process. In Brazil, we have two territorial public policies for sustainable development, the National Program for the Sustainable Development of Rural Territories (Pronat) and the Program Territories of Citizenship (PTC) which aim at balance of power relations between social actors of each Rural Territory ruled by a Collegiality for Territorial Development (CODETER). This research is based on the hypothesis that the rural territories subjected to public policies Pronat and PTC have experienced a balancing of the social game in terms of power relations of social actors engaged in CODETER. We start from the idea that territorialized social processes are complex phenomena that can be understood by the systemic approach.
This research aims at the development of a method of modeling and simulation of the power relations between the institutions of the socio-territorial systems for the evaluation of public policies of local development. The research approach is based on systems theory, complexity, complex systems theory, limited rationality and recursive modeling supported by one or more social theories. The starting point is the Soclab approach, which formalized the Sociology of Collective Organized Action for the analysis of power relations between social actors. The proposed approach will be validated by its application on two case studies, the Rural Territory of Sergipe and the Rural Territory of Lower São Francisco, examples of a Brazilian public territorial policy which aims to balance the power between social actors in rural areas.

Related papers



  • Dynamic Management of Coupled Self-adaptive Systems: Application to Ambient Socio-technical Systems, PhD ("en alternance") prepared by Fabrice Crasnier, under the supervision of Marie-Pierre Gleizes, Pierre Glize and Jean-Pierre Georgé. Start: 09/2015.

Abstract:   In recent years, the number of connected objects continues to grow in our personal and professional environment. Some of them exponentially colonize the industrial world and urban cities under the aegis of the digital transformation that is now called Industry 4.0 and smart cities. The goal of this anthill of connected objects is to facilitate our immersion in an environment equipped with sensors and actuators so as to make our everyday life more pleasant while respecting a certain ethic of eco-citizenship.
Defining an approach to local real time adaptation for systems allowing a collective convergence means immersing in a dynamic environment of systems having capacities of real time learning in order to adapt to the evolutions of the environment not known during the design phase. Ambient sociotechnical systems are particularly relevant because they have many devices immersed in the human environment to facilitate their activities while reducing the cognitive load. Moreover, they contribute to the emergence of the notion of well-being felt by a human being depending on their state of equilibrium.
The learning of the environment related to the well-being is carried out in our research by a learning technique relying on self-adaptive multi-agent systems representing the four characteristics of human physical comfort ie thermal comfort, visual comfort, olfactory comfort or comfort auditory (called "Comfort MAS"). These will initially be responsible for learning about the thermal, luminous, olfactory or auditory environments and, in a second phase, for providing their criticality levels to the multi-agent systems representing the ambient socio-technical systems of the sensors and effectors of the environment (called "Devices MAS"). The criteria can be determined either by observing the actions of the users that will be recorded in the form of a range of acceptable data that can be assimilated to the notion of satisfaction, or by the standards and norms resulting from the studies made on each type of comfort [ASHRAE Standard 55 - Thermal Comfort]. Comforts influencers are, in turn, the MAS Devices that produce changes in the environment by the different actions they perform.
Finally, to account for the optimal energy consumption and the gains made by the implementation of such digital objects, it is necessary that these daily new tools can know their own consumption in order to better evaluate their relevance to produce actions in their environment. The "Consumption MAS" is the self-adaptive ambient sociotechnical system able to learn these consumptions in real time.

Related papers


  • Integration of the Social Dimension into Processes, PhD prepared by Hanane Ariouat, under the supervision of Chihab Hanachi and Eric Andonoff. Start:09/2015.

Abstract:   A process is conventionally represented using three perspectives (informational, organizational and behavioral). The focus is often put on the behavioral aspect which specifies the coordination of the tasks composing the processes. It can be defined mechanistically (established order), systemically (event-driven) or in an emergent way (seen as a system of cooperative actors). This thesis will rather be in the emerging context. The idea is then to produce a set of mechanisms (methods, models, tools...) that facilitates the cooperation of these actors while taking into account the social perspective as a first-class citizen dimension. This dimension should integrate interpersonal relationships (power, responsibility, networks influence, social context ...) as well as their modes of interaction. This social dimension may be inspired by cooperation techniques derived from distributed artificial intelligence or groupware and will include recommendations concerning the social dimension but also the three above perspectives.
The ANR GéNéPi project, which addresses the problem of building a process aware response to natural crisis such as floods ("Flood of the Loire River" as an example) will serve as a support for the illustration of our proposals.

Related papers



  • Evolution and Perturbation of Social Networks, PhD prepared by Audren Bouadjio Boulic under the supervision of Chihab Hanachi and Frédéric Amblard. Start: 10/2014.

Abstract:   Le projet GenStar vise à proposer un outil fournissant une population synthétique, spatialisé dans une ville, et interconnecté. Les réseaux sont une représentation pratique et lisible des interactions entre entités.
Dans ce cadre, la génération de réseaux sociaux réalistes est donc utilisée pour interconnecter cette population synthétique. L'objet de la thèse est de fournir des modèles de génération fournissant des réseaux réalistes, et capable d'évoluer avec le temps ou après avoir subi des perturbations.
L'un des modèles de génération consiste à utiliser une simulation agents, en dotant ces agents de comportements de construction topologique sur le réseau. L'interaction entre les agents, sous certaines contraintes, va alors créer un certain type de réseau. On explore les paramètres initiaux à l'aide d'un algorithme génétique pour trouver les conditions de création pour le type de réseau souhaité.

Related papers


Phd Theses Recently Defended


The complete list of all the PhD thesis (and HdRs) defended in our team, accessible from the IRIT's publications database, is here or you can explore this database from this form.


  • Automatic Analysis of User Traces for Continuous Testability and Enhancement of Complex Software by Adaptive Multi-agent Systems, CIFRE PhD in collaboration with Berger-Levrault, defended on 19/06/2020 (also on YouTube) by Florent Mouysset, under the supervision, for SMAC, of Marie-Pierre Gleizes and Frédéric Migeon.

Abstract:   The software for public services are always more complex as the regulation evolves constantly and the user requirement continues to refine. Thus, it is sometimes difficult to maintain and use these applications.
To guarantee an acceptable quality level, we explore, in this thesis, the possible exploitation of user traces. A user trace is a footprint left by a user when using a software. At short term, the objective is providing feedback for the production teams. For example, it we could allow us to confront the acceptance test traces to the actual usage. At long term, a smart assistant could be designed to help the user to perform some tricky tasks or repetitive actions. To do so, it is necessary to model the software from user traces.
Currently, the existing methods are limited, particularly when the noise level is too high or when the traces are too complex. The adaptive multi-agents systems have already proved their ability to manage complex data. Hence, this approach seems appropriate to our problem. Now, after defining the business domain vocabulary, we work on the agent behaviors.

Related papers


  • Self-Organization of Real-time, Multi-objective and Multi-criteria Traffic in a Virtual World, in collaboration with Sopra Steria, CIFRE PhD defended on 03/06/2020 (also on YouTube) by Augustin Degas, under the supervision, for SMAC, of Marie-Pierre Gleizes and Elsy Kaddoum.

Abstract:  A virtual world is composed of numerous mobile entities. The traffic that results from their behavior is characterized by two main levels : (1) The first one is the micro level, which is composed of those mobile entities. Those entities have their own properties and physical constraints, like a maximal speed, a shape, a position, some differential constraints and so on, amongst some other constraints or goals, for exemple a destination or the need to avoid other entities. (2) Resulting from the interactions between the entities, the second level is the macro level. This level characterizes properties of the traffic, from straightforward properties like density or mean speed, to other properties like a number of conflicst created or avoided, or the adaptation of the entities to a sudden modification of their environment.
The goal of this PhD thesis is to produce a realistic traffic that satisfies a set of constraints and objectives on both of those levels. This traffic should also allow humans to interact with the simulation, and adapt to those interactions.

Related papers


Abstract:   Production forecasting has become an increasingly important tool for integrating renewable energy sources. Forecasting wind power is especially complex because it depends on several heterogeneous factors: weather, electromechanics, topography, etc.
This research work (related to the meteo*swift project) aims at developing an hour or day ahead forecasting system of electrical production from meteorological data, for a wind turbine or a wind farm. Since some factors which impact performance may evolve over time (surface roughness, new buildings, noise limitation...), it becomes mandatory to provide an adaptive forecasting system which has no need to be explicitly reprogrammed when input conditions evolve.
Therefore, using multi-agent systems will enable this self-adaptation and also taking into account more parameters than a standard power curve which uses only wind speed to calculate electrical power. Furthermore, other factors such as wind direction, turbulent kinetic energy or air density will be taken into account.

Report  Related papers


Abstract:   The main purpose of this research is to provide a brand new set of tools, algorithms and models called AMAS4BigData that make the use of Adaptive Multi-Agent Systems (AMAS) as generic as possible allowing anyone to use this approach for its own data analysis task without any knowledge of the AMAS theory, especially when the data sets grow fast or change in real-time. To do so, AMAS4BigData has to be able to:

  • Analyze the dynamic data flow that comes either continuously or in batches to the database, or changing over time;
  • Use just-in-time analysis and compare to the database in order to pre-filter the data before integration;
  • Manage incomplete and uncertain data;
  • Analyze properties, clusters and characteristics so as to provide dynamic visualizations;
  • Raise an alert about inconsistent data or types of events/patterns.

These functionalities are Big Data’s main challenges. They can potentially be dealt with the Adaptive Multi-Agent Systems which our team is one of the few in the world to master. Thus, to illustrate its genericity AMA4BigData might be applied on several areas like:

  • High risk situations detection and emergency detection in medical supervision, especially maintaining elderly people at home using sensors;
  • Energy resources optimization and more widely ambient systems optimization;
  • Shed light on interesting properties about travels and services at town scale using Open Data.

Report   Related papers   Poster   Demos


Abstract:   Cells of a higher organism are diverse and coordinate their actions by constantly exchanging chemical messages. Despite numerous methodological and technological advances, it is still very difficult to characterize this dynamic state in vivo or in vitro. However, many questions are still open as to the nature of intercellular communication. Indeed, even though most chemical messengers (proteins in general) are now well characterized, their action on the cells seems in many cases to depend on the context of the cells that receive these messages. In some extreme cases the same messenger may have an action or its opposite on a cell depending on the situation of the cell. It is possible to explain this type of phenomenon if the messengers are not univocally linked to a single action but rather participates in the elaboration of a message formed by several messengers which together determine the action that the cell must perform. This explanation also has the advantage of going in the direction of nature which tends to minimize the resources used for a given task. Indeed, DNA encodes the amino acid sequences of the proteins using 4 bases grouped by 3 rather than 20 bases each encoding for a specific amino acid. Similarly, the proteins are all composed from an "alphabet" of 20 amino acids rather than having different complex molecules for each function. It is therefore reasonable to assume that this strategy also applies to cell-to-cell communication and rather than having a messenger for each possible action of the cell, a signal combining system will reduce the number of messenger types to produce.
However, it is difficult experimentally to follow dynamically and simultaneously the concentration of different messengers around a single cell. Biology is still largely a statistical science that draws its lessons from the average over large cell populations.
To validate this hypothesis of combining signals it may therefore be advantageous to simulate a simplified cellular system and to observe the conditions for the emergence of communication between the cells and the nature of this communication. If for a simple system, the communication is already of contextual nature it becomes very probable that this is also the case for the real cells. In this case the study of real systems would be enriched by this concept.

Report   Related papers  Poster   Demos


 

Abstract:   Cette thèse porte sur la conception de systèmes multi-agents ambiants pour faciliter l’autonomie et l’accessibilité aux espaces publics, en particulier aux personnes âgées.
Le vieillissement de la population et son corollaire, l’accroissement des situations de handicap liées à la progression en âge, appellent une approche nouvelle de l’accessibilité des espaces publics urbains. Actuellement, certaines difficultés, à l’exemple des déficiences cognitives légères (notamment amnésiques), rendent ardu l’accès à la ville des personnes âgées concernées et favorisent leur (auto) confinement à domicile.
Dans ce contexte, cette thèse propose la conception d’un dispositif outillé d’assistance aux personnes âgées dans leurs activités quotidiennes à l’extérieur. Pour permettre son acceptabilité, ce dispositif sociotechnique ambiant s’appuie sur une démarche de conception interdisciplinaire et collaborative. La complexité de ce travail réside dans la prise en compte, la compréhension et la modélisation des différents points de vue sociotechniques corrélés (organisationnel, collaboratif et technique) et de l’environnement ouvert dans lequel le dispositif doit s’insérer.
L’utilisation des Systèmes Multi-Agents permet d’appréhender le dispositif visé et son interaction avec l’environnement avec le bon niveau d'abstraction. L’apprentissage de la collaboration inter-agents est facilité par l’utilisation de la théorie des AMAS.
Les contributions sont les suivantes : (1) La définition d’une approche de conception centrée utilisateur et basée sur les scénarios, pour laquelle nous avons défini un cycle de vie et fourni un méta-modèle de scénarios. (2) Une spécification et conception d’un système multi-agent adaptatif, doté d’apprentissage par renforcement. (3) Un prototype de ce dispositif avec quelques expérimentations.

Report   Related papers


  • A Cooperative Architecting Procedure for Systems of Systems based on Self-adaptive Multi-Agent Systems, PhD defended on 30/11/2017 by Teddy Bouziat, under the supervision of Pierre Glize, Valérie Camps and Stéphanie Combettes.

Abstract:   Depuis la fin de seconde guerre mondiale, l’ingénierie des systèmes a permis le développement de méthodologies et d’outils pour contrôler le développement de systèmes et de projets de plus en plus complexes. Ce domaine de recherche interdisciplinaire continue de se développer de nos jours. Cependant, en 1990, la chute de l'URSS a provoqué un changement de doctrine militaire aux Etats-Unis en passant d'une confrontation bipolaire à une mondialisation des conflits comportant une grande variété de menaces. Sa nouvelle doctrine était de réutiliser et de faire collaborer ses systèmes de défense existants pour produire un système de défense de haut niveau, décentralisé, adaptable et composé de systèmes indépendants et complexes. C'est l'apparition du concept de Système de Systèmes (SdS). Après 2000, ce concept s’est étendu au domaine civil tel que la gestion de crise ou les systèmes logistiques. Plus précisément, un SdS est un système complexe caractérisé par la nature particulière de ses composants: ces derniers, qui sont des systèmes, ont tendance à être autonomes en terme opérationnel et en terme de gestion ainsi que géographiquement distribués. Cette caractérisation spécifique a conduit à repenser les domaines de recherche de l’ingénierie des systèmes, tels que la définition, la taxonomie, la modélisation, l'architecture, etc. Ainsi par exemple, l’architecture des SdS se concentre sur la façon dont les composants indépendants d'un SdS peuvent être structurés de manière dynamique et peuvent changer de manière autonome leurs interactions de manière efficace pour atteindre l'objectif du SdS et pour faire face à la forte dynamicité de l'environnement.
Cette thèse de doctorat se concentre principalement sur deux domaines de recherche des SdS. Le premier concerne leur modélisation formelle et le second leur architecture. Dans le premier, nous proposons un nouveau modèle de SdS appelé le modèle SApHESIA (SoS Architecting HEurIstic based on Agent). Nous avons utilisé la théorie des ensembles ainsi qu’un modèle basé agent afin de prendre en compte les caractéristiques communes des SdS trouvées dans la littérature. Deuxièmement, nous proposons une nouvelle méthodologie d'architecture pour les SdS basée sur l'approche par AMAS (Adaptive Multi-Agent System) qui préconise une coopération complète entre tous les composants d’un SdS à travers le concept de criticité. La criticité est une métrique qui représente la distance entre l'état actuel d'un composant et ses objectifs. Dans cette méthodologie, l'architecture du SdS évolue d’elle-même au fil du temps en s’auto-organisant pour s'adapter à la dynamicité de l'environnement dans laquelle il est plongé, tout en tenant compte des objectifs locaux respectifs de ses composants. Enfin, nous mettons en avant ce modèle ainsi que cette méthodologie à travers 4 exemples provenant de différents domaines (militaire, logistique et exploratoire) et validons la faisabilité, l'efficacité, l'efficience et la robustesse de notre méthodologie d'architecture que nous avons développée et proposée.

Related papers

 


Abstract:   This thesis is about artificial learning in the context of complex systems. Complex systems are characterized by some properties. In particular, they are open, heterogeneous, and their dynamics are non-linear and have retro-action cycles. In addition, they often feature a large number of entities and interactions. Hence, learning (ie generating a model) in such environments is very difficult for artificial learning algorithms.
Given the challenges presented by complex systems, this thesis proposes to approach the problem of learning in a decentralized and bottom-up way, by using the multi-agent systems paradigm. To develop a multi-agent system capable of acting as a learning system, we use the Adaptive Multi-Agent System (AMAS) approach. It proposes a set of mechanisms to allow agents to interact effectively, and to promote the self-organization of the multi-agent system.
We propose and describe AMOEBA (for Agnostic MOdEl Builder by self-Adaptation), a supervised artificial learning system based on this AMAS approach. It is made of autonomous agents. Each of these agents constructs a local and personal representation of a part of the solution space. This activity is carried out following successive feedback of the environment. Agents seek to maintain a cooperative state, which is characterized by mutually beneficial interactions. When this state is lost, the agents modify their internal representations and their interactions, in order to return to a cooperative state. The global activity of the agents makes it possible to construct a global representation of the observed world, that is to say a model, adaptive and in permanent readjustment, which can be exploited to carry out tasks of control, forecasting or decision support.
AMOEBA is evaluated on several experiments, revealing interesting properties, in particular in terms of scaling, reactivity and simplicity of implementation.

Report   Related papers


Abstract:   Building the best plan in product treatment, the best schedule to a building construction or the best route for a salesman in order to visit a maximum of cities in the time allowed while taking into account different constraints (economic, temporal, humans or meteorological): in all of those variety of applications, optimizing the planning is a complex task including a huge number of heterogeneous entities in interaction, the strong dynamics, multiple contradictory objectives, etc.
Mission planning for constellations of satellites is a major example: a lot of parameters and constraints, often antagonists must be integrated, leading to an important combinatorial search space. Currently, in Europe, plans are built on ground, just before the satellite is visible by the ground stations. Any request coming during the planning process must wait for the next period. Moreover, the complexity of this problem grows drastically: the number of constellations and satellites increases, as the number of daily requests. Current approaches have shown their limits. To overcome those drawbacks, new systems based on decentralization and distribution inherent to this problem, are needed.
The adaptive multi-agent systems (AMAS) theory and especially the AMAS4Opt (AMAS For Optimization) model have shown their adequacy in complex optimization problems solving. The local and cooperative behavior of agents allows the system to self-adapt to highly dynamic environments and to quickly deliver adequate solutions.
In this thesis, we focus on solving mission planning for satellite constellations using AMAS. Thus, we propose several enhancement for the agent models proposed by AMAS4Opt. Then, we design the ATLAS dynamic mission planning system. To validate ATLAS on several criteria, we rely on huge sets of heterogeneous data. Finally, this work is compared to an operational and standard system on real scenarios, the value of our system.

Report   Related papers


Abstract:   Electricity plays an increasingly important role in our society. Indeed, we are moving toward the era of "everything electric". The needs evolving, it is mandatory to rethink the way electricity is produced and distributed. This then introduces the concept of an autonomous and intelligent power system called the Smart Grid. The Smart Grid is a concept of electrical network able to support autonomously any changes and faults that may occur.
Obviously, the geographical distribution of electrical networks and the environment (weather conditions, ...) make it impossible to predict events that will occur. To do this, this study proposes an innovative agent-based framework as well as the design and implementation of cooperative agents behaviors aiming at solving common power systems related problems: the Load Flow analysis and the State Estimation.
hese issues have been addressed by the mean of Adaptive Multi-Agent Systems. These systems are known to be efficient to solve complex problems and have the ability to adapt their functioning to the evolutions of their environment. The results obtained show the relevance of using such self-adaptive systems to solve the issues inherent to the Smart Grid.

Report   Related papers


Abstract:  This thesis is about learning the behavior of the aircrafts in the sky. With those behaviors the goal is to generate traffic in an autonomous and flexible way into a simulation.
The current methods of air traffic simulation need to prepare the scenario before the simulation and the interventions of humans during the simulation to make the traffic realistic. Traffic generation is a complex task because the behaviors of the planes depends on many variables and several actors : the air traffic controller decide what trajectory to follow among many possibilities, then the pilot react , more or less promptly, to this order in a, more or less rigorous, way.
An adaptive multi-agent system monitors trajectories of real aircrafts to learn how the planes behave in the real sky. The agents involved in this process cooperate and update the links between them to create a network representing the global behavior of all aircrafts. This network can then be queried by an aircraft agent in a simulation to know what it should do according to its current situation.
We present EVAA (Self-Adaptive Virtual Environment) able to learn the behavior of aircrafts and to generate air traffic by using those behaviors in a autonomous way.

Report   Related papers


Abstract:   The AMAS (Adaptive Multi-Agent Systems) theory proposes to solve complex problems for which there is no known algorithmic solution by self-organization. The self-organizing behaviour of the cooperative agents enables the system to self-adapt to a dynamical environment to maintain the system in a functionality adequate state. In this thesis, we apply the theory to the problematic of control in ambient systems, and more particularly to service robotics.
Service robotics is more and more taking part in ambient environment, we talk of ambient robotics. Ambient systems have challenging characteristics, such as openness and heterogeneity, which make the task of control particularly complex. This complexity is increased if we take into account the specific, changing and often contradictory needs of users. This thesis proposes to use the principle of self-organization to design a multi-agent system with the ability to learn in real-time to control a robotic device from demonstrations made by a tutor. We then talk of learning from demonstrations. By observing the activity of the users, and learning the context in which they act, the system learns a control policy allowing to satisfy users.
Firstly, we propose a new paradigm to design robotic systems under the name Extreme Sensitive Robotics. The main proposal of this paradigm is to distribute the control inside the different functionalities which compose a system, and to give to each functionality the capacity to self-adapt to its environment.
To evaluate the benefits of this paradigm, we designed ALEX (Adaptive Learner by Experiments), an Adaptive Multi-Agent System which learns to control a robotic device from demonstrations. The AMAS approach enables the design of software with emergent functionalities. The solution to a problem emerges from the cooperative interactions between a set of autonomous agents, each agent having only a partial perception of its environment. The application of this approach implies to isolate the different agents involved in the problem of control and to describe their local behaviour. Then, we identify a set of non-cooperative situations susceptible to disturb their normal behaviour, and propose a set of cooperation mechanisms to handle them. The different experimentations have shown the capacity of our system to learn in real- time from the observation of the activity of the user and have enable to highlight the benefits, limitations and perspectives offered by our approach to the problematic of control in ambient systems.

Report (URL towards UPS' Phd Thesis)   Related papers


Abstract:   Due to increased connected objects, multiscale systems are more and more widespread. Those systems are highly distributed, heterogeneous, dynamic and open. They can be composed of hundreds of software components deployed into thousands of devices.
Deployment of software systems is a complex post-production process that consists in making software available for use and then keeping it operational. For multiscale systems, deployment plan expression just as deployment realization and management are tasks impossible for a human stakeholder because of heterogeneity, dynamics, number, and also because the deployment domain is not necessarily known in advance.
The purpose of this thesis, in relation with the Income project, is to study and propose solutions for the deployment of distributed multiscale software systems. Firstly, we provide an up-to-date terminology and de?nitions related to software deployment, plus a state of the art on automatic deployment of distributed software systems. The rest of the contribution lies in the proposition of : (1) a complete process for autonomic deployment of multiscale systems ; (2) a domain specific language, MuScADeL, which simplifies the deployment conceptor task and allows the expression of deployment properties such as informations for the domain state probing ; (3) and a middleware, MuScADeM, which insures the automatic generation of a deploy- ment plan according the domain state, its realization and finally the maintenance in an operational condition of the system.

Report (More details)   Related papers


Abstract:   Component-based Software Architectures and Multi-Agent Systems: Mutual and Complementary Contributions for Supporting Software Development In this thesis, we explore the various aspects of the mutual and complementary contributions that multi-agent systems (MASs) and component-based software architectures (CBSAs) can provide to each other. In a way, this work is the study of how both worlds can be integrated together, either by supporting MAS implementation using component-based abstractions or by supporting CBSA construction and adaptation using self-adaptive MASs.
As a pragmatic starting point, we study how MAS development is currently done in the field and propose an understanding of the general methodology of development of MASs from an architecturally-oriented point of view. This results in the distinction between two main activities in MAS development. The first one, which we call macro-level design, is concerned with requirements and design choices tackled by multi-agent approaches as a way to decompose the solution in terms of agents and their interactions. The second one, which we call micro-level design, is concerned with requirements and design choices that accompany and more importantly support the result of the first activity to bridge the gap between design and implementation. From this conclusion, we infer that it is necessary to support this micro-level architectural activity with adequate abstractions that favour reuse, separation of concern and maintenance. In particular, an abstraction called "species of agent" is introduced for this purpose. It integrates with traditional component-oriented abstractions and acts both as an architectural abstraction and as an implementation. We define, illustrate, analyse and discuss a component model (SpeAD), an architectural description language (SpeADL) and a design method (SpEArAF) that ease and guide the description and the implementation of MASs using species of agents. This complete answer to the question of MAS development, which is supported by a tool (MAY) to exploit SpeADL with Java, has been applied to many applications in our research team.
Then, by setting back such a solution in the context of the CBSA field, we show how MASs differ and relate to traditional means of development in terms of structural abstractions.
To complete this study, we explore through various experiments how self-adaptive MASs can be used to support the building and the adaptation of CBSAs. Here, the agents and their continuous reorganisation act on one hand as the engine of the construction and of the dynamic adaptation of the architecture, and on the other hand as the runtime container that actually connects these elements together and maintains the architecture alive and working. This makes such an approach, even though it is exploratory and prototypal, a completely integrated solution to architecture building, execution and adaptation. This work opens several interesting research paths to build tools to support the development and the evolution of software architectures at design and at runtime.

Report   Related papers