En s’inscrivant dans une approche énactive et interdisciplinaire de la coordination interpersonnelle (Bourbousson, 2015), cette thèse visait à mieux comprendre la manière dont des rameurs expérimentés en aviron (co-) régulaient leur activité collective en temps réel en relation avec leur bateau. Trois études de cas sur des équipages en deux de pointe sans barreur composent cette thèse. L’Étude 1 pointe que (a) les deux rameurs faisaient rarement simultanément l’expérience de leur action conjointe, (b) certains coups de rame étaient cependant simultanément vécus comme efficaces ou non-efficaces, et (c) les rameurs régulaient activement leur activité collective en s’ajustant mutuellement aux comportements de leur partenaire (i.e., (co- )régulation interpersonnelle). L’Étude 2 montre qu’à l’issue du programme d’entraînement (a) la proportion du nombre d’expériences simultanément vécues par les rameurs relatives à leur action conjointe avait significativement augmentée, et (b) les rameurs régulaient activement leur activité collective en s’ajustant aux variations dynamiques de leur environnement matériel commun, le bateau (i.e., (co-)régulation extra-personnelle). L’Étude 3 pointe que les rameurs modifiaient la nature de leurs ajustements mutuels en relation avec différentes contraintes de cadence imposées. En outre, les adaptations comportementales des rameurs ont suggéré l’existence d’une propriété de « dégénérescence » (Araujo & Davids, 2016) dans le système social que constituent les rameurs. Enfin, les expériences vécues rapportées par les rameurs étaient concomitantes des moments saillants d’ajustements mutuels suggérant des formes de « participatory sense-making » dans les instants de co-régulation (Di Paolo & De Jaegher, 2010). / By adopting an enactive and interdisciplinary approach to interpersonal coordination (Bourbousson, 2015, De Jaegher & Di Paolo, 2007), this thesis aimed to better understand the way in which experienced rowers in rowing (co-)regulated their collective activity in time in relation to the boat. Three case studies of coxless-pair crews composed this thesis. Study 1 points out that (a) the two rowers rarely experienced simultaneous joint action at the same time, (b) there were simultaneously experienced oar strokes as effective or detrimental, and (c) suggested that rowers actively regulated their collective activity by adjusting to each other's behaviors (i.e., interpersonal (co-)regulation). Study 2 shows that at the end of the training program (a), the proportion of the number of experiences simultaneously lived by the rowers relative to their mutual coordination significantly increased, and (b) suggested that rowers actively regulated their collective activity by adjusting to boat behavior (i.e., extra-personal (co-)regulation). Study 3 points out that the rowers modified the nature of their mutual adjustments in relation to different imposed cadence constraints. In addition, behavioral adaptations of rowers suggested the existence of a "degeneration" property (Araujo & Davids, 2016) in the social system constituted by the rowers. Finally, the lived experiences reported by the rowers were concomitant with the salient moments of mutual adjustment, as observed in the behavioral data, suggesting participatory sense-making forms in the moments of coregulation (Di Paolo & De Jaegher, 2010).

La modélisation et la simulation des systèmes complexes constitue une solution idéal pour comprendre ces systèmes. En effet, l'expérimentation virtuelle permet, par rapport à l'expérimentation réelle dans le champ d'étude considéré, d'apporter des réponses plus rapides aux questions posées sur ces systèmes, ce qui donne la possibilité de proposer des solutions en un temps adapté au contexte réel. Ce travail traite la question de la représentation de l'action humaine en prenant en compte sa dimension temporelle et spatiale aux échelles individuelle et collective. Cette question a déjà été traitée dans le domaine de l'intelligence artificielle, en général, et celui des systèmes agricoles, en particulier, qui constitue le domaine d'application de cette thèse. Les modèles proposés jusqu'à présent se basaient principalement sur la théorie de l'action planifiée en ne prenant en compte que la dimension temporelle de l'action. Les limites majeures de ces modèles résident dans leur complexité dans la mesure où il est difficile de pouvoir prédire l'ensemble des changements futurs dans l'environnement de l'acteur. Cela conduit à la nécessité de re-planifier fréquemment les actions afin d'obtenir des résultats cohérents. La deuxième limite réside dans l'écart qu'il peut y avoir entre les résultats des actions simulées et la réalité observée. En effet, un acteur ne réalise pas systématiquement les actions qu'il prévoit selon les situations réelles dans lesquelles il se trouve. Afin de pallier aux limites des modèles de l'action planifiée, nous avons développé un modèle de l'action humaine qui se base sur la théorie de l'action située. L'action est vue comme un processus doté d'une épaisseur temporelle émergent des situations créées par l'interaction entre l'acteur et son environnement dans le temps et dans l'espace. Notre modèle combine le concept d'affordance, le concept de stigmergie ainsi que la notion d'émergence. Nous proposons donc un système multi-agents dans lequel l'espace est explicitement représenté et partitionné en un ensemble de places. Le pilotage de chaque place est attribué à un agent abstrait. Celui-ci représente un observateur capable de détecter à tout instant les affordances émergentes sur sa place ainsi que de déclencher l'action appropriée. Les acteurs sont représentés comme des entités de l'environnement au même titre que les objets passifs. Ces entités de l'environnement portent un ensemble d'informations sur leurs capacités à exécuter ou subir des actions. Ces informations permettent aux agents, grâce aux méta-connaissances qu'ils détiennent de détecter les affordances. Celles-ci, une fois détectées, sont réifiées dans l'environnement et utilisées par les agents grâce à un mécanisme de sélection d'actions pour déterminer l'action qui sera finalement exécuter. La coordination des actions au niveau collectif se fait par stigmergie : les agents communiquent de façon implicite en utilisant un ensemble de marques qui sont une métaphore des phéromones des colonies de fourmis. Afin de montrer la pertinence du modèle proposé, un prototype appliqué au domaine des systèmes de production agricoles a été implémenté en utilisant la plateforme AnyLogic. / Simulation modelling of complex systems nowadays is an ideal solution to get a good understanding of these systems. In effect, compared with real experiments in the field of studies considered, virtual experiments allow one to quickly answer questions about these systems and provide solutions within a delay well adapted to their actual context. This thesis deals with the issue of human action representation, accounting with its temporal and spatial dimensions at individual and collective levels. This question has already been addressed in the field of Artificial intelligence in general and in the one of Agricultural systems in particular, the latter being the application domain of this thesis. The models proposed to date were mainly based upon the theory of planned action, explicitly accounting with the temporal dimension of action only. The main limits of these models lie in their complexity, because the ability to predict all future changes in actors' behaviors is far too difficult. This difficulty leads to the need of frequently re-planning the course of actions in order to get consistent results. The second drawback lies in the discrepancy that may arise between the results of simulated actions and actual observations. In effect, real actors do not realize systematically the actions they forecast according to the situations they actually encounter. In order to overcome the limits of planning models, we developed a model of human action based on the theory of situated action. Action is there viewed as a process endowed with a temporal thickness and emerging from the situations created by the interaction, through time and space, between the actor and its environment. Our model combines the concepts of affordance and stigmergy as well as the notion of emergence. Therefore we propose a multi-agents system within which space is explicitly represented and partitioned into a set of “places”. The control of each place is left to an abstract agent standing for an observer capable of detecting the affordances occurring on its place and trigger appropriate actions. Actors as well as passive objects are represented as “environmental entities”. These entities carry information about their capacity of performing or undergoing actions. This information allows the agents to detect affordances thanks to the meta-knowledge they hold. Once detected, these affordances are reified in the environment to be used to determine the action that will eventually be executed. Coordination of actions, at the collective level, is performed through stigmergy: the agents communicate implicitly between them using a set of marks as a metaphor of pheromons in ant colonies. To prove the relevance of the proposed model, a software prototype, applied to the domain of agricultural production systems, has been implemented with the simulation platform AnyLogic.

In ant societies, and, more in general, in insect societies, the activities of the individuals, as well as of the society as a whole, are not regulated by any explicit form of centralized control. On the other hand, adaptive and robust behaviors transcending the behavioral repertoire of the single individual can be easily observed at society level. These complex global behaviors are the result of self-organizing dynamics driven by local interactions and communications among a number of relatively simple individuals. The simultaneous presence of these and other fascinating and unique characteristics have made ant societies an attractive and inspiring model for building new algorithms and new multi-agent systems. In the last decade, ant societies have been taken as a reference for an ever growing body of scientific work, mostly in the fields of robotics, operations research, and telecommunications. Among the different works inspired by ant colonies, the Ant Colony Optimization metaheuristic (ACO) is probably the most successful and popular one. The ACO metaheuristic is a multi-agent framework for combinatorial optimization whose main components are: a set of ant-like agents, the use of memory and of stochastic decisions, and strategies of collective and distributed learning. It finds its roots in the experimental observation of a specific foraging behavior of some ant colonies that, under appropriate conditions, are able to select the shortest path among few possible paths connecting their nest to a food site. The pheromone, a volatile chemical substance laid on the ground by the ants while walking and affecting in turn their moving decisions according to its local intensity, is the mediator of this behavior. All the elements playing an essential role in the ant colony foraging behavior were understood, thoroughly reverse-engineered and put to work to solve problems of combinatorial optimization by Marco Dorigo and his co-workers at the beginning of the 1990's. From that moment on it has been a flourishing of new combinatorial optimization algorithms designed after the first algorithms of Dorigo's et al., and of related scientific events. In 1999 the ACO metaheuristic was defined by Dorigo, Di Caro and Gambardella with the purpose of providing a common framework for describing and analyzing all these algorithms inspired by the same ant colony behavior and by the same common process of reverse-engineering of this behavior. Therefore, the ACO metaheuristic was defined a posteriori, as the result of a synthesis effort effectuated on the study of the characteristics of all these ant-inspired algorithms and on the abstraction of their common traits. The ACO's synthesis was also motivated by the usually good performance shown by the algorithms (e.g., for several important combinatorial problems like the quadratic assignment, vehicle routing and job shop scheduling, ACO implementations have outperformed state-of-the-art algorithms). The definition and study of the ACO metaheuristic is one of the two fundamental goals of the thesis. The other one, strictly related to this former one, consists in the design, implementation, and testing of ACO instances for problems of adaptive routing in telecommunication networks. This thesis is an in-depth journey through the ACO metaheuristic, during which we have (re)defined ACO and tried to get a clear understanding of its potentialities, limits, and relationships with other frameworks and with its biological background. The thesis takes into account all the developments that have followed the original 1999's definition, and provides a formal and comprehensive systematization of the subject, as well as an up-to-date and quite comprehensive review of current applications. We have also identified in dynamic problems in telecommunication networks the most appropriate domain of application for the ACO ideas. According to this understanding, in the most applicative part of the thesis we have focused on problems of adaptive routing in networks and we have developed and tested four new algorithms. Adopting an original point of view with respect to the way ACO was firstly defined (but maintaining full conceptual and terminological consistency), ACO is here defined and mainly discussed in the terms of sequential decision processes and Monte Carlo sampling and learning. More precisely, ACO is characterized as a policy search strategy aimed at learning the distributed parameters (called pheromone variables in accordance with the biological metaphor) of the stochastic decision policy which is used by so-called ant agents to generate solutions. Each ant represents in practice an independent sequential decision process aimed at constructing a possibly feasible solution for the optimization problem at hand by using only information local to the decision step. Ants are repeatedly and concurrently generated in order to sample the solution set according to the current policy. The outcomes of the generated solutions are used to partially evaluate the current policy, spot the most promising search areas, and update the policy parameters in order to possibly focus the search in those promising areas while keeping a satisfactory level of overall exploration. This way of looking at ACO has facilitated to disclose the strict relationships between ACO and other well-known frameworks, like dynamic programming, Markov and non-Markov decision processes, and reinforcement learning. In turn, this has favored reasoning on the general properties of ACO in terms of amount of complete state information which is used by the ACO's ants to take optimized decisions and to encode in pheromone variables memory of both the decisions that belonged to the sampled solutions and their quality. The ACO's biological context of inspiration is fully acknowledged in the thesis. We report with extensive discussions on the shortest path behaviors of ant colonies and on the identification and analysis of the few nonlinear dynamics that are at the very core of self-organized behaviors in both the ants and other societal organizations. We discuss these dynamics in the general framework of stigmergic modeling, based on asynchronous environment-mediated communication protocols, and (pheromone) variables priming coordinated responses of a number of ``cheap' and concurrent agents. The second half of the thesis is devoted to the study of the application of ACO to problems of online routing in telecommunication networks. This class of problems has been identified in the thesis as the most appropriate for the application of the multi-agent, distributed, and adaptive nature of the ACO architecture. Four novel ACO algorithms for problems of adaptive routing in telecommunication networks are throughly described. The four algorithms cover a wide spectrum of possible types of network: two of them deliver best-effort traffic in wired IP networks, one is intended for quality-of-service (QoS) traffic in ATM networks, and the fourth is for best-effort traffic in mobile ad hoc networks. The two algorithms for wired IP networks have been extensively tested by simulation studies and compared to state-of-the-art algorithms for a wide set of reference scenarios. The algorithm for mobile ad hoc networks is still under development, but quite extensive results and comparisons with a popular state-of-the-art algorithm are reported. No results are reported for the algorithm for QoS, which has not been fully tested. The observed experimental performance is excellent, especially for the case of wired IP networks: our algorithms always perform comparably or much better than the state-of-the-art competitors. In the thesis we try to understand the rationale behind the brilliant performance obtained and the good level of popularity reached by our algorithms. More in general, we discuss the reasons of the general efficacy of the ACO approach for network routing problems compared to the characteristics of more classical approaches. Moving further, we also informally define Ant Colony Routing (ACR), a multi-agent framework explicitly integrating learning components into the ACO's design in order to define a general and in a sense futuristic architecture for autonomic network control. Most of the material of the thesis comes from a re-elaboration of material co-authored and published in a number of books, journal papers, conference proceedings, and technical reports. The detailed list of references is provided in the Introduction.

Die vorliegende Dissertation hat einen technologischen und einen anwendungsbezogenen Schwerpunkt. Technologisch ordnen sich die präsentierten Forschungsergebnisse in das Gebiet der "Swarm Intelligence" (dt.: Schwarm-Intelligenz) ein. Swarm Intelligence ist ein Teilbereich der Informatik, der sich an der Überschneidung zwischen der Multi-Agenten Systeme Forschung der Künstlichen Intelligenz und dem Forschungsgebiet "Artificial Life" (dt.: Künstliches Leben) befindet. Im Gegensatz zur Swarm Intelligence im allgemeinen, überträgt der spezielle Ansatz "Synthetic Ecosystems" (dt.: synthetische Ökosysteme) nicht nur Koordinationsmechanismen aus biologischen Multi-Agenten Systemen, wie zum Beispiel Insekten Kolonien, in den Entwurf künstlicher Systeme. Vielmehr sollen die grundlegenden Prinzipien "natürlich" entstandener komplexer Systeme, also auch zum Beispiel einer Aktienbörse, übernommen werden. Als anwendungsbezogener Hintergrund der Dissertation wurde die verteilte Steuerung moderner industrieller Fertigungsanlagen gewählt. Die Fertigungssteuerung ist ein geeignetes Anwendungsfeld für die Technologien, die im Rahmen der Forschungsarbeiten entwickelt wurden. Damit dient die Präsentation eines synthetischen Ökosystems für die Fertigungssteuerung der Demonstration des neuartigen Ansatzes zum Entwurf, Realisierung und Evaluierung komplexer, industriell relevanter Systeme. Gleichzeitig leistet die vorgestellte Architektur der Fertigungssteuerung und die darin verwandten Koordinationsverfahren einen Beitrag zur Weiterentwicklung holonischer Produktionssysteme. Der holonische Ansatz zur Produktionsplanung und -steuerung genießt derzeit große Aufmerksamkeit sowohl in der Forschung als auch in der Industrie. Als Teilgebiet der Entwicklung intelligenter Fertigungssysteme (engl.: IMS - Intelligent Manufacturing Systems), propagiert der holonische Ansatz eine Abkehr von der traditionell zentralistischen und hierarchischen Planung und Steuerung hin zu selbst-organisierenden Systemen autonom (inter-)agierender Individuen ("Holone"). Bei der praktischen Umsetzung holonischer Systeme werden sehr häufig Technologien aus der Multi-Agenten Systeme Forschung angewandt. Mit dieser Dissertation rücken auch synthetische Ökosysteme in das Blickfeld holonischer Systeme. Natürliche Agentensysteme im allgemeinen und Kolonien sozialer Insekten im besonderen faszinieren durch ihre Robustheit, ihre Flexibilität und ihre Anpassungsfähigkeit. Solche Systeme bestehen häufig aus sehr vielen, sehr einfachen Individuen und doch weisen sie ein komplexes und koordiniertes Gesamtverhalten auf. Es gibt mehrere Zweige in unterschiedlichen Wissenschaften, zum Beispiel in der Biologie, Physik, Ökonomie oder in der Informatik, die sich mit verteilten Systemen lokal interagierender Individuen beschäftigen. Ihre Erforschung resultiert in einer Reihe wiederholt beobachteter grundlegender Eigenschaften. Um künstlich erschaffene Systeme mit ähnlichen Eigenschaften auszustatten werden Entwurfsprinzipien für das Design von Multi-Agenten Systemen in dieser Dissertation vorgeschlagen. Jedes Entwurfsprinzip wird systematisch eingeführt, motiviert und in seinen Konsequenzen für Anwendungen in der Fertigungssteuerung diskutiert. Stigmergie ist ein grundlegendes Konzept der Koordination einer großen Anzahl von Individuen unter anderem in Kolonien sozialer Insekten. Die Formulierung dieses Konzepts ist auf den Biologen Grassè zurückzuführen, welcher in der Mitte des zwanzigsten Jahrhunderts das Schwarmverhalten von Termiten untersuchte. Stigmergie beruht auf der Tatsache, daß das Verhalten eines jeden Individuums durch die aktuelle Konfiguration seiner lokalen Umwelt bestimmt wird. Die Umwelt wiederum, wird durch die Aktivitäten der Individuen verändert. Diese Wechselwirkung führt in Verbindung mit entsprechend ausgelegten individuellen Verhaltensmustern zur Emergenz einer global koordinierten Erfüllung der anstehenden Aufgaben der Kolonie. Im Detail wird sematektonische von marker-basierter Stigmergie unterschieden, wobei bei sematektonischer Stigmergie der Zustand der Aufgabenerfüllung selbst (z.B. Stand des Nestbaus) das Individualverhalten beeinflußt, während marker-basierte Stigmergie aufgabenunabhängige Marker (z.B. Pheromone) in der Umwelt platziert. Multi-Agenten Systeme finden ihre Realisierung in Software, welche gegebenenfalls an physische Aktuatoren gekoppelt ist. Im allgemeinen besteht diese Software aus einer Laufzeitumgebung und den darin ausgeführten Agenten. Die vorliegende Dissertation präsentiert eine Erweiterung von Laufzeitumgebungen um eine anwendungsunabhängige Pheromon Infrastruktur (PI). Die PI ermöglicht es den Softwareagenten des jeweiligen synthetischen Ökosystems, künstliche Pheromone als Datenstrukturen in einem virtuellen Raum abzulegen und wahrzunehmen. Diese Datenstrukturen dienen als Marker in stigmergetischen Koordinationsmechanismen. Die Algorithmen der PI operieren auf diesen künstlichen Pheromonen und emulieren die natürlichen Vorgänge der räumlichen Ausbreitung und Verdunstung von Pheromonen auf abstrakter Ebene. Zusätzlich wird das natürliche Vorbild um eine automatische Aufbereitung von Informationen erweitert. Die Funktionalität der PI wird in dieser Dissertation spezifiziert. Des weiteren wird ein formales Modell erstellt, welches die Grundlage einer numerischen Analyse der Eigenschaften der PI bildet. Die Analyse liefert Vorhersagen für das Entstehen von räumlichen Mustern von Pheromonkonzentrationen in der PI. Diese Vorhersagen können dann in der Feineinstellung und der Evaluierung von Koordinationsmechanismen verwendet werden. Außerdem dient das formale Modell als Grundlage für den Beweis der globalen Stabilität der PI. Damit ist gesichert, daß unabhängig von der gewählten räumlichen Struktur und den von der jeweiligen Anwendung generierten Pheromonen die Konzentrationen der Pheromone immer in ihrer Stärke begrenzt sind. Der Beweis der globalen Stabilität ist eine wichtige Voraussetzung für die Verwendung der PI in praktischen Anwendungen. Die Spezifikation einer verteilten Realisierung der PI bildet den Abschluß der allgemeinen Betrachtung. Die Agenten, welche die (virtuelle) räumliche Struktur der PI widerspiegeln, werden im Detail spezifiziert. Auf der Basis dieser Spezifikation ist im Rahmen der Dissertation ein Prototyp der PI realisiert worden. Dieser Prototyp diente dem Nachweis des vorhergesagten Verhaltens der Infrastruktur und der späteren Evaluierung des entwickelten Fertigungssteuerungssystems. Im weiteren Verlauf der vorliegenden Dissertation wird ein neuartiger Ansatz zur Fertigungssteuerung betrachtet. Die absehbaren Veränderungen der äußeren Bedingungen der industriellen Produktion, ausgelöst durch den globalen Übergang von Anbieter- zu Verbrauchermärkten, erfordert die Fertigung immer komplexerer und variantenreicherer Produkte in ständig schwankenden Stückzahlen und deutlich verkürzten Lebenszyklen bei gleichzeitig sinkenden Kosten. Zur Erfüllung dieser Anforderungen in der Massenproduktion wandelt sich die traditionell starr verkettete Strangfertigung (z.B. Transferstraßen) zur flexiblen Fließfertigung (z.B. flexible Bearbeitungszentren). Die Steuerung einer flexiblen Fließfertigung erfordert neue Herangehensweisen. In einer holonischen Fertigung, zum Beispiel, organisiert sich die Produktionsplanung und Produktionssteuerung selbst um die Erfüllung der aktuellen Aufträge. Dabei werden in der Steuerung verteilte, reaktive Verfahren verwendet, welche eine deutlich gesteigerte Robustheit und Flexibilität gegenüber Störungen und Veränderungen aufweisen. Der Übergang zur flexiblen Fließfertigung bedeutet die Einführung von Flexibilität in der Bearbeitung aber auch im Transport des Materials. Es ist eine grundlegende Eigenschaft dieser Fertigungssysteme, daß zu einem beliebigen Zeitpunkt eine Reihe möglicher Transportwege und damit eine Vielzahl möglicher Muster im Materialfluß zur Verfügung stehen. Dabei führt aber nur eine kleine Menge dieser Muster zu einer bestmöglichen Erfüllung der globalen Produktionsziele (z.B. hoher globaler Durchsatz). Es ist also die Aufgabe der Fertigungssteuerung in jeder Situation das bestmögliche Materialflußmuster zu erreichen. Ist ein verteilter Ansatz für die Steuerung gewählt worden, so muß diese Optimierung nach globalen Produktionszielen in die lokalen Steuerungsentscheidungen integriert werden, ohne die Autonomie der lokalen Einheiten zu verletzen. Die Dissertation präsentiert ein sogenanntes geführtes Fertigungssteuerungssystem (GFSS), welches einen verteilten und reaktiven Steuerungsansatz mit einer Flußoptimierung unter Beachtung globaler Produktionsziele in neuartiger Weise verbindet. Der Entwurf des GFSS folgte den vorgeschlagenen Prinzipien für synthetische Ökosysteme und die Agenten im GFSS werden mit Hilfe der Pheromon Infrastruktur koordiniert. Die Agenten und Pheromone des GFSS werden detailliert spezifiziert und in einem realistischen Beispiel aus der Automobilindustrie evaluiert. In der Evaluierung wird von den Ergebnissen der Analyse der PI Gebrauch gemacht. Die dabei gewählte numerische Beschreibung des Einzelverhaltens und die darauf aufbauende Betrachtung des emergierenden Gesamtverhaltens weist den Weg zu einer systematischen Evaluierung von emergenten Systemeigenschaften in synthetischen Ökosystemen. In einem abschließenden Kapitel werden die drei inhaltlichen Schwerpunkte der Dissertation noch einmal betrachtet. Vor dem Hintergrund des GFSS werden die vorgeschlagenen Entwurfsprinzipien für synthetische Ökosysteme systematisch auf ihre Anwendbarkeit und praktische Bedeutung hin überprüft. Außerdem wird die allgemeine Verwendung der PI für den Austausch von Informationen zwischen Agenten untersucht. Und schließlich wird die Fertigungssteuerung aus der Sicht abstrakter Zustandsräume diskutiert. Die vorliegende Dissertation weist den Weg für eine Reihe weiterführender Forschungsarbeiten. So werden zum einen detaillierte Konzepte für die Erweiterung des GFSS um eine automatische Strategiebewertung und -generierung und um ein Visualisierungssystem vorgestellt. Zum anderen werden aber auch notwendige Ergänzungen der Entwurfsprinzipien und mögliche Verbesserungen der PI und des darauf basierenden Evaluierungsansatzes vorgeschlagen. / The synthetic ecosystems approach attempts to adopt basic principles of natural ecosystems in the design of multiagent systems. Natural agent systems like insect colonies are fascinating in that they are robust, flexible, and adaptive. Made up of millions of very simple entities, these systems express a highly complex and coordinated global behavior. There are several branches in different sciences, for instance in biology, physics, economics, or in computer science, that focus on distributed systems of locally interacting entities. Their research yields a number of commonly observed characteristics. To supply engineered systems with similar characteristics this thesis proposes a set of principles that should be observed when designing synthetic ecosystems. Each principle is systematically stated and motivated, and its consequences for the manufacturing control domain are discussed. Stigmergy has shown its usefulness in the coordination of large crowds of agents in a synthetic ecosystem. Sign-based stigmergy through synthetic pheromones is supported by an extension to runtime environments for software agents called the pheromone infrastructure. In this thesis the operation of the pheromone infrastructure is specified, formally modeled and analyzed, and an implementation is presented. The guided manufacturing control system for flexible flow shops is designed following the proposed principles and it uses the pheromone infrastructure to coordinate its agents. It comprises two subsystems. The control (sub)system, which enables production, is distributed and reactive. The advisory (sub)system observes the operation of the control system and advises the manufacturing execution under global considerations. This thesis specifies the guided manufacturing control system and evaluates its operation in a simple but realistic example adapted from the automotive industry. The applicability of the design principles, the usage of the pheromone infrastructure, and the operation of manufacturing control in abstract state spaces are considered on the basis of the guided manufacturing control system.

Cette thèse explore les capacités d’une intelligence ambiante équipée d’un réseau de capteurs de pression au sol. Elle traite le problème de la perception d’un environnement au travers un réseau de capteurs de basse résolution. Les difficultés incluent l’interpretation des poids dispersés pour des objets avec multiples supports, l’ambiguïté de poids entre des objets, la variation du poids des personnes pendant les activités dynamiques, etc. Nous introduisons des nouvelles techniques, partiellement inspirées du domaine de la vision par l’ordinateur, pour la détection, le suivi et la reconnaissance des entités qui se trouvent sur le sol. Nous introduisons également des nouveaux modes d’interaction entre les environnements équipés de tels capteurs aux sols, et les robots qui évoluent dans ces environnements. Ceci permet l’interprétation non-intrusive des événements qui ont lieu dans des environnements dotés d’une intelligence ambiante, avec des applications dans l’assistance automatisée à domicile, l’aide aux personnes âgées, le diagnostic continu de la santé, la sécurité, et la navigation robotique / This thesis explores the capabilities of an ambient intelligence equipped with a load-sensing floor. It deals with the problem of perceiving the environment through a network of low-resolution sensors. Challenges include the interpretation of spread loads for objects with multiple points of support, weight ambiguities between objects, variation of persons’ weight during dynamic activities, etc. We introduce new techniques, partly inspired from the field of computer vision, for detecting, tracking and recognizing the entities located on the floor. We also introduce new modes of interaction between environments equipped with such floor sensors and robots evolving inside them. This enables non-intrusive interpretation of events happening inside environments with embedded ambient intelligence, with applications in assisted living, senile care, continuous health diagnosis, home security, and robotic navigation

In ant societies, and, more in general, in insect societies, the activities of the individuals, as well as of the society as a whole, are not regulated by any explicit form of centralized control. On the other hand, adaptive and robust behaviors transcending the behavioral repertoire of the single individual can be easily observed at society level. These complex global behaviors are the result of self-organizing dynamics driven by local interactions and communications among a number of relatively simple individuals.<p><p>The simultaneous presence of these and other fascinating and unique characteristics have made ant societies an attractive and inspiring model for building new algorithms and new multi-agent systems. In the last decade, ant societies have been taken as a reference for an ever growing body of scientific work, mostly in the fields of robotics, operations research, and telecommunications.<p><p>Among the different works inspired by ant colonies, the Ant Colony Optimization metaheuristic (ACO) is probably the most successful and popular one. The ACO metaheuristic is a multi-agent framework for combinatorial optimization whose main components are: a set of ant-like agents, the use of memory and of stochastic decisions, and strategies of collective and distributed learning.<p><p>It finds its roots in the experimental observation of a specific foraging behavior of some ant colonies that, under appropriate conditions, are able to select the shortest path among few possible paths connecting their nest to a food site. The pheromone, a volatile chemical substance laid on the ground by the ants while walking and affecting in turn their moving decisions according to its local intensity, is the mediator of this behavior.<p><p>All the elements playing an essential role in the ant colony foraging behavior were understood, thoroughly reverse-engineered and put to work to solve problems of combinatorial optimization by Marco Dorigo and his co-workers at the beginning of the 1990's.<p><p>From that moment on it has been a flourishing of new combinatorial optimization algorithms designed after the first algorithms of Dorigo's et al. and of related scientific events.<p><p>In 1999 the ACO metaheuristic was defined by Dorigo, Di Caro and Gambardella with the purpose of providing a common framework for describing and analyzing all these algorithms inspired by the same ant colony behavior and by the same common process of reverse-engineering of this behavior. Therefore, the ACO metaheuristic was defined a posteriori, as the result of a synthesis effort effectuated on the study of the characteristics of all these ant-inspired algorithms and on the abstraction of their common traits.<p><p>The ACO's synthesis was also motivated by the usually good performance shown by the algorithms (e.g. for several important combinatorial problems like the quadratic assignment, vehicle routing and job shop scheduling, ACO implementations have outperformed state-of-the-art algorithms).<p><p>The definition and study of the ACO metaheuristic is one of the two fundamental goals of the thesis. The other one, strictly related to this former one, consists in the design, implementation, and testing of ACO instances for problems of adaptive routing in telecommunication networks.<p><p>This thesis is an in-depth journey through the ACO metaheuristic, during which we have (re)defined ACO and tried to get a clear understanding of its potentialities, limits, and relationships with other frameworks and with its biological background. The thesis takes into account all the developments that have followed the original 1999's definition, and provides a formal and comprehensive systematization of the subject, as well as an up-to-date and quite comprehensive review of current applications. We have also identified in dynamic problems in telecommunication networks the most appropriate domain of application for the ACO ideas. According to this understanding, in the most applicative part of the thesis we have focused on problems of adaptive routing in networks and we have developed and tested four new algorithms.<p><p>Adopting an original point of view with respect to the way ACO was firstly defined (but maintaining full conceptual and terminological consistency), ACO is here defined and mainly discussed in the terms of sequential decision processes and Monte Carlo sampling and learning.<p><p>More precisely, ACO is characterized as a policy search strategy aimed at learning the distributed parameters (called pheromone variables in accordance with the biological metaphor) of the stochastic decision policy which is used by so-called ant agents to generate solutions. Each ant represents in practice an independent sequential decision process aimed at constructing a possibly feasible solution for the optimization problem at hand by using only information local to the decision step.<p>Ants are repeatedly and concurrently generated in order to sample the solution set according to the current policy. The outcomes of the generated solutions are used to partially evaluate the current policy, spot the most promising search areas, and update the policy parameters in order to possibly focus the search in those promising areas while keeping a satisfactory level of overall exploration.<p><p>This way of looking at ACO has facilitated to disclose the strict relationships between ACO and other well-known frameworks, like dynamic programming, Markov and non-Markov decision processes, and reinforcement learning. In turn, this has favored reasoning on the general properties of ACO in terms of amount of complete state information which is used by the ACO's ants to take optimized decisions and to encode in pheromone variables memory of both the decisions that belonged to the sampled solutions and their quality.<p><p>The ACO's biological context of inspiration is fully acknowledged in the thesis. We report with extensive discussions on the shortest path behaviors of ant colonies and on the identification and analysis of the few nonlinear dynamics that are at the very core of self-organized behaviors in both the ants and other societal organizations. We discuss these dynamics in the general framework of stigmergic modeling, based on asynchronous environment-mediated communication protocols, and (pheromone) variables priming coordinated responses of a number of ``cheap' and concurrent agents.<p><p>The second half of the thesis is devoted to the study of the application of ACO to problems of online routing in telecommunication networks. This class of problems has been identified in the thesis as the most appropriate for the application of the multi-agent, distributed, and adaptive nature of the ACO architecture.<p><p>Four novel ACO algorithms for problems of adaptive routing in telecommunication networks are throughly described. The four algorithms cover a wide spectrum of possible types of network: two of them deliver best-effort traffic in wired IP networks, one is intended for quality-of-service (QoS) traffic in ATM networks, and the fourth is for best-effort traffic in mobile ad hoc networks.<p><p>The two algorithms for wired IP networks have been extensively tested by simulation studies and compared to state-of-the-art algorithms for a wide set of reference scenarios. The algorithm for mobile ad hoc networks is still under development, but quite extensive results and comparisons with a popular state-of-the-art algorithm are reported. No results are reported for the algorithm for QoS, which has not been fully tested. The observed experimental performance is excellent, especially for the case of wired IP networks: our algorithms always perform comparably or much better than the state-of-the-art competitors.<p><p>In the thesis we try to understand the rationale behind the brilliant performance obtained and the good level of popularity reached by our algorithms. More in general, we discuss the reasons of the general efficacy of the ACO approach for network routing problems compared to the characteristics of more classical approaches. Moving further, we also informally define Ant Colony Routing (ACR), a multi-agent framework explicitly integrating learning components into the ACO's design in order to define a general and in a sense futuristic architecture for autonomic network control.<p><p>Most of the material of the thesis comes from a re-elaboration of material co-authored and published in a number of books, journal papers, conference proceedings, and technical reports. The detailed list of references is provided in the Introduction.<p><p><p> / Doctorat en sciences appliquées / info:eu-repo/semantics/nonPublished