Cette thèse propose une méthode pour coordonner flexiblement des Systèmes Multi-Agents (SMA). Plus en détails, nous étudions comment influencer des agents artificiels afin que, collectivement, ils atteignent des objectifs complexes et/ou dynamiques dans des environnements eux-aussi complexes et dynamiques (ex: un groupe de robots pour secourir les victimes lors d'un désastre, qui peut s'adapter à une grande variété de dangers, conditions climatiques, état des victimes).Dans ce but, nous avons d'abord étudié pourquoi, dans les sociétés humaines, les humains parviennent à coordonner relativement flexiblement mais pas leurs contreparties artificielles (agents des SMA). Cette opposition peut être grandement expliquée à l'aide d'un facteur clef : la culture. Les humains qui partagent un même bagage culturel se coordonnent flexiblement plus facilement, car ils ont une idée commune de ce que "travailler ensemble" veut dire. A contrario, les agents n'ont pas ce bagage et leurs échecs pour travailler ensemble s'apparente souvent à des chocs culturels.Ainsi, notre objectif consiste à répondre à la question suivante: peut-on utiliser une culture semblable à celle des humains comme un outil coordonner les SMA (et si oui, comment) ? Pour répondre à cette question, il nous faut d'abord expliquer : comment intégrer une culture semblable à celle des humains dans un SMA? Cette seconde question en soulève une troisième à étudier en premier : comment est-ce que la culture influence la manière dont la coordination se passe dans les sociétés humaines ?1- Nous montrons que de manière générale, la culture influence les décisions individuelles prises en situation d'interaction (ex: au travers d'attentes, de manière d'agir et de raisonner). Cette influence mène à l'occurrence de schémas d'interaction abstraits, récurrent et cohérents, qui, généralement, améliorent la performance collective. Ensuite, nous spécifions comment les principaux mécanismes l'influence connue de la culture (ex: importance culturelle accordée au pouvoir, aux règles) appliquent spécifiquement en situation de coordination (ex: la culture influence si les dirigeants donnent des ordres vs. des propositions à leurs subordonnés).2-Nous montrons comment répliquer les mécanismes l'influence de la culture sur la coordination dans les SMA. Tout d'abord, puisque la culture est fondée dans les décisions individuelles, nous mettons en avant un mécanisme de décision humain clef qui, à la fois, est sensible à la culture et influence la coordination. Ce mécanisme se trouve dans les valeurs, ce que les gens considèrent comme "bien" ou "important" (ex: honnêteté, discipline, autonomie). Ensuite, nous intégrons ces valeurs dans une architecture agent capable de prendre des décisions en situation de coordination. Enfin, nous illustrons que notre architecture peut en effet reproduire l'influence de la culture sur la coordination à travers de deux simulations qui répliquent des phénomènes culturels en situation de coordination connus.3-Nous étudions comment ces valeurs, inspirées des valeurs humaines, peuvent être utilisées coordonner des SMA. Tout d'abord, nous étudions pour quels problèmes les valeurs offrent un moyen opérationnel pour soutenir la coordination. A l'instar des sociétés humaines, les valeurs sont particulièrement offrent un haut niveau de flexibilité, quand les agents doivent raisonner eux-même pour établir une coordination. Puis, nous étudions les détails techniques à considérer pour utiliser en pratique des valeurs pour coordonner flexiblement des SMA (ex: quelles valeurs choisir ? Comment les représenter ?).En résumé, cette thèse met en évidence que les principaux mécanismes de l'influence de la culture sur la coordination (en particulier, grâce à l'influence de la culture sur les valeurs) peuvent être répliquées au sein des SMA. De plus, nous montrons que ces mécanismes peuvent être manipulés dans le but de coordonner des SMA. / This thesis proposes a method for supporting flexible coordination in multi-agent systems (MASs). In other words, we aim at influencing societies of artificial agents such that they can handle complex or evolving environments and collective goals (e.g. robots providing an emergency support capable of handling various hazards, climatic conditions, status of victims).Towards achieving this goal, we first investigated why in human societies, for which MASs can be seen as an ``artificial" counterpart, humans manage to coordinate relatively flexibly comparatively with artificial agents in MASs. We discovered that culture is a key factor of this relative success. Briefly, when humans share a cultural background, they manage to coordinate more flexibly because they share a common idea about what ``working together'' means. Conversely, artificial agents miss this aspect, leading in turn to coordination failures that can are similar to cultural clashes.The lack raises our goal: we want to better understand how culture can be integrated within and used for coordinating artificial societies. This goal raises the following research question: (how) can human-like culture be used as a tool for supporting coordination in artificial societies? As a preliminary step for answering this question, we need first to answer this question: (how) can the influence human-like cultures be integrated within artificial societies? In turn, this question raises a third one to be answered first: how does culture influence coordination in human societies?As a first step, we expand general theories of culture for conceptualizing its influence in the context of coordination. From a generic perspective, we explain that culture influences individual decisions that support matching expectations and coherent interaction patterns, leading in turn to (generally) better collective performance. From a more specific perspective, we specify how the core acknowledged patterns of the influence of culture (e.g. cultural importance given to power status, to rules) apply in the context of coordination (e.g. culture influences the likeliness that leaders are (made) responsible for making decisions for subordinates vs. proposing alternatives).As a second step, we study how to replicate human-like influences of culture on coordination within artificial societies. First, since culture is grounded within individual decisions, we investigate the core culturally-sensitive decision aspects that impact the most (flexible) coordination in human societies. We discover that values, what people consider as ``good'' or ``important'' (e.g. honesty, obedience, autonomy), constitute such an aspect by deeply supporting a wide range of (interaction-related) decisions. Then, for illustrating how to replicate influence of culture within artificial societies, we build an value-sensitive agent decision architecture that can make coordination-related decisions. Finally, we illustrate that our architecture can replicate the influence of culture on coordination through two simulations that replicate known coordination-related cultural phenomena.As a third step, we study how human-like values can be used for supporting coordination in artificial societies. First, we investigate the range of coordination problems for which values can offer an operational means for supporting coordination. As in human societies, values are particularly adequate for problems with complex and dynamic environments, requiring agents to make coordination-related decisions. Then, towards concretely implementing values, we study the technical details to consider when using values for supporting flexible coordination (e.g. how to concretely design values and integrating them within decision processes).

Resource allocation in control and communication systems constitutes the distribution of (finite) system resources in a way that achieves maximum system functionality and or cost effectiveness. Specific resource allocation problems in subband coding, Discrete Multi-tone modulation based systems and autonomous multi-agent control are addressed in this thesis. In subband coding, the number of bits used (out of a target bit budget) to code a sub- band signal are allocated in a way that minimizes the coding distortion. In Discrete Multi-tone modulation based systems, high bit rate streams are split into several parallel lower rate streams. These individual data streams are transmitted over different subchannels. Given a target bit rate, the goal of resource allocation is to distribute the bits among the different subchannels such that the total transmitted power is minimized. The last problem is achieving stable control of a fleet of autonomous agents by utilizing the available communication resources (such as transmitted Power and bandwidth) as effectively as possible. We present an efficient bit loading algorithm that applies to both subband coding and single-user multicarrier communication system. The goal is to effect an optimal distribution of B bits among N subchannels (subbands) to achieve a minimum transmitted power (distortion error variance) for multicarrier (subband coding) systems. All the algorithms in literature, except a few (which provides a suboptimal solution), have run times that increase with B. By contrast, we provide an algorithm that solves the aforementioned problems exactly and with a complexity (given by O(N log(N)),) which is dependent only on N. Bit loading in multi-user multicarrier systems not only involves the distribution of bit rates across the subchannels but also the assignment of these subchannels to different users. The motivation for studying suboptimal bit allocation is underscored by implicit and explicit claims made in some of the papers which present suboptimal bit loading algorithms, without a formal proof, that the underlying problem is NP-hard. Consequently, for no other reason than the sake of completeness, we present a proof for NP-hardness of the multiuser multicarrier bit loading problem, thereby formally justifying the search for suboptimal solutions. There has been a growing interest in the area of cooperative control of networks of mobile autonomous agents. Applications for such a set up include organization of large sensor networks, air traffic control, achieving and maintaining formations of unmanned vehicles operating under- water, air traffic control etc. As in Abel et al, our goal is to devise control laws that, require minimal information exchange between the agents and minimal knowledge on the part of each agent of the overall formation objective, are fault tolerant, scalable, and easily reconfigurable in the face of the loss or arrival of an agent, and the loss of a communication link. A major drawback of the control law proposed in Abel et al is that it assumes all agents can exchange information at will. This is fine if agents acquire each others state information through straightforward sensing. If however, state information is exchanged through broadcast commu- nication, this assumption is highly unrealistic. By modifying the control law presented in Abel et al, we devise a scheme that allows for a sharing of the resource, which is the communication channel, but also achieves the desired formation stably. Accordingly we modify the control law presented in [23] to be compatible with networks constrained by MAC protocols.

Le travail présenté dans cette thèse définit une nouvelle approche pour la modélisation et la conceptualisation des processus métiers dans les entreprises afin de construire des outils logiciels d'assistance intelligents qui prennent en charge ces processus. L'approche proposée définit une ontologie dédiée à l'étude des organisations, nommée K-CRIO. Elle est décrite à l'aide du langage de représentation des connaissances OWL. Afin d'illustrer nos travaux, un système d'assistance a été implanté sur la base des résultats issus de la modélisation et de la conceptualisation d'un processus métier spécifique avec l'ontologie K-CRIO. Ce système prend la forme d'une application Web qui intègre et exploite pleinement les avantages des systèmes multiagents. L'ontologie K-CRIO est une ontologie dédiée à l'étude des organisations et à l'analyse organisationnelle des processus métiers qu'elles mettent en oeuvre. Plus précisément, elle est utilisée pour comprendre, analyser et raisonner sur ces organisations. Les organisations visées sont celles composées d'acteurs humains impliqués tout au long de la conception de produits et, pour ce faire, organisés selon un processus métier. L'éventail de ce type d'organisations est assez large. Nous avons donc limité notre étude aux organisations qui produisent des logiciels comme objectif final du processus. Dans ce contexte, l'ontologie K-CRIO peut être utilisée pour modéliser la structure organisationnelle du processus et les activités qui en résultent. Cette ontologie peut ensuite être exploitée afin de concevoir des outils d'assistance à la mise en oeuvre des processus ciblés au sein des organisations décrites. Plus précisément, l'ontologie fournit des moyens de raisonnement, d'annotation des ressources, et de suivi des processus de conception, permettant des recherches et de proposer pro-activement des conseils et des contenus appropriés. Afin d'illustrer l'utilisation de K-CRIO, nous appliquons K-CRIO sur deux processus différents: le modèle en cascade et la méthodologie Scrum. Ces exemples sont des processus de développement de logiciels classiques. En outre, pour le processus Scrum, qui est un processus agile de développement de logiciel, largement utilisé dans les entreprises de logiciels, nous avons conçu et développé un outil d'assistance intelligent. Cet outil contribue principalement à aider les Scrum Masters en leur fournissant des indicateurs pour les assister dans leurs prises de décisions ainsi que par la constitution d'une base de connaissances sur les activités / The work presented in this PhD thesis defines a new approach for the modeling and the conceptualization of enterprise business processes in the perspective of building intelligent assistance software tools to support these processes. The proposed approach defines an organizational ontology, named K-CRIO. Its description is based on the Ontology Web Language. To illustrate our work, an intelligent assistance system has been designed and implemented according to the result from the modeling and conceptualization of a specific business process with the K-CRIO Ontology. It is a web-based application that integrates and takes full advantage of multi-agent systems.The K-CRIO Ontology is an Ontology dedicated to the study of organizations and the analysis of business processes adopting an organizational point of view. Specifically, it is used to understand, analyze and reason about organizations and the processes they implement. The targeted organizations are those composed of entities involved throughout products' design and, to do so, following a defined business process. The range of this type of organizations is quite wide. We have thus limited our study to organizations that produce software as the final process goal, specifically IT enterprises delivering software products or services. In this context, the K-CRIO ontology could be used to model structure of the considered organizations and model human activities appearing in their business processes. This ontology could be used to support process assistance within the described organizations. More specifically, the ontology could provide means for reasoning, annotating resources, monitoring design processes, enabling searches and pro-actively proposing tips and proper content. In order to illustrate the usage of K-CRIO, we apply K-CRIO on two different processes: the Waterfall Model and the Scrum methodology. These examples are both classical software-development processes. Moreover, for Scrum, the famous agile software-development process widely used in software enterprises, we have designed and developed an intelligent assistance tool. This tool mainly helps Scrum Masters to make decision by monitoring Scrum project teams' activities within their various projects and collecting knowledge about these activities.

Engineering complex systems and New Product Development (NPD) are major challenges for contemporary engineering design and must be studied at three levels of: Products, Processes and Organizations (PPO). The science of complexity indicates that complex systems share a common characteristic: they are robust yet fragile. Complex and large scale systems are robust in the face of many uncertainties and variations; however, they can collapse, when facing certain conditions. This is so since complex systems embody many subtle, intricate and nonlinear interactions. If formal modelling exercises with available computational approaches are not able to assist designers to arrive at accurate predictions, then how can we immunize our large scale and complex systems against sudden catastrophic collapse? This thesis is an investigation into complex product design. We tackle the issue first by introducing a template and/or design methodology for complex product design. This template is an integrated product design scheme which embodies and combines elements of both design theory and organization theory; in particular distributed (spatial and temporal) problem solving and adaptive team formation are brought together. This design methodology harnesses emergence and innovation through the incorporation of massive amount of numerical simulations which determines the problem structure as well as the solution space characteristics. Within the context of this design methodology three design methods based on measures of complexity are presented. Complexity measures generally reflect holistic structural characteristics of systems. At the levels of PPO, correspondingly, the Immunity Index (global modal robustness) as an objective function for solutions, the real complexity of decompositions, and the cognitive complexity of a design system are introduced These three measures are helpful in immunizing the complex PPO from chaos and catastrophic failure. In the end, a conceptual decision support system (DSS) for complex NPD based on the presented design template and the complexity measures is introduced. This support system (IMMUNE) is represented by a Multi Agent Blackboard System, and has the dual characteristic of the distributed problem solving environments and yet reflecting the centralized viewpoint to process monitoring. In other words IMMUNE advocates autonomous problem solving (design) agents that is the necessary attribute of innovative design organizations and/or innovation networks; and at the same time it promotes coherence in the design system that is usually seen in centralized systems.

A Large Scale Robot Colony (LSRC) is a complex artifact comprising of a significant population of both mobile and static robots. LSRC research is in its literary infancy and it is therefore necessary to rely upon external fields for the appropriate framework, Multi Agent Systems (MAS) and Large Scale Systems (LSS). At the intersection of MAS, LSS and LSRC exist near identical issues, problems and solutions. If attention is paid to coherence then solution portability is possible. The issue of Self-Reliability is poorly addressed by the MAS research field. Disparity between the real world and simulation is another area of concern. Despite these deficiencies, MAS and LSS are perceived as the most appropriate frameworks. MAS research focuses on three prime areas, cognitive science, management and interaction. LSRC is focused on Self-Sustainability, Self-Management and Self-Organization. While LSS research was not primarily intended for populations of mobile robots, it does address key issues of LSRC, such as effective sustainability and management. Implementation of LSRC that is based upon the optimal solution for any one or two of the three aspects will be inferior to a coherent solution based upon all three. LSRC’s are complex organizations with significant populations of both static and mobile robots. The increase in population size and the requirement to address the issue of Self-Reliance give rise to new issues. It is no longer sufficient to speak only in terms of robot intelligence, architecture, interaction or team behaviour, even though these are still valid topics. Issues such as population sustainability and management have greater significance within LSRC. As the size of a robot populations increases, minor uneconomical decisions and actions inhibit the performance of the population. Interaction must be made economical within the context of the LSRC. Sustainability of the population becomes significant as it enables stable performance and extended operational lifespan. Management becomes significant as a mechanism to direct the population so as to achieve near optimal performance. The Self-Sustainability, Self-Management and Self-Organization of LSRC are vastly more complex than in team robotics. Performance of the overall population becomes more significant than individual or team achievement. This thesis is a presentation of the Cooperative Autonomous Robot Colony (CARC) architecture. The CARC architecture is novel in that it offers a coherent baseline solution to the issue of mobile robot Self-Reliance. This research uses decomposition as a mechanism to reduce problem complexity. Self-Reliance is decomposed into Self-Sustainability, Self-Management, and Self-Organization. A solution to the issue of Self-Reliance will comprise of conflicting sub-solutions. A product of this research is a set of guidelines that manages the conflict of sub-solutions and maintains a coherent solution. In addressing the issue of Self-Reliance, it became apparent that Economies of Scale, played an important role. The effects of Economies of Scale directed the research towards LSRC’s. LSRC’s demonstrated improved efficiency and greater capability to achieve the requirements of Self-Reliance. LSRC’s implemented with the CARC architecture would extend human capability, enabling large scale operations to be performed in an economical manner, within real world and real time environments, including those of a remote and hostile nature. The theory and architecture are supported using published literature, experiments, observations and mathematical projections. Contributions of this work are focused upon the three pillars of Self-Reliance addressed by CARC: Self-Sustainability, Self-Management and Self-Organization. The chapter on Self-Sustainability explains and justifies the relevance of this issue, what it is, why it is important and how it can be achieved. Self-Sustainability enables robots to continue to operate beyond disabling events by addressing failure and routine maintenance. Mathematical projections are used to compare populations of non-sustained and sustained robots. Computer modeling experiments are used to demonstrate the feasibility of Self-Sustainability, including extended operational life, the maintenance of optimal work flow and graceful physical degradation (GPD). A detailed explanation is presented of Sustainability Functions, Colony Sites, Static Robot Roles, Static Robot Failure Options, and Polymorphism. The chapter on Self-Management explores LSS research as a mechanism to exert influence over a LSRC. An experimental reactive management strategy is demonstrated. This strategy while limited does indicate promising potential directions for future research including the Man in the Loop (MITL) strategy highly desired by NASA JPL for off world command and control of a significant robot colony (Huntsberger, et. al., 2000). Experiments on Communication evaluate both Broadcast Conveyance (BC) and Message Passing Conveyance (MPC). These experiments demonstrate the potential of Message Passing as a low cost system for LSRC communication. Analysis of Metrics indicates that a Performance Based Feedback Method (PBFM) and a Task Achievement Method (TAM) are both necessary and sufficient to monitor a LSRC. The chapter on Self-Organization describes a number of experiments, algorithms and protocols on Reasoning Robotics, a minor variant of Reactive Robotics. Reasoning Robotics utilizes an Event Driven Architecture (EDA) rather than a Stimulus Driven Architecture (SDA) common to Reactive Robotics. Enhanced robot performance is demonstrated by a combination of EDA and environmental modification enabling stigmergy. These experiments cover Intersection Navigation with contingency for Multilane Intersections, a Radio Packet Controller (RPC) algorithm, Active and Passive Beacons including a communication protocol, mobile robot navigation using Migration Decision Functions (MDF’s), including MDF positional errors. The central issue addressed by this thesis is the production of Self-Reliance guidelines for LSRC’s. Self-Reliance is perceived as a critical issue in advancing the useful and productive applications for LSRC’s. LSRC’s are complex with many issues in related fields of MAS and LSS. Decomposition of Self-Reliance into Self-Sustainability, Self-Management and Self-Organization were used to aid in problem understanding. It was found that Self-Sustainability extends the operational life of individual robots and the LSRC. Self-Management enables the exertion of human influence over the LSRC, such that the ratio of humans to robots is reduced but not eliminated. Self-Organization achieves and enhances performance through a routine and reliable LSRC environment. The product of this research was the novel CARC architecture, which consists of a set of Self-Reliance guidelines and algorithms. The Self-Reliance guidelines manage conflict between optimal solutions and provide a framework for LSRC design. This research was supported by literature, experiments, observations and mathematical projections.

This thesis presents an approach to modeling Multi Agent Systems (MAS). A framework and its implementation are presented as an extension to .NET. A number of definitions of agents are evaluated for the purpose of a broad understanding of the term software agent. Software agent has been defined in MAS context and its characteristics are identified and implemented. Motivation factors for building framework for MAS have been discussed. A number of existing technologies are discussed and evaluated. A number of agent systems previously developed are also being discussed in the middle part of the thesis. A model software agent has been defined and its characteristics are divided in two basic categories essential and optional. Its implementation has been distributed into different components throughout the MAS framework. Some of these characteristics are jointly implemented by a number of components and others responsibility rest on the individual components. Detail working of the MAS framework (i.e. what to do, when to do) is explained as guide to develop MAS using MAS framework. The protocols followed by the framework components to make communication possible between them are discussed at components level. The required information for developing MAS using MAS framework are also discussed. It answers the why, when and how questions in regards to using MAS framework A case study on Dynamic Truck Scheduling (DTS) system is discussed, designed and implemented using the MAS framework. DTS System has been used as a prototype application to test and evaluate the framework. DTS also represents a model problem that can be answered by using MAS; complete in-depth details about the problem statement are discussed. It also discusses the design and implementation of the solution along with the test results of the framework. Possible future expansion is presented in light of a number of limitations known of the MAS framework. The code working behind the different components of the MAS framework is given in appendices. Some important standards of XML that are used to pass information between agents and MAS framework components are also given in the format of tables.

對於現有文獻中討論的預測市場模型，嘗試加入風險與聲譽變數，觀察與分析其成效，並參考文獻中的代理人系統實驗方法，對論文中相關部分進行修正、設計並模擬之預測市場模型。 / In this research, we proposed two variables that could be incorporated with prediction markets: Reputation and Risk. Instead of attracting new players, The reputation system could stop losing bankrupted player, Player willing to help bankrupted player will gain reputation, and bankrupted player will lose reputation. Previous works suggest longshot bias is related to the risk-neutrality of players. Our approach is to experiment dierent risk distribution. We observe the impact of these variables in an agent-based model of prediction markets. We use zero-intelligence agents, where human qualities such as maximizing prot, learning or obeserving are missing. We further discuss the result, and the impact of risk and reputation.

La conception préliminaire avion est un problème d'optimisation multi-disciplinaire et multi-objectif qui consiste à trouver les valeurs des paramètres de description d'un avion et de ses performances en fonction de contraintes. Ces valeurs sont calculées grâce à des connaissances disciplinaires. Les interdépendances entre les paramètres et les non linéarités complexifient les calculs pour lesquels les méthodes classiques se montrent insuffisantes. Pour pallier ce problème, nous proposons une solution basée sur les systèmes multi-agents adaptatifs dans lesquels les agents représentent les disciplines, les paramètres de conception et les performances de l'avion. Les agents en utilisant des connaissances disciplinaires et par un comportement coopératif trouvent collectivement les valeurs des paramètres de conception qui satisfont les contraintes et les performances. Nous montrons que cette approche permet aussi d'intégrer plusieurs questions telles l'étude de sensibilité des paramètres, les fronts de Pareto ou la co-conception en temps réel.

L'approche à base d'agents s'intéresse à la modélisation et la simulation de systèmes complexes. Un des aspects importants dans le processus de conception est lié à la mise au point des paramètres du modèle. En effet, ces modèles sont généralement caractérisés par de nombreux paramètres qui déterminent la dynamique globale du système simulé. L'espace des paramètres peut être alors gigantesque. De plus, le comportement de ces systèmes complexes est souvent chaotique. L'approche que nous suggérons est de considérer le problème de calibrage des modèles à base d'agents comme un problème d'optimisation. La validation peut alors être reformulée comme l'identification d'un jeu de paramètres qui optimise une fonction, par exemple une mesure de distance entre le modèle artificiel que nous simulons et le système réel. Nous avons proposé trois approches complémentaires dans le cadre de la thèse. La première approche se fonde sur l'application directe d'un algorithme d'optimisation où le modèle est vu comme une boîte noire dont les entrées sont les valeurs de paramètres et la sortie la valeur d'une fonction objectif évaluée après la simulation du modèle. La deuxième approche consiste ensuite à explorer différentiellement l'espace des paramètres en le découpant de manière adaptative en sous-espaces d'autant plus finement découpés que les paramétrages correspondants sont a priori intéressants. Enfin, la troisième approche consiste à effectuer une seule simulation où les agents sont modifiés en ligne.

L'instrumentation de la supervision passe par la compréhension de son utilité et de sa faisabilité. Les indicateurs fournissent au tuteur des informations lui permettant de réguler les activités d'apprentissage au niveau Cognitif, Pédagogique, Social et Technique. Cette thèse se propose de répondre aux questions portant sur l'identification des types d'indicateurs utiles à chaque rôle de régulation du tuteur et sur la proposition d'une méthodologie permettant de réutiliser un ensemble d'indicateurs, de ces types, dans un environnement de supervision.<br />Le modèle CAS de supervision propose une taxonomie permettant au tuteur d'avoir une perception de l'activité d'apprentissage. Une enquête auprès de tuteurs en ligne a montré que les indicateurs Cognitifs permettent de jouer des rôles pédagogique et cognitif, les indicateurs Sociaux le rôle social. L'utilité des indicateurs d'Activité (Parcours et Technique) n'en est pas clairement ressortie. L'enquête révèle également un besoin de personnalisation de la supervision. <br />A la problématique d'instrumentation de la réutilisation, nous répondons en proposant un formalisme de Patron d'Indicateur Réutilisable qui permet de capitaliser les savoir-faire en terme de définition d'indicateur, ainsi qu'une Architecture Multi-Agent Ouverte permettant de les réutiliser dans de nouveaux contextes de supervision. Ces savoir-faire, modélisés en termes de fonction, peuvent dans l'architecture proposée se déployer et se combiner pour informer le tuteur sur l'activité d'apprentissage. Les opérations d'Agrégation et de Composition sont définies sur les fonctions «indicateur». L'environnement EM-AGIIR, implémentant l'architecture proposée est ouvert, évolutif et permet au tuteur de personnaliser son activité de supervision. Un ensemble d'indicateurs sur les traces du projet MATES du réseau européen Kaleidoscope y ont été réutilisés.