Autonomous aerial transportation has multiple potential applications including emergency cases and rescue missions where ground intervention may be difficult. In this context, the following work will address the control of multi-agent Vertical Take-off and Landing aircraft (VTOL) transportation system. We develop a decentralized method. The advantage of such a solution is that it can provide better maneuverability and lifting capabilities compared to existing systems. First, we consider a cooperative group of VTOLs transporting one payload. The main idea is that each agent perceive the interaction with other agents as a disturbance while assuming a negotiated motion model and imposing certain magnitude bounds on each agent. The theoretical model will be then validated using a numerical simulation illustrating the interesting features of the presented control method. Results show that under specified disturbances, the algorithm is able to guarantee the tracking with a minimal error. We describe a toolbox that has been developed for this purpose. Then, a system of multiple VTOLs lifting payloads will be studied. The algorithm assures that the VTOLs are coordinated with minimal communication. Additionally, a novel gripper design for ferrous objects is presented that enables the transportation of ferrous objects without a cable. Finally, we discuss potential connections to human in the loop transportation systems.

Management solutions for current and future Infrastructure-as-a-Service (IaaS) Data Centers (DCs) face complex challenges. First, DCs are now very large infrastructures holding hundreds of thousands if not millions of servers and applications. Second, DCs are highly heterogeneous. DC infrastructures consist of servers and network devices with different capabilities from various vendors and different generations. Cloud applications are owned by different tenants and have different characteristics and requirements. Third, most DC elements are highly dynamic. Applications can change over time. During their lifetime, their logical architectures evolve and change according to workload and resource requirements. Failures and bursty resource demand can lead to unstable states affecting a large number of services. Global and centralized approaches limit scalability and are not suitable for large dynamic DC environments with multiple tenants with different application requirements. We propose a novel fully distributed and dynamic management paradigm for highly diverse and volatile DC environments. We develop LAMA, a novel framework for managing large scale cloud infrastructures based on a multi-agent system (MAS). Provider agents collaborate to advertise and manage available resources, while app agents provide integrated and customized application management. Distributing management tasks allows LAMA to scale naturally. Integrated approach improves its efficiency. The proximity to the application and knowledge of the DC environment allow agents to quickly react to changes in performance and to pre-plan for potential failures. We implement and deploy LAMA in a testbed server cluster. We demonstrate how LAMA improves scalability of management tasks such as provisioning and monitoring. We evaluate LAMA in light of state-of-the-art open source frameworks. LAMA enables customized dynamic management strategies to multi-tier applications. These strategies can be configured to respond to failures and workload changes within the limits of the desired SLA for each application.

A qualidade do serviço oferecido por Provedores do Serviço de Internet (Internet Service Provider - ISPs) depende diretamente da quantidade de recursos disponíveis naquele momento. Nas últimas décadas, essa qualidade tem sido afetada por frequentes e intensos ataques que consomem tais recursos, como é o caso dos ataques de Negação de Serviço Distribuídos (Distributed Denial of Service - DDoS). Com o objetivo de tornar a rede dos ISPs mais resiliente aos diferentes tipos de ataques DDoS, foram desenvolvidas técnicas contra tais ataques ao longo dos últimos anos. Com o objetivo de contribuir com a melhoria de tais mecanismos, esta dissertação apresenta um método autônomo reativo para detecção e mitigação de ataques DDoS, utilizando um sistema multiagentes (SMA), em redes de ISPs. A propriedade principal do método proposto é identificar padrões de tráfego característicos de um ataque, como um grande fluxo de pacotes direcionados para um serviço ou equipamento, dentro da rede do ISP. Com os agentes posicionados nas prováveis vítimas e nos pontos da rede com maior fluxo de pacotes, o processo de mitigação inicia-se automaticamente após uma quantidade de pacotes, excedente ao tráfego padrão, passar por qualquer um dos nós monitorados. Como o tráfego entrante na rede do ISP é dinâmico, seja ele legítimo ou malicioso, a utilização de agentes tende a facilitar o processo de definição da rota de ataque, conforme mostram os resultados experimentais obtido com o sistema proposto. / The quality of service offered by the Internet Service Provider (ISP) depends directly on the amount of resources available at that time. In recent decades, this quality has been affected by the frequent and intense attacks that consume these resources, such as the Distributed Denial of Service (DDoS) attacks. In order to make the ISPs network more resilient to different types of DDoS attacks, techniques have been developed against such attacks over the past few years. Aiming to contribute to the improvement of such mechanisms, this dissertation presents a reactive autonomous method for detecting and mitigating DDoS attacks using a Multi-Agent system (MAS), in networks of ISPs. The main property of the proposed method is to identify characteristic traffic patterns of an attack, such as a large stream of packets directed to a service or equipment within the ISP network. With agents positioned on likely victims and at points of the network with the highest packet stream, the mitigation process starts automatically after a number of packets exceeding the traffic pattern, go through any of the monitored nodes. Since the incoming traffic on the network of any ISP is dynamic, whether legitimate or malicious, the using of agents tends to facilitate the process of defining the route of attack, as shown by the experimental results obtained with the proposed system.

Esta dissertação propõe uma Ontologia Funcional de Reputação para agentes. Tal ontologia representa uma parte do conhecimento científico sobre reputação provido tanto pelas Ciências Humanas quanto pelos trabalhos em Inteligência Artificial. Seu objetivo é oferecer uma perspectiva funcional para a representação e análise da reputação como mecanismo de controle social em sociedades de agentes, de forma a sustentar a implementação de modelos de reputação para agentes. A Ontologia Funcional de Reputação foi construída a partir das categorias de conhecimento da Ontologia Funcional do Direito, proposta por Valente (1995), e utiliza a categorização de conceitos definida para o mundo jurídico para a representação do mundo social. A ontologia foi codificada em OWL DL, uma linguagem formal baseada em lógica descritiva. Uma vez construída a ontologia, os conceitos utilizados por diversos modelos e sistemas de reputação foram utilizados na sua avaliação. A utilização de um motor de inferência permitiu comparar estes conceitos, representados por meio de classes OWL, com as classes da ontologia. Tal comparação permite avaliar, de modo preliminar, a adequação do uso da Ontologia Funcional de Reputação como uma possível interlingua entre diversos agentes heterogêneos, cada qual utilizando um modelo de reputação diferente, que necessitam interoperar. / This work presents a Functional Ontology of Reputation for agents. This ontology represents the broad knowledge about reputation produced in some areas of interest such as Social Sciences and Artificial Intelligence. Its goal is to provide a functional perspective both to represent and analyze reputation as a social control mechanism for agent’s societies, in order to support the implementation of reputation model for agents. The Functional Ontology of Reputation employs the primitive categories of knowledge used in the Functional Ontology of Law proposed by Valente (1995). The idea is that the concepts of the legal world can be used to model the social world, through the extension of the concept of legal rule to social norm and the internalization of social control mechanisms in the agent's mind, so far externalized in legal institutions. The Functional Ontology of Reputation contains five main categories that have been borrowed from or inspired by the Functional Ontology of Law: Reputative Knowledge, Responsibility Knowledge, Normative Knowledge, World Knowledge and Common Sense Knowledge.As in the Functional Ontology of Law, the distinction among the categories of the reputation ontology are accomplished according to a functional perspective, in which each component of the reputation system, embedded in the social system, exists to perform a specific function in the effort to achieve social objectives, such as trust, reciprocity and social cooperation. The Functional Ontology of Reputation was implemented in OWL, a description logic language. This ontology was evaluated by using several concepts related to reputation, included in different reputation models and reputation systems. These concepts were defined as OWL classes and a reasoner was used in order to produce the comparison between these concepts and the ontology classes. This comparison allows evaluating, in a preliminary way, the Functional Ontology of Reputation utilization as a possible interlingua between several heterogeneous agents that need to interoperate, despite the utilization of different reputation model.

Nesta dissertação, o autor aborda a dificuldade de prever-se o desempenho dos recursos humanos em um processo de desenvolvimento de software em um ambiente de Programação Extrema (XP) (BECK, 2000) e propõe uma solução com potencial para minimizar esse problema. Especificamente, o problema, a ser tratado neste trabalho, consiste em melhorar as previsões dos gerentes de projeto – no âmbito do ambiente mencionado - com relação ao desempenho dos recursos humanos na geração de valor para o negócio. Tal valor para o negócio é alcançado através da implementação, por parte dos programadores, das diversas funcionalidades de um sistema de software. Para a construção da solução proposta neste trabalho, o autor analisou um sistema XP de desenvolvimento de software (composto por ambiente, pessoas e processo), conforme o processo de modelagem proposto por Streit (2006) e apoiado na revisão da literatura relevante. Em seguida, o autor estruturou esse sistema em um modelo conceitual para, finalmente, desenvolver um modelo computacional do sistema analisado, baseado em múltiplos agentes inteligentes modelados conforme a arquitetura Beliefs-Desires-Intentions (BDI), ou Crenças-Desejos-Intenções. O modelo computacional da simulação multi-agente foi desenvolvido com o apoio da ferramenta SeSAm (KLÜGL, 2006). Testado através da experimentação estatística 2k Fatorial (LAW e KELTON, 2000), o modelo de simulação multi-agente de processos de desenvolvimento de software, para ambientes de Programação Extrema, demonstrou eficácia e aplicabilidade prática sobre o problema em questão. / In this research, the author adresses the difficulty to forecast the performance of the human resources in a software development process in an Extreme Programming (XP) (BECK, 2000) environment and proposes a solution that may be suitable to minimize this problem. Specifically, the main problem consists on how to improve the assumptions of the project managers - in the aforementioned environment - related with the human resources performance in generating value for the business. This value generation is reached through the implementation, by programmers, of the various functionalities of a software system. To build the solution proposed in this research, the author analysed a XP software development system (composed of environment, people and process) considering the modeling process proposed by Streit (2006) and also the relevant related works. This system was later structured in a conceptual model and, in sequence, in a computational model based on the Beliefs-Desires-Intentions (BDI) architecture of intelligent agents. The computational model of the multi-agent simulation was build with the support of the SeSAm (KLÜGL, 2006) tool. The tests of the multi-agent simulation of XP software develoment process model used the 2k Factorial statistical experimentation (LAW e KELTON, 2000) and their results demonstrated the effectiveness and practical applicability of the model for the research problem.

In direct reference to the saying “time is money”, nowadays scenario simulations play a key role in the tasks people perform. Optimizing the time length of tasks and synchronizing them properly is essential to increased profits in any line of business. In this thesis the Bergkvist-Insjön sawmill yard process will be computer simulated. As the number of trucks arriving at the sawmill is unknown, the unexpected arrival of trucks would produce a high pressure on internal resources and handling operations. The aim of this paper is to optimize the usage of the resources in the Bergkvist-Insjön sawmill, by running three different scenarios built based on the real system simulation. Scenario number three, in which a log stacker only has the tasks to unload the trucks and supply the measurement station, has been found most efficient. In the simulation of this scenario, the number of logs processed by the sawmill is the highest one. Also, the time spent by the log stackers between their tasks is the shortest one from all scenarios. The results of this thesis revealed that the most efficient improvement of the sawmill yard would be gained by a different tasks’ priority for the operating log stackers.

A qualidade do serviço oferecido por Provedores do Serviço de Internet (Internet Service Provider - ISPs) depende diretamente da quantidade de recursos disponíveis naquele momento. Nas últimas décadas, essa qualidade tem sido afetada por frequentes e intensos ataques que consomem tais recursos, como é o caso dos ataques de Negação de Serviço Distribuídos (Distributed Denial of Service - DDoS). Com o objetivo de tornar a rede dos ISPs mais resiliente aos diferentes tipos de ataques DDoS, foram desenvolvidas técnicas contra tais ataques ao longo dos últimos anos. Com o objetivo de contribuir com a melhoria de tais mecanismos, esta dissertação apresenta um método autônomo reativo para detecção e mitigação de ataques DDoS, utilizando um sistema multiagentes (SMA), em redes de ISPs. A propriedade principal do método proposto é identificar padrões de tráfego característicos de um ataque, como um grande fluxo de pacotes direcionados para um serviço ou equipamento, dentro da rede do ISP. Com os agentes posicionados nas prováveis vítimas e nos pontos da rede com maior fluxo de pacotes, o processo de mitigação inicia-se automaticamente após uma quantidade de pacotes, excedente ao tráfego padrão, passar por qualquer um dos nós monitorados. Como o tráfego entrante na rede do ISP é dinâmico, seja ele legítimo ou malicioso, a utilização de agentes tende a facilitar o processo de definição da rota de ataque, conforme mostram os resultados experimentais obtido com o sistema proposto. / The quality of service offered by the Internet Service Provider (ISP) depends directly on the amount of resources available at that time. In recent decades, this quality has been affected by the frequent and intense attacks that consume these resources, such as the Distributed Denial of Service (DDoS) attacks. In order to make the ISPs network more resilient to different types of DDoS attacks, techniques have been developed against such attacks over the past few years. Aiming to contribute to the improvement of such mechanisms, this dissertation presents a reactive autonomous method for detecting and mitigating DDoS attacks using a Multi-Agent system (MAS), in networks of ISPs. The main property of the proposed method is to identify characteristic traffic patterns of an attack, such as a large stream of packets directed to a service or equipment within the ISP network. With agents positioned on likely victims and at points of the network with the highest packet stream, the mitigation process starts automatically after a number of packets exceeding the traffic pattern, go through any of the monitored nodes. Since the incoming traffic on the network of any ISP is dynamic, whether legitimate or malicious, the using of agents tends to facilitate the process of defining the route of attack, as shown by the experimental results obtained with the proposed system.

Avec le développement des objets connectés, les agents embarqués déployés dans des environnements physiques et les applications multi-agents qui les impliquent deviennent de plus en plus populaires. Ces systèmes multi-agents sont amenés à partager le même environnement physique. Cette cohabitation d'agents de systèmes différents, qui n'ont pas nécessairement été prévus pour interagir entre eux par les concepteurs, les amène cependant à se solliciter. Un agent peut alors participer à la réalisation d'objectifs incompatibles avec les siens ou ceux de ses collectifs. Pour éviter ces situations, nous proposons un modèle d'agent basé sur les identités pour l'aider à décider de sa participation ou non à des actions collectives. / Embedded agents deployed in physical environments are increasingly interoperable, and are likely to coexist with agents of others systems in a same physical space. So, an agent needs to be able to cooperate with agents from other systems and to form coalitions with unfamiliar teammates. However, before committing to cooperate with others, an agent must take into account that it may get involved in the achievement of objectives that are incompatible with its own, with the global objectives of its system, or with those derived from its previously joined coalitions. To avoid such situations, we propose an identity-based cooperation mechanism. This mechanism involves creating and sustaining the agent identity, and a commitment process to reason about identities when solicited to participate in a collective trans-MAS action.

This dissertation studies the application of several event-driven control schemes in multi-agent systems. First, a new cooperative receding horizon (CRH) controller is designed and applied to a class of maximum reward collection problems. Target rewards are time-variant with finite deadlines and the environment contains uncertainties. The new methodology adapts an event-driven approach by optimizing the control for a planning horizon and updating it for a shorter action horizon. The proposed CRH controller addresses several issues including potential instabilities and oscillations. It also improves the estimated reward-to-go which enhances the overall performance of the controller. The other major contribution is that the originally infinite-dimensional feasible control set is reduced to a finite set at each time step which improves the computational cost of the controller. Second, a new event-driven methodology is studied for trajectory planning in multi-agent systems. A rigorous optimal control solution is employed using numerical solutions which turn out to be computationally infeasible in real time applications. The problem is then parameterized using several families of parametric trajectories. The solution to the parametric optimization relies on an unbiased estimate of the objective function's gradient obtained by the "Infinitesimal Perturbation Analysis" method. The premise of event-driven methods is that the events involved are observable so as to "excite" the underlying event-driven controller. However, it is not always obvious that these events actually take place under every feasible control in which case the controller may be useless. This issue of event excitation, which arises specially in multi-agent systems with a finite number of targets, is studied and addressed by introducing a novel performance measure which generates a potential field over the mission space. The effect of the new performance metric is demonstrated through simulation and analytical results.

This dissertation analyzes the contracting problem between a firm and the research employees in its R&D department. The dissertation consists of two chapters. The first chapter addresses a simplified problem in which the R&D unit has only one agent. The second chapter studies a scenario in which the R&D unit consists of a team. In the first chapter, I look at problem in which a principal hires an agent to do a multi-stage R&D project. The transition from one stage to the next is modeled by a Poisson-type process, whose arrival rate depends on the agents choice of effort. I assume that effort choice is binary and unobservable by the principal. To overcome the repeated moral-hazard problem, the principal offers the agent a long-term contract which specifies a flow of payments based on his observation of the outcome of the project. The optimal contract combines rewards and punishments: the payment to the agent decrease over time in case of failure and jumps up to a higher level after each success. I also show that the optimal contract can be implemented by using a risky security that has some of the features of the stocks of these firms, thereby providing a theoretical justification for the wide-spread use of stock-based compensation in firms that rely on R&D. In the second chapter, I look at a scenario in which the R&D unit consists of a team, which I assume, for simplicity, comprises two risk-averse agents. Now, the Poisson arrival rate is jointly determined by the actions of both agents with the action of each remaining unobservable by both the principal and the other agent. I assume that when success in a phase occurs the principal can identify the agent who was responsible for it. In this model, incentive compatibility means that each agent is willing to exert effort conditional on his coworker putting in effort, and thus exerting effort continuously is a Nash-equilibrium strategy played by the agents. In this multiagent problem, each agents payment depends not only on his own performance, but is affected by the other agents performance as well. Similar to the single-agent case, an agent is rewarded when he succeeds, and his payment decreases over time when both agents fail. Regarding how an agents payment relates to his coworkers performance, I find that the optimal incentive regime is a function of the way in which agents efforts interact with one another: relative-performance evaluation is used when their efforts are substitutes whereas joint-performance evaluation is used when their efforts are complements. This result sheds new light on the notion of optimal incentive regimes, an issue that has been widely discussed in multi-agent incentive problems.