The creation of expressive styles for digital art is one of the primary goals in non-photorealistic rendering. In this paper, we introduce a swarm-based multi-agent system that is capable of producing expressive imagery through the use of multiple digital images. At birth, agents in our system are assigned a digital image that represents their 'aesthetic ideal'. As agents move throughout a digital canvas they try to 'realize' their ideal by modifying the pixels in the digital canvas to be closer to the pixels in their aesthetic ideal. When groups of agents with different aesthetic ideals occupy the same canvas, a new image is created through the convergence of their competing aesthetic goals. We use our system to explore the concepts and techniques from a number of Modern Art movements and to create an interactive media installation. The simple implementation and effective results produced by our system makes a compelling argument for more research using swarm-based multi-agent systems for non-photorealistic rendering. / Graduate

The objective of this thesis is to develop decentralized methods for building robust multi-agent networks through self-organization. Multi-agent networks appear in a large number of natural and engineered systems, including but not limited to, biological networks, social networks, communication systems, transportation systems, power grids, and robotic swarms. Networked systems typically consist of numerous components that interact with each other to achieve some collaborative tasks such as flocking, coverage optimization, load balancing, or distributed estimation, to name a few. Multi-agent networks are often modeled via interaction graphs, where the nodes represent the agents and the edges denote direct interactions between the corresponding agents. Interaction graphs play a significant role in the overall behavior and performance of multi-agent networks. There- fore, graph theoretic analysis of networked systems has received a considerable amount of attention within the last decade. In many applications, network components are likely to face various functional or structural disturbances including, but not limited to, component failures, noise, or malicious attacks. Hence, a desirable network property is robustness, which is the ability to perform reasonably well even when the network is subjected to such perturbations. In this thesis, robustness in multi-agent networks is pursued in two parts. The first part presents a decentralized graph reconfiguration scheme for formation of robust interaction graphs. Particularly, the proposed scheme transforms any interaction graph into a random regular graph, which is robust to the perturbations of their nodes/links. The second part presents a decentralized coverage control scheme for optimal protection of networks by some mobile security resources. As such, the proposed scheme drives a group of arbitrarily deployed resources to optimal locations on a network in a decentralized fashion.

In a multi-agent system (MAS), the overall performance is greatly influenced by both the design and the control of the agents. The physical design determines the agent capabilities, and the control strategies drive the agents to pursue their objectives using the available capabilities. The objective of this thesis is to incorporate control strategies in the early conceptual design of an MAS. As such, this thesis proposes a methodology that mainly explores the interdependency between the design variables of the agents and the control strategies used by the agents. The output of the proposed methodology, i.e. the interdependency between the design variables and the control strategies, can be utilized in the requirement analysis as well as in the later design stages to optimize the overall system through some higher fidelity analyses. In this thesis, the proposed methodology is applied to a persistent multi-UAV surveillance problem, whose objective is to increase the situational awareness of a base that receives some instantaneous monitoring information from a group of UAVs. Each UAV has a limited energy capacity and a limited communication range. Accordingly, the connectivity of the communication network becomes essential for the information flow from the UAVs to the base. In long-run missions, the UAVs need to return to the base for refueling with certain frequencies depending on their endurance. Whenever a UAV leaves the surveillance area, the remaining UAVs may need relocation to mitigate the impact of its absence. In the control part of this thesis, a set of energy-aware control strategies are developed for efficient multi-UAV surveillance operations. To this end, this thesis first proposes a decentralized strategy to recover the connectivity of the communication network. Second, it presents two return policies for UAVs to achieve energy-aware persistent surveillance. In the design part of this thesis, a design space exploration is performed to investigate the overall performance by varying a set of design variables and the candidate control strategies. Overall, it is shown that a control strategy used by an MAS affects the influence of the design variables on the mission performance. Furthermore, the proposed methodology identifies the preferable pairs of design variables and control strategies through low fidelity analysis in the early design stages.

This thesis relates to the usage of trust modelling in multi-agent systems - environments in which there are interacting software agents representing various users (for example, buyers and sellers exchanging products and services in an electronic marketplace). In such applications, trust modelling may be crucial to allow one group of agents (in the e-commerce scenario, buyers) to make effective decisions about which other agents (i.e., sellers) are the most appropriate partners. A number of existing multi-agent trust models have been proposed in the literature to help buyers accurately select the most trustworthy sellers. Our contribution is to propose several modifications that can be applied to existing probabilistic multi-agent trust models. First, we examine how the accuracy of the model can be improved by limiting the network to a portion of the population consisting of the most trustworthy agents, such that the less trustworthy contributions of the remaining agents can be ignored. In particular, we explore how this can be accomplished by either setting a maximum size for a buyer's advisor network or setting a minimum trustworthiness threshold for agents to be accepted into that advisor network, and develop methods for appropriately selecting the values to limit the network size. We demonstrate that for two models, both the Personalized Trust Model (PTM) developed by Zhang as well as TRAVOS, these approaches will yield significant improvements to the accuracy of the trust model, as opposed to using an unrestricted advisor network. Our final proposed modification is to use an advisor referral system in combination with one of the network-limiting approaches. This would ensure that if a particular agent within the advisor network had not met a specified level of experience with the seller under consideration, it could be replaced by another agent that had greater experience with that seller, which should in turn allow for a more accurate modelling of the seller's trustworthiness. We present a particular approach for replacing advisors, and show that this will yield additional improvements in trust-modelling accuracy with both PTM and TRAVOS, especially if the limiting step were such that it would yield a very small advisor network. We believe that these techniques will be very useful for trust researchers seeking to improve the accuracy of their own trust models, and to that end we explain how other researchers could apply these modifications themselves, in order to identify the optimal parameters for their usage. We discuss as well the value of our proposals for identifying an "optimal" size for a social network, and the use of referral systems, for researchers in other areas of artificial intelligence.

In this work, reinforcement learning algorithms are studied with the help of potential field methods, using robosoccer simulators as test beds. Reinforcement Learning (RL) is a framework for general problem solving where an agent can learn through experience. The soccer game is selected as the problem domain a way of experimenting multi-agent team behaviors because of its popularity and complexity.

This thesis addresses the problem of learning in physical heterogeneous multi-agent systems(MAS) and the analysis of the benefits of using heterogeneous MAS with respect tohomogeneous ones. An algorithm is developed for this task; building on a previous work on stability in distributed systems by Tad Hogg and Bernardo Huberman, and combining two phenomena observed in natural systems, task partition and hierarchical dominance. This algorithm is devised for allowing agents to learn which are the best tasks to perform on the basis of each agent's skills and the contribution to the team global performance. Agents learn by interacting with the environment and other teammates, and get rewards from the result of the actions they perform. This algorithm is specially designed for problems where all robots have to co-operate and work simultaneously towards the same goal. One example of such a problem is role distribution in a team of heterogeneous robots that form a soccer team, where all members take decisions and co-operate simultaneously. Soccer offers the possibility of conducting research in MAS, where co-operation plays a very important role in a dynamical and changing environment. For these reasons and the experience of the University of Girona in this domain, soccer has been selected as the test-bed for this research. In the case of soccer, tasks are grouped by means of roles.One of the most interesting features of this algorithm is that it endows MAS with a highadaptability to changes in the environment. It allows the team to perform their tasks, whileadapting to the environment. This is studied in several cases, for changes in the environment and in the robot's body. Other features are also analysed, especially a parameter that defines the fitness (biological concept) of each agent in the system, which contributes to performance and team adaptability.The algorithm is applied later to allow agents to learn in teams of homogeneous andheterogeneous robots which roles they have to select, in order to maximise team performance. The teams are compared and the performance is evaluated in the games against three hand-coded teams and against the different homogeneous and heterogeneous teams built in this thesis. This section focuses on the analysis of performance and task partition, in order to study the benefits of heterogeneity in physical MAS.In order to study heterogeneity from a rigorous point of view, a diversity measure is developed building on the hierarchic social entropy defined by Tucker Balch. This is adapted to quantify physical diversity in robot teams. This tool presents very interesting features, as it can be used in the future to design heterogeneous teams on the basis of the knowledge on other teams.

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.

The thesis solves a challenging problem in military logistics for tasks such as transportation scheduling and combinatorial auctions. A conceptual model has been developed that captures the organisational business processes involved and an effective implementation suitable for computer software agents. The protocol facilitates planning and task allocation among organisations in decentralised, dynamic and open environments.

Η παρούσα εργασία πραγματεύεται ζητήματα που αφορούν σε συστήματα αυτομάτου ελέγχου με ιδιαίτερη έμφαση στα συστήματα κατανεμημένου ελέγχου. Σε αυτό το πλαίσιο, πραγματοποιείται στην εργασία καταρχάς μια σύντομη ανασκόπηση σε σχέση με τα συστήματα ελέγχου και με το πώς αυτά τα συστήματα εισήχθησαν για χρήση στη βιομηχανία. Στη συνέχεια, παρουσιάζονται τα κατανεμημένα συστήματα ελέγχου, τα βασικά χαρακτηριστικά τους αλλά και οι βασικές μεθοδολογίες σχεδίασης ενός τέτοιου συστήματος. Γίνεται επίσης σημαντική αναφορά στα πολυπρακτορικά συστήματα (Multi Agent Systems – MAS) τα οποία αποτελούν το μέλλον της ανάπτυξης των κατανεμημένων συστημάτων. Τέλος, συνοψίζονται τα βασικά προβλήματα, οι τάσεις αλλά και η κατάσταση στην αγορά αναφορικά με αυτά τα συστήματα. / -

Dentro da Educação a Distância (EaD), os softwares de apoio como os Ambientes Virtuais de Aprendizagem (AVA) são considerados recursos que favorecem a comunicação entre os atores envolvidos, permitindo a troca de informação. Atribuir Inteligência Artificial a esses AVAs, utilizando Sistemas Multiagentes (SMA) e uma forma de procurar que os mesmos tenham um bom desempenho e que seus recursos facilitem o processo de aprendizagem. Esse trabalho cont em um estudo sobre os principais AVAs existentes e sobre os métodos alternativos de integração de AVA com SMA. Ao analisar o estado da arte dos AVAs pode-se observar que todos trabalham como ferramentas de auxílio ao aluno, por em nenhum deles trabalha aspectos da gestão da EaD dando suporte aos aspectos relevantes da legislação dessa modalidade. Por esse motivo, essa Tese tem por objetivo a criação de um modelo de integração AVA-SMA que possa tornar o AVA MOODLE capaz de auxiliar os gestores da EaD em suas diferentes tarefas, com base na incorporação, ao mesmo, de um modelo de representação de legislação. Para realizar essa integração do modelo AVA-SMA orientado a Legislação foi desenvolvido um espec co modelo organizacional de Sistema Multiagente. Por m com base em um estudo de caso, ser a realizado simulações para veri car as funcionalidades do Modelo de Sistema AVA-SMA orientado a Legislação, proposto nesta Tese. / In Distance Learning (EaD), supporting software such as Virtual Learning Environments (VLE) are considered resources that favor communication between the actors involved, allowing the exchange of information. Assigning Arti cial Intelligence to these VLEs, using Multi-Agent Systems (MAS) is a way of ensuring they have a good performance and that its resources facilitate the learning process. This work contains a study on the major existing VLEs and on alternative methods to integrate VLE with MAS. When analyzing the state of the art of the VLEs it is possible to see that all of them work as aid tools for students, but none of them work on management aspects of distance learning that support the relevant aspects of the legislation for this type of education. Therefore, this thesis aims to create a VLE-MAS integration model that can make the VLE MOODLE able to help distance learning managers in their di erent tasks, based on incorporating a legislation representation model to it. To accomplish this integration of the legislation-oriented VLE-MAS model, a speci c Multi-Agent System organizational model was developed. At last, based on a case study, simulations will be conducted to verify the functionalities of the VLE-MAS System Model oriented to legislation, proposed in this thesis.