textabstract“Agent-based Inter-organizational Systems (ABIOS) in Advanced Logistics Operations” explores the concepts, the design, and the role and impact of agent-based systems to improve coordination and performance of logistics operations. The dissertation consists of one conceptual study and three empirical studies. The empirical studies apply various research methods such as a multiple-case study research, coordination mechanism design, and predictive analytics using big data. The conceptual study presents a theoretical exploration and synthesis explaining the demand for inter-organizational systems (IOS) and the corresponding IOS functionalities. The first empirical study presents a multiple-case study exploring real-life ABIOS implementations in the warehousing and transportation business. The second empirical study provides an auction based coordination mechanism design for the container’s pick-up/delivery appointment reservation problem that involves the seaports and drayage operators. The third empirical study presents a seaport service rate prediction system that could help drayage operators to improve their predictions of the duration of the pick-up/delivery operations at a seaport by using the subordinate trucks’ trajectory data. Based on these studies, the dissertation offers new insights on the role of inter-organizational systems in mitigating coordination equivocality and uncertainties; the interplay among the ABIOS applications, required structural adjustments, and the potential of business performance improvement opportunities; and the development of two ABIOS prototypes: an auction based coordination mechanism and a predictive analytics application based on big data.

technology acceptance grows, there is a need for software engineering approaches to deal with the design of large, complex multiagent applications. Currently, existing approaches work well for small systems but are not well suited for large, complex applications. Thus, we propose extending multiagent approaches with service-oriented principles to simplify the design of such systems. Our approach composes small multiagent-based services to build larger, more complex applications. This paper introduces the key concepts required to design reusable multiagent services and shows how the services can be composed to create complex multiagent applications.

systems composed of highly similar, semi- autonomous agents are being developed in several application domains. An example of such multi-agent systems is a fleet, or "constellation" of satellites. In constellations o f satellites, each satellite is commonly treated as a distinct autonom ous agent that must cooperate to achieve higher-level constellatio n goals. In previous work, we have shown that modeling a constellation of satellites or spacecraft as a product line of agent s (where the agents have many shared commonalities and a few key differences) enables reuse of software analysis and design assets. We have also previously developed efficient safety analysis techniques for product lines. We now propose the use of Bi-Directional Safety Analysis (BDSA) to aid in system certification. We extend B DSA to product lines of multi-agent systems and show how t he analysis artifacts thus produced contribute to the software' s safety case for certification purposes. The product-line approach lets us reuse portions of the safety analysis for multiple agents , significantly reducing the burden of certification. We motivate and illustrate this work through a specific application, a product -line, multi- agent satellite constellation.

ions defined in the DBCM to ’talk about’ the actual resources. The actual resources are the actual ’things’ that are exchanged between the coordinating entities. In the context of this work, the interaction objects set, O, is considered as the actual resources set. The link between the resource nodes and actual resources is made by the control functions2. A control function is a Boolean function that returns true when the actual resource conforms the standard that has been established between the producer and consumer, and false otherwise. So, for each resource node a control function is associated in order to control if the actual resource corresponds to the resource node. Still, some resource nodes do not require an exchange of actual resources. In fact, some resource nodes model a prerequisite relationship among the activities without a concrete exchange of an actual resource. For these particular resource nodes, a particular value (null or zero) is associated as a control function. The resource nodes that belong to the roles-cut must always be mapped to non-null control functions. In fact, in a distributed system the communication between the distributed activities always implies the explicit exchange of interaction object. The resource recognition mapping, κ, is introduced as an assignment of control functions to resource nodes in a DBCM: Definition 4.5.6. Let G = (A,R, P,C) be a coordination model and O the set of actual resources. The resource control mapping κ of the coordination model G is a function from the set of resource nodes to the set of control functions or a null value: κ : R→ (O → {>,⊥}) ∪ {0} The roles-cut structure imposes that only the resources belonging to the role-cut are observable by an external observer. So, the mapping κ is said to satisfy the constraints of the roles-cut c iff: ∀r ∈ R { κ(r) = 0 if r / ∈ c κ(r) 6= 0 if r ∈ c The recognition of an actual resource sequence m = m0m1...mn ∈ O∗ according to a resource nodes pattern p = r0r1...rm ∈ R∗ and a mapping κ is performed as follows: 1. each resource node in the resource pattern p is replaced by its associated control function defined by κ. This produces a sequence of control functions f = f0f1...fj ∈ ((O → {>,⊥}) ∪ {0})∗. 2. the sequence of functions is normalized by removing all the zero (or null) elements. The normalized sequence is noted f ′ = ||f ||. 3. the sequence of functions ’is applied’ to the sequence of actual resources m. This is done by replacing each i-th element of m (mi) by the application of the i-th function of f ′ to mi. So, the result of this operation is a sequence of Boolean values. Malone and Crowston works have already introduced the concept of control function when they were speaking about the usability dependency. te l-0 01 42 84 3, v er si on 1 23 A pr 2 00 7 108 Chapter 4 : Coordinating the Autonomous Agents 4. if the resulting sequence contains only > values, this means that the actual resources sequence has been validated by the resource pattern and the recognition mapping κ; otherwise, it has been invalided. To simply the notations, the fact that a resource pattern p recognizes an actual resource sequence m according to the mapping κ is noticed by a logical predicate: validate(p, c, κ). For each pattern of resource nodes (that is element of R∗) and a mapping κ, we can define the set of all actual resources (elements of O∗) which are recognized. Within the MAS paradigm the O set is interpreted as interaction objects. So, the sequence of actual resources is defined as a conversation and the set of all recognized conversations is defined as the set of valid conversations. Definition 4.5.7. Let O be the set of interaction objects; G = (A,R, P,C) a DBCM associated to the resource recognition mapping κ and a roles-cut d such that the mapping κ satisfies the constraints of the roles-cut d. The set of valid conversations generated by G and the axiom x0 ∈ R∗ is defined as the set of all the sequences of interaction objects that are validated by at least one resource pattern of G according to the axiom x0: ∀c ∈ O∗, c ∈ conversations(G,κ,d,x0) ⇐⇒ ∃p ∈ r patterns(G,x0) : validate(p, c, κ) Example 4.5.2. For this example, we assume that the set of actual resources contains two binary values O = {0, 1}. The resources of Gping-pong are linked to this set by the following mapping3 κ: κ :  κ(start) 7→ 0 κ(ping) 7→ λ(x):(x = 0) κ(pong) 7→ λ(x):(x = 1) Now, let us follow the presented steps to show that the actual resources sequence, s = 0.1.0.1 is validated by the resource pattern r = ’start.ping.pong.ping.pong’. 1. each element of r is replaced by its associated recognition function: f = κ(start).κ(ping).κ(pong).κ(ping).κ(pong) = 0.λ(x) : (x = 0).λ(x) : (x = 1).λ(x) : (x = 0).λ(x) : (x = 1) 2. normalization of the recognition functions sequence: this is done by removing all the zero elements from the sequence of functions: ||f || = λ(x) : (x = 0).λ(x) : (x = 1).λ(x) : (x = 0).λ(x) : (x = 1) 3. apply the sequence of recognition functions to the sequence of the actual resources: ||f ||(s) = λ(x) : (x = 0)(0).λ(x) : (x = 1)(1).λ(x) : (x = 0)(0).λ(x) : (x = 1)(1) = (0 = 0).(1 = 1).(0 = 0).(1 = 1) = >.>.>.> 4. we can conclude that the sequence s have been validated by the pattern r. Furthermore, since the pattern r belongs to r patterns(Gping-pong,start), then the sequence of actual resources s belongs to the valid conversations set defined by this coordination model. 3 The sign = has to be interpreted as the logical comparison. te l-0 01 42 84 3, v er si on 1 23 A pr 2 00 7 4.6 Validating the Conversations According to a Coordination Protocol 109 4.6 Validating the Conversations According to a Coordination Protocol The previous section has presented a formal link between the coordination model and the generated conversations. Still, the presented results show only that this link exists and do not give the means in order to decide if a conversation belongs or not to the valid conversations of a coordination protocol. This section presents a method that uses the Queue Petri Nets (QPN) in order to validate the conversations in an incremental manner. 4.6.1 Informal Presentation of Petri Nets The Petri Net is a well-known formalism, invented by Carl Adam Petri, to express concurrent activities. This formalism has already been used to express directly the conversation protocols in several works (cf. §4.2.4, page 96). First, let us present informally the structure of a Petri Net and give an extension used to validate the conversations. The Petri Net is a digraph composed of transitions (graphically represented as rectangles); places (graphically represented as circles); and tokens. The places contain zero or more tokens. The transitions are linked to places by two kinds of arrows: ˆ production arrow: this arrow is directed from a transition to a place. When a transition is fireable, the tokens are produced in each place linked to this transition by the production arrows. ˆ consumption arrow: this arrow is directed from a transition to a place. A transition is said to be fireable when all the places linked to this transition by a consumption arrow contains at least one token. When the transition is fired, the tokens of consumption places are decremented by one. A more complete and detailed presentation of the Petri Net theory is given in Annex B. 4.6.2 The Queue Petri Net Theory In order to validate the conversations among the agents, we have extended the original Petri Net theory that is presented Annex B to define the Queue Petri Nets (QPN). The QPN is a structure composed of a Petri Net and a queue of interaction objects. To be fireable, the transitions of the QPN do not depend only on the tokens but also on the queue. When a transition is fired, it consumes the tokens and one object found on the front of the queue. The QPN are formally built by following the same steps as for simple Petri Nets. Queue Petri-Net Structure Definition 4.6.1. Let O be the set of interaction objects; O∗ the set of ordered sequences of interaction objects. The Queue Petri Net structure is a six tuple C = {P, T, I, O, F, q}, P = {p1, p2, ..., pn} is a finite set of places n ≥ 0 and T = {t1, t2, ..., tm} is a finite set of transitions m ≥ 0. P ∩ T = ∅. I : T → P∞ is defined as the input function, a mapping from transitions to bags of places and O : T → P∞ is defined as the output function, a mapping from transitions to bags of places; F : T → {0} ∪ (O → B), is defined as the message control te l-0 01 42 84 3, v er si on 1 23 A pr 2 00 7 110 Chapter 4 : Coordinating the Autonomous Agents function, a mapping from transitions to control functions or zero; q ∈ O∗ is defined as the message queue. That is an ordered sequence of elements belonging to O. Queue Petri Net Marking The marking of the QPN is similar to simple Petri Nets (see Annex B). Execution Rules for QPN The execution of the QPN is totally controlled by the tokens distributed over places and the objects of the queue. The QPN executes by firing transitions. For the QPN, two kinds of transitions are identified: 1. transitions that are associated to zero (or null) as control function: these transitions fire as normal Petri Net transitions by consuming the tokens from their input places and producing new tokens distributed over there output places. 2. transitions that are associated to non-null control functions: these transitions are enabled only if there are enough tokens in their input places and the associated control function validates the first object found in the queue. The firing of these transitions consumes the tokens of the input places and produces new tokens in the output places. Besides, the first object of the queue is removed. Definition 4.6.2. Transition tj ∈ T in a marked Queue Petri Net C = (P, T, I, O, F, q) with marking μ is enabled iff: 1. for all pi ∈ P : μ(pi) ≥ #(pi, I(tj)) 2. if F (tj) 6= 0 then: F (tj)(front(q)) = > Definition 4.6.3. Let tj be a fireable trans

i* is a framework for goal-oriented requirements modeling. This framework contains an organizational modeling language defined by actors and dependencies among them. Although i* has been widely used, it is still hard to work with it due to its diversity of variants. In this paper, we propose a unified metamodel that presents elements of two well-known variants of i*. We show how OCL restrictions can be applied this metamodel in order to generate a specific variant. In doing so, we intend to unify the efforts of i* community. This metamodel can be useful when developing generic tools for i*, as well as to reason about i* language. Moreover, it makes it easier to extend the i* language. Thus, our proposal is an attempt towards the unification on i* variants by considering similarities and differences among them.

are different types of information system architectures, but there has not been a structural comparison between the different types. I develop in this paper a structured approach to assess the three main Information System architectures, Enterprise Resource Planning, Service-Oriented Architecture and Multi Agent Systems, based on a structured approach. Furthermore, I provide in this paper an extensive analysis of the key characteristics of the three main IS architectures. In this paper, I develop a systematic and structured approach to pair wise compare and assess these key characteristics of the main architectures based on the IS development framework as developed by Zachman. The aim of this paper are twofold, first, to develop a framework which is suitable for comparing IS architectures and second, comparing the three architectures. The framework I developed is based on the Zachman framework and the views of the original framework are grouped together in due to the complexity of comparing the original framework for the three architectures, this resulted in a new 6x1 framework. The results of the comparison show that there are several key difference between the architectures in the areas of what kind of data is stored, the ability to keep working with legacy systems, the amount of complexity and uncertainty in which these architectures can operate and how the network which corresponds with these architectures is arranged. KeywordsResource Planning (ERP), Service-Oriented Architecture (SOA), Multi Agent Systems (MAS), Information Systems Architecture (ISA), framework, characteristics.

[ES]Hasta el momento no se conocen modelos de tiempo real especificamente desarrollados para su uso en sistemas abiertos, como las Organizaciones Virtuales de Agentes (OVs). Convencionalmente, los modelos de tiempo real se aplican a sistemas cerrados donde todas las variables se conocen a priori. Esta tesis presenta nuevas contribuciones y la novedosa integracion de agentes en tiempo real dentro de OVs. Hasta donde alcanza nuestro conocimiento, este es el primer modelo especificamente disenado para su aplicacion en OVs con restricciones temporales estrictas. Esta tesis proporciona una nueva perspectiva que combina la apertura y dinamicidad necesarias en una OV con las restricciones de tiempo real. Esto es una aspecto complicado ya que el primer paradigma no es estricto, como el propio termino de sistema abierto indica, sin embargo, el segundo paradigma debe cumplir estrictas restricciones. En resumen, el modelo que se presenta permite definir las acciones que una OV debe llevar a cabo con un plazo concreto, considerando los cambios que pueden ocurrir durante la ejecucion de un plan particular. Es una planificacion de tiempo real en una OV. Otra de las principales contribuciones de esta tesis es un modelo para el calculo del tiempo de ejecucion en el peor caso (WCET). La propuesta es un modelo efectivo para calcular el peor escenario cuando un agente desea formar parte de una OV y para ello, debe incluir sus tareas o comportamientos dentro del sistema de tiempo real, es decir, se calcula el WCET de comportamientos emergentes en tiempo de ejecucion. Tambien se incluye una planificacion local para cada nodo de ejecucion basada en el algoritmo FPS y una distribucion de tareas entre los nodos disponibles en el sistema. Para ambos modelos se usan modelos matematicos y estadisticos avanzados para crear un mecanismo adaptable, robusto y eficiente para agentes inteligentes en OVs. El desconocimiento, pese al estudio realizado, de una plataforma para sistemas abiertos que soporte agentes con restricciones de tiempo real y los mecanismos necesarios para el control y la gestion de OVs, es la principal motivacion para el desarrollo de la plataforma de agentes PANGEA+RT. PANGEA+RT es una innovadora plataforma multi-agente que proporciona soporte para la ejecucion de agentes en ambientes de tiempo real. Finalmente, se presenta un caso de estudio donde robots heterogeneos colaboran para realizar tareas de vigilancia. El caso de estudio se ha desarrollado con la plataforma PANGEA+RT donde el modelo propuesto esta integrado. Por tanto al final de la tesis, con este caso de estudio se obtienen los resultados y conclusiones que validan el modelo.

[EN]Component reuse is one of the techniques that most clearly contributes to the evolution of the software industry by providing efficient mechanisms to create quality software. Reuse increases both software reliability, due to the fact that it uses previously tested software components, and development productivity, and leads to a clear reduction in cost. Web services have become are an standard for application development on cloud computing environments and are essential in business process development. These services facilitate a software construction that is relatively fast and efficient, two aspects which can be improved by defining suitable models of reuse. This research work is intended to define a model which contains the construction requirements of new services from service composition. To this end, the composition is based on tested Web services and artificial intelligent tools at our disposal. It is believed that a multi-agent architecture based on virtual organizations is a suitable tool to facilitate the construction of cloud computing environments for business processes from other existing environments, and with help from ontological models as well as tools providing the standard BPEL (Business Process Execution Language). In the context of this proposal, we must generate a new business process from the available services in the platform, starting with the requirement specifications that the process should meet. These specifications will be composed of a semi-free description of requirements to describe the new service. The virtual organizations based on a multi-agent system will manage the tasks requiring intelligent behaviour. This system will analyse the input (textual description of the proposal) in order to deconstruct it into computable functionalities, which will be subsequently treated. Web services (or business processes) stored to be reused have been created from the perspective of SOA architectures and associated with an ontological component, which allows the multi-agent system (based on virtual organizations) to identify the services to complete the reuse process. The proposed model develops a service composition by applying a standard BPEL once the services that will compose the solution business process have been identified. This standard allows us to compose Web services in an easy way and provides the advantage of a direct mapping from Business Process Management Notation diagrams.

Workshop on Agent-based Approaches for the Transportation Modeling and Optimization.- Workshop on Agent-Based Solutions for Manufacturing and Supply Chain.- Workshop on User-Centric Technologies and Applications.- Workshop on Conflict Resolution in Decision Making.- Workshop on Multi-Agent System Based Learning Environments.- Workshop on Multi-agent based Applications for Sustainable Energy Systems.- Workshop on Agents and multi-agent Systems for AAL and e-Health.

With the rapid development of applications in open distributed environments such as eCommerce, privacy of information is becoming a critical issue. Today, many online companies are gathering information and have assembled sophisticated databases that know a great deal about many people, generally without the knowledge of those people. Such information changes hands or ownership as a normal part of eCommerce transactions, or through strategic decisions that often includes the sale of users' information to other firms. The key commercial value of users' personal information derives from the ability of firms to identify consumers and charge them personalized prices for goods and services they have previously used or may wish to use in the future. A look at present-day practices reveals that consumers' profile data is now considered as one of the most valuable assets owned by online businesses. In this thesis, we argue the following: if consumers' private data is such a valuable asset, should they not be entitled to commercially benefit from their asset as well? The scope of this thesis is on developing architecture for privacy payoff as a means of rewarding consumers for sharing their personal information with online businesses. The architecture is a multi-agent system in which several agents employ various requirements for personal information valuation and interaction capabilities that most users cannot do on their own. The agents in the system bear the responsibility of working on behalf of consumers to categorize their personal data objects, report to consumers on online businesses' trustworthiness and reputation, determine the value of their compensation using risk-based financial models, and, finally, negotiate for a payoff value in return for the dissemination of users' information.

With the growing complexity of multi-agent applications and environments in which they are deployed, there is a need for development techniques that would allow for early testing and validation of application design and implementation. This is particularly true in cases where the developed multi-agent application is to be closely integrated with an existing, real-world system of multi-agent nature. Drawing upon our previous experiences with development of complex multi-agent applications, we propose simulation-aided design of multi-agent systems (SADMAS), a methodology tightly integrating simulations of the target system into the MAS application development process. In its heart lies the use of mixed-mode simulation, a simulation where parts of the deployed application operate in the target environment and parts remain simulated. We argue, that employing SADMAS process contributes to reduction of risks involved in development of complex MAS applications, as well as it helps to accelerate the process. Besides describing the capstones of the SADMAS approach and consequences of its application, we also illustrate it's use on a case-study of a next-generation decentralised air traffic management system.

Wireless sensor networks can be a very useful technology for monitoring remote and hostile environments. In this paper, we firstly report on our experience with landslide monitoring, and analyze the issues and the challenges we had to face in programming and deploying a suitable and useful wireless sensor network infrastructure. Following, we discuss how, within the CASCADAS project, we are contributing to the development of a novel component-based framework to facilitate the design and development of autonomic and situation-aware communication services for the use in modern network scenarios. Such a framework can become a useful tool to facilitate the development of easy-to-deploy, robust, and flexible sensor-network-based monitoring systems and, in particular, of landslide monitoring systems.

Wireless communication and increased miniaturisation of computing devices has catalysed the development of so-called mobile applications. Mobile applications are deployed in wireless networks that consist of nodes that can move freely in space. Mobile applications are characterised by (1) their need to take into account their physical environment (usually called context) explicitly; and (2) their need to deal with dynamics and unexpected events originating from their context. This dissertation first proposes two programming abstractions, views and roles, to simplify mobile application development with respect to those two needs. Second, two alternative implementation strategies for middleware that supports the abstractions are designed and evaluated. Third, the abstractions’ usability is evaluated in a real world application, automatic guided vehicle control. The first abstraction, a view, is an automatically up-to-date collection of data objects, that are copies or representations of data objects available on a set of nodes in the network. The middleware automates gathering the data objects from a set of nodes, and maintains the view in the face of dynamically changing availability of the data objects. The second abstraction, a role, encapsulates the behaviour of a component of the application engaging in a protocol. The middleware automates the setup and maintenance of an interaction session between a number of participating components in the mobile network, in the face of a continuously changing number

Wireless Sensor Networks (WSN) are rapidly gaining attention since enabling a broad spectrum of novel applications ranging from vineyard monitoring to space exploration. WSNs may consist of up to thousands of nodes, which typically have limited recourses (computational, memory, energy, etc.) and are deployed in a dynamic environment where they have to continuously adapt to new conditions. Given the resource constrains of typical WSN nodes, the amount of functionality that can be realized at each node is also highly limited. Therefore, cooperation between nodes is needed in order to accomplish more complex tasks. These facts turn design of applications for WSNs into a challenging endeavor. A promising approach how to cope with this is using the emergent self-organization metaphor. Unfortunately most of the current designs using this metaphor are developed employing the ad hoc and trial-and-error method, which is inherently unsystematic, as well as, inefficient. In this paper we present an approach how to systematically design emergent self-organization for massively distributed embedded systems like WSN.

WikiCrimes is a collaborative system where users can store information about crime occurrences in a georeferenced database. Anyone can access WikiCrimes to consult or add information to the database. In this paper we describe the further work done to extend the model and implementation of the WikiCrimes collaborative system as an Open Institutional Environment. The prototype implements an environment where commitments are used to define norms, and together with reputation bring social order to the system.

Whilst the Multi-Agent System (MAS) paradigm has the potential to enable complex heterogeneous information systems to be integrated, there is a need to represent and specify the nature of qualitative conceptual transactions in order that they are adequately comprehended by a goal-directed MAS. Using the Transaction Agent Model (TrAM) approach we examine the use of Conceptual Graphs to model an extension to an existing MAS in the community healthcare domain, whereby the existing agent capabilities are augmented with a robust set of behaviours that provide emergency healthcare management. We illustrate how TrAM serves to enrichen the requirements gathering process, whilst also supporting the definition and realisation of quantitative measures for the management of qualitative transactions.

Whilst several agent-oriented modelling languages have been developed by independent research groups, it is now appropriate to consider a consolidation of these various approaches. There are arguably three things that need consolidation and future standardization: individual symbols, the underpinning metamodel, and the diagram types. Here we address only the third issue by extending an earlier analysis that resulted in recommendations for various diagram types for the modelling of a multiagent system (MAS). Here, we take each of these previously recommended diagram types and see how each is realized in a wide variety (over 20) of current agent-oriented software engineering (AOSE) methodologies. We also take the opportunity to express, as exemplars, some of these diagram types using the recently published FAML notation.

Whilst Multi-Agent System (MAS) architectures appear to offer a more flexible model for designers and developers of complex, collaborative information systems, implementing real-world business processes that can be delegated to autonomous agents is still a relatively difficult task. Although a range of agent tools and toolkits exist, there still remains the need to move the creation of models nearer to code generation, in order that the development path be more rigorous and repeatable. In particular, it is essential that complex organisational process workflows are captured and expressed in a way that MAS can successfully interpret. Using a complex social care system as an exemplar, we describe a technique whereby a business process is captured, expressed, verified and specified in a suitable format for a healthcare MAS.

While today's agent oriented software engineering facilitates the development of complex, distributed systems, fundamental problems remain. One of the problems is that the developer sees only part of the system, or one side of the business, but not the whole context. The Business Process Modeling Notation BPMN is considered a remedy here, but while suitable for modelling some aspects of agenthood, there are others for which BPMN does not go very well. While well-suited for modelling behaviours and especially interactions at a high level of abstraction, BPMN is too cumbersome for modelling every detail of a process and too rigid for modelling dynamic behaviours. Therefore, BPMN has to be embedded into the agent engineering process such that its strengths can be exploited while avoiding its weaknesses. In this paper, BPMN is used mainly for modelling the agents' high-level behaviour and communication, leaving other aspects to established agent programming techniques. The approach is applied for the JIAC multi-agent framework, for which a respective mapping of concepts is given. Finally, the prospects of such an approach are discussed, especially regarding the usage of BPMN for modelling dynamic aspects in agent systems.

While there is ample evidence that Multiagent Systems and Technologies (MAS&T) are vigorous as a research area, it is unclear what practical application impact this research area has accomplished to date. In this chapter, we describe method and results of a survey aiming at a comprehensive and up-to-date overview of deployed examples of MAS&T. We collected and analyzed 152 applications, covering important perspectives, such as ownership, maturity, vertical sectors, and usage of programming languages and agent platforms. We conclude that MAS&T have been successfully deployed in a significant number of applications, though mostly in what could be called niche markets. Off the spotlights of mass markets and current funding buzzwords, it appears that MAS&T is useful for various sectors. Note: This is a preliminary authors’ copy of a paper to appear in the forthcoming book Agent-Oriented Software Engineering edited by Onn Shehory and Arnon Sturm, to be published in the Springer Software Engineering book series, see http://www.springer.com/computer/swe/book/978-3-642-54431-6. Please use the citation information provided on the official Springer site to refer to this paper.