Alexander Helleboogh

Modeling dynamic environments in multi-agent simulation

Real environments in which agents operate are inherently dynamic—the environment changes beyond the agents’ control. We advocate that, for multi-agent simulation, this dynamism must be modeled explicitly as part of the simulated environment, preferably using concepts and constructs that relate to the real world. In this paper, we describe such concepts and constructs, and we provide a formal framework to unambiguously specify their relations and meaning. We apply the formal framework to model a dynamic RoboCup Soccer environment and elaborate on how the framework poses new challenges for exploring the modeling of environments in multi-agent simulation.

The MACODO middleware for context-driven dynamic agent organizations

One of the major challenges in engineering distributed multiagent systems is the coordination necessary to align the behavior of different agents. Decentralization of control implies a style of coordination in which the agents cooperate as peers with respect to each other and no agent has global control over the system, or global knowledge about the system. The dynamic interactions and collaborations among agents are usually structured and managed by means of roles and organizations. In existing approaches agents typically have a dual responsibility: on the one hand playing roles within the organization, on the other hand managing the life-cycle of the organization itself, for example, setting up the organization and managing organization dynamics. Engineering realistic multiagent systems in which agents encapsulate this dual responsibility is a complex task. In this article, we present a middleware for context-driven dynamic agent organizations. The middleware is part of an integrated approach, called MACODO: Middleware Architecture for COntext-driven Dynamic agent Organizations. The complementary part of the MACODO approach is an organization model that defines abstractions to support application developers in describing dynamic organizations, as described in Weyns et al. [2010]. The MACODO middleware offers the life-cycle management of dynamic organizations as a reusable service separated from the agents, which makes it easier to understand, design, and manage dynamic organizations in multiagent systems. We give a detailed description of the software architecture of the MADOCO middleware. The software architecture describes the essential building blocks of a distributed middleware platform that supports the MACODO organization model. We used the middleware architecture to develop a prototype middleware platform for a traffic monitoring application. We evaluate the MACODO middeware architecture by assessing the adaptability, scalability, and robustness of the prototype platform.

The MACODO organization model for context-driven dynamic agent organizations

Todays distributed applications such as sensor networks, mobile multimedia applications, and intelligent transportation systems pose huge engineering challenges. Such systems often comprise different components that interact with each other as peers, as such forming a decentralized system. The system components and collaborations change over time, often in unanticipated ways. Multiagent systems belong to a class of decentralized systems that are known for realizing qualities such as adaptability, robustness, and scalability in such environments. A typical way to structure and manage interactions among agents is by means of organizations. Existing approaches usually endow agents with a dual responsibility: on the one hand agents have to play roles providing the associated functionality in the organization, on the other hand agents are responsible for setting up organizations and managing organization dynamics. Engineering realistic multiagent systems in which agents encapsulate this dual responsibility is a complex task. In this article, we present an organization model for context-driven dynamic agent organizations. The model defines abstractions that support application developers to describe dynamic organizations. The organization model is part of an integrated approach, called MACODO: Middleware Architecture for COntext-driven Dynamic agent Organizations. The complementary part of the MACODO approach is a middleware platform that supports the distributed execution of dynamic organizations specified using the abstractions, as described in Weyns et al. [2009]. In the model, the life-cycle management of dynamic organizations is separated from the agents: organizations are first-class citizens, and their dynamics are governed by laws. The laws specify how changes in the system (e.g., an agent joins an organization) and changes in the context (e.g., information observed in the environment) lead to dynamic reorganizations. As such, the model makes it easier to understand and specify dynamic organizations in multiagent systems, and promotes reusing the life-cycle management of dynamic organizations. The organization model is formally described to specify the semantics of the abstractions, and ensure its type safety. We apply the organization model to specify dynamic organizations for a traffic monitoring application.

How to get multi-agent systems accepted in industry?

With many researchers in the Multi-Agent System (MAS) community, we share the opinion that too much of the quality and relevant research in the area of MAS is underrepresented in the development of complex distributed systems in practice today. In our experience, a Babylonic mismatch is a crucial factor in this fact – research on MAS profiles itself as an isolated community and, as such, may create artificial thresholds to convince mainstream software developers of its merits. We argue that integrating the concepts and techniques from agent-based software engineering within mainstream software engineering provides opportunities to amplify the industrial adoption of MAS. To ground this position, we discuss MAS engineering from the perspective of the software engineering area that we are most familiar with: software architecture.

An Architectural Approach to Support Online Updates of Software Product Lines

Despite the successes of software product lines (SPL), managing the evolution of a SPL remains difficult and error-prone. Our focus of evolution is on the concrete tasks integrators have to perform to update deployed SPL products, in particular products that require runtime updates with minimal interruption. The complexity of updating a deployed SPL product is caused by multiple interdependent concerns, including variability, traceability, versioning, availability, and correctness. Existing approaches typically focus on particular concerns while making abstraction of others, thus offering only partial solutions. An integrated approach that takes into account the different stakeholder concerns is lacking. In this paper, we present an architectural approach for updating SPL products that supports multiple concerns. The approach comprises of two complementary parts: (1) an update viewpoint that defines the conventions for constructing and using architecture views to deal with multiple update concerns, and (2) a supporting framework that provides an extensible infrastructure supporting integrators of a SPL. We evaluated the approach for an industrial SPL for logistic systems providing empirical evidence for its benefits and recommendations.

The agent environment in multi-agent systems: A middleware perspective

Interaction is at the core of multi-agent systems. We use agent environment as a general term to denote the medium for agent interaction. Over the last years, the agent environment has been subject of active research. In this paper, we reflect on the role of the agent environment in multi-agent systems from a middleware perspective. Our study yields the following observations: (1) multi-agent system engineers consider distributed middleware (RMI, CORBA, etc.) as the basic platform for developing multi-agent systems, (2) common middleware services (security, persistency, etc.) are only minimally considered in multi-agent systems, (3) domain-specific middleware for multi-agent systems such as communication services and support for stigmergic coordination are typically developed as stand-alone services and as such difficult to compose with other services. From these observations, we derive a number of challenges for research on environments in multi-agent systems: (1) to amplify reuse, application-specific services should be further consolidated into domain-specific services, (2) the problem of integration must be tackled, i.e. horizontal integration among domain-specific services for multi-agent systems, and vertical integration of domain-specific services upwards with the agents, and downwards with the common middleware services and the underlying distributed platform, (3) to support dynamic changing requirements of the system at hand, flexible composition and dynamic adaptation of services must be supported by the agent environment.

Characterizing Relations between Architectural Views

It is commonly agreed that an architectural description (AD) consists of multiple views. Each view describes the architecture from the perspective of particular stakeholder concerns. Although views are constructed separately, they are related as they describe the same system. A thorough study of the literature reveals that research on relations between views is fragmented and that a comprehensive study is hampered by an absence of common terminology. This has become apparent in the discussion on inter-view relational concepts in the revision of IEEE 1471 as ISO/IEC 42010 (Systems and Software Engineering -- Architectural Description). This paper puts forward a framework that employs a consistent terminology to characterize relations between views. The framework sheds light on the usage, scope and mechanisms for relations, and is illustrated using several representative approaches from the literature. We conclude with a reflection on whether the revision of ISO 42010 aligns with our findings.

The Packet-World: A Test Bed for Investigating Situated Multi-Agent Systems

Research on situated multi-agent systems investigates how to model a distributed application as a set of cooperating autonomous entities (agents) which are situated in an environment. Many fundamental issues remain unrevealed in this research area. A profound understanding of these issues, however, is necessary before situated multi-agent systems can be applied to industry-strength applications. We use the abstract application called the Packet-World quite extensively as a test bed for investigating, experimenting and evaluating fundamental concepts and mechanisms. Examples are active perception, decision making of situated agents, synchronization of simultaneous actions and indirect coordination. The Packet-World has direct connections with real-world applications, such as the decentralized control of a warehouse transportation system through unmanned vehicles. In this article, we describe the Packet-World and we give an overview of our research for which we have used the Packet-World as a test bed.

Using code analysis tools for architectural conformance checking

Architectural conformance checking verifies whether a system conforms to its intended architecture, which is essential to safeguard the quality attributes of the system. Due to the size of many systems, performing conformance checking by means of manual code inspections is often practically infeasible. Code analysis tools can be used to automatically check architectural conformance. In this paper, we investigate several code analysis tools that offer support for Java and compare them on their usefulness for architectural conformance checking: Architecture Rules, Macker, Lattix DSM, SonarJ, Structure101 and XDepend.

Managing Agent Interactions with Context-Driven Dynamic Organizations

Organizations are at the heart of multi-agent systems. To deal with the ongoing dynamics and changes in the system, organizations have to adapt. Typically, agents are responsible to deal with the complexity of organization dynamics. In this paper, we present an approach for context-driven dynamic organizations in which the agent environment takes the burden of managing organization dynamics. Driven by the context, the agent environment manages the evolution of organizations and actively advertises roles to the agents, supporting the necessary collaborations between agents needed in the current context. We introduce a conceptual model for context-driven dynamic organizations and present a software architecture that supports the model in a distributed setting. The proposed approach separates the management of dynamic evolution of organizations from the actual functionality provided by the agents playing roles in the organizations. Separating these concerns makes it easier to understand, design, and manage organizations in multi-agent systems. We show how we have applied context-driven dynamic organizations in a concrete case of monitoring traffic jams. In this case, camera agents associated with traffic monitoring cameras collaborate in organizations. Depending on the context, camera agents play different roles, with responsibilities ranging from simple measurement to data aggregation. When a traffic jam covers the viewing range of multiple cameras, organizations are dynamically merged, assuring cameras detecting the same traffic jam can collaborate. Vice versa, when a traffic jam dissolves, the organization is dynamically split up. Test results indicate that context-based dynamic organizations is a promising approach to support decentralized traffic monitoring.