Catálogo de publicaciones - libros

The emerging “Computational Grid” infrastructure poses many new opportunities for the developing science of large scale multi-agent simulation. The ability to migrate agent experiments seamlessly from simple, local single-processor development tools to large-scale distributed simulation environments provides valuable new models for experimentation and software engineering: first develop local, flexible prototypes, then as they become more stable progressively deploy and experiment with them at larger scales. Currently this kind of progressive scalability is hard for both practical and theoretical reasons: Practically, most agent platforms are designed for just one environment of operation. Smooth scalability is more than a matter of increasing agent numbers. Smooth scaling requires clear integration and consistent alignment between a variety of MAS system and simulation architectures and differing underlying infrastructures. This paper reports on recent progress with our experimental platform MACE3J, which now simulates MAS models seamlessly across a variety of scales and architecture types, from single PCs, to Single System Image (SSI) multicomputers, to heterogeneous distributed Grid environments.

Multi-Agent Systems (MASs) provide a valuable tool for handling increasing software complexity and supporting rapid and accurate decision making. Various environments for testing, analyzing and developing MASs have been developed. This paper describes an approach to integrating agents into distributed simulations. Using the JADE toolkit and the HLA (High Level Architecture), a general architecture is obtained, where both the high level agent specific services and the underlying middleware comply with international standards. In this paper, we show how an MAS may be used to represent entities in a simulation, focusing on the issue of agent to agent communication, as this is one of the key characteristics of MASs. The causality problem in agent communication is described, and conditions for ensuring consistency are identified. A prototype system has been implemented to demonstrate the feasibility of our solution and some experimental results are presented.

The efficient simulation of multi-agent systems presents particular challenges which are not addressed by current parallel discrete event simulation (PDES) models and techniques. While the modelling and simulation of agents, at least at a coarse grain, is relatively straightforward, it is harder to apply PDES approaches to the simulation of the agents’ environment. In conventional PDES approaches a system is modelled as a set of logical processes (LPs). Each LP maintains its own portion of the state of the simulation and interacts with a small number of other LPs. The interaction between the LPs is assumed to be known in advance and does not change during the simulation. In contrast, the environment of a MAS is read and updated by agent and environment LPs in ways which depend on the evolution of the simulation. As a result, MAS simulations typically have a large state which is not associated with any particular agent or environment LP. In [1] we proposed a new approach to the distributed simulation of MAS in which the shared state is maintained by a tree of additional logical processes called Communication Logical Processes (CLP). In this paper we refine this model by giving precise definitions of a set of operations which allow agent and environment LPs to interact with the shared state and briefly outline how these operations could be implemented by a CLP.

Time management is essential when simulating multi-agent systems (MASs) as it allows consistent and repeatable simulation runs. So far, time management lacks support to express the timing requirements of a simulation explicitly and at an abstraction level appropriate for MAS developers. Moreover, integrating time management into a MAS requires the developer to alter the design of the MAS. In this paper, we first propose to capture timing requirements that reflect the semantics of MAS activities in an explicit model. Second, we present a time management infrastructure that starts from a semantic duration model description to integrate all time management functionality into a MAS transparently, i.e. without requiring the developer to alter the design of the MAS. We use aspect-oriented programming technology as it allows , a crucial software engineering requirement. As a case, we apply our approach to the Packet-World.

Here we present AKIRA, a framework for Agent-based cognitive and social simulations. AKIRA is an open-source project, currently developed mainly at ISTC-CNR, that exploits state-of-the-art techniques and tools. It gives to the programmer a number of facilities for building Agents at different levels of complexity (e.g. reactive, deliberative, layered). Here we describe the main architectural features (i.e. Hybridism of the Agents and the Energy Model) and the theoretical assumptions that motivate it. We also present some simulations.

This paper presents a multi-agent based simulation framework for cognitive systems engineering of socio-technical systems. Comprehensive design analysis of socio-technical systems requires modeling of various aspects of the work environment and of behavior and performance of humans. This framework provides a distinct focus on the work-environment, specifying it as a coherent collection of declarative models spanning its multiple aspects. Compared to traditional methods in cognitive systems engineering, aggregating multiple aspects allows greater detail and scale in modeling socio-technical systems. This also addresses design issues that cut across individual aspects, thus enabling a comprehensive what-if analysis of the system. Humans are computationally modeled as proactive and interactive agents operating within their work-environment. Emergent behavior of the system in response to design changes in both humans and their work environment can thus be simulated. The framework is illustrated through an example in air traffic control. The framework can be applied to problems in enterprise re-engineering, organizational structuring, etc.

For an integrated simulation such as the natural environment affected by human society, it is indispensable to provide an integrated simulator that incorporates multiple computational models. We proposed a multi-layer socio-environmental simulation by layering the social interaction scenario on environmental simulation. For this simulation, we connect two different systems. One is a scenario description language , which is suitable for describing social interactions. Another is CORMAS, which models interactions between a natural environment and humans. The key idea is to realize a mapping between agents in different systems. This integration becomes possible by the salient feature of : users can write scenarios for controlling legacy agents in other systems. Moreover, we find that controlling the flow of information between the two systems can create various types of simulations. We also confirm the capability of CORMAS/, in the well-known Fire-Fighter domain.

Several tag models with intriguing properties have been advanced recently. But currently there is little detailed understanding of the underlying processes. Specifically it is not know what (if any) are the necessary conditions for tag systems to produce high levels of cooperation. We identify, for the first time, what appears to be a necessary condition that previous tag models contained. It appears that, in general, for tag-based systems to support high levels of cooperation tags must mutate faster than strategies because cooperative tag groups need to spread (by mutation of tags) before free riders (by mutation on strategies) invade the group. We test this theory with simulation.

High performance computing clusters have been a critical resource for computational science for over a decade and have more recently become integral to large-scale industrial analysis. Despite their well-specified components, the aggregate behavior of clusters is poorly understood. The difficulties arise from complicated interactions between cluster components during operation. These interactions have been studied by many researchers, some of whom have identified the need for holistic multi-scale modeling that simultaneously includes network level, operating system level, process level, and user level behaviors. Each of these levels presents its own modeling challenges, but the user level is the most complex due to the adaptability of human beings. In this vein, there are several major user modeling goals, namely descriptive modeling, predictive modeling and automated weakness discovery. This study shows how multi-agent techniques were used to simulate a large-scale computing cluster at each of these levels.

Organizations depend on regular meetings to carry out their everyday tasks. When carried out successfully, meetings offer a common medium for participants to exchange ideas and make decisions. However, many meetings suffer from unfocused discussions or irrelevant dialogues. Within Social Science sometimes general, informal meeting guidelines are formulated. To study meetings in detail, we first formalize general properties for meetings and a generic meeting protocol for the role interactions in meetings that is coherent with such guidelines. In the context of a case study, an example meeting is simulated based on this protocol. The properties are verified in this simulated trace. These properties are also validated by verifying them against a formalisation of empirical data of a real meeting in the same context. A comparison of the two traces reveals that a real meeting is more robust in the sense by exception violations of the protocol may occur, and these exceptions are handled effectively without damaging the success of the meeting. Given this observation, a more refined protocol is specified that includes exception-handling strategies. Based on this refined protocol a meeting is simulated that closely resembles the real meeting.