1 Multi-Agent Systems and Simulation: A Survey from the Agent Community’s Perspective
Fabien Michel, Jacques Ferber, and Alexis Drogoul
Introduction • M&S for MAS: The DAI Case • MAS for M&S: Building Artificial Laboratories • Simulating MAS: Basic Principles • The Zeigler’s Framework for Modeling and Simulation • Studying MAS Simulations Using the Framework for M&S • Conclusion
2 Multi-Agent Systems and Simulation: A Survey from an Application Perspective
Klaus G. Troitzsch
Simulation in the Sciences of Complex Systems • Predecessors and Alternatives • Unfolding, Nesting, Coping with Complexity • Issues for Future Research: The Emergence of Communication
3 Simulation Engines for Multi-Agent Systems
Georgios K. Theodoropoulos, Rob Minson, Roland Ewald, and Michael Lees
Introduction • Multi-Agent System Architectures • Discrete Event Simulation Engines for MAS • Parallel Simulation Engines for MAS • Issues for Future Research
Simulation studies have accompanied the development of multi-agent systems from the beginning. Simulation has been used to understand the interaction among agents and between agents and their dynamic environment. The focus has been on test beds, and the description and integration of agents in dynamic virtual environments. Micro and individual-based simulation approaches also became aware of the new possibilities that the agent metaphor and the corresponding methods offer. The area of social simulation played a key role, being enriched and equally challenged by more detailed models of individuals and contributing itself to a better understanding of the effects of cooperation and coordination strategies in multi-agent environments. During the first decade the focus of research has been on the modeling layer. Only gradually, the need to take a closer look at simulation took hold, e.g., how to ensure efficient and repeatable simulation runs.
The chapter by Fabien Michel, Jacque Ferber, and Alexis Drougol on “Multi-agent systems and simulation: A survey from the agent community’s perspective” gives a historical overview of the methodological developments at the interface between multi-agent systems and simulation from an agent’s perspective. The role of test-beds in understanding and analyzing multi-agent systems in the 1980s, the development of abstract agent models, the role of social simulation in promoting research in multi-agent systems and simulation, and the challenges of describing agents and their interactions shape the first decade of research. With the environment of agents becoming an active player, the questions about timing move into focus and with them traditional problems of simulator design, e.g., how to handle concurrent events. For in-depth analysis simulation questions like validity of models, design and evaluation of (stochastic) simulation experiments need to be answered, but also new one emerge in the context of virtual, augmented environments.
The significant impact of social science on multi-agents research is reflected in the realm of simulation. In the chapter “Multi-Agent Systems and Simulation: A Survey from an Application Perspective,” Klaus Troitzsch traces the first simple agent-based models back to the 1960s. Particularly, analyzing the micro and macro link of social systems, i.e., the process of human actions being (co–) determined by their social environment and at the same time influencing this social environment, permeates agent-based simulation approaches from the beginning, despite the diversity of approaches which manifests itself in varying level of details, number of agents, interaction patterns (e.g., direct or in-direct via the environment), and simulation approach. The aim of these simulation studies is to support or falsify theories about social systems. However, in doing so, they also reveal mechanisms that help to ensure certain desirable properties in a community of autonomous interacting entities and as such can be exploited for the design of software agent communities as proposed by the “socionics” initiative.
A long neglected area of research has been the question of how to execute multi-agent models in an efficient and correct manner. This question is addressed in the chapter by Georgios Theodoropolous, Rob Minson, Roland Ewald, and Michael Lees on “Simulation Engines for Multi-Agent Systems”. Often agent implementations were translated into discrete stepwise “simulation” with no explicit notion of simulation time. However, the need to associate arbitrary time with the behavior of agents and synchronize the behavior of agents with the dynamics of the environment led to discrete event simulation approaches. As the simulation of multiple heavy weight agents require significant computation effort, sequential discrete event simulators are complemented by parallel discrete ones and help an efficient simulation of multi-agent systems. Interestingly, in the opposite direction we find the agent approach exploited to support the distributed simulation of latency simulation systems. Simulation systems are interpreted as agents and the problem of interoperability and synchronization of these simulation systems is translated into terms of communication and coordination.
