An Environment-Based Methodology to Design Reactive Multi-agent Systems for Problem Solving
Lecture Notes in Computer Science
Even if the multi-agent paradigm has been evolving for fifteen years, the development of concrete methods for problem solving remains a major challenge. This paper focuses on reactive multi-agent systems because they provide interesting properties such as adaptability and robustness. In particular, the role of the environment, which is effectively where the system computes and communicates, is studied. From this analysis a methodology to design or engineer reactive systems is introduced. Our
... roach is based on the representation of the problem's constraints considered as perturbations to stabilize. Agents are then defined, in the second place, as a means of regulating the perturbations. Finally, the relevancy of our proposition is justified through the development of two solving models applied to real and complex problems. the system computes, builds and communicates. In the problem-solving framework, it is clear that one reactive agent can neither handle a representation of the problem nor compute its solution. The resolution is obtained from numerous agent-agent and agent-environment interactions    . Agent interactions are reactions to perceptions, they participate directly in the solving processes, but they do not provide a means to express the problem. So, the representation of the problem can only be defined through the environment model. In this paper we re-examine the role played by each element in collective systems, by focusing on the environment. This work is motivated by the necessity of clarifying the common points used in different environment-based techniques and reactive agent-based MAS. Thus we present a synthetic view on reactive systems by considering existing collective solving systems such as the pheromone-based approach and the eco-resolution model. This analysis allows us to propose a methodology aimed at building environment-based solving systems. The proposed methodology establishes the link between the representation of the problem, expressed as environmental constraints, and agent behaviors, which are regulation items of the environmental perturbations. This method contrasts with classical approaches that involve defining agents and interactions by following the expected organization (as proposed in  ). In our case, agents are defined in the second place, and build as regulation processes depending on the problem model. The environment is clearly defined as a first-class entity of the multiagent system ( as shown the importance of such an approach in multiagent conception). The paper is structured as follows. Sect.2 presents a re-examination of reactive MAS from an automatic control point of view and classical collective models are analyzed. In Sect.3 the four main points of the methodology are introduced, first with a general point of view and then in detail considering a concrete use. This section ends with a comparison to related work. Section 4 illustrates the methodology through two examples of applications: the satisfaction-altruism model for decentralized cooperation between situated agents and a Physics based model for localization and target tracking. Finally, in Sect.5, we conclude on the proposed methodology and present some future work.