Dynamics of multi-agent systems by RasID learning

Presents a model and learning method for controlling multi-agent systems that are composed of plural agents. Generally speaking multi-agent systems are recognized as the systems where the group behavior of many autonomous subjects or individuals is different from the one obtained by just accumulating the behavior of each agent. In the paper, the dynamic of each agent is described by the equation including parameters for controlling the interaction of each agent, and these parameters are trained to meet the specification of the multi-agent dynamics. The dynamics of the agents can be represented by the distance between the agents and the following or dispersal parameters that express the interaction among the agents. These parameters are calculated by fuzzy inference whose antecedent part is the function of the distance between the agents. By training means and variances of the membership function of the antecedent part and consequent part, we can obtain the required dynamics of the agents. The criterion of this learning is a maximum Lyapunov exponent because of it´s suitability for describing stable or chaotic movements