Task oriented functional self-organization of mobile agents team: Memory optimization based on correlation feature

Abstract

We developed a new optimization algorithm for multiagent coordination based on indirect and unsupervised communication. The mobile agents team task is simply searching and collecting "food items". The global coherent behavior is emergent, meaning that despite the fact that agents have no global map of the environment and do not directly communicate with each other they coordinate their behavior to achieve a global "goal". The coordinated response of the agents is the result of indirect communication via local changes in the environment. Each agent records the encountered objects in a memory register and by appropriate weighting of local perception the agent tries to estimate the global spatial distributions of the objects in the environment. The range of spatial and temporal indirect coupling among the agents is controlled via a "memory radius". We developed an optimized an algorithm that adapts the "memory radius" according to environment changes to minimize the computational time required to achieve the "goal" (piling the objects of the same kind together). Our optimization procedure is based on the correlation feature of the emergent pattern. The maximum speed of feature decreases leads to an optimized dependence of the "memory radius" on simulation time step. We derived also an analytic relationship between the "memory radius" and the time step based on the intermediate steady-state assumption. Numerical simulations confirmed that our analytic relationship coincides with the numerical optimization criterion based on the correlation feature. © 2004 Springer-Verlag.