Simulating a competition for foods between ant colonies as a coordinated model of autonomous agents
M. Kubo, Yukinori Kakazu · 2002
In this study, a simple and powerful methodology for realizing a distributed and autonomous system is proposed. It is inspired by the ecological nature of ant colonies. Under the set assumption that ants do not use one to one communications for problems very much, the proposed methodology improves the colony's total activity through a learning process. An artificially self-organizing system requires such a learning property, therefore each ant is regarded as an autonomous agent composed of a stochastic learning automaton (SLA). Because of the stochastic trial-and-error method introduced by SLA, the model is expected to exhibit good performances through learning without using one to one communications; an agent can decide on an action suitable to the system to which it belongs. To observe the ability and the limitations of this model a typical coordinated motion planning problem observed in ant society-a dynamical resource allocation problem was taken for the authors' experiment. The results shown were as expected.>