Optimal Resource Allocation of Communicating Multi-Agent System Using Genetic Algorithm

Abstract

The artificial ant problem [1], [2] describes ants searching for food pellets on a grid using limited knowledge of the local environment. We generalize this model by means of a multi-agent system of communicating ants with intelligence evolved from genetic algorithm. The objective is to find the most food pellets with given energy constraint. A smart ant can ignore the broadcast if it has already collected plenty of food locally, but has received few broadcasts from its teammates lately. On the other hand, if an ant cannot find any food locally, yet some of its teammates are sending out a lot of food broadcast elsewhere, then it may be wise to follow the broadcast and escape the current no-food region. We model this decision strategy on the response to broadcast using genetic algorithm and the result shows that the performance of multiple-ant team in fixed-total-energy search is improved. Since total energy consumed by the team of ants is constant, the number of steps per ant used will be smaller for team with more member, we find that there exists optimal number of team members from simulation. The result depends on both the resource allocated to the team and the food distribution. We distribute food uniformly over an annulus of radius r at the rim of a disk with a bigger radius R, where the ants start their search in the center of the disk. This food distribution provides both a control on the average food density, and a density gradient, while avoiding anisotropic food distribution. This provides a first step to model general food distribution for real application.