Routing Based on Evolved Agents

In this paper we propose adaptive routing algorithms for two dimensional regular grids using intelligent agents. The routing algorithms are evolved by a genetic algorithm optimizing the behavior of the agents whose task is to find the shortest paths from source nodes of the grid to assigned target nodes, e. g., to transport messages. The whole task is also known as multiple target searching. The agents have only a local view of their neighbor nodes, therefore the routing algorithms, that are defined by the behavior of the agents, are local. The multi agent system is modeled as cellular automata. The results show that our technique produces robust algorithms and that the optimal number of cells is between 512-768 for 256 agents.