Evolving 6-State Automata for Optimal Behaviors of Creatures Compared to Exhaustive Search

摘要

We applied an Island Model Genetic Algorithm (GA) to a Multi-Agent System (MAS) modeled in Cellular Automata (CA) in order to find the optimal behavior of the agents. The agents' task is to visit all free cells in a cellular grid containing obstacles as fast as possible. For this investigation we used a previously defined set of five different environments. The agents are controlled by a finite state machine with a restricted number of states and outputs (actions of the agents). Finite state machines with 4 to 7 states have been evolved by the GA. We compared the effectiveness (quality of solutions) and efficiency of the GA to an exhaustive search of all possible solutions. A special hardware (FPGA logic) has been used to enumerate and evaluate all 6-state finite state machines. The results show that the GA is much faster but almost as effective as the exhaustive search.