A Robust System for Testing Evolutionary Agent Interaction in Multiple Environments.

Author: Amber Fischer


This research presents a method for the creation of a robust system for testing evolutionary agents in multiple environments.  The architecture consists of several interacting modules, the Genetic Algorithm (GA) module, the GA-to-Agent Mapping module, the Agent module, and the Environment module.  The Genetic Algorithm module provides the user interface point, and controls the execution of all other modules.  Designed to be problem independent, the GA parameters can be user-defined.  The GA-to-Agent Mapping module contains the mapping details needed to create a working agent, such as a neural network, from the genetic code of a single chromosome.  The Agent module, created from the GA-to-Agent module, can interact dynamically with the Environment module, returning to the main Genetic Algorithm module a fitness evaluation for the given Agent (chromosome). The system is designed for an investigation of Evolutionary Neural Network mapping techniques applied in multiple dynamic environments.  The multiple environments will indicate the robustness of the evolutionary neural network agent, looking for specific intelligent behavior; such as pattern matching and classification abilities, scalability, prediction ability, control, and decision making.

Diagram of System:




Last updated: 23 September, 2001 07:10:02 PM
Please refer all questions about the site to
Copyright © 2001 UMR's SESL. All Rights reserved.                                 SESL Home