Evaluating the Performance of E-coli with Genetic Learning From Simulated Testing

摘要

This paper addresses the problem of finding the techniques of performance evaluation for elementary agents. From an evolutionary standpoint, the robust navigational algorithms were used by even the simplest of biological systems because the systems were able to learn how to evaluate their performance. The objective of this paper is to study one of the simplest biological, yet intelligent systems, an E. coli cell, and see how this could be of benefit to the design of control strategies for the single-agent intelligent systems. The robot is equipped with sensors and actuators, has a rudimentary knowledge representation system and is capable of conducting search, i.e. is equipped by the means of decision making. The robot itself is looked upon from a two-dimensional perspective and is analyzed in a computer-simulated environment. We present a design of the Variable Structure Controller (VSC) that combines the properties of any two structures or strategies from the ten initially available to our robot. VSC equipped robot should be able to come up with its own strategies of motion, without human intervention. The system under consideration supports the rudimentary learning subsystems that could be envisioned. The idea of using Genetic Programming (GP) is not introduced here for the sake of finding the best controller but rather for the purpose of demonstrating that improved functionality can be achieved via on-line or simulated learning.