Michael C. Moed

Reducing the search time of a steady state genetic algorithm using the immigration operator

An examination is made of the fundamental trade-off between exploration and exploitation in a genetic algorithm (GA). An immigration operator is introduced that infuses random members into successive GA populations. It is theorized that immigration maintains much of the exploitation of the GA while increasing exploration. To test this theory, a set of functions that often require the GA to perform an excessive number of evaluations to find the global optimum of the function is designed. For These functions, it is shown experimentally that a GA enhanced with immigration (1) reduces the number of trials that require an excessive number of evaluations and (2) decreases the average number of evaluations needed to find the optimum function.<<ETX>>

Robotic Task Planning Using A Connectionist/Symbolic System

An evaluation system called the Associative Rule Memory (ARM) is described that operates with an interactive or automatic planner in a robot-based world. The ARM is constructed from a neural network model called a Boltzmann Machine,and ranks alternative robotic actions based on the probability that the action works as expected in achieving a desired effect. The system is experience-based and can predict the probability of achieving a desired effect for robotic actions that have not been explicitly tested in the past. By providing the ARM with a desired effect, such as a goal of a plan, the ARM will quickly and efficiently find a set of robotic actions that have a high probability of achieving that goal.

Planning under uncertainty in the NASA FTS environment

An evaluation system called the associative rule memory (ARM) that operates with an interactive or automatic planner in a robot-based world, such as the world of the NASA Flight Telerobotic Servicer (FTS), is described. The ARM is constructed from a neural network model called a Boltzmann Machine, and ranks alternative robotic actions based on the probability that the action works as expected in achieving a desired effect. The system is experience-based, and can predict the probability of achieving a desired effect for robotic actions that have not been explicitly tested in the past. The ARM is designed to quickly and efficiently find high probability of effect for robotic actions for a given desired effect. This paper details the construction of the ARM for the NASA FTS robotic environment. Examples are also provided that demonstrate the use of the ARM within a current NASA symbolic planning system.