Publication Date
2012
Document Type
Thesis
Committee Members
John Gallagher (Advisor), Andrew Hsu (Other), Michael Raymer (Committee Member), Mateen Rizki (Committee Member)
Degree Name
Master of Science (MS)
Abstract
Wing and airframe damage to insect scale micro air vehicles potentially cause significant losses in pose and position control precision. Although one can imagine many possible means of adapting the flight controllers to restore precise pose and position control, severe limits on computational resources available on-board an insect sized vehicle render many of them impractical. Additionally, limits on sensory capability degrade any such vehicle's ability to critique its own performance. Any adaptive solutions one would propose to recover flight trajectory precision, therefore, would require a resource light implementation, preferably without need for relatively expensive floating-point operations, along with the capability to assess control quality via relatively infrequent and possibly imprecise point estimates of vehicle pose and position.
This thesis will expand on previous work that employed an adaptive oscillator as a component of an altitude controller inside a simulated insect-scale flapping-wing micro air vehicle. In that work, it was demonstrated that an adaptive oscillator could learn novel wingbeat patterns unique to the capability of specific, possibly damaged, wings. These wingbeat patterns would restore the relationships between control outputs and wing motion; and thus; restore correct whole-vehicle action. In that earlier work, the core of the adaptive oscillator was an evolutionary algorithm (EA) that operated as a mutation driven stochastic hill climber. In this work, we explore the use of and the potential benefits of an EA variant that operates as a crossover driven schema/hyperplane sampler. For this work we selected the Compact Genetic Algorithm (cGA), as it possess an efficient hardware-level implementation and is clearly a crossover-driven hyperplane sampler. The thesis will present experimental results from which one may assess the relative performance of this style of genetic search as well as speculate upon its potential utility for more complex flight control problems.
Page Count
63
Department or Program
Department of Computer Science
Year Degree Awarded
2012
Copyright
Copyright 2012, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.
