Date of Award
Thesis - Open Access
Master of Science in Unmanned and Autonomous Systems Engineering
Electrical, Computer, Software, and Systems Engineering
Dr. Keith Garfield
First Committee Member
Dr. Richard Stansbury
Second Committee Member
Dr. William Barott
Goals of this research were to develop a conceptual algorithm that can optimize execution time for generating a solution set and demonstrate that a solution set of one sub-problem can be applied to another sub-problem within the same problem set. To achieve the proposed goals, GloPro was developed to generate rule sets for different sub-problems within a problem set, as well as identifying which rule sets are to be utilized for a given instance of the problem. The algorithm was to be robust, as to be applicable to a wide array of problems without radical re-design per problem. This idea was fueled by the concept of Structure-Mapping Theory, where a set of knowledge is mapped from one domain to another based on the shared baseline characteristics. Utilizing a Genetic Algorithm (GA), plus A* with a classifier hybrid, the algorithm includes a period of supervised learning followed by execution in an operational environment. Progressive learning occurred through application of the algorithm to multiple sub-problems, each having unique characteristics. The algorithm was applied to a simulated robotic agent in a maze environment as a proxy for other problems. This problem is well known, but still an active problem in the field of robotics. The experimental results indicate that the hybrid GA with A* technique is feasible, and that progressive learning is enhanced through application of previous learning results to a period of learning. In addition, the evolved solutions were unique to the sub-problems, indicating that this technique can be used to develop robust solutions across sub-problems.
Scholarly Commons Citation
Ghelarducci, Leo Austin, "Robust Optimization and Groundwork for Problem Mapping" (2017). PhD Dissertations and Master's Theses. 328.