College of Graduate Studies: Theses & Dissertations

Term of Award

Spring 2026

Degree Name

Master of Science in Computer Science (M.S.)

Document Type and Release Option

Thesis (restricted to Georgia Southern)

Copyright Statement / License for Reuse

Digital Commons@Georgia Southern License

Department

Department of Computer Science

Committee Chair

Ryan Florin

Committee Member 1

Andrew Allen

Committee Member 2

Vijayalakshmi Ramasamy

Abstract

Modern robotic warehouse systems operate in dynamic and disruption-prone environments where maintaining connectivity is essential for operational continuity. While Digital Twin (DT) technology has emerged as a promising paradigm for monitoring and analysis of cyber-physical systems, many existing implementations remain limited to passive observation and lack integrated decision-support capabilities. In particular, selecting effective bypass actions under structural disruptions, where multiple feasible alternatives exist, remains insufficiently addressed. This thesis proposes an integrated Digital Twin-oriented framework for adaptive decision-making in robotic warehouse environments, focusing on restoring connectivity under path node disruptions. The warehouse is modeled as a fully connected directed graph, where obstacles are represented as node failures that break connectivity. The objective is to restore rack connectivity by converting a minimal set of rack nodes into path nodes under structural constraints. The proposed methodology consists of three integrated components. First, a deterministic bypass generation algorithm constructs feasible recovery candidates based on structural graph analysis, ensuring all solutions satisfy connectivity and directional constraints. Second, a simulation framework evaluates these candidates under controlled and repeatable conditions, using metrics such as coverage, path length, spatial deviation, and rack conversion cost to capture operational trade-offs. Third, a Learning-to-Rank (LtR) model prioritizes candidates by learning preference patterns from simulation-generated data, enabling data-driven decision-making without relying on fixed heuristic rules. The framework is mapped onto the MAPE-K (Monitor–Analyze–Plan–Execute–Knowledge) architecture, where algorithmic generation corresponds to analysis; simulation supports planning through what-if evaluation, and ranking refines decision selection. Experimental evaluation using simulation-based testing demonstrates that the learning-assisted approach improves decision quality compared to baseline strategies by effectively capturing multi-objective trade-offs. Overall, this work presents a unified framework that integrates algorithmic planning, simulation-based evaluation, and learning-assisted ranking within a Digital Twin paradigm. The proposed approach advances Digital Twins from passive monitoring systems toward adaptive decision-support systems capable of prioritizing alternative actions under dynamic conditions.

Research Data and Supplementary Material

No

Download

Available for download on Wednesday, April 16, 2031

Share

COinS