Dissertation Title: An Insight into Graph-based Optimization Approaches to Robot Navigation and Mapping with Human Autonomy Teaming-based Strategy
When: February 13, 2025
Where: Simrall 228 (in-person) and Microsoft Teams (virtual) – Meeting Link
Candidate: Timothy Sellers
Degree: Doctor of Philosophy in Electrical and Computer Engineering
Committee Members: Dr. John Ball, Dr. Ryan Gree, Dr. Umar Iqball, Dr. Chaomin Luo
Abstract:
The field of autonomous robotics has witnessed significant advancements, with robots increasingly deployed in critical applications such as search and rescue, environmental monitoring, precision agriculture, and defense operations. These systems must navigate highly dynamic and obstacle-dense environments, demanding enhanced adaptability, efficiency, and decision-making capabilities. This research presents a comprehensive integration of graph-based techniques, nature inspired optimization algorithms, and deep learning to address these challenges. By combining methodologies such as adaptive graph traversal, spatial decomposition, and task coordination, the study aims to advance navigation, mapping, and control mechanisms while ensuring safety and reliability in diverse operational scenarios.
A Dynamically Constrained Delaunay Triangulation (D2T) algorithm is developed to enable adaptive real-time navigation in changing environments. The Improved Node Selection Algorithm (iNSA) enhances graph traversal by optimizing path selection in dense obstacle regions. A Human-Autonomy Teaming (HAT) framework combines the Node Optimization Protocol (NOP) and a bio-inspired neural network with an Adaptive Window Strategy (BNN-AWS) to facilitate responsive adjustments based on human input, environmental changes, and critical waypoints for robust navigation in search and rescue scenarios. A middle point cell decomposition model is presented, utilizing vertical cell decomposition and dynamically regulated middle points with an enhanced optimization algorithm to achieve smoother paths and increased computational efficiency. Deep learning augments precision agriculture and multi-robot collaboration, enabling seamless data collection and navigation between aerial and ground vehicles. A Self-Organizing Map (SOM) neural network, combined with advanced formation control strategies, enhances task allocation and coordination in multi-robot systems. This integration ensures efficient spatial organization and reliable performance, particularly in large-scale environmental assessments and remote sensing tasks. Simulation and comparison studies validate the effectiveness and robustness of these methodologies, establishing new benchmarks for autonomous robotic systems in complex, real-world applications. Simulation, comparison studies and on-going experimental results of optimization algorithms applied for autonomous robot systems demonstrate their effectiveness, efficiency and robustness of the proposed methodologies.
