Mechanical Engineering Commons^™

Data-Driven Predictive Modeling To Enhance Search Efficiency Of Glowworm-Inspired Robotic Swarms In Multiple Emission Source Localization Tasks, Payal Nandi

Mechanical & Aerospace Engineering Theses & Dissertations

In time-sensitive search and rescue applications, a team of multiple mobile robots broadens the scope of operational capabilities. Scaling multi-robot systems (< 10 agents) to larger robot teams (10 – 100 agents) using centralized coordination schemes becomes computationally intractable during runtime. One solution to this problem is inspired by swarm intelligence principles found in nature, offering the benefits of decentralized control, fault tolerance to individual failures, and self-organizing adaptability. Glowworm swarm optimization (GSO) is unique among swarm-based algorithms as it simultaneously focuses on searching for multiple targets. This thesis presents GPR-GSO—a modification to the GSO algorithm that incorporates Gaussian Process Regression (GPR) based data-driven predictive modeling—to improve the search efficiency of robotic swarms in multiple emission source localization tasks. The problem formulation and methods are presented, followed by numerical simulations to illustrate the working of the algorithm. Results from a comparative analysis show that the GPR-GSO algorithm exceeds the performance of the benchmark GSO algorithm on evaluation metrics of swarm size, search completion time, and travel distance.