Author ORCID Identifier

https://orcid.org/0009-0008-1088-7801

Semester

Spring

Date of Graduation

2025

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Mechanical and Aerospace Engineering

Committee Chair

Christopher Griffin

Committee Co-Chair

Hailin Li

Committee Member

Mario Perhinschi

Committee Member

Xueyan Song

Abstract

Unmanned aerial vehicles (UAVs), particularly quadrotor platforms, have proven to be indispensable tools for search and rescue (SAR) teams due to their maneuverability, rapid deployment, and affordable operation. The current SAR use cases of quadrotors span from aerial surveillance to mission planning and strategic deliveries. However, most SAR applications remain passive or semi-autonomous, such that they rely on human-operated control or data assessment. Advancements in autonomous control strategies bring about a future where a single SAR operator can program and deploy several UAVs. This thesis details the development and simulation of a fully autonomous, AI-powered quadrotor framework capable of detecting missing persons, navigating toward them, and dynamically avoiding obstacles within a realistic search environment. The autonomous system developed is platform independent, mission-driven, and requires no human control between deployment and SAR recovery. The simulation environment was constructed in Unreal Engine 5 leveraging a high-resolution terrain model. The mesh for the model landscape was generated using QL2 United States Geological Survey (USGS) LiDAR point clouds. The quadrotor model is controlled through MATLAB/Simulink code via co-simulation plugins that link the two software suites. The proposed UAV framework has three core subsystems for mission success: dynamic 3D Dubins path planning, active potential field obstacle avoidance, and onboard computer vision guidance. The performance of the system was evaluated by comparing the results of ten search scenarios. In simulations with a Full HD resolution camera and no line-of-sight occlusion, the UAV regularly detected ground targets at a Euclidean distance of over 30 meters while flying at an average mission speed of 3.9 meters per second. In an open terrain mission, 10 seconds from takeoff, the system was able to autonomously avoid two trees and identify a target located at a distance of over 78 meters. The framework was constructed using modular architecture: the computer vision engine, sensor models, and control logic are all independently interchangeable creating several opportunities for unique future search and rescue missions. This architecture enables a single operator to deploy and monitor multiple UAVs from a singular ground station significantly enhancing the scalability and effectiveness of searching operations.

Download

Share

COinS