Graduation Semester and Year

Spring 2026

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering

Department

Electrical Engineering

First Advisor

Yan Wan

Second Advisor

Qilian Liang

Third Advisor

Ahmet Taha Koru

Abstract

Emergency response operations such as wildfire monitoring, post-disaster search and rescue, and hazardous-material incidents frequently take place in environments where conventional communication infrastructure has been compromised and Global Positioning System (GPS) signals are unreliable, degraded, or completely unavailable. In these conditions, autonomous robotic systems that support first responders must treat localization and communication as coupled system-level problems rather than as independent components. This dissertation argues that GPS-denied emergency-response autonomy is best achieved by jointly designing localization, communication, and robotic platform integration as a unified autonomy stack, rather than by optimizing these functions as separate modules. To support this claim, the dissertation presents four integrated contributions: an on-demand emergency aerial communication system, a directional aerial communication framework with simultaneous relative localization, a dense feature-based UAV geolocalization framework, and an autonomous ground-vehicle support system that assists the aerial platform during emergency operations.

The first contribution presents the development of Aerial Communication using Directional Antennas (ACDA), a UAV-carried on-demand communication system designed to establish long-range wireless links during emergency response. ACDA is built on a cyber- physical system co-design concept in which the antenna control aligns the directional antennas to enable a robust communication channel and the strength of the communication signal serves as a measurement that informs the alignment. The platform progressed through five iterative versions (Prototype, Alpha, Beta, Gamma, and Delta) and was validated through field demonstrations with first-responder partners. The second contribution extends the ACDA platform of the first contribution to address long-distance UAV-to-UAV communication during emergency-response missions. A three- dimensional Received Signal Strength Indicator (RSSI) model is derived within the antenna half-power beamwidth and combined with visual–inertial odometry (VIO) and barometric altitude in a sliding-window factor graph. The factor graph estimates the relative position between two UAVs and supplies the desired antenna heading to a constrained model predictive controller that drives a one-degree-of-freedom directional antenna. The framework is implemented on a Robot Operating System 2 (ROS 2) platform with PX4 flight control and validated through outdoor flight experiments at the Maverick Autonomous Vehicle Research Center. The third contribution targets GPS-denied UAV geolocalization against geo-referenced satellite imagery, addressing emergency-response scenarios in which a single UAV must self- locate over a wide disaster area without a partner-UAV reference. A frozen DINOv3 backbone with a learned lightweight projection head produces dense cross-view features, from which a continuous, pose-conditioned observation likelihood is constructed by sampling the satellite feature map under each pose hypothesis. The likelihood is fused with a VIO motion prior in a Stein Variational Gradient Descent particle filter (SVGD-PF) that transports particles along gradients of the posterior, avoiding the sample impoverishment of standard sequential importance resampling. The pipeline is deployed on a Jetson Orin NX as a ROS 2 v package and evaluated on an outdoor flight covering approximately 270 m. The fourth contribution documents the design and integration of an autonomous ground vehicle (UGV), built on a Clearpath Jackal platform, that serves as a ground-support node for the emergency-response UAV system: it provides ground-level scouting ahead of the re- mote UAV, acts as a mobile communication relay between the local UAV and the ground command station when terrain blocks the air-to-ground link, and supplies a ground-derived prior against which the UAV’s localization estimate can be cross-checked. The system fuses LiDAR, visual, and inertial odometry through an Extended Kalman Filter, combines slam_toolbox-based map localization with Nav2 global and Model Predictive Path Integral local planners, and adds a dynamic-object-aware SLAM front end built on a 2D LiDAR + RGB-D + YOLO pipeline whose YOLO inference is offloaded to an NVIDIA Jetson Orin NX co-processor through ROS 2 to remove first responders and evacuating vehicles from the LiDAR scan before it is consumed by the SLAM stack. The platform is tested in both Gazebo simulation and physical drive trials. Across these four contributions, the dissertation treats localization and communication as jointly designed components of a single autonomy stack tailored to emergency-response operations, and reports the implementation, integration, and experimental observations gathered during the supporting field studies and demonstrations with first-responder partners.

Keywords

UAV, Communication, Localization, Robotics

Disciplines

Controls and Control Theory | Robotics | Systems and Communications

License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download

Available for download on Friday, May 05, 2028

Share

COinS