Date of Award


Document Type


Degree Name

Master of Engineering (ME)


Electrical Engineering

Committee Chair/Advisor

Javad Velni

Committee Member

Fatemeh Afghah

Committee Member

Kuang-Ching Wang


This paper presents a culmination of work utilizing custom-designed autonomous vehicles and learning based simulations applied to real world applications across three distinct research domains. The first focuses on deploying autonomous drones in an agricultural setting for plant phenotyping. A telescoping arm is used to deploy a camera into the plant foilage for under-the-canopy imaging. The system utilizes on-board image processing to identify early-stage issues such as diseases, bacteria, and pests in crop plants to provide data for crop loss mitigation. The system takes advantage of the maneuverability of drones allowing for quick random sampling when traversing large areas.

In the second domain, the research aims to develop a trust inference model to address scenarios where agents in a swarm collaborate to achieve a coverage control task. Empirical data from human subjects is used for the probabilistic model development where collection is done through simulation tools and user interfaces. A trained single-agent model is extended to create the multi-agent model. This model utilize a dynamic Bayesian networks and produce stochastic predictions. For validation, the model is applied to a Voronoi-based area coverage problem in real time, where agents adjust their behavior to maximize the team performance and hence human trust. As a result of this research, multi-agent teams will be able to increase their individual trust levels thereby enhancing team performance and efficiency.

The third domain delves into collaborative target tracking between a drone and a ground vehicle, employing deep reinforcement learning (DRL) and a fiducial marker vision system. Custom modular platforms leverage the aerial advantages of drones and the differential control capabilities of ground vehicles. The drone, acting as the centralized controller, collaborates with the ground vehicle, featuring wheel encoders and orientation sensors. Proximal Policy Optimization (PPO) in a Unity-based simulated environment refines the DRL model, with real-world experiments demonstrating effective target tracking. Results showcase improved stability and accuracy, affirming the practical applicability of this collaborative system in scenarios such as surveillance and reconnaissance. Together, these research endeavors contribute to advancing autonomous systems, offering innovative solutions in plant monitoring, swarm collaboration, and collaborative target tracking for real-world applications.

Available for download on Tuesday, December 31, 2024