Semester
Summer
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 Member
Mario Perhinschi
Committee Member
Patrick Browning
Abstract
This thesis presents the development and evaluation of a neural network-based optimal controller for asymmetrically loaded quadrotor unmanned aerial systems (UAS). Traditional control strategies such as PID are typically designed under symmetry assumptions and often degrade in performance when faced with significant loading asymmetries. To address this, a six-degree-of-freedom quadrotor model incorporating rotor dynamics and center-of-mass offsets was developed. A trajectory optimization framework using MATLAB’s fmincon solver generated over 50,000 energy-optimal trajectories across symmetric and asymmetric conditions. These were used to train a range of feedforward neural network architectures in a full-factorial study.
The best-performing controller was identified, having five hidden layers with 250 neurons per hidden layer. This decision was based on validation error and ANOVA analysis. Simulation results demonstrated that, under symmetric conditions, the neural network controller outperformed a conventionally tuned PID controller with up to 80% reduction in settling time, 70% reduction in overshoot, and 91% reduction in undershoot, while increasing total energy consumption by only 2%. Under asymmetric loading, similar transient benefits were observed at a marginal cost in energy consumption of 3%. These findings showcase improved performance under neural network based control for asymmetrically loaded quadrotor UAS.
Limitations of this work include its simulation-based evaluation, lack of formal stability guarantees, and absence of hardware deployment. Future work includes hardware-in-the-loop testing, dataset refinement, stability analysis, and expanded baseline comparisons to further evaluate and extend the proposed approach.
Recommended Citation
O'Hara, Ross, "End-to-End Neural Network Based Optimal Control for Asymmetric Quadrotor UAS" (2025). Graduate Theses, Dissertations, and Problem Reports. 13014.
https://researchrepository.wvu.edu/etd/13014