ORCID Number

0009-0005-4922-3461

Date of Award

Fall 2025

Access Type

Thesis - Open Access

Degree Name

Master of Science in Aeronautical Engineering

Department

Aerospace Engineering

Committee Chair

Richard Prazenica

Committee Chair Email

prazenir@erau.edu

First Committee Member

Kadriye Merve Dogan

First Committee Member Email

dogank@erau.edu

Second Committee Member

Cagri Kilic

Second Committee Member Email

kilicc@erau.edu

College Dean

James W. Gregory

Abstract

This thesis presents the development of vision-based estimation and model predictive control (MPC) strategies to enable an Unmanned Aerial Vehicle (UAV) to land autonomously on an Unmanned Surface Vessel (USV) subjected to wave-induced motion. An innovative Adaptive-Covariance Extended Kalman Filter (AEKF) implementation was developed for the estimation of the 6 degree-of-freedom USV states using GPS and vision-based measurements of AprilTag markers on the USV landing platform. The AEKF employs an uncontrolled 6 degree-of-freedom nonlinear model augmented with second-order harmonic wave-induced motion dynamics. The AEKF implements two correction techniques: an adaptive covariance adjustment and an artificial covariance inflation regulated by a Normalized Innovation Square (NIS) check. These corrections ensure the stability and adaptivity of the filter in the presence of multiple added wave-induced estimation states or in scenarios where the covariance matrices are unknown. The AEKF was implemented in combination with a scheduled MPC algorithm to generate the control actions required for the UAV to perform safe landing maneuvers. The scheduled MPC was designed and implemented in three distinct phases: Approach and Follow, Descent, and Landing. An MPC cost function was designed specifically for each flight phase, and in each case, the optimal control is computed to minimize the cost function subject to constraints and a linearized UAV dynamics model. The performance of the combined AEKF and MPC algorithm was evaluated via simulation studies of a UAV landing on an USV under smooth, slight, and moderate sea conditions. In the simulations, a linear model of the sea state was employed based on Airy Theory. As part of these studies, the AEKF parameters, as well as the MPC prediction and control horizons, were optimized for the different flight phases and sea states. The simulation results show that the AEKF provides accurate estimation of the wave-induced USV states, but that higher wave disturbances require a larger number of states in the wave dynamics model along with a lower NIS threshold to ensure efficient filter performance. The results also demonstrate that, once the prediction and control horizons have been properly selected, the MPC controller results in a high landing success rate. It was also found that phases that rely on estimates of the vertical position and velocity, such as Descent and Landing, require smaller prediction and control horizons. This is due to small disturbances on the vertical axis states, which reduce the accuracy of the AEKF estimates. Finally, the constrained states and controls used to maintain the linearization of the UAV during the MPC optimization limit the controller to be feasible only in scenarios with static or slow moving USVs.

Download

Share

COinS