Trident Scholar Abstracts 2024

Christopher A. Civetta
Midshipman First Class
United States Navy

Probabilistic Object Tracking Using Quantified Camera Uncertainty Parameters in a Binocular System

The need for improved optically-driven systems increases as the Navy becomes increasingly autonomous and attempts to lower its radio frequency (RF) signature. Specifically, the Office of the Chief of Naval Operations (OPNAV) has approved the development of an autonomous aerial refueling system. As part of this effort, the USNA Vision Integration In Polymanual and Experimental Robotics (VIPER) lab is conducting research to enhance and quantify the performance of computer vision techniques as a feedback mechanism for this system. In support of these efforts, this research explores uncertainty propagation in tracking systems leveraging multi-view cameras. By establishing the probabilistic uncertainty of tracked objects, this work provides a novel metric for use in evaluating and fielding systems leveraging camera-based feedback. By leveraging ground truth data provided by the VIPER lab, a binocular system’s uncertainty parameters are experimentally estimated, and a tracked object’s three-dimensional (3D) position uncertainty is approximated. Unlike traditional tracking methods, this process produces a probability “cloud” around the object of interest. This research supports autonomous aerial refueling and generalized applications of multi-view camera feedback to autonomous systems in two ways: (1) enabling tracking systems to use probability distributions to produce confidence intervals to address risk in decision-making; (2) augmenting existing automated image labeling processes with probability information to increase further the effectiveness of training and evaluation of artificial neural networks.

---

Infections caused by bacteria of the Mycobacterium genus, such as tuberculosis (TB), pose a growing threat to global health as they become increasingly antibiotic resistant, requiring novel treatments to heal and protect the warfighter at home and abroad. A promising target for new antibiotics is the L,D-transpeptidase (Ldt) class of enzymes which play a crucial role in the construction, remodeling, and maintenance of bacterial cell walls; however, little is known about the specific function of the various classes of Ldt. Thus, we sought to facilitate the study of Ldts by utilizing an endogenous cell wall biosynthetic pathway to produce putative Ldt substrates. Additionally, we began to elucidate the functions of the different classes of Ldts by using these putative substrates to conduct substrate specificity tests. Overall, we have made notable progress towards our objectives: we have purified and solubilized all the enzymes needed to complete the biosynthetic pathway and have confirmed the activity of all but two of these enzymes. We have successfully increased the activity of the crucial first enzyme in the biosynthetic pathway, MurA. Finally, we have begun the production of putative Ldt substrates and have initiated Ldt substrate specificity studies with commercially available substrates and the putative Ldt substrates.

---

Institutions and their operations are becoming chronically affected by climatebased disruptions. Current efforts to model these disruptions emphasize probabilistic models, notably Bayesian networks. Bayesian networks are adept at capturing the interconnectedness of flood events, infrastructure failure, and operational disruptions better than other probabilistic and deterministic models. This paper presents a framework using the U.S. Naval Academy’s infrastructure and sea level flooding to assess operational risk. Leveraging the open-source Python package, pyAgrum, we visualized probabilistic relationships for operational risk assessment.

Two frameworks are proposed: focused and holistic. The focused approach utilizes existing data on road infrastructure, incorporating historical NOAA tide data and road heights, with assumptions on road flow and usability. Sea-level projection models show the change in impact to road operations over time. Conversely, the holistic model utilizes comprehensive data collection to better represent sea level, multiple infrastructure elements, and institutional missions. Using Bayesian updating and observed flooding impacts, the model can adapt to current data as institutions observe the effects of flood impacts.

This project underscores the efficacy of Bayesian networks in operational risk assessment amidst present and anticipated flood impacts, offering insights for institutions seeking to improve their resilience modeling techniques.

FACULTY ADVISOR
Professor William Traves
Mathematics Department

---

Andrew S. Kolesar
Midshipman First Class
United States Navy

---

One tie linking nearly all the top Fortune 500 companies in the world is that billions of dollars are allocated each year using linear programming (LP). Its applications are also ubiquitous to military applications such as optimal sensor placement. LP is a mathematical modeling tool that models real-world optimization problems with an objective function and series of linear inequalities. Considering the implications of LP, accurate solutions are crucial. Top-commercial LP solvers generally perform well for the vast majority of instances, but decades of literature have shown that they suffer from inaccuracies in 3-5% of models due to improper floating-point (fixed-precision) calculations. These solvers trade the risk of inaccuracy for efficiency as exact solutions take substantially more time to execute. The best available exact solver is SoPlex, which validates the optimality of LPs by using exact rational factorizations to solve key linear systems. These rational factorizations are the bottleneck of SoPlex, occupying over 90% of the runtime in the worst case, and making this solver undesirable. This project addresses these drawbacks and vastly improves our ability to solve LPs exactly. Specifically, Lourenco et. al. developed a suite of algorithms that exactly solve linear systems faster than rational arithmetic. The project implements these improved solvers into the current state-of-the-art SoPlex; thereby significantly accelerating the solver as well as increasing its ability to handle problems of larger dimensions. Additionally, we test the new solver against the base version of SoPlex using real-world LPs from the NetLib repository. In doing so, we show that our new algorithm outperforms baseline SoPlex by a factor of 2.5 on average and 1.7 in geometric mean in run time. Altogether, our new variant of SoPlex expedites the exact solver; allowing larger and more difficult LP problems to be solved correctly.

FACULTY ADVISOR
Assistant Professor Christopher Lourenco
Mathematics Department

---

James B. Margeson
Midshipman First Class
United States Navy

Structure of the X-Ray Continuum Emission Region in the Gravitationally Lensed Quasar SDSS J1339+1310

We analyze 14 seasons of existing optical imagery and 4 epochs of new X-ray monitoring data of the gravitationally lensed quasar SDSS J1339+1310 to place empirical constraints on the size and structure of the X-ray continuum emission region surrounding that system’s supermassive black hole (SMBH). SDSS J1339+1310 (hereafter SDSS 1339) is a distant quasar at redshift z_s = 2.24 that is doubly imaged by a foreground galaxy at z_l = 0.61. Since the light from both of the quasar’s images passes through some portion of the lens galaxy, individual stars that are orbiting within the lens galaxy also lens the images causing time-variable magnification in a phenomenon known as “microlensing.” The amplitude and timescale of this microlensing variability is a function of the size of the source itself, so we analyze the variability to extract constraints on the size of the quasar central engine.

Using a Bayesian Monte Carlo method we successfully constrained the optical source size log(r_opt,1/2/cm) = 15.65 (+0.26,-0.28) in the r-band, corresponding to ∼76 r_g for a 4.0×10⁸ M black hole. We found a 2σ upper limit on the full band (0.2 − 8.0 keV) X-ray continuum emission region of log(r_full,1/2/cm) ≤ 14.88 (∼13 r_g), roughly double the innermost stable circular orbit (ISCO). We also found a hard band (2.1 − 8.0 keV) upper bound log(r_hard,1/2/cm) ≤ 15.78. We are unable to obtain a size measurement for the soft band (0.2 − 2.1 keV). The small X-ray source sizes, compared to the system’s SMBH, support models that express X-ray creation as a factor of the physical properties of a quasar’s central engine and suggest a SMBH rotating at a significant fraction of its maximum angular speed.

FACULTY ADVISOR
Professor Christopher Morgan
Physics Department

---

With recent advances in submarine technology, U.S. Forces must have an effective national strategy to (1) detect and (2) neutralize undersea threats. Regarding (1), we create a strategic optimization model to effectively place sensors to maximize the expected number of detected targets based on historical threat traffic data and the availability and capability of surveillance resources. Our model accounts for overlapping coverage by multiple sensors, introducing nonlinearity into the formulation. We linearize our model using a logarithmic transformation followed by tangent line approximations that guarantee a solution within a specified error tolerance. Regarding (2), we formulate a tactical optimization model to route search teams in response to detected threats, maximizing the expected value of detected targets based on probabilistic information about target locations and estimated strategic values of targets. We account for logistical aspects of the search process, including waterspace constraints and various capabilities of search assets such as search speed and likelihood of detecting a present target. Working in tandem, these two models help our forces to be more efficient in the context of open ocean search and detection of enemy submarines.

FACULTY ADVISORS
Assistant Professor Anna Svirsko
Mathematics Department 

Associate Professor Daphne Skipper
Mathematics Department 

CDR Bradley Alaniz, USN (Ret.)
Undersea Warfighting Development Center

Dr. Michael Kopp
Undersea Warfighting Development Center

Associate Professor Esra Buyuktahtakin Toy
Virginia Tech 

---

Adrien D. Richez
Midshipman First Class
United States Navy

Computer Vision for Uncrewed Aerial System Guidance Navigation and Control

The United States Navy (USN) increasingly relies on uncrewed aerial system (UAS) assets to enhance operational effectiveness and reduce costs. However, a key challenge is the need for a certified guidance, navigation, and control (GNC) system for UAS assets in Global Positioning System (GPS) and Radio Frequency (RF) denied environments. This gap hampers fleet operations in contested areas related to the peer adversaries of the United States (US). To address this challenge, the US Naval Academy (USNA) is collaborating with the Office of Naval Research (ONR), the Naval Air Warfare Center Aircraft Division (NAWCAD), and George Washington University (GWU) to explore innovative solutions.

Supported by ONR’s Shipboard Autonomous Launch and Recovery (SALR) research initiative, the USNA Waterfront Readiness Department has retrofitted a yard patrol (YP) craft with a flight deck, emulating the operational environment of a Guided Missile Destroyer (DDG) at a quarter scale and a fraction of the operating cost. This report presents initial efforts by USNA to integrate computer vision for autonomous UAS launch and recovery operations on this modified YP. Results demonstrate that UAS pose estimation relative to fixed fiducial markers located on and near the YP’s landing zone can be achieved by employing a comprehensive data package. This payload comprises an onboard companion computer and a depth-sensing camera, allowing for future investigation of more advanced and emergent computer vision methods.

Furthermore, the system has enabled exploration into the integration of machine learning models, allowing the UAS to continue operations within the ship’s terminal area, even at ranges where fiducial markers are no longer practical. The project establishes equipment, software, procedures, and recommendations for future researchers aiming to certify autonomous Naval UAS GNC systems for operations in contested shipboard environments.

FACULTY ADVISORS
CDR Donald Costello, USN
Weapons, Robotics and Control Engineering Department

Professor Robert Niewoehner
Aerospace Engineering Department

---

Samuel S. Shin
Midshipman First Class
United States Navy

A Fast First-Order Method to Calculate the Lovász Theta Function

The Lovász theta function of a graph ϑ(G) is a graph invariant that bounds the Shannon capacity Θ(G), the rate at which a confusability graph can communicate information with zero risk of error. Despite its interesting applications, Θ(G) is hard to compute and remains unknown for even simple graphs like the heptagon. A well-known approximation is ϑ(G), which is theoretically computable in polynomial time as the optimal value of a semidefinite program. However, due to the problem’s dramatic runtime scaling, practical implementations that can calculate ϑ(G) for instances of more than a few hundred vertices do not exist.

We address this issue by developing an accelerated projected gradient descent method for ϑ(G). We first approximate ϑ(G) as a min-max problem that explicitly enforces equalities and penalizes positive semidefiniteness over compact primal and dual spaces. We smooth this approximation with prox-functions, allowing us to find efficient formulas for the gradient and duality gap. We then apply Nesterov accelerated gradient to our approximation and exploit its feasible region’s specific structure to construct a linear projection that rounds ∈-saddle-points for the approximation to ∈-optimal solutions for ϑ(G). The result is an accelerated first-order algorithm that decomposes ϑ(G), a complicated optimization problem, into a series of smaller, easily computable subroutines that calculate ϑ(G) to a specified accuracy. Lastly, we present a MATLAB framework for our algorithm.

FACULTY ADVISORS
Professor Nelson Uhan
Mathematics Department 

Dr. David J. Phillips
University of New Mexico

---

Ψ Chloe M. Skogg
Midshipman First Class
United States Navy

Formulation of Barnacle-Inspired Underwater Adhesives

Barnacles produce a sub-micron layer of proteinaceous adhesive that enables their sessile lifestyle in intertidal environments with dynamic temperatures and hydration. Unlike synthetic adhesives used today, barnacles use a tightly folded, extreme form of protein aggregate structure that has been specifically evolved for underwater applications. The natural glue used by these tenacious, biofouling organisms is reliant on an ability to cure in place, maintain insolubility when cured, and bond surfaces. Yet, how these amyloid formations perform such functions remains unclear. Abundant proteins generated as byproducts of agriculture have been shown to be capable of forming barnacle-like materials through changes in pH, temperature, or other means of denaturation. This presents an opportunity to develop new bioinspired underwater curing material systems at practical scales; however, only few proteins have been developed into underwater adhesives with limited knowledge of the thermodynamic and chemical phenomenon occurring in these systems. Little is understood about the role of denaturants naturally occurring in seawater and how they impact the strength and durability of these glues. To better understand these properties, we developed a standard of denaturation for four commercial protein systems using both the denaturants found in seawater and a series of extreme denaturants. Whey protein derivatives, Bovine Serum Albumin, and α-Lactalbumin were used to successfully create two distinct, tunable underwater adhesives with the above denaturants. Investigation of the material and adhesive properties of these two protein-based adhesives was done using gel inversion tests, rheometry, differential scanning calorimetry, infrared spectroscopy, lap shear testing, and additional adhesive tests. These results indicate the ability to adjust the workable window of these adhesives in curing time, bond strength, and viscoelastic properties. Application of these glues occurred in varying environments, exhibiting success in artificial laboratory settings as well as brackish river water and Atlantic Ocean samples. Degradation and longevity testing showed these adhesives were viable candidates for underwater repair for over one year. Our efforts developed new scalable amyloid-like adhesives using commercial protein systems that achieve robust bonding via underwater curing. This work provides new understanding in the formulation, development, and deployment of underwater adhesives in a field that has been relatively untouched.

FACULTY ADVISORS
Associate Professor Elizabeth Yates
Chemistry Department 

Dr. Christopher So
Naval Research Laboratory

---

Katelyn M. Villa
Midshipman First Class
United States Navy

Morphological Comparison of CN and OH in Comet Coma

Comets are small, icy bodies that preserve information about the formation conditions of the early solar system. As comets are heated by the Sun their ices sublimate and create an extended coma whose morphology can be used to study properties of the nucleus. In this project, we use coma morphology to investigate comets’ natal properties.

This is the first large-scale comparison of dust, OH, and CN morphology across a range of comets. We address to what extent CN morphology matches that of OH, and similarly, to what extent OH morphology matches that of dust, as well as the consistency of these comparisons across dynamical classes. Correlated CN and OH implies that their likely parent molecules are intermixed in the comet, while uncorrelated CN and OH suggests inhomogeneities in the nucleus, and points to parts of the comet having formed in very different places in the early solar system. We also consider correlation between OH and dust as an indicator of ‘icy grains’ in the dust tail that significantly contribute to the OH signature.

We have identified 23 comets, imaged from 2008-2023, for which we possess images of sufficient quality acquired with both CN and OH comet narrowband filters, as well as dust images in the R-band. These images were acquired over approximately 100 nights using three professional telescopes. We have analyzed the spatial distribution of dust, OH, and CN in each comet.

Our results indicate that the majority of comets show evidence of icy grains, suggesting that OH morphology is likely a result of ice sources in the dust tail in addition to the traditional nuclear source. Our comparison of CN and OH morphology has yielded an approximately even spread of correlated vs. uncorrelated comets, pointing to a diversity of formation scenarios — some likely significantly more heterogeneous than others.

FACULTY ADVISORS
Assistant Professor Matthew Knight
Physics Department 

Professor Jeffrey Larsen
Physics Department