114333 research outputs found
Sort by
Interpreting FMS and SSQ Cybersickness Ratings via User Tolerance in Virtual Reality.
No full textCybersickness remains a persistent challenge limiting the usability of virtual reality (VR), yet commonly used subjective measures such as the Simulator Sickness Questionnaire (SSQ) and the Fast Motion Sickness (FMS) scale are often difficult to interpret in terms of user outcomes. In particular, there is limited guidance on how specific values on these scales relate to user tolerance or the likelihood that users will discontinue a VR experience due to discomfort. This work addresses this gap by calibrating cybersickness severity interpretations against early termination of a VR experience. In Study 1 (N = 183), participants played an immersive VR game for up to 20 minutes and were free to terminate the experience at any time. Using a Behavioral Risk Banding (BRB) framework, logistic regression linked a modified 0-10 version of the FMS (FMS-10) and post-exposure SSQ scores to dropout probability, yielding behaviorally grounded risk severity bands (mild, moderate, severe, extreme). Survival analyses then showed how these bands mapped onto tolerance over time, revealing a transitional region marked by heightened individual variability. Study 2 (N = 304) evaluated the robustness of these categories across variation in VR content and mitigation context by applying the Study 1 thresholds without re-estimation. Although absolute dropout risk varied, the ordering and behavioral meaning of severity categories were preserved. Together, these results provide a practical, behaviorally grounded framework for interpreting cybersickness ratings in terms of user tolerance and usability across VR contexts.This preprint is published as Kelly, J., Dorneich, M., & Gilbert, S. (2026, January 20). Interpreting FMS and SSQ Cybersickness Ratings via User Tolerance in Virtual Reality. Jan 20, 2026. https://doi.org/10.31234/osf.io/p3xy4_v
Curating the world’s largest biodiversity dataset for AI
No full textThis thesis presents BioTrove, the largest publicly accessible, AI-ready dataset for biodiversity applications. Curated from the iNaturalist platform and filtered to include only research-grade content, BioTrove contains over 161.9 million captioned images spanning approximately 366,000 species across the Animalia, Plantae, and Fungi kingdoms. Each image is paired with detailed taxonomic metadata, enabling a wide range of machine learning applications in ecology, agriculture, and conservation.
To demonstrate its utility, we release BioTrove-CLIP, a suite of vision-language foundation models trained on a 40-million image subset focused on seven taxonomic groups critical to biodiversity and agriculture. These models achieve strong zero-shot and few-shot performance across several challenging benchmarks, including rare species recognition and life stage classification. The dataset and models provide a scalable solution to long-standing challenges in ecological AI—namely, the need for fine-grained classification, taxonomic diversity, and robust generalization to unseen species.
This work has received substantial recognition, including a \textbf{Spotlight presentation at NeurIPS 2024} and national-level coverage by the \textbf{USDA National Institute of Food and Agriculture (NIFA)}. These acknowledgements highlight the broader significance of BioTrove in supporting AI development for food security, ecosystem preservation, and climate change mitigation.
Looking forward, BioTrove is designed for continuous growth. The dataset can be updated as new species observations and annotations become available. Moreover, its rich structure enables extensions to a variety of tasks beyond image classification, such as species detection, life stage prediction, trait analysis, and geotemporal ecological modeling. We anticipate that BioTrove will remain a valuable resource for both the AI and biodiversity science communities in addressing urgent global challenges
Design, synthesis, and structural studies of chiral and topologically complex macrocycles
No full textMacrocyclic and topologically complex molecules represent an important frontier in modern organic synthesis due to their unique stereochemical properties and functional potential in catalysis, pharmaceutical chemistry, and materials science. However, efforts to discover and optimize new macrocycles are often impeded by limitations in synthetic access to this class of compounds. To enable the more generalized macrocycle synthesis, a common macrocyclization strategy with tolerance of functional groups for the preparation of many different types of macrocycles are needed. This dissertation describes the development of new synthetic strategies to construct stereochemically defined macrocyclic and lasso-like architectures, focusing on the control and understanding of planar chirality, axial atropisomerism, and topology of macrocyclic complex.
In the first study, a catalytic enantioselective approach was established for the synthesis of planar chiral macrocyclic metacyclophanes via a Pd-catalyzed macrocyclization. This method provides access to a diverse range of enantioenriched macrocycles incorporating 2,3,4-trisubstituted pyridines, affording meta-, meta–ortho-, and meta–para-cyclophanes in high yields and enantiomeric excess. Structural and conformational analyses through X-ray crystallography, NMR spectroscopy, and computational studies revealed distinct syn/anti preferences and demonstrated that both enantiomers can be obtained from two regioisomer precursors using the same chiral ligand.
The second part of this work focuses on the synthesis and characterization of atropdiastereomeric macrocyclic metacyclophanes. These achiral macrocycles, containing a 1,2,6-trisubstituted benzene core, exhibit restricted rotation around aryl–carbon and aryl–amide bonds, leading to the formation of observable and isolable in and out stereoisomers. This study highlights how conformational constraints in macrocyclic frameworks can give rise to new forms of stereoisomerism even in the absence of classical chiral centers.
In the third study, a modular synthetic strategy was developed to construct lasso-type molecules, whose mechanically interlocked, three-dimensional structures are maintained by steric and coordination-based locking mechanisms. Copper complexation was employed to facilitate the threading process during azide–alkyne cycloaddition and Copper removal yielded stable lasso structures. The methodology allows systematic variation of ring size, loop, and tail sequences, enabling the creation of a library of lasso molecules with tunable topological features.
Together, these studies expand the synthetic toolbox for constructing architecturally and stereochemically complex macrocycles. The approaches developed herein provide new insights into the design, synthesis, and stereochemical control of macrocyclic and topologically entangled molecules with potential applications in catalysis, material, and pharmaceutical chemistry
Understanding the structure and properties of microgravity vs. terrestrial solders for in-space applications
No full textThis dissertation provides quantitative correlations between solidification and flow characteristics, processing conditions, microstructure, and porosity for solder joints produced in terrestrial vs. microgravity conditions, while also evaluating the reliability of 40Pb–60Sn, off-eutectic 50Pb–50Sn, and lead-free 95.5Sn–3.8Ag–0.7Cu solders under extreme cryogenic and elevated temperatures mimicking space exploration (-157ºC to +121ºC outside the ISS).
This dissertation includes results from microstructural characterization of ISSI microgravity, ISSI terrestrial, and a third set of freshly made terrestrial 40wt%Pb-60wt%Sn solder to demonstrate the effects of the absence of buoyancy and natural convection on void formation in microgravity. ISSI microgravity wire wrap 40Pb-60Sn solder had ~13× more voids than ISSI terrestrial 40Pb-60Sn solder, while ISSI microgravity wire feed 40Pb-60Sn solder had 4× more voids than ISSI wire wrap 40Pb-60Sn solder. In microgravity, the presence of large free surfaces (e.g., voids larger than 1 mm) induced Marangoni convection, which drove the accumulation of primary dendrites and smaller voids toward regions that had been at higher temperatures during solidification. Under cryogenic conditions, the effects of β-to-α phase transformation in Sn below 13ºC and ductile to brittle transition in various Sn-based solders were examined. Evidence of DBT in eutectic and off-eutectic Pb-Sn, and 95.5Sn-3.8Ag-0.7Cu solders was observed from in-situ SEM micropillar experiments. In-Situ SEM micro-pillar compression demonstrated that plastic strain at instability dropped from ɛ = 0.06 at RT to ɛ = 0.03 at -85°C and lower temperatures for 95.5Sn-3.8Ag-0.7Cu solder and 0.109 at RT to 0.062 at -85 °C and lower temperatures for 40Pb-60Sn solder, signaling a ductile-to-brittle transition around this temperature. This work contributes directly to NASA’s goals of improving in-space manufacturing and repair capabilities while establishing a strong foundation for future innovations in soldering processes tailored to the harsh conditions of deep-space exploration
Resource-efficient AI for real-world applications: Model compression, sample-efficient learning, and inference optimization across traffic safety, quantum systems, and sequential decision making
No full textThis dissertation addresses the fundamental challenge of deploying powerful artificial
intelligence systems under real-world resource constraints. While modern AI has achieved
remarkable capabilities in laboratory settings, practical deployment is often hindered by three
critical bottlenecks: excessive computational demands, labeled data scarcity, and prohibitive
inference costs. This work develops a unified framework of resource-efficient AI techniques and
demonstrates their effectiveness across four interconnected research contributions spanning
diverse application domains.
The first contribution (Chapter 3) develops a novel pruning framework for Mamba state-space
models, achieving 70% parameter reduction with only 3-9% performance degradation across
language modeling, long-range understanding, and time-series forecasting benchmarks. This
enables deployment of powerful sequence models on edge devices with limited computational
resources. The second contribution (Chapter 4) introduces physics-informed self-supervised
learning for quantum systems through Hamiltonian-Masked Autoencoding (HMAE), reducing
labeled data requirements by 3-5× and achieving 85.3% accuracy in phase classification with only
10 labeled examples—addressing the critical challenge of expensive quantum simulations. The
third contribution (Chapter 5) presents a meta-learned caching framework for reinforcement
learning with large language model priors, reducing inference costs by 3.8-4.7× while maintaining
96-98% of uncached performance and achieving practical latencies of 85-93ms on consumer
hardware. The fourth contribution (Chapter 6) develops HybridMamba, a specialized architecture
for fine-grained temporal localization in traffic surveillance, achieving 1.50-second mean absolute
error (72% improvement over existing methods) while maintaining real-time processing at 7.8
FPS on 2,500 traffic videos from the Iowa Department of Transportation.
Collectively, this research demonstrates that principled approaches to efficiency—grounded in
domain knowledge, theoretical guarantees, and adaptive optimization—can unlock transformative
AI capabilities in practical settings where resource constraints would otherwise prove prohibitive.
The contributions span methodological innovations (gradient-aware pruning, physics-informed
masking, meta-learned caching, hierarchical temporal processing), theoretical advances (stability
analysis for pruned SSMs, quantum information-theoretic masking, KL-divergence bounds for
cached posteriors), and empirical validation across 9 distinct benchmarks. The findings have
direct applications in traffic safety, quantum materials discovery, robotics, and autonomous
systems, with broader implications for deploying AI in resource-constrained environments
A Kalman filter for handheld probe tracking
No full textAccurate registration of Non-Destructive Evaluation (NDE) data is critical for digital systems
that seek to integrate this data for locating damage and other structural problems. One way to
register NDE data is to track the probe that records the data. There are many devices that can
be used to track locations such as GPS, inertial measurement units (IMUs), and rotation and
offset trackers such as cameras and fiber-optic sensors. These tracking devices have their own
strengths and weaknesses that can limit their quality if used alone. For example, IMUs have
relatively low noise levels, but are prone to drift. GPS, cameras, and fiber-optic sensors are stable
but noisy. If multiple tracking devices are combined together and integrated properly, then a
tracking system can be created that incorporates the strengths of each of these tracking devices
while leaving out the weaknesses. One such integration tool that can be used is the Kalman filter.
The Kalman filter, when tuned properly, can take information from a properly-informed dynamic
model, as well as the tracking devices, and make more accurate and certain estimates about the
location of the probe. The objective of this study is to track the pose (orientation and offset) of
an eddy-current probe through the integration of a highly-nonlinear dynamic model, fiber-optic
pose tracker, and IMU using an unscented Kalman filter (UKF)
Investigating energetic demands of the scallop visual system
No full textScallops possess a distributed visual system consisting of tens to hundreds of morphologically similar eyes, yet the energetic implications of maintaining such a system remain largely unexplored. While previous studies on vertebrates and insects have highlighted the high metabolic demands of paired-eyed systems, little is known about how eye number affects metabolism in organisms that possess tens to hundreds of eyes. In this thesis, I investigate the relationship between eye number and metabolic rate in the bay scallop, Argopecten irradians, to understand how the number of eyes contributes to overall metabolic demands. Using oxygen consumption as a proxy for metabolic rate, I found that individuals with more eyes exhibit higher rates of oxygen consumption, independent of body mass. Additionally, eye number explains a substantial portion of variation in metabolic rate, whereas body mass contributes very little compared to eye number. These results suggest that maintaining large number of eyes imposes measurable metabolic demands in the scallop visual system. This research demonstrates that eye number is a major predictor of variation in metabolic rate in scallops, highlighting the physiological demands associated with maintaining distributed visual systems
A direct solution to the J2 perturbed Lambert's problem
No full textThe development of a novel solution to the J2 perturbed Lambert’s problem is presented by reformulating the first-order generalized Kepler’s Equation into a time-of-flight equation. Recent developments in the study of Lambert’s problem have led to an exact solution, which is a function of contour integrals. By developing a time-of-flight equation that includes the effects of Earth’s oblateness, we can use the exact solution method to directly solve for the semi-major axis of the transfer orbit and the initial and final velocity magnitudes. Traditional methods of solving this problem involve iterative methods that incrementally correct the solution to the unperturbed problem to account for perturbations. In the classical Lambert’s problem, the velocities are assumed to be co-planar; however, when J2 perturbations are considered, this is not the case. Therefore, a method is derived to determine the initial and final orbital planes using the average time rate of change of the classical orbital elements. With the semi-major axis and orbital planes, the full velocity vectors that satisfy the boundary conditions can be computed. Monte Carlo analysis is performed to assess the accuracy of this solution relative to several different propagators. The error in the terminal position is less than 10 [km] on average when compared to first-order propagation
Sampling-based optimized adaptive discretization and its applications in robotics
No full textDiscretization is critical in robotics because real robotic systems operate in continuous, high-dimensional state spaces that are intractable for most computational methods. Converting these continuous spaces into compact finite representations enables efficient planning, learning, and safety analysis. However, existing discretization methods often prioritize approximation accuracy while neglecting explicit control of partition size. This imbalance causes state representations to grow exponentially with dimension, which limits the scalability of downstream robotic applications.
This thesis presents a principled framework for optimal discretization that explicitly balances accuracy and representation complexity. Building upon theoretical conditions that characterize an optimal partition structure, we introduce two sampling-based adaptive discretization algorithms: GOAD and λ-OAD. GOAD maximizes split gain to better approximate the optimal solution with highly compact partitions, while λ-OAD enforces a dynamically adapting split gain threshold that balances accuracy with reduced computational cost.
Extensive experiments across a variety of synthetic datasets and robotic tasks demonstrate the effectiveness of the proposed methods compared to state-of-the-art discretization techniques. Both GOAD and λ-OAD achieve substantially fewer partitions under the same approximation accuracy, and λ-OAD further improves efficiency by significantly reducing computation time. These results highlight the scalability and practical advantages of the proposed methods for robotics applications requiring accurate yet computationally efficient discretization
Robust and efficient methods to design platinum-free electrocatalysts for hydrogen production.
No full textThe challenge of designing electrode materials for hydrogen production using chemically-disordered compounds is an amalgamation of four separate challenges. We address each of these challenges piecemeal through the development of novel methods that are efficient, robust and modular. Compounds of interest within modern catalysis, such as the Transition Metal Phosphide family with the stoichiometry M2P, have multiple symmetry-distinct surface terminations. The first method
we developed was to ensure robust calculations of surface energies using data fusion, such that we can select models of the pristine surface of these compounds. Moreover, the coexistence of metallic and non-metallic atoms at the surface of these compounds has been shown to lead to charge and site heterogeneity across the pristine surface. The second method we developed was a model of adsorption on a symmetry-distinct surface site under continuous increase in coverage which we
call Symmetry-Informed Lateral Interactions or SymILI. This method efficiently describes steady-state coverage, lateral (adsorbate-adsorbate) interactions, and the evolution of these descriptors under applied potentials such that we may use it to interpret observed activity of M2P facets. While SymILI can quantifying the strength of such interactions, it does not provide an underlying mechanistic understanding. Complimentary electronic descriptors can help reveal the spatial and energetic character of these interactions. The third method we developed was based on the density of d-orbitals perpendicular to the surface, which was found to correlate with charge transfer resistance for M2P facets, suggesting that this descriptor may be used to efficiently infer activation barriers instead of relying on expensive calculations of minimum energy pathways (MEPs).
Fourthly, we provided a command-line utility to generate reliable models of systems with chemical disordering of crystals, such that finite (non-zero) Short-Range Order is supported and optimal solutions are achieved within seconds. As a popular strategy to optimize the performance of materials, this method of generating Supercell Random APproximateS (or SCRAPs) will help the community efficiently explore and exploit the design space available for chemically disordered systems. The
four methods developed as part of this thesis can be readily and synergistically combined to ensure reliable structural representations of surfaces and chemical disorder are utilized within DFT, and that accurate inference of catalytic performance does not become so computationally expensive that it becomes challenging to perform high-throughput materials design and screening