Each year, UB recognizes the academic excellence of highly accomplished students, honoring their achievements and commitment to scholarly success.
This award recognizes outstanding Student Showcase projects from each decanal area that achieve superiority in presentation, content and scope, and which therefore merit acknowledgement as efforts worthy of university-wide distinction.
Laibah Ahmed, Computer Science BS
FAIR Train: Annotating Tensile Testing Data Using Ontology
Matthew Drummond, Engineering Science
Jordan Ngamaleu-Kemta, Electrical Engineering
Ghazal Vasseghi, Environmental Engineering PhD
Cellulose Acetate Microfiber Release from Cigarette Filters in Agitated Water
The Student Showcase features student research and creative projects completed under the mentorship of UB faculty.
CNN-BASED 3D VESSEL RECONSTRUCTION FROM SPARSE-VIEW DYNAMIC DSA WITH MARKOV CHAIN-INSPIRED APPROACH
Ahmad Rahmatpour
Three-dimensional digital subtraction angiography (3D-DSA) is widely used for evaluating intracranial aneurysms, but standard reconstruction requires more than 100 projections and increases radiation and contrast exposure. This study investigates whether a convolutional neural network (CNN) can recover high-quality 3D vascular volumes from sparse-view dynamic DSA data. A Markov chain-inspired, three-stage sequential 2D U-Net framework was developed to progressively refine truncated-view reconstructions from 30 to 50 to 70 projections and ultimately approximate full-angle image quality. Retrospective data from 156 aneurysm cases across three hospitals were used, including 118 rotational DSA scans and 38 reference 3D volumes. Sparse-view datasets were simulated from 108-view CBCT acquisitions reconstructed with the Feldkamp-Davis-Kress algorithm using ASTRA and DICOM-defined geometry. Across a 15-case test cohort, the model improved median Dice from 0.16 in truncated input volumes to 0.84 after refinement, demonstrating strong agreement with reference reconstructions and promising potential for safer intraoperative neurovascular imaging.
COMPARATIVE EVALUATION OF GROUND GLASS AND ENGINEERED DIFFUSERS FOR VOLUMETRIC SPECKLE-MODULATING OPTICAL COHERENCE TOMOGRAPHY OF THE OUTER RETINAL LAYERS
Austin Yetter
Optical coherence tomography (OCT) is an imaging technique limited by speckle noise inherent to the imaging process. We implemented a mounting system to compare a ground glass and an engineered diffuser that reduce speckle noise and improve image quality in human subject retinal imaging. Volumetric scans were acquired as repeated frames with independent speckle patterns, which were motion corrected and averaged. The averaged frames were used to generate volumetric and en face visualizations. Fine structural details and potential biomarkers on the outer retina were revealed that are otherwise obscured by speckle noise. The diffuser and imaging methods could be easily implemented into commercially available OCT systems, increasing diagnostic abilities.
PREDICTING DELAYED CEREBRAL ISCHEMIA BASED ON ANGIOGRAPHIC IMAGES OF SUBARACHNOID HEMORRHAGE
George Dimopoulos
Subarachnoid hemorrhage (SAH), demands precise imaging for effective treatment. This study introduces an approach integrating a 3D neurovascular atlas with 2D Digital subtraction angiography (DSA) images to allow targeted, region-specific quantitative analysis, thus enhancing diagnostic accuracy during interventions. DSA images were aligned, or co-registered with the atlas to allow for a descriptive segmentation of the image. These segment were analyzed using machine learning to predict disease outcomes. The co-registration process revealed that deformable registrations were essential to achieve precise overlays of the 3D atlas projections with the 2D DSA. This approach enabled the extraction of targeted quantitative angiography parameters, essential for detailed vascular assessment in SAH treatments. The integration of the 3D atlas registration with the 2D DSA projections provides labelling of affected arterial territories, potentially improving the accuracy of diagnostics and supporting better informed clinical decisions at the time of intervention.
REFLECTIVITY ANALYSIS OF HUMAN PHOTORECEPTOR LAYERS IN OPTICAL COHERENCE TOMOGRAPHY
Jada Beckford
Optical coherence tomography (OCT) uses near infrared light to generate cross-sectional images of the retina. Each image is composed of multiple 1-dimensional plots (A-scan) and represents different optical brightness levels. Using the software ImageJ, we adapted and optimized the analytical method to assess varying reflectivity around the ellipsoid zone (EZ), or the inner segment - outer segment junction in human retina. Our preliminary results show subtle variations of reflectivity profile between eyes with younger age and older age, which can provide insights into age-related structural photoreceptor variations.
DEVELOPING A PATIENT-DERIVED IPSC ENDOTHELIAL MODEL TO INVESTIGATE MITOCHONDRIAL DYSFUNCTION IN MELAS
Lucas Davis and Karl Swanson
Mitochondrial encephalomyopathy, lactic acidosis, and stroke-like episodes (MELAS) is the most common mitochondrial disease and is characterized by recurrent stroke-like episodes (SLE) in children and young adults. These episodes cause progressive neurological disability and are the strongest predictor of reduced life expectancy. There are currently no disease-modifying therapies. One major barrier to therapeutic development is the lack of reliable human vascular models to study how mitochondrial dysfunction leads to SLE. Endothelial cells, which line all blood vessels, are increasingly implicated in MELAS pathogenesis. Mitochondrial defects in these cells are thought to impair energy production, increase reactive oxygen species, reduce nitric oxide availability, and promote inflammation, all as hallmarks of endothelial dysfunction that may precipitate SLE. However, this hypothesis remains insufficiently tested due to the absence of robust human models. This project seeks to establish a patient-specific vascular model of MELAS using induced pluripotent stem cell (iPSC) technology. Skin-derived iPSCs from healthy individuals have successfully been differentiated into induced endothelial cells (iECs) using a rapid 4-day protocol and validated by endothelial marker expression and functional flow alignment. MELAS patient-derived iPSCs will be converted into MELAS-iECs and examined for mitochondrial and endothelial abnormalities. MELAS-iECs will be assessed for ATP production, reactive oxygen species generation, nitric oxide bioavailability, and inflammatory activation.
EMG-BASED TASK AND PERFORMANCE CLASSIFICATION USING NONLINEAR AUTOREGRESSIVE EXOGENOUS (NLARX) MODELLING FOR REHABILITATION ASSESSMENT
Reshma Katharin Biju and Julian Martinez
This work presents a machine learning approach for analyzing electromyography (EMG) signals to classify motor tasks and predict performance in a rehabilitation setting. Data were collected from a single subject performing flexion, extension, and alternating cursor-control tasks. EMG signals from flexor and extensor muscles were processed into activation envelopes and used to construct a cursor proxy representing motor intent. A nonlinear autoregressive model with exogenous inputs (NLARX) was developed to capture neuromuscular dynamics, with model residuals used as features for classification. The pipeline included signal preprocessing, downsampling, nonlinear system identification, and feature extraction. Task classification achieved 60.2% accuracy using linear discriminant analysis (LDA), while pass/fail prediction was effective only for flexion (68.2%). Future work focuses on multi-subject data pooling to evaluate subject-independent EMG-performance relationships and assess cross-subject generalization of the NLARX framework.
OSTEOPULSE: DEVELOPMENT OF A LOCALIZED VIBRATION DEVICE TO ACCELERATE HEALING IN BONE FRACTURES
Alex Barletta, Emma Gillebaard, Riley Wymer and Evan Yandricha
Standard fracture management involves immobilization by casting, but there is increasing interest in supplementary therapies to accelerate fracture healing. Localized mechanical vibration in the frequency range of 20 - 90 Hz for <20 min/day has been explored as a potential approach to treating musculoskeletal conditions, understood to provide anabolic stimulation to tissues similar in nature to exercise but with minimal exertion. Our group developed a cast-attachable mechanical stimulation device which delivers localized low-intensity high-frequency vibrations (LIHFV), providing a low-cost, minimally invasive, at-home care solution to accelerate fracture healing. While further research is needed to identify the most effective parameters, our device will be equipped to deliver any LIHFV regimen.
INVESTIGATING CELLULAR INTERACTIONS USING CUSTOM 3D PRINTED MODEL SYSTEMS
Aidren Arul Alphonse, Mahmud Amin and Ansh Kumar
Cellular interactions regulate many biological and pathophysiological processes. To study these interactions, culture inserts are commonly used for in vitro assays including proliferation, migration, and invasion. However, commercially available inserts are often expensive, limited in design flexibility, and restrict experimental customization. This project aims to develop low-cost, customizable cell culture inserts using 3D printing. Parametric computer-aided design (CAD) models were used to design inserts with adjustable parameters including height, wall diameter, thickness, and inter-compartment spacing. These parameters determine compatibility with multi-well plates and influence media volume and cell-loading capacity. Inserts were fabricated using fused deposition modeling (FDM) with biocompatible thermoplastics.The inserts were evaluated for their ability to support co-culture systems and transwell-style migration and invasion assays. Preliminary functional testing for in vitro validation under process Additive manufacturing provides a scalable platform for producing customizable cell culture inserts, improving accessibility for academic and resource-limited research laboratories.
ENABLING POLYOLEFIN CIRCULARITY: DISSOLUTION MODELING FOR SOLVENT-BASED RECYCLING
Ali Ghasemi
Solvent resistance in semicrystalline polyolefins is critical for product durability but hinders solvent-based recycling. This study employs a combined experimental and modeling approach to resolve the fundamental mechanisms governing the dissolution of high-density polyethylene (HDPE) and polypropylene (PP). We present a validated phenomenological model that captures the processes governing the dissolution of semicrystalline polymers, i.e., solvent diffusion, transformation from crystalline to amorphous domains, specimen swelling, and polymer chain untangling. Through parametric sensitivity analysis, the impact of film thickness or particle radius, initial degree of crystallinity, temperature, and solvent type on dissolution kinetics is evaluated. This work offers insights on the interplay of decrystallization and polymer chain disentanglement during the time-course of HDPE and PP dissolution. Further, this work facilitates the design and optimization of dissolution-precipitation recycling process that can unlock value from the million tons of polyolefins annually that are currently being landfilled or incinerated following their use.
ROLE OF GLYCOSYLATION IN THE MIGRATION OF PANCREATIC CANCER
Arun Singh
Pancreatic ductal adenocarcinoma (PDAC) has a 13% 5-year survival rate, hampered by late detection and immune evasion. Aberrant glycosylation, driven by Warburg effect-induced hexosamine pathway flux and elevated O-GlcNAcylation, promotes tumor progression and immunosuppression. Surface glycans, such as sialylated ITGA3/ITGB1 engaging Siglec-10 on macrophages, exemplify this axis. We hypothesize additional PDAC glycans mediate evasion. Using CRISPR screens on PANC-1/FG COLO 357 lines, we identify glycogene knockouts restoring phagocytosis and inhibiting metastasis. Validated targets offer novel immunotherapeutic avenues.
BIODEGRADABLE ZWITTERIONIC POLYMERS FOR MULTIDRUG CO-DELIVERY
Cassie Witt
Poly(lactic acid) (PLA) is an FDA approved synthetic biodegradable polymer for biomedical application. It is not antigenic, meaning that humans will not produce antibodies against it. Functionalized lactide monomers, such as an allyl or acetylenyl monomer, can be used to produce PLA with reactive groups. These reactive groups allow for PLA to be modified with multiple anti-cancer drugs, water-solubility enhancing zwitterions, and other moieties like dyes. This modified polymer has promising potential as a next-generation ultra biocompatible drug delivery system.
AN OPTIMIZATION FRAMEWORK FOR THE TRANSITION TOWARD A CIRCULAR PLASTICS ECONOMY
Matthew Bablin and Cassidy Shafer
Global plastic demand is projected to double by 2050, along with proportional increases in greenhouse gas emissions and pollution from mismanaged waste. Currently, over 99% of the plastics produced in the U.S. are made from fossil sources, while most plastic waste is either landfilled (75%) or incinerated (16%). Transitioning toward a circular plastics economy requires novel solutions to minimize waste and maximize resource recovery. Therefore, integrating sustainable feedstocks (e.g., bio-based materials) and advanced recycling technologies can help achieve this goal. This project will develop an optimization framework to evaluate pathways for transitioning to a circular plastics economy. Using mathematical optimization methods, we will assess pathways including conventional fossil-based plastics, bio-based alternatives, and emerging recycling technologies, including hybrid systems (such as mechanical combined with solvent-based recycling). Specifically, we will propose a multi-objective optimization model that considers economic, environmental, and circularity metrics. The framework will be applied to representative product categories, including single-use packaging, recyclable containers, and durable goods.
SUSTAINABLE SOLVENT BASED SEPARATION OF POLYOLEFINS FOR ADVANCED PLASTICS RECYCLING
Jameson Bonnar and Shikha Solanki
Plastics are widely used because of their diverse and desirable properties; however, these properties are often achieved through additives such as plasticizers and flame retardants. Additives can potentially migrate from plastics and lead to human exposure via, e.g., food/drinks in plastic packaging. Further, additives can propagate into products having recycled content. In this way, plastic use and recycling become vectors for spreading chemicals of concern. Our research targets dissolution/precipitation which is a low-energy and low-greenhouse gas (GHG) emission methodology for recycling waste plastic that cannot be mechanically recycled. This method not only separates specific polymer types from mixtures or blends, but also purifies plastics from additives or fillers, without negatively affecting the properties of the original polymers. This paper will address the removal of dyes from polyolefins using dissolution/precipitation with switchable hydrophilicity solvents.
A COMPUTATIONAL FRAMEWORK TO QUANTIFY UNCERTAINTY IN PROFITABILITY AND GREENHOUSE GAS EMISSIONS FOR PLASTIC RECYCLING PATHWAYS
Matthew Bablin and Will Reid
Over 400 million tonnes of plastic are produced worldwide each year, and less than 10% of this is recycled. The rest ends up in landfills or the open environment, causing environmental hazards. When selecting technologies for plastics recycling, two critical variables to consider are profitability and greenhouse gas (GHG) emissions. This work proposes a computational framework that uses Monte Carlo simulation to quantify uncertainty and help inform the economic and environmental performance of plastic recycling pathways, including pyrolysis, solvent-based processes, and mechanical recycling. Results indicate that mechanical processes and solvent-based recycling with temperature reduction for precipitation have the highest potential to reduce GHG emissions, with up to 77% savings. This solvent-based variant offers the highest profit margins, with an estimated annual revenue of 47 million USD. The proposed framework can help inform investors, policymakers, and industry stakeholders when selecting technologies and designing facilities to advance a circular plastics economy.
EFFECT OF GAMMA RADIATION ON THE MECHANICAL PROPERTIES OF SEISMIC PROTECTIVE DEVICES
Ajaykumar Patel
Advanced reactor developers are considering seismic isolators and dampers to mitigate the effects of earthquake shaking on safety-related equipment. Seismic isolators and dampers installed near a radiation source, without shielding, may experience changes in their mechanical properties, which could affect the design-basis response of safety-related equipment. The U.S. Department of Energy funded project is characterizing the effects of gamma radiation on seismic isolators and dampers in collaboration with Idaho National Laboratory (INL). Pre-irradiation mechanical testing at the University at Buffalo (UB) established baseline properties of the seismic isolators and their components. The isolators were then irradiated using a Cobalt-60 gamma irradiator at Idaho National Laboratory (INL), followed by post-irradiation mechanical testing at UB to characterize changes in their properties. The outcomes of this project will help advanced nuclear designers develop maintenance protocols and shielding requirements for seismic isolators to achieve the intended performance throughout the design life of the structure.
TOWARDS A REMOTE SENSING SOLUTION TO QUANTIFY NITROUS OXIDE EMISSIONS BY INTEGRATING SHORTWAVE AND THERMAL INFRARED BANDS
Ayesha Riaz
Nitrous oxide (N2O) is a potent greenhouse gas whose emissions are dominated by natural and agricultural soils and are highly heterogeneous and episodic, yet existing observational techniques lack the spatial coverage and near-surface sensitivity needed to resolve this variability. In this study, we evaluate a remote sensing framework that integrates shortwave infrared (SWIR) and thermal infrared (TIR) spectral bands to enhance detectability of column-integrated mixing ratio (XN2O). To implement this, we expand the capacity of the SPLAT-VLIDORT radiative transfer model to jointly simulate both spectral regions and apply linear sensitivity analysis to quantify XN2O measurement error and vertical sensitivity under realistic environmental conditions and instrumental designs. The joint SWIR-TIR setting achieves single-sounding measurement error of approximately 3.2 ppb for an airborne instrument with a ground footprint size of 20 m and 1.1 ppb for spaceborne instrument with a footprint size of 0.7 km, while retaining sensitivity to the near-surface layers.
CELLULOSE ACETATE MICROFIBER RELEASE FROM CIGARETTE FILTERS IN AGITATED WATER
Ghazal Vasseghi
Cigarette butts (CBs), among the most common littered items globally, are composed of cellulose acetate (CA) fibers that degrade slowly and can release microfibers (MFs) in aquatic environments. This study quantifies MF release from CBs under three agitation levels (0, 80, and 200 rpm) over 10-day controlled experiments and extrapolates findings to estimate environmental impacts in New York State (NYS). CBs exhibited an initial rapid release of loose fibers upon immersion, followed by slower, sustained shedding. High agitation (200 rpm) significantly increased MF release (p < 0.001), yielding 1.69 times more fiber than stagnant conditions. Fiber length was not significantly affected by time or agitation (p > 0.1). Extrapolation of first-day results estimates that CB litter in NYS may release 7.15E+07-1.04E+09 MFs per day. Improved disposal practices and waste management could reduce these emissions.
SCREENING CU(I)-BASED ELECTROCHEMICAL CARBON CAPTURE SORBENTS FOR STABILITY
Nicole Gerdes
Excessive carbon dioxide (CO₂) in the atmosphere continues to drive severe climate challenges of great concern. This calls for further development of energy- and cost-efficient technologies to capture CO₂ out of the atmosphere for long-term sequestration in, for instance, deep saline aquifers. Electrochemical carbon capture, which uses electricity to drive the CO₂ capture reactions, has low energy and cost, but current electrochemical approaches are often unstable in the presence of oxygen gas. One promising but air-unstable electrochemical approach uses the Cu(II)/Cu(I) redox couple to control a CO₂ capture absorbent, the substance binding to CO₂. This research explores how effective different CO₂ absorbents are at increasing the Cu(II)/Cu(I) reduction potential, a measure of how unstable the Cu(I) state is to oxygen. The findings presented in this research provide insight into which absorbents (histidine, histamine, or imidazole-acetic acid) and buffer combination(s) will stabilize the Cu(I) system most effectively.
EYE-TRACKING DIFFERENCES IN ANXIOUS AND NON-ANXIOUS DRIVERS: A VR AND GAZE ENTROPY STUDY OF YOUNG DRIVERS
Daisha Cardenas-Sanchez and Hayden Radel
Crashes in motor vehicles are a leading cause of death in young adults, which has been credited to young drivers lack of experience and impulsivity behind the wheel Although, driving anxiety, which impacts cognitive behavior, has been studied little. This work uses virtual reality (VR) and eye-tracking to quantify driver behavior including scanning patterns and scan complexity. 31 young drivers (age 18-25) were grouped into anxiety and non-anxiety drivers based on the Driving Cognitions Questionnaire and completed five scenarios in VR. There were four unique scenarios, and the fifth was a repetition, including: car waiting to turn from a side road (repeated), pedestrian crossing at crosswalk, work zone blocking right lane, garbage truck blocking right lane. During each scenario, driver eye movements were recorded at 200 Hz. These eye behaviors were quantified using Areas of Interest (AOIs) to classify what object the participant was looking at in each instant. Then, by looking at the order of fixations in the AOIs, Standard Gaze Entropy (SGE) and Gaze Transition Entropy (GTE) were calculated. In the pedestrian crossing at crosswalk scenario, non-anxious drivers were found to exhibit lower GTE, reflecting more predictable and stable scanning patterns than the anxious drivers who exhibited less predictable scanning patterns (p=0.0206). SGE was found to not differ significantly (p=0.3227), which indicates that overall spatial distribution was relatively equal. This work demonstrates the impact of driving anxiety on visual attention behaviors. It also supports the use of VR for understanding driving behavior in situations where on-road tests may not be appropriate or practical. These findings may inform future approaches to detecting driver anxiety based on eye behavior and enhancing road safety for all road users through advanced driver-warning systems.
OPTIMIZING THE SOLAR SALES FUNNEL: AI GUIDED AUTOMATED VERIFICATION VS. MANUAL OUTREACH
Max Clarke and DJ Ruszkowski
SolarScope, developed by TriTech Labs, addresses high customer acquisition costs in the solar industry by replacing manual "gut-feeling" sales tactics with a data-driven scoring engine. By engineering a proprietary algorithm that evaluates financial readiness and site suitability, the platform automates lead verification to prioritize high-conversion prospects. Currently in pilot testing with four industry partners, SolarScope streamlines the transition from sales to engineering, demonstrating how algorithmic prioritization reduces "soft costs" and accelerates renewable energy adoption. This research highlights a scalable, automated solution for optimizing the B2B solar sales funnel.
SEARCHING DNA DATABASES ON MOBILE DEVICES
Andrew Mikalsen
The key component to many forms of DNA analysis is the reference database of all known genomes. In an effort to enable DNA analysis on mobile devices, it has become desirable to search this database in resource constrained settings. This introduces new systems challenges, both from the mobile computing and databases perspectives. Our talk will discuss our work on mobile DNA analysis, including our approaches to indexing massive DNA databases, memory management, and I/O scheduling.
LIFESTYLE AS DATA: UNPACKING YOUTUBE "DAY IN THE LIFE" VIDEOS THROUGH LARGE-SCALE ANALYSIS
Daniel Nyamollo
"Day in the Life" (DITL) videos are a form of documentary-style content that purport to provide a behind-the-scenes, everyday look at a person's daily routine, showcasing a stylized view of particular lifestyles, occupations, and/or locations. These videos, popular on multiple social media platforms including YouTube and Tiktok, are responsible for billions of views, and thus have the potential to (re)shape perspectives on what "real life" looks like. However, little is understood about the messages conveyed in DITL videos for particular social identities, nor what factors are associated with a particular DITL video going viral. To this end, my work presents an initial investigation into the universe of DITL videos, assessing what exactly they tend to focus on, and what factors are associated with video popularity. Specifically, using computational methods, I analyze around 20,000 DITL videos on YouTube to study factors associated with view counts and prevalent themes in video titles and descriptions. The results show the variety of ways these videos are produced and presented, and how they may impact perceptions of critical career paths and locations around the globe.
OLLI REVIVAL
Marco Bianco
This project focused on restoring functionality to a previously autonomous electric bus through reverse engineering and low-level system redevelopment. The original system was non-operational, requiring a comprehensive understanding of its electrical, control, and drive-by-wire subsystems. The objective was to design and implement a new low-level control architecture capable of enabling safe and reliable vehicle operation. Over the course of one year, our team enabled the high-voltage, low-voltage, DC-DC converter, and driving systems through reverse engineering of the vehicle's electrical architecture, subsystem testing, and low-level software development. As a result, the bus was successfully transitioned from an inoperable state to a functional vehicle capable of controlled operation. This work demonstrates the feasibility of restoring complex autonomous vehicle systems through reverse engineering and highlights the importance of hardware software integration in electric vehicle design.
NO BLIND SPOTS: TURNING SHORT DRIVE-THROUGH VIDEOS INTO INTERACTIVE COMPLETE 3D MODELS OF VEHICLES FROM ROOF TO UNDERCARRIAGE
Nitin Kulkarni
Used-vehicle purchases often depend on limited photos, while undercarriage inspection remains slow, uncomfortable, and rarely documented for on-line buyers. We present a unified, distortion-native pipeline that generates a 3D model of the vehicle, capturing everything from the roof down to the undercarriage. SAM 3 instance masks with motion gating isolate the target vehicle in cluttered dealerships and enforce rigidity by excluding rotating wheels during SfM. RoMA~v2 then produces dense correspondences on raw wide-FOV frames using mask-guided sampling and filtering. These verified matches drive a rig-aware SfM optimization with calibrated intrinsics and measured/CAD rig pose priors to recover consistent poses and sparse geometry. Finally, 3DGUT Gaussian Splatting renders interactive 3D models for walk-around viewing and underbody inspection. On 25 dealership vehicles, exterior reconstructions achieve 28.66 dB PSNR, 0.89 SSIM, and 0.21 LPIPS on held-out views.
FPGA-BASED REAL-TIME AUDIO PROCESSING FOR VOICE-ACTIVATED SYSTEMS
Opeyemi Omoboye
Voice-activated technology improves accessibility for individuals with disabilities, yet traditional software-based processing is often power-intensive and restricted by sequential execution. This research explores using Field Programmable Gate Arrays (FPGAs) as a high-performance hardware platform for real-time audio processing to address these limitations. Using Verilog, the project implements a specialized audio pipeline. Signals are loaded into Block RAM and conditioned via a Finite Impulse Response (FIR) filter to remove noise. A Fast Fourier Transform (FFT) then converts these signals into the frequency domain, enabling Mel-frequency cepstral coefficient (MFCC) feature extraction. These features are classified through FPGA logic to identify commands. By leveraging hardware parallelism, this FPGA-based approach aims to outperform software methods in both latency and energy efficiency. This research advances the design of responsive, accessible voice-activated systems, ultimately enhancing independence and quality of life for individuals with physical or visual impairments.
BUFFALO BYTE V2.0: ENABLING AUTONOMY IN SMALL-SCALE MOBILE ROBOTS
Le Quoc Viet Pham
Buffalo Byte v2.0 is a small-scale autonomous robot with a mass of 17g, with a volumetric footprint of 54.4 mm x 30.1 mm x 19.4 mm. This work builds upon the original Buffalo Byte platform, and version 2.0 enhances the computation capability, sensing, and wireless communication. This upgrade allows the robot to process a larger amount of data onboard, using a dual-core system on a chip (Soc). The robot integrates a 9-axis inertial measurement unit (IMU) and an 8 pixel by 8 pixel ToF depth image sensor to enable closed-loop motion control, obstacle avoidance, navigation, and ultimately SLAM (Simultaneous Localization And Mapping). In future work, these robots will work together in swarm applications, exchanging information with external devices or peer robots for distributed tasks.
UB LUNABOTICS AUTONOMOUS MOON ROVER
Sujal Dattarao Bhakare
The University at Buffalo Lunabotics team is developing a fully autonomous lunar excavation rover for the NASA Lunabotics Competition. The rover navigates a simulated lunar environment, excavates regolith, and deposits it in a designated zone without human intervention under strict time constraints. The system integrates a Jetson-based edge AI computer running ROS 2 for perception and autonomy, alongside real-time microcontrollers for motor control and sensor management. Using stereo vision, depth sensing, IMU data, and encoder feedback, the rover performs localization, obstacle avoidance, and path planning in uneven terrain. Designed for reliability and modularity, the platform combines a reinforced mechanical chassis, an optimized excavation mechanism, and a custom power distribution system to ensure stable high-current operation. The project demonstrates scalable autonomous surface operations for future lunar resource utilization.
CUSTOMER ATLAS
Nishanth Reddy Bonikela, Long Cao, Dharmi Khadela and Milind Kumar
Customer Atlas is an interactive geographic analytics platform developed to replace static data tracking with real-time, map-driven outreach visualization. Designed for researchers, organizations, and investors, the platform transforms raw engagement data into actionable geospatial insights. This semester, the team established a robust technical foundation, including a scalable Postgres/PostGIS database, a dynamic map interface with advanced filtering, and automated data pipelines. Key achievements include the integration of Supabase for secure authentication, token-based registration workflows, and a responsive HTML email system. The resulting MVP successfully demonstrates regional engagement trends and streamlines data management compared to traditional spreadsheets. With a functional infrastructure for data intake and visualization now in place, future development will focus on global scaling, performance optimization, and the implementation of AI-powered predictive analytics. Customer Atlas provides a professional, scalable solution for identifying outreach gaps and making informed, data-driven strategic decisions.
ROBUST DETECTION OF SYNTHETIC GUNSHOT AUDIO IN REAL-WORLD SCENARIOS
JaeHyeong Chang and Chengzhe Sun
This project addresses the emerging threat of AI-generated environmental deepfakes used to trigger false emergency responses or mask actual crimes. While AI voice cloning is well-studied, the synthesis of high-impact environmental sounds-such as gunshots, glass breaking, and explosions-presents a critical security vulnerability. We will develop a robust classifier capable of distinguishing between authentic acoustic signatures and those produced by state-of-the-art Text-to-Audio (TTA) and Audio-to-Audio (ATA) generators.
PERFORMATIVE ANALYSIS OF ELECTRIC VEHICLE (EV) CHARGING STRATEGIES
Jordan Ngamaleu-Kemta
This research examines how electric vehicle (EV) charging can contribute to a more sustainable energy future. We hypothesize that smart charging with reactive power support would be most effective in mitigating voltage drops. Five strategies are considered: (i) uncoordinated charging, where EVs charge without regard for grid conditions; (ii) smart charging, which schedules EV charging based on voltage constraints; (iii) vehicle-to-grid (V2G), enabling EVs to discharge energy back into the grid; (iv) smart charging + reactive power support;(v) V2G + reactive power support where EV inverters provide reactive power to mitigate the adverse effects of high EV penetration. The strategies are implemented and tested on a 33-bus distribution network over a 24-hour horizon. Results highlight the trade-offs between grid performance and EV charging flexibility. It is evident that smart charging with reactive power and vehicle-to-grid charging with reactive power are the best strategies that minimize voltage drops.
DESIGN OF DUCTED HEAT EXCHANGER FOR FUEL CELL POWERED ELECTRIC AIRCRAFT
Jingye Guo
This project will discover required MW-level heat rejection system on fuel cell powered electrical aircraft. We use a computational model to calculate heat flux, drag, specific cooling power and coolant pumping power during taxi, takeoff, climb, acceleration and cruise. The effect of altitude and aircraft speed on ducted heat exchanger (DHEX) will be explained. The calculated results enable engineers to optimize the ratio of duct inlet size to radiator core size according to specific parameters, such as drag, cooling power, and volume of ducted heat exchanger. While DHEX is effective for cooling in most flight phases, it can't provide enough cooling power during takeoff due to limited air intake and high ground temperatures. This insufficient cooling capacity necessitates the use of a hybrid power system combining FC and turboelectric generator (TEG) for takeoff.
FPGA IMPLEMENTATION OF SHORT FFTS
Margot Barry
A method for implementing a short 16-point FFT on a low-cost 7-series FPGA is presented with the goal of optimizing device area, number of hardware multipliers used, and number of complex multiplications required. The memory required for the input samples is only 4-deep and does not require complete bit reversal, and the design retains a complexity of O(n*log(n)).
AUTOMOTIVE TRANSMISSION CONDITION ASSESSMENT USING DATA DRIVEN DIAGNOSTICS
Reilly Popa
Assessing automatic transmission condition in inspections is challenging due to limited testing time, varying operating conditions and sensor placement constraints. This work focuses on the development and evaluation of practical testing procedures designed for operational use, including a stationary shift test and a controlled drive test. These procedures enable synchronized acquisition of on-board diagnostic parameters along with audio and vibration measurements during short, repeatable test sequences. Data was collected across approximately 200 vehicles, including those with known transmission issues, and the procedures were iteratively refined to ensure consistency and feasibility in real inspection environments. Preliminary analysis primarily focused on on-board diagnostic data to explore relationships between collected signals and reported transmission conditions. Building on this foundation, ongoing work investigates vibration signals as a potential additional source of condition information.
CHARACTERIZING MOS2 AND SNS2 FROM FLAKES TO FUNCTION
Moumiza Hasan, Pranati Reddy Kuntla and Lunfu Qi
MoS2 and SnS2 are promising 2D semiconducting materials for nanoelectronic devices, and their performance strongly depends on flake thickness, transfer quality, and fabrication conditions. This project presents a step-by-step workflow for MoS₂ and SnS₂ device fabrication and characterization, beginning with manual mechanical exfoliation of flakes from bulk crystals. Suitable flakes are identified and transferred onto cleaned chips, then dry-transferred onto a marker-patterned silicon chip for device fabrication. After device formation, electrical characterization is performed using output and transfer measurements (ID/VD and ID/VG), which are used to evaluate current modulation and extract basic device metrics.The material verification is carried out using Raman Spectroscopy where Raman peak positions are identified in order to confirm the material identity, thickness and uniformity. This project overall covers the fabrication choices for measure devices providing a practical workflow for MoS2 and SnS2 device development.
SPIR PROJECT - RF LIDAR BASED PROPAGATION MODEL
Jonah Hakobian-Leone, Labid Hossain, Krish Puwar, Alvi Rashid and Nicholas Toutoundjian
As the frequencies utilized for wireless signal transmissions continue to increase with next generation cellular networks, these signals will suffer heavy attenuation due to obstacles blocking the line-of-sight (LOS) path from transmitter to receiver. These obstructions may be detrimental to network performance, therefore it will be increasingly important to accurately model this behavior before a prospective network is deployed. Through the use of light detection and ranging (LiDAR) data, a geographical digital twin of an area can be constructed that captures relevant obstacles. This digital twin can then be used as input to a propagation modeler, allowing the model to operate with a high degree of environmental awareness, allowing for greater simulation accuracy.
ANALYZING PRELIMINARY K-12 STUDENT SKETCHES WITH ELABORATION RUBRIC
Nischal Sunar
The preliminary sketches represent several ideas from the same problem conducted in K-12 setting via MODs platform is a central activity in the design process (Ghahremani et al., 2024; Nikolic et., 2019). Students engage in the web-based front-end platform for designing projects focused on Earth Science or environmental topics such as water conservation and micro-plastics covering eight lessons. This study was organized at the Mid-Atlantic University to rising 9th grade students for an hour with the consent form signed from the participants by their parents. There were 24 different sketches collected from the experiments and analyzed using the methods of categorizing drawing into five distinct phases: Holistic, Component, Details, and Interconnected. These results were presented into stack bar char and Correlation Heatmap to have more descriptive understanding. Rubric has been developed from this idea for analyzing sketches that helps researchers to distinguish drawing holistically.
THE TROUBLE WITH TROUBLESHOOTING: INVESTIGATING UNDERGRADUATE ENGINEERING STUDENTS' CHALLENGES WITH TROUBLESHOOTING VIA PROCESS AND KNOWLEDGE TYPE APPLICATION
Christopher Romeo
Undergraduate engineering students often struggle with troubleshooting technical issues that naturally arise during essential class activities, like lab experiments and course projects. However, existing research on this topic has not explored why, or in what ways, troubleshooting is such a challenge for them. My dissertation study addresses this gap by trying to understand the difficulties that undergraduate engineering students face while troubleshooting. This qualitative case study will utilize observations, problem-solving interviews, process artifacts, and reflexive thematic analysis methods to identify and characterize the predominant ways in which electrical engineering students struggle with troubleshooting a circuit, and the accompanying reasons why. It will leverage theories of problem complexity and ill-structuredness, troubleshooting process, knowledge type application, and differences in expert-novice troubleshooting strategies and practices. The knowledge gained from this study will be used to develop research-based instructional scaffolds that support the design and facilitation of dedicated troubleshooting exercises for students in engineering classes.
DESIGNING FOR BELONGING: THE ROLE OF CURRICULUM-BASED INTERVENTIONS IN FOSTERING STUDENT ENGAGEMENT IN ENGINEERING
Monica Perez
Existing engineering education typically depends on short interventions that lack theoretical grounding, often overlooking the intricate processes of student belonging and the development of identity. This research evaluates sustained, multimodal pedagogical intervention within introductory engineering courses. Using a mixed-methods framework, the study employs a longitudinal quasi-experimental design to track belonging and identity shifts across two semesters, complemented by qualitative interviews and journey-mapping to analyze lived student experiences. The findings highlight the critical necessity of replacing isolated activities with long-term, classroom-level support structures. By isolating the "active ingredients" within curriculum design, this study provides actionable empirical evidence to improve student retention and empower students in constructing their professional engineering identity, offering a scalable, theory driven model for educators to foster sustainable inclusion.
INTEGRATING AI LITERACY INTO ENGINEERING EMPLOYABILITY: EXAMINING GAPS ACROSS STAKEHOLDERS
Muhammad Ali Sajjad
Artificial intelligence (AI) is rapidly transforming engineering practice, yet existing employability frameworks do not adequately account for AI-related competencies. This study examines the intersection of AI literacy and engineering employability by investigating how students, educators, and employers perceive the importance of AI literacy and the preparedness of engineering graduates. Drawing on literature, AI literacy is conceptualized across five dimensions i.e. technical understanding, application of AI tools, critical evaluation, ethics and societal impact, and meta cognition. A mixed-methods approach is employed. The quantitative phase uses a survey to compare stakeholder perceptions of importance and preparedness across these dimensions, enabling identification of misalignment. The qualitative phase involves interviews to further explore how stakeholders conceptualize AI literacy within employability. The findings aim to inform curriculum design and support the development of an adapted employability framework that integrates AI literacy competencies.
TRANS-FORMING THE SPACES OF ENGINEERING
Roland Orselli
This project examines the spaces in which undergraduate transgender and gender nonconforming (TGNC) engineers feel comfortable expressing themselves or being their most authentic selves. Queer students in Science, Technology, Engineering, and Mathematics (STEM) often face a "chilly" environment due to issues like under-representation and a dominant masculine culture. With a rise in TGNC students entering higher education, there is a need for educators and staff to prioritize their needs and create safe spaces. Preliminary work shows that TGNC engineers find comfort in physical and virtual spaces, with a majority of our participants having a deep connection to artistic spaces and forms of expression. The results of this analysis have implications of expanding knowledge on safe spaces and TGNC joy, as well as improving visibility of TGNC individuals' needs in higher education.
EXPLORING HOW ACCESSIBILITY AND ETHICS PLAY A ROLE IN DEVELOPING A SUSTAINABLE MINDSET FOR UNDERGRADUATE ENVIRONMENTAL ENGINEERS THROUGH PEDAGOGY
Valerie Sullivan
This project addresses a crucial gap in undergraduate Environmental Engineering education. The lack of intentional integration of ethics, equity, and accessibility in course design. Traditional engineering instruction often centers on technical problem-solving while neglecting the historical, social, and cultural systems that shape environmental challenges, such as colonialism and capitalism. Through a qualitative study of Environmental Engineering courses, this project explores how instructors incorporate (or omit) inclusive and ethical pedagogical practices and how these choices influence students' understanding of their relationship with the environment. The findings will inform actionable strategies to build more equitable and sustainability-oriented curricula that support diverse student populations, including neurodivergent learners.
WINNY'S CLOSET RECONSTRUCTION
Akosua Adu, Hayley FitzPatrick, Charlie Goodell, Huda Mahul and Renicia Roper
The Delavan Grider Community Center is in the heart of the East Side of Buffalo, commonly providing a safe space for residents to engage in recreational activities, events, and support with necessary supplies. Ms. Winny, a community advocate, helped lobby for a new community center, which is now the Delavan Grider Community Center. In dedication to her advocacy, a donation closet named Winny's Closet was made. This closet originally started out as a job attire initiative, now serving First Generation Americans and people within the community. The community center is seeking a solution that helps them monitor and manage their inventory within the closet. Through extensive research into safety standards, inventory management principles, and process improvement methodologies, we are working towards developing a solution that adheres to the closet's requirements. We have decided to implement methods such as zoning, lean six sigma, and digitized inventory management.
SECOND PAIR OF EYES': SURGEONS' VISUAL ORIENTATION STRATEGIES AND PERCEPTIONS FOR GAZE GUIDANCE IN LAPAROSCOPIC CHOLECYSTECTOMY
Taylor Quinn
Artificial intelligence (AI) tools offer new opportunities to support human perception and provide gaze guidance in surgery. However, there remains a need to better understand surgeons' perceptions of these tools and to identify desired features across levels of expertise. This study explored surgeons' perceptions of AI-based gaze guidance to determine how such tools can best meet surgeons' needs while recognizing their limitations. Seventy-one surgeons (38 attendings, 33 residents) watched laparoscopic cholecystectomy videos and responded to open-ended questions about their perceptions of AI for gaze guidance in surgery. Responses were analyzed using latent content analysis to identify themes and desired AI features. Each response was also categorized as positive, negative, or neutral to assess overall perception. Attendings (92%) and residents (82%) were positive about AI for visual gaze guidance. Neutral responses (9.8%) reflected uncertainty or limited familiarity with AI. Desired features included directing attention, identification, highlighting and visualization, feedback, decision support, educational tools, communication and collaboration, and planning and mapping.
FAIR TRAIN: ANNOTATING TENSILE TESTING DATA USING ONTOLOGY
Laibah Ahmed
Experimental data is often stored in semi-structured formats which limits their interoperability, reusability, and integration into AI-workflows. This project presents an ontology-based annotation pipeline that can annotate data, namely tensile testing data, and transform it into machine-readable knowledge graphs. We used tensile test data collected from student workshops, and mapped data measurements and properties to standardized ontology terms. Python scripts were developed to parse the CSV files of the data and create RDF triples that connect them to instances of ontology terms and relationships. The triples were then serialized into TTL and JSONLD formats, ready to be interpreted by machines. This project demonstrates how ontology-based approaches can make data more open, enhancing FAIR data practices and streamlining future data reuse and analysis.
MODELING QCM DOPANT UPTAKE VIA THE DIFFUSION EQUATION
William Dove
This project provides a theoretical basis for observed dopant uptake time profiles in swollen pBTTT films. Dopant uptake was monitored using quartz crystal microbalance (QCM) methodology, with frequency shift converted to mass uptake via the Sauerbrey relation. Experimental uptake profiles were compared with analytical solutions using one-dimensional Fickian diffusion. The uptake profiles deviate from simple analytic solutions, suggesting a variable diffusion coefficient or nonlinear behavior. Further comparisons with solutions for step function, and cubic basis spline diffusion coefficients imply a spatially variant diffusion coefficient in one dimension is not sufficient to explain the empirical results. These results indicate that dopant transport in swollen pBTTT films is governed by nonlinear or coupled processes beyond simple Fickian diffusion, relating to swelling and tortuosity of the film.
ACOUSTIC FIELD DRIVEN IN SITU MIXING FOR UNIFORM FILLER DISPERSION IN ADDITIVE MANUFACTURING OF POLYMER COMPOSITES
Dilasha Thapa
This work presents a proof of concept for acoustic streaming assisted in-situ mixing to address the fundamental challenge of maintaining uniform filler dispersion during composite vat photopolymerization. A ring-shaped PZT transducer was integrated into a customized DLP platform to induce recirculating flow in the resin vat for continuous mixing of filler particles during printing. Quantitative particle tracking experiments showed that particle velocity increased with both applied voltage and particle concentration. In printing trials, acoustic streaming mitigated sedimentation and maintained homogeneous particle distribution throughout the printed height across different layer thicknesses. Mechanical testing further demonstrated a ~65% reduction in anisotropy relative to the non-streaming condition. Together, these results establish acoustic streaming as an effective in-situ mixing method for improving filler homogeneity in composite DLP printing.
COMPUTATIONAL ASSESSMENT OF A PLUME-DRIVEN SCALING LAW FOR MIXTURE FRACTION INHOMOGENEITY IN COMPARTMENT FIRES
Maximus Wunderlich
This study computationally evaluates a plume-driven scaling theory relating mixture fraction variance to fire intensity and ventilation conditions. Large-eddy simulations (LES) are performed using Fire Dynamics Simulator (FDS) for a canonical ventilated methane enclosure to isolate turbulent mixing between fuel and air within the buoyant plume. A mesh convergence study using high-performance computing resources ensures grid independence. A strong scaling analysis complemented this effort, optimizing parallel processing efficiency and the distribution of the computational workload across the high-performance computing nodes. A parametric test matrix spanning fire intensity and ventilation conditions is then used to assess agreement with the proposed scaling behavior. The results connect fundamental plume dynamics to macroscopic combustion behavior and support improved modeling of efficiency and pollutant formation in compartment fires, with potential applications in cleaner and more efficient combustion system design.
DESIGN AND DEPLOYMENT OF A HIGH-ALTITUDE BALLOON PAYLOAD FOR ATMOSPHERIC POLLUTION MONITORING OVER WESTERN NEW YORK
Michael De La Cruz
Urban and industrial emissions are increasingly impacting air quality, making it vital to understand pollution distribution across altitudes. Traditional ground-based systems have limited resolution which leads to difficulties in identifying the pollutants in the atmosphere. To address this, our mission was to use a high-altitude balloon to gain a more detailed understanding of pollution distribution across altitudes. We developed a payload controlled by an Arduino Uno R4 Minima microcontroller. To accurately measure the atmospheric pollution across different altitudes, we needed to measure the UV penetration, temperature, atmospheric pressure, altitude, and ozone concentration throughout the flight. We used UV sensors, a barometer, a temperature sensor, and a parts per million sensor to collect our data. All data was time-stamped and stored on a micro-SD and the system ran autonomously throughout the flight. The high-altitude balloon payload provided results that we used to model the atmospheric pollution. Its flight was determined by lower atmospheric winds and jet stream winds. Though constrained by wind-determined trajectories, balloon missions gather high resolution atmospheric data beyond the reach of ground-based systems.
PHYSICAL REALIZATION OF A TUNED MASS DAMPER SYSTEM
Nyah Cardona
Machines exhibit erratic vibrations when the operating frequency matches the natural frequency of the machine, a phenomenon defined as resonance. These resonant vibrations cause irreversible damage and impairs the lifespan of these machines. To mitigate these resonant vibrations, Tuned Mass Dampers Systems (TMDs) are implemented, which absorb vibrations from the main body of the machine at critical frequency, by adding a tuned secondary component. Through developing an undamped lumped parameter system with a point mass and a stiffness element TMD model that operates at lower frequencies, the motion of these systems can be analyzed for data acquisition. This research aims to find significant changes in the motion of the model without special equipment and illustrate the vibration absorption mechanism. The model will aid as a tool for data collection and educational purposes. With further modifications, other systems can be modeled and measured, allowing further understanding of the applications of vibrations.
NOZZLE-ASSISTED CONTINUOUS 3D PRINTING OF MESOSCALE MULTIMATERIAL STRUCTURES
Uma Bhattacharjee
Digital Light Projection (DLP) has recently become a powerful additive manufacturing method for printing complex 3D structures with great precision. However, there are challenges regarding the current techniques involving the material, design complexity and other manufacturing constraints. To address these issues, a hybrid DLP-based process named Nozzle-Assisted Continuous Printing was developed. This process uses nozzle-based material deposition and continuous solidification to reduce printing times and to maintain high part quality. One focus of this project is the use of computational fluid dynamics to determine the optimum single nozzle orientation and multi-nozzle gradient during the continuous material deposition into the resin tank. Three 2D models were created in ANSYS Fluent to simulate three different input nozzle angles (outward, straight, inward). Additionally, seven 3D models were used to determine the relationship between nozzle gradient and flow rate. Another focus of the project was to print mesoscale structures with the specified nozzle choices.
UB LUNABOTICS AUTONOMOUS MOON ROVER
Ethan Kenney
The UB Lunabotics team has spent the past year developing a fully autonomous lunar excavation rover for the NASA Lunabotics Competition. The rover navigates a simulated lunar environment, excavates regolith, and deposits material autonomously under strict time constraints, integrating ROS 2-based autonomy, edge AI perception, embedded control, and a custom power distribution system.
HYDROGEN NUCLEATION DYNAMICS DURING WATER ELECTROLYSIS
Sriram Metta and Nicholas Schuck
Electrochemical gas evolution governs the efficiency of green hydrogen production. Although progress in electrode development and cell design has been made, gas bubble formation on electrode surfaces remains limiting, as accumulated bubbles block active sites and increase ohmic resistance. This research investigates bubble nucleation during water electrolysis. Using the Young-Laplace equation, accounting for contact angle, surface tension, and supersaturation ratio, the critical cavity radius for nucleation was calculated. Nickel electrodes were engineered with controlled nanoscale cavity arrays based on this radius, enabling systematic study of nucleation site density beyond natural surface roughness. Polarization curves showed electrodes with higher roughness and indentation had a 1.57% reduction in overpotential across 0-0.5 A/cm², indicating micro/nanostructures add nucleation sites with lower energy barriers. However, excessive indentation caused overcrowding and reduced performance at high current densities, highlighting the need to optimize nucleation site density to improve electrolysis efficiency.
INFRASTRUCTURE-GUIDED CONNECTIVITY-ENHANCED ROAD CRACK DETECTION AND ESTIMATION
Yamini Ramesh, Rishabh Shukla, Swarat Sarkar and Haosong Xiao
In this work, we report the world's first infrastructure-guided communication-enhanced road crack detection pipeline that is effective and implementable on passenger vehicles. We first design a customized communication protocol to transmit the region of interest from the infrastructure to the vehicle. With proper camera image processing (e.g., dynamic cropping and frame selection), the focused images are provided to the crack detection model. Leveraging state-of-the-art crack detection model backbones and a carefully prepared dataset comprising a forward-facing view with a crack, we train the model to improve crack-detection performance. We demonstrate the full detection pipeline on an experimental vehicle platform and showcase the detection effectiveness, and project future research directions.
DESIGNING A WATERPROOF CASE TO CAPTURE THE ULTRASONIC CALLS OF THE BENGAL SLOW LORIS
Sara Alabyadh, Ariana Chatlani, Kevin Lin and Raymond Zeng
Working with Pointdexter Lab and the School of Engineering and Applied Sciences's (SEAS) DREAM lab, this project focused on the design of a waterproof case to house a MicroMoth acoustic logger. This case is to be deployed in rainforests within Thailand and Vietnam to further research being done by Pointdexter Labs to record and analyze ultrasonic vocalizations of an endangered primate species, the Bengal slow loris. This presentation describes the student engineering team's efforts to follow a standard design approach, develop technical requirements based on client needs, ensure the design meets IP-67 waterproof standards and optimizes acoustic performance. This project highlights differences between engineering principles in theory and their practical application.
A DIGITAL TWIN FRAMEWORK FOR AUTONOMOUS UAV TRACKING WITH GAZEBO-BASED SIMULATION AND SIM-TO-REAL VALIDATION
Awais Ahmad, Aditya Bhatt, Soham Mehra, Urvish Shah and Yaswanth Thota
Autonomous tracking of flying aerial objects has become fundamental capability for next next-generation aviation whose implications affect both civilians and defense sectors. A learning-based YOLO object detector is employed to estimate the target UAV's image-plane location, which is subsequently used by an Image-Based Visual Servoing (IBVS) controller to generate real-time pursuit commands. While civilian applications leverage this technology in pursuit of drifting subjects during search and rescue operations, and to facilitate infrastructure following behaviors during inspection missions, analogous capabilities are required within the defense sector to enable effective threat mitigation. In such high-stake operations, human-in-the-loop control latencies renders manual intervention impractical, necessitating fully autonomous on-board systems with millisecond-level reaction capabilities. Validating these high-frequency control loops directly in the physical environment is logistically untenable due to hardware fragility and inherent safety risks. Furthermore, modern learning-based methodologies demand extensive training episodes that far surpasses the operational constraints imposed by the hardware such as battery endurance and field deployment. These limitations necessitate the adoption of Digital Twin framework as primary testbed, enabling safe, exhaustive algorithmic optimization prior to deployment in real-world settings. To bridge the gap between theory and development, we implement this high-fidelity Digital Twin using the Gazebo simulation environment, which is particularly well suited for UAV research due to its seamless integration with Robotics Operating System (ROS). Unlike other alternative simulators, Gazebo provides full interface parity with physical flight experiments by exposing identical communication pathways through MAVROS and pymavlink, including the same topics, services, and message definitions as with a real flight controller equivalent. This fidelity facilitates robust Software-in-the-Loop (SITL) validation, enabling the direct deployment of tracking algorithms onto real-world UAVs with minimal reconfiguration. This accurate interface replication is pivotal for minimizing Sim-to-Real transfer errors. By validating against realistic constraints early in the design cycle, the Digital Twin offers strong assurance that theoretical performance will be preserved in actual flight scenarios. Thus, the Gazebo-based Digital Twin provides reliable testbed for developing and validating autonomous UAV algorithms prior to real-world deployment.