We present a system for detecting user fear in virtual reality (VR) environments using motion-based behavioral data. Motion data from VR headsets was collected from participants playing various VR applications and processed to extract meaningful features for a machine learning model. Our model was trained to identify when a user is likely or unlikely to be experiencing fear, achieving 80% accuracy on unseen users. Integrating the model inside VR applications enables real-time fear detection as a safety mechanism for those who are sensitive to distressing stimuli. For example, the system can support adaptive interventions such as warning notifications or adjustments to the virtual environment to mitigate discomfort. To the best of our knowledge, this work is among the first to predict users’ emotional states directly from VR headset motion data.

Dynamic control of mid-infrared optical properties is essential for thermal management, chemical sensing, and infrared camouflage, yet achieving large modulation depths with broad spectral coverage remains challenging. We present an electrically tunable metafilm absorber operating across the 8 to 14 micrometer atmospheric window that exploits voltage-controlled epsilon-near-zero physics in an InAs p-n junction. The device integrates an ultrathin 7 nm n-type layer at the metal-insulator-metal cavity interface where gap plasmon fields concentrate, enabling efficient carrier modulation in the optically active region. Electron diffusion creates a graded carrier distribution that produces distributed epsilon-near-zero conditions coupling to different wavelengths for broadband response. A thick 520 nm p-type layer provides dual functionality as both p-n junction component and non-resonant spacer, leveraging asymmetric effective masses in III-V semiconductors so that ENZ-enhanced absorption occurs selectively in the electron-modulated region. Under reverse bias from 0 to -10 V, the device achieves 86.1 percentage point absorption modulation at 12.2 micrometers with 88% average relative modulation across the atmospheric window. This work extends voltage-controlled ENZ concepts into the true mid-infrared by combining a p-n junction architecture with gradient ENZ physics for broadband reconfigurability, with principles applicable to other wavelength ranges and material platforms.

Cardiovascular disease (CVD) accounts for 31% of global mortality, with higher death rates in developing regions. Early detection of interrelated CVD subtypes, e.g., myocardial infarction (MI) and heart failure (HF), is critical, yet conventional laboratory testing remains inaccessible and single-biomarker-based in many point-of-care settings. Here, we present a rapid, low-cost, and multiplexed diagnostic platform based on a deep learning-enabled dual-mode vertical flow assay (xVFA) for detecting 3 critical cardiac biomarkers: cTnI, NT-proBNP, and CK-MB within 23 minutes using 50 µL of serum. The proposed system integrates colorimetric and chemiluminescence sensing modalities within a paper-based xVFA, enabling both high sensitivity and a wide dynamic range. A neural network is employed to analyze assay signals from a Raspberry Pi-based mobile reader and yields robust quantification performance (Pearson’s r > 0.96 vs. reference assays) in blind testing on 92 patient serum samples. This deep learning-enhanced xVFA platform offers a promising solution for accessible, high-performance point-of-care diagnostics, particularly in resource-limited settings, with the potential to significantly improve early CVD detection, cross-biomarker level-based risk stratification, and patient outcomes, while offering shorter turnaround times and lower per-test costs.

This study investigates cerebrospinal fluid (CSF) transport in the brain’s glymphatic system using CFD simulations in COMSOL Multiphysics with watertight anatomical CAD models and a proprietary CSF fluid property. Time-dependent simulations were conducted under varying effective longitudinal flow parameters (ELFP) to analyze wave-driven CSF dynamics, comparing full-scale CAD and reduced-order models (ROMs) in MATLAB and Excel. The ROM approach preserved high accuracy while reducing computational runtime from hours to seconds, enabling efficient analysis of velocity and pressure distributions. Building on these results, we are developing large language models (LLMs) to predict CSF flow behavior across different ELFP values, creating a framework that accelerates flow modeling beyond conventional CFD simulations.

Minimum-time trajectories set the theoretical limit for aggressive quadrotor maneuvers. Hehn (2012) presented an indirect method, based on Pontryagin's Minimum Principle (PMP), for calculating minimum-time trajectories with switching-time optimization. However, the switching times were arbitrary, and the subsequent boundary value problem was computationally expensive. Moreover, the PMP approach was hard to generalize to more complicated quadrotor aerodynamics. In this project, I present a direct collocation method for solving minimum-time 2D quadrotor trajectory problems and compare the results with those of an indirect method. The results show that direct collocation recovers the bang-bang and bang-singular structure with much less computational effort. Future work includes generalizing the solver to 3D and incorporating unsteady aerodynamics. The codes and example notebooks are compiled into an open-source Julia package, MinTimeQuadTraj.jl, for convenient computations of minimum-time trajectories.