In-situ carbon mineralization has the potential to store large quantities of carbon dioxide (CO2), supporting the transition towards a net-zero society. To maximize carbon storage within resource and time constraints, comprehensive research is necessary to understand the parameters that influence carbon mineralization kinetics and thermodynamics. In particular, the temperature and pH conditions during mineral dissolution are hypothesized to affect mineral reactivity and, consequently, carbon storage efficiency. This research project analyzes how varying temperature and pH affect mineral reactivity via mineral dissolution experiments that systematically vary temperature and pH, and solution characterization via ICP-OES. By observing how strongly temperature and pH affect reactivity, insights are gained regarding the key parameters that affect mineral dissolution and can limit in-situ carbon mineralization. Results show a strong effect of temperature on the mineral reactivity of key cation species (Ca2+, Mg2+) for two host mineral compositions during CO2 dissolution experiments. pH-stat (constant pH) dissolution experiments also display similar trends, with increasing temperature from 23°C to 70°C increasing the concentration of Ca2+ and Mg2+. These findings improve understanding of the sensitivity of mineral reactivity to temperature and pH, and provide insights aiding the development of more efficient and scalable carbon mineralization strategies.

Limbal Stem Cell Deficiency (LSCD) is a disease of the eye characterized by the breakdown of stem cells in the cornea. A diagnosis method for the disease applies anterior segment optical coherence tomography (ASOCT) imaging for the purpose of measuring the thickness of the epithelial layer of the cornea, as the thickness of the cornea’s epithelial layer correlates to the stage of severity of LSCD. Currently, annotations from clinically labeled ASOCT images are manually extracted one at a time and employed into an in-development novel machine learning tool for the purpose of improving efficiency of diagnosis of LSCD patients. Within these images, there are columns of light due to light refracting errors when the images are taken. Discrepancies in the layers prevent accurate clinician measurements of the epithelial layer and disrupts related models that rely on segmenting the epithelial layer. Inpainting refers to the process of completing missing portions of an image with predictive pixel values so that the newly added portions match the rest of the image. Our goal is to remove these light columns and replace them with the appropriate surrounding textures using generative artificial intelligence inpainting models.

Membranes encapsulate all cells, demarcating the boundary between a cell and its environment. Selective permeability of membranes regulates the exchange of ions, signaling molecules, and macromolecules making membrane integrity essential for cellular function and survival. As research and therapeutics advance, the ability to controllably modulate membrane permeability for intracellular delivery has become increasingly important. A range of physical methods have been developed to transiently permeabilize cell membranes to exogenous molecules by applying external fields. Such membrane disrupting methods also minimize unwanted immune responses and chemical reactions compared to viral or chemical mediated delivery. Despite the advantages of higher throughput and delivery efficiency, physical poration techniques often compromise cell viability due to irreversible pore formation. Furthermore, the extent that external fields interact with cellular membranes to create pores remains to be studied. To address these challenges, we aim to apply our expertise in electrified Cryo-EM to elucidate the molecular mechanisms of membrane opening, healing, and degradation during electroporation. Preliminary tests successfully visualized electric field-induced membrane damage in Pseudomonas Aeruginosa (PA) through confocal microscopy. Ongoing work focuses on quantifying pore size after electric field applications on PA membranes. By defining thresholds for electroporation, we hope to advance the ability to controllably modulate cell membranes.

Navigating complex city streets with robots usually requires a human to be constantly at the wheel, which is exhausting and prone to error. While AI can help, current systems often force the human and the AI to fight over the same controls, increasing the mental burden on the person. We propose Assistive Urban Robot Autonomy (AURA), a framework that changes how humans and robots collaborate. Instead of micromanaging every movement, the human provides high-level directions (like "turn left at the park") while the AI handles the complex low-level steering and obstacle avoidance. To build AURA, we created UrbanWalks, a massive dataset of real-world navigation and descriptive language. Our results in both simulations and real-world tests show that AURA follows instructions accurately and adapts to new environments. Most importantly, this division of labor reduces the amount of manual work a human has to do by over 75% compared to traditional methods. This research demonstrates a more natural, efficient way for humans to supervise autonomous systems in our daily urban environments.

Depth perception, facilitated by head movement and binocular vision, is a vital yet overlooked component of robotic surgery. Current systems often decouple camera control from surgical limb manipulation, which increases cognitive load and heightens the risk of spatial errors during high-stakes procedures. The Teleoperated Enhanced Robotic Surgical Assistant (TERESA) addresses this limitation by integrating an immersive virtual reality (VR) interface directly into the surgical control loop. The system utilizes Unity to process stereoscopic camera feeds, which are streamed to a Meta Quest 3 headset via a low-latency wireless protocol. This configuration provides a distinct view for each eye to simulate natural depth perception while simultaneously mapping the surgeon’s head movements to robotic actions. Performance analysis reveals that while the current Unity-based framework provides a modular infrastructure, it introduces processing overhead that complicates meeting the 100ms latency threshold required for safe teleoperation. Consequently, a C++-based architecture is proposed to minimize overhead and ensure consistent real-time responsiveness. By maintaining the spatial relationship between surgeon movement and robotic feedback, TERESA demonstrates that consumer-grade technology and open-source tools can offer a viable path toward accessible, high-fidelity surgical systems. This research is significant as it provides a scalable model for reducing surgical risk and democratizing access to advanced medical robotics.