Current direct air capture (DAC) literature contains gaps in CO₂ absorption at very low concentrations. This poses challenges for scaling DAC systems, as there is limited equilibrium data available under ambient conditions. This study investigates equilibrium CO₂ absorption in dilute NaOH solutions at CO₂ concentrations of 300 and 400 ppm, across NaOH concentrations of 0.004% to 4% and temperatures of 5°C, 20°C, and 40°C. Absorption is quantified using DIC analysis and titration to measure carbonate formation, and results are compared to thermodynamic models in Aspen Plus to evaluate the accuracy of modelling tools at dilute CO₂ concentrations. CO₂ absorption increases significantly with NaOH concentration, whereas temperature has a more modest effect. Aspen Plus captures general trends but tends to overpredict CO₂ uptake at higher NaOH concentrations. This study addresses gaps in existing literature and assesses the reliability of thermodynamic modelling tools under ambient conditions, both of which are essential for the industrial scale-up of DAC systems.

The ferrobotic platform offers a decentralized solution for automated diagnostic testing, yet its reliable operation depends on chemically transparent surface coatings and high-performance magnetic nanoparticles. Current reliance on commercial superhydrophobic sprays introduces contamination risks like calcium assay interference; furthermore, robust platform performance requires synthesizing ultrafine, surfactant-coated nanoparticles to ensure colloidal stability while maximizing magnetic response for high-force settings. This paper details the optimization of lab-synthesized components via two protocols: fabricating superhydrophobic surfaces using stoichiometric silanization of octadecyltrichlorosilane (OTS) and water, and synthesizing superparamagnetic iron oxide nanoparticles (SPIONs) via controlled coprecipitation. The synthesized OTS coating achieved water contact angles of 149 ± 4°, creating a hierarchical nanofiber structure essential for low-friction droplet transport, though testing revealed a contact angle decay over ten minutes. Optimizing SPION synthesis through strict kinetic and oxidation controls improved particle yield. Building upon these components, the platform's performance was benchmarked by executing off-chip (NanoDrop) and on-chip diagnostic assays (Basic Metabolic Panel, T4 ELISA). Finally, calibration curves were generated across varied magnetic nanoparticle suspensions to successfully quantify human blood analytes, demonstrating the platform's viability for reliable and precise diagnostic testing.

Traditional rigid robots struggle to traverse irregular terrain, interact with human users, and navigate tight spaces, making them infeasible for search-and-rescue and most human-robot interaction tasks. Therefore, we developed SPROUT, the Soft Pneumatic Robot with Omnidirectional Unfurling Tendrils, which uses multiple tendril-like legs to dynamically change shape and propel itself. The tendrils are based on vine-robots, soft, pneumatic robots made of fabric that grow via tip eversion. Central motors and pneumatic valves control the motion of each tendril leg. Different mechanisms for tendril motion, including cable, pneumatic muscle, and pre-formed shape systems, are being tested for functionality. Future work will enable basic controlled motion and full mobility.

Parkinson’s disease is an incurable chronic condition which manifests itself in a range of symptoms including gait abnormalities. As current clinical assessment methods do not accurately represent the patient’s current state, there is a need for wearable system that monitor disease progression. In this work, a wearable smart magnetoelastic insole was developed for plantar pressure mapping and gait monitoring. The device consists of a 3D printed thermoplastic styrene elastomer (TPS) body that integrates magnetoelastic coupling (MC) units positioned beneath a flexible circuit board with integrated Hall sensors. The device uses the giant magnetoelastic effect of soft systems to convert biomechanical motion to electrical signals. To optimise the sensitivity of MC units, the influence of NdFeB nanoparticle concentration on the Magnetoelastic coupling factor (MCF) was investigated. It was found that the magnetic response of the MC units was highly dependent on nanoparticle concentration, the 80 wt% sample showed the highest MCF with a value of -16.6 mT/MPa closely followed by the 50 wt% sample (-11 mT/MPa). Ultimately, the 50% MC units were selected for device integration because of their lower material cost. The fabricated insole demonstrates strong capability of pressure mapping and gait monitoring. Further work aims to develop the flexible PCB, calibrate the device and extract the relevant biomarkers from the subject’s gait signals.

Achieving conformal passivation in sub-micron, lateral high–aspect-ratio geometries is a critical challenge for advanced semiconductor packaging. This work investigates thermal atomic layer deposition (ALD) of Al₂O₃ for Silicon Interconnect Fabric (Si-IF) applications, focusing on whether limited TMA precursor transport restricts uniform Cu passivation at 10x10 mm^2 dielets. The study hypothesizes ALD transitions from a reaction-limited to a transport-limited regime under typical conditions, where insufficient precursor residence time before purge limits deep-region coverage. Experimental inputs, including uHAST oxidation data and EDX cross-sectional measurements, are used to estimate penetration depth and validate transport behavior. A physics-based framework combining Knudsen diffusion and residence time analysis is applied by comparing diffusion times with precursor exposure times. To address insufficient penetration, a modified multi-pulse ALD sequence incorporating a stop-flow hold step is proposed to extend precursor residence time. Preliminary analysis indicates baseline conditions are insufficient for full penetration in 10×10 mm^2 dielets, requiring increased effective exposure (~8 s hold per half-cycle). A design of experiments (DOE) is developed to systematically evaluate effects of dose, flux, temperature, gap height, and dosing strategy on conformality. This work provides a scalable framework for achieving reliable passivation, directly addressing a key bottleneck in reliable 3D heterogeneous integration.

Vision-Language-Action (VLA) models have demonstrated strong potential for predicting semantic actions in navigation tasks, demonstrating the ability to reason over complex linguistic instructions and visual contexts. However, they are fundamentally hindered by visual-reasoning hallucinations that lead to trajectory deviations. Addressing this issue has conventionally required training external critic modules or relying on complex uncertainty heuristics. In this work, we discover that monitoring a few attention heads within a frozen VLA model can accurately detect path deviations without incurring additional computational overhead. We refer to these heads, which inherently capture the spatiotemporal causality between historical visual sequences and linguistic instructions, as Navigation Heads. Using these heads, we propose an intuitive, training-free anomaly-detection framework that monitors their signals to detect hallucinations in real time. Surprisingly, among over a thousand attention heads, a combination of just three is sufficient to achieve a 44.6% deviation detection rate with a low false-positive rate of 11.7%. Furthermore, upon detecting a deviation, we bypass the heavy VLA model and trigger a lightweight Reinforcement Learning (RL) policy to safely execute a shortest-path rollback. By integrating this entire detection-to-recovery pipeline onto a physical robot, we demonstrate its practical robustness. All source code will be publicly available.

Existing RTL-based FFT libraries lack flexibility and portability to be useful practically, while current High-Level Synthesis (HLS) FFT libraries aren’t performant. Therefore, we aim to build a high-performance FFT library using HLS to provide both flexible and performant FFT designs. In addition, due to the vast amount of possible FFT configurations, we first develop a semantic-first compiler for the Cooley-Tukey algorithm, followed by building an LLM-based agentic workflow to help us automate the creation of new compilers for different FFT algorithms.