This dataset contains the original drone video segments captured during the collaborative multi-drone experiment conducted by KAIST and EPFL in the Songdo International Business District, South Korea, from October 4–7, 2022. These video segments represent the pre-processed, stable hovering footage used to generate the publicly available Songdo Traffic (DOI:10.5281/zenodo.13828384) and Songdo Vision (DOI:10.5281/zenodo.13828408) datasets.

Localizing gas sources is a challenging task due to the complex nature of gas dispersion. Informative Path Planning (IPP) plays a crucial role in guiding robots to sample at high-information positions, thereby accelerating the estimation process. Existing probabilistic gas source localization methods often require robots to halt at sampling positions, averaging gas measurements over time. Consequently, when selecting the next sampling position, information gains are usually computed precisely through computationally heavy procedures, limiting evaluations to a small set of potential positions. In our previous work, we introduced a sense-inmotion strategy that eliminates the need for prolonged stops at sampling points, therefore allowing the incorporation of measurements taken during robot movement. Building upon this advancement, we propose to extend information gain evaluation in a more continuous manner, from a point evaluation to a path evaluation. However, existing IPP methods are too computationally expensive when transitioning from goal-based to region-based evaluations. To address this challenge, we first assess three lightweight information extraction metrics. Based on the selected metrics, we propose a novel IPP algorithm that computes cumulative information gain along the robot's path and dynamically prioritizes exploration or exploitation based on the uncertainty of the source estimation. The proposed method is extensively evaluated through both high-fidelity simulations and physical experiments. Results show that our proposed method consistently outperforms a benchmark state-of-theart method, achieving a 40% increase in source localization success rate and halving the experimental time in challenging environments.

Digital Twin (DT) technology has emerged as a transformative tool in urban flood risk management (UFRM), enabling real-time data integration, predictive modeling, and decision support. This systematic review synthesizes existing literature to evaluate the scientific impact, technological advancements, and practical applications of DTs in UFRM. Using the PRISMA 2020 framework, we retrieved 1085 records (Scopus = 85; Web of Science = 1000), merged and deduplicated them using DOI and fuzzy-matched titles, screened titles/abstracts, and assessed full texts. This process yielded 85 unique peer-reviewed studies published between 2018 and 2025. Key findings highlight the role of remote sensing (e.g., satellite imagery, IoT sensors) in enhancing DT accuracy, the integration of machine learning for predictive analytics, and case studies demonstrating reduced flood response times by up to 40%. Challenges such as data interoperability and computational demands are discussed, alongside future directions for scalable, AI-driven DT frameworks. This review identifies key technical and governance challenges while recommending the development of modular, AI-driven DT frameworks, particularly tailored for resource-constrained regions.

Emerging technologies and the ubiquitous processing of all kinds of information – personal and private, governmental and public – have created new problems and dilemmas in society, thus requiring fresh approaches in both theory and observation. The 2016 SITE Mini-Track begins with a session on "IT, Trust and Strategy." The second session is entitled "Cloud Pricing, Innovation Tournaments and the Sharing Economy." It showcases a range of contemporary issues that have been discussed over the years in this mini-track. gainst the retailers. The third session is on "Connectivity, Social Networks and Third-Party Payer Business Models." The mini-track will conclude with an open discussion and debate about the highlights, ideas, and solutions that were presented and discussed over the course of the mini-track sessions. Each year, we engage in this process to encourage next steps and further study of important business, social and policy issues, as well as to collaboratively identify new directions for research.

Understanding animal behavior is a critical goal for neuroscience. Computer vision contributes to neuroscience by providing automated, scalable analysis for animal movement and interaction. In controlled laboratory settings, pose estimation has proven highly effective in capturing the abstract structure and movement of animal bodies. This has led to the development of multi-animal pose estimation models to support research on social behavior. We present a novel architecture DLCRNet for multi-animal pose estimation on four multi-animal benchmarks in laboratory settings, targeting social interactions where animals are frequently occluded. Furthermore, we introduce BUCTD, a novel two-stage method that achieves state-of-the-art performance on human and animal benchmarks designed for crowded scenarios. While these advancements improve our understanding of behavior in controlled settings, studying animals in the wild require individual-level behavioral understanding over time and space, where re-identification (ReID) becomes crucial for tracking individuals across diverse environments and time periods. To address this, we develop PoseSwin, a novel ReID architecture that incorporates bodypart-aware features to distinguish individual Alaskan brown bears for longitudinal monitoring across years, habitats and viewpoints. Finally, to bridge the gap between foundational research and real-world application, we apply computer vision techniques to zebrafish welfare assessment. We aim to build a reliable fish behavioral dataset and a robust model for automated welfare monitoring. Although full behavioral dataset is still being collected, we present a data pre-processing pipeline and tool for annotation and conduct simulation experiments for abnormal behavior detection on a public dataset. These efforts provide practical insights into data collection and foundational tools for developing supervised methods in future studies.

Three-dimensional cell culture models (3D-CCMs), such as organoids and spheroids, are powerful tools for studying organ development and drug testing. Histology is a valuable imaging technique for its ability to reveal tissue architecture and biomarker localization at arbitrary depths within 3D-CCMs. However, conventional histological workflows are often low-throughput and labor-intensive, primarily due to the random embedding of 3D-CCMs within hydrogel blocks. To address these limitations, this thesis developed and validated two micromanipulation platforms based on gravitational and acoustic principles to enable precise spatial organization of 3D-CCMs during embedding. These platforms successfully meet the need for high-content histological analysis and are ready for implementation in laboratory, where they can support more efficient and standardized workflows. We first developed the HistoBrick platform, a structured hydrogel block containing microwells that passively aligned 3D-CCMs of homogeneous sizes on a single plane via sedimentation. The HistoBrick also supported the co-embedding of multiple experimental conditions, significantly increasing throughput of histological workflows and reducing reagent use. A major contribution of this work was the development and validation of a novel composite hydrogel composed of poly(ethylene glycol) diacrylate (PEGDA) and gelatine, specifically designed for cryosectioning. This PEGDA-gelatine hydrogel enabled seamless fabrication of the HistoBrick while preserving fragile tissue integrity, such as photoreceptors in retinal organoids. By improving sample preparation, the HistoBrick enabled long-term analysis of photoreceptors in retinal organoids, which was previously too time- and labor-intensive using conventional embedding methods. While the HistoBrick platform effectively aligned homogeneous 3D-CCMs, the reliance on passive sedimentation limits equatorial alignment in heterogeneous samples. To address this limitation, we developed a second platform that uses acoustic levitation to achieve precise equatorial alignment of 3D-CCMs for enhanced histological analysis. The platform levitated 3D-CCMs within compartmentalized wells, supporting the co-embedding of multiple experimental conditions within a single block. As a result, the platform enabled planar alignment of 3D-CCMs by their center of mass with higher precision than its sedimentation-based counterpart, significantly enhancing the information content of histological sections for size-heterogeneous 3D-CCMs beyond the capabilities of the HistoBrick. After enabling precise 3D-CCM positioning for histology, the scope of this work was extended to explore broader applications of acoustic micromanipulation. To this end, a third platform was developed using a novel transducer: piezoelectric micromachined ultrasonic transducer (PMUT) array. PMUT arrays generated spatially programmable bulk acoustic waves in fluid. We demonstrated, for the first time, deterministic 3D levitation of particles in water, at rest and under continuous flow, by generating standing acoustic waves across the height of the chamber. Additionally, we leveraged spatiotemporal modulation of the acoustic field for continuous planar transport of microparticle aggregates. Future developments, including phased-array operation, are expected to increase the resolution of PMUT arrays and open new opportunities for assembling complex 3D-CCM systems to study multi-organ interactions.

The transition from fossil fuels to renewable energy sources is hindered by the intermittent nature of renewables, necessitating efficient energy storage solutions. H2, a carbon-free energy carrier with the highest mass-energy density (33.3 kWh.kg-1), is a promising option. However, its direct use is limited by safety concerns, including a wide flammability range (4-75 vol%) in air, high flame speed, elevated flame temperature (over 2100 °C), and, consequently, NOx emissions. Therefore, in all the H2 applications, maintaining its concentration below the lower flammability limit in air is crucial. Catalytic H2 combustion (CHC), therefore, emerges as a promising alternative to overcome these challenges. State-of-the-art CHC catalysts are based on Pt and Pd, which can initiate the reaction at room temperature. However, at high H2 concentrations, the rapid water formation rate will extinguish the reaction. This operational drawback, combined with cost and resource scarcity, underscores the need for alternative catalysts based on earth-abundant transition metals (TMs). Replacing Pt and Pd while addressing their limitations remains a central challenge, necessitating a deeper understanding of reaction kinetics, catalytic mechanisms, and nanoscale engineering of metal particles. To ensure reliable catalytic activity comparisons, we developed a method to calculate metal dispersion that considers nanoparticles' (NP) geometry and crystal structure. It is demonstrated that an incorrect geometry assumption (such as spherical nanoparticles) would introduce errors in dispersion and turnover frequency (TOF) calculations, ultimately leading to unreliable assessments of catalytic activity. We then synthesised a series of TM-Al2O3 catalysts (TM = Pt, Ru, Co, Ni, Mo) and evaluated their CHC activity. Owing to the high activity of the Pt-Al2O3 catalyst, a new low-temperature plug-flow reactor is designed and built. By integrating the multi-ion detection mode of a quadrupole mass spectrometer (QMS) with an analogue inputted thermocouple, a high data acquisition rate is achieved, which is essential for reliably determining the kinetic parameters. Moreover, it is found that the Ru-Al2O3 and Co-Al2O3 catalysts exhibited similar CHC activity with long-term stability. To optimise Ru utilisation and maximise the H2 conversion rate, we tuned the Ru NPs' size to modulate the metal-support interaction (MSI). Advanced spectroscopic and microscopy methods revealed that NP's size influences Ru dispersion, MSI and the ratio between Ru-O and Ru0, which plays a critical role in the CHC activity. Furthermore, using in-situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS), the role of OH groups in the CHC reaction mechanism is identified, a finding further validated by density functional theory (DFT) calculations. Lastly, we investigate the effect of operational parameters on the CHC stability of the Pt-Al2O3 catalyst to control the water-induced deactivation. Using IR-thermography, we found that a higher GHSV mitigates the deactivation. We also studied the dynamic heat evolution during catalyst reduction and CHC reaction propagation, contributing to a deeper understanding of CHC stability. In summary, this thesis advances the fundamental understanding of CHC mechanisms and kinetics, supporting the development of cost-effective, stable, and efficient non-Pt and Pd catalysts and paving the way for their integration into practical H2-based technologies.

Laser Powder Bed Fusion (LPBF) is a promising metal additive manufacturing (AM) technology that enables high-precision fabrication of high-performance parts with unprecedented design flexibility. However, its widespread adoption remains hindered by the occurrence of various defects, which significantly degrade the mechanical properties and reproducibility of fabricated parts. These defects are inherently linked to the complex and dynamic nature of melt pool behavior, governed by a combination of laser-material interaction, heat transfer, fluid flow, and phase transformations.

This doctoral research integrates in-situ X-ray experiments and high-fidelity computational fluid dynamics (CFD) modeling to investigate the underlying mechanisms of melt pool dynamics and laser absorption during LPBF. The work focuses on understanding fluid flow behaviors, laser energy distributions, and pore-related phenomena under various processing conditions, with particular emphasis on the role of the complex Marangoni effect.

The melt flow in LPBF of SS316L under nearly pore-free regimes was statistically quantified with a high resolution of ~10 µm by employing tungsten tracer particles during in-situ synchrotron X-ray imaging, revealing inward Marangoni convection due to surface active elements. This convection was observed to increase the conduction-keyhole threshold, thereby expanding the pore-free process window. Complementary in-situ X-ray diffraction (XRD) of laser melting provided additional insight into temperature evolution, and a new approach was proposed to estimate the melt pool width from the amorphous intensity of the liquid phase.

A high-fidelity CFD model at the powder scale was developed with the OpenFOAM framework and validated against experimental data for both Ti-6Al-4V and SS316L. By incorporating advanced physical modules such as multi-element vaporization, multiple laser reflections, temperature-dependent absorptivity, and an enhanced powder bed model, the simulation can accurately predict keyhole and melt pool morphologies, replicate measured laser energy absorption, and reproduce melt flow and pore formation observed experimentally. Moreover, offset effects of inward Marangoni flow on keyhole formation observed experimentally were captured by simulations and are shown to be valid at practical laser scanning speeds of several hundred mm/s.

A systematic parameter study across different materials and melt pool regimes using the high-fidelity CFD model provided quantitative insights into melt pool dynamics and laser energy absorption under various processing conditions, leading to the derivation of universal scaling laws for laser absorption. In addition, the model was extended to LPBF with copper, which is difficult to process using near-infrared lasers due to its high reflectivity, offering insights into strategies to improve its printability by increasing the laser absorptivity of powder and substrate.

Overall, this work advances the fundamental understanding of melt pool dynamics in LPBF, emphasizing the impact of the complex Marangoni effect and the importance of accurately calibrating the temperature-dependent coefficient of surface tension (CST) in LPBF modeling. The high-quality experimental datasets, validated CFD model, and scaling laws provide a predictive framework for defect mitigation and process optimization in metal additive manufacturing.

The proteome of a cell determines its structural and functional features and needs to be tightly regulated to maintain proper cellular function. The level of each protein in a cell is set by its rate of synthesis and decay. The interplay between these rates determines protein turnover, which can vary strongly between different cell types. These rates exhibit intrinsic variability in vivo and fluctuate according to nutriment availability and other environmental cues. It is known that the inhibition of protein decay and the subsequent accumulation of misfolded proteins triggers the Integrated Stress Response that ultimately represses protein expression. Meanwhile, much less is known about how a change in protein synthesis impacts protein decay in mammalian cells.

This PhD thesis aims to quantitatively demonstrate and dissect the coordination of protein synthesis and decay in mammalian cells. To this end, we utilize well-controlled cellular and perturbation models, state-of-the-art quantitative live-cell imaging techniques and analysis pipelines, a novel Bayesian inference algorithm, and dynamic SILAC. To comprehensively address the research question, we have introduced a novel theoretical understanding of the concept of protein turnover, proposed new analytical tools, and demonstrated how this novel approach may lead to a reconsideration of previously published data and results.

We show that the adaptation of the protein decay to a change in protein synthesis is primarily mediated by a core passive adaptation mechanism unable to maintain protein levels, yet buffering them. Using a simple mathematical model, we were able to quantitatively predict this protein turnover adaptation and protein level fold-change at steady-state. We demonstrate that protein decay adapts to protein synthesis in the 5-10 hours timescale. Moreover, we find that in mouse embryonic stem cells, a facultative mTOR-mediated adaptation adds up to the core passive adaptation, ensuring protein level maintenance.

This work shed light on the dynamic and the intertwining of protein synthesis and protein decay. It also highlights the impossibility of fully disentangling these two fundamental processes of cell biology.