The ongoing energy transition introduces challenges in designing energy systems, particularly at the low-voltage level, where additional loads like electric vehicle charging points and heat pumps are predominantly connected. Obtaining accurate grid data for modeling and analysis is hindered by availability and confidentiality concerns, especially for low-voltage distribution grids. Unlike high voltage grids, low-voltage grids lack standardized handling due to their large size and heterogeneity.This research addresses an essential gap in the availability of grid data, offering a methodological base to generate synthetic low-voltage grids based on geospatial data. This includes the consideration of multiple feeders per transformer, greenfield and brownfield transformer positioning, and variable dimensioning of equipment components. Based on data clustering methods and an algorithmic graph-based approach, the tool provides a scalable solution that is applicable to large areas such as cities, states, or even countries with hundreds of thousands of grids.The derived representative synthetic grid topologies support the development of more resilient and efficient future energy systems by enabling research on specific issues like grid planning and reinforcement while considering regional constraints, such as grid congestion or voltage violations. This supports a comprehensive understanding of low-voltage energy systems in the context of the energy transition.

The OpenFlexure Microscope is an accessible, 3Dprinted robotic microscope, with sufficient image quality to resolve diagnostic features including parasites and cancerous cells. As access to lab-grade microscopes is a major challenge in global healthcare, the OpenFlexure Microscope has been developed to be manufactured, maintained and used in remote environments, supporting point of care diagnosis. The steps taken in transforming the hardware and software from an academic prototype towards an accepted medical device include addressing technical and social challenges, and are key for any innovation targeting impact in low-resource healthcare.

This paper presents a proof of concept demonstrating how thousands of training episodes in deep reinforcement learning (DRL) can be condensed into just a few episodes. The central focus is to highlight a refined version of the algorithm that was foundational to my PhD several years ago, further developed and improved upon in this work. The refined algorithm, detailed at https://codeocean.com/capsule/0436858/tree/v1 , achieves significant acceleration in learning by addressing a key insight in DRL: the learning process can be expedited by providing the agent with ample and diverse information. This is accomplished by slicing complex strategies into manageable steps and distributing these steps across multiple artificial neural networks (ANNs). Each ANN receives both plentiful and varied data, enhancing exploration and allowing for rapid convergence to a stable and effective policy.

Unmanned Aerial Vehicles (UAVs), due to their low cost and high flexibility, have been widely used in various scenarios to enhance network performance. However, the optimization of UAV trajectories in unknown areas or areas without sufficient prior information, still faces challenges related to poor planning performance and low distributed execution. These challenges arise when UAVs rely solely on their own observation information and the information from other UAVs within their communicable range, without access to global information. To address these challenges, this paper proposes the Qedgix framework, which combines graph neural networks (GNNs) and the QMIX algorithm to achieve distributed optimization of the Age of Information (AoI) for users in unknown scenarios. The framework utilizes GNNs to extract information from UAVs, users within the observable range, and other UAVs within the communicable range, thereby enabling effective UAV trajectory planning. Due to the discretization and temporal features of AoI indicators, the Qedgix framework employs QMIX to optimize distributed partially observable Markov decision processes (Dec-POMDP) based on centralized training and distributed execution (CTDE) with respect to mean AoI values of users. By modeling the UAV network optimization problem in terms of AoI and applying the Kolmogorov-Arnold representation theorem, the Qedgix framework achieves efficient neural network training through parameter sharing based on permutation invariance. Simulation results demonstrate that the proposed algorithm significantly improves convergence speed while reducing the mean AoI values of users. The code is available at https://github.com/UNIC-Lab/Qedgix.

The design, fabrication, and demonstration of a novel Silicon Carbide Low Gain Avalanche Detector (4H-SiC LGAD), exhibiting an ultra-fast time response and excellent time resolution, are reported. The use of field plates is proposed to suppress the high electric field caused by the negative bevel-etched angle in 4H-SiC LGADs based on TCAD simulations. Experimental measurements confirm that the field plate significantly increases the breakdown voltage of the 4H-SiC LGADs. Gain and time resolution are measured by using the ultraviolet transient current technique (UV-TCT) showing that 4H-SiC LGADs possess excellent timing performance, with a time resolution better than 35 ps for the injected laser signal with single minimum ionizing particle (MIP) charges generation at room temperature. Additionally, the gain suppression effect in a 4H-SiC LGAD is observed for the first time.

Accurate glomerular segmentation is crucial for clinical diagnosis and pathological analysis in renal histology research. To address the challenge of blurred boundaries between glomeruli and surrounding tissues in nephropathology analysis, this paper proposes EdgeAttenNet, an innovatively improved model based on the BGNet architecture that enhances the recognition accuracy of glomerular boundaries. The model primarily consists of three enhanced core modules improved from BGNet: Attention-Guided Boundary Module (AGBM), Edge-guided Feature Enhancement Module (EFEM), and Adaptive Fusion Context Module (AFCM). The AGBM first captures glomerular boundary information through multi-scale attention mechanisms. The EFEM enhances boundary feature representation at different scales by incorporating the extracted edge semantic information. Subsequently, the AFCM integrates these multi-scale features through attentional feature fusion to strengthen the model's global feature characterization capabilities. Experimental results demonstrate that EdgeAttenNet achieves superior glomerular segmentation performance over the baseline BGNet with a 1.76% improvement in mIoU, reaching 95.14% and 94.32% mIoU on the KPIs2024 and HuBMAP datasets, respectively. These results validate our model's excellent performance and robustness under complex pathological image conditions. The findings demonstrate EdgeAttenNet's great potential in clinical applications and provide new insights into addressing blurred boundaries in medical image segmentation.

Predictive Control optimization typically relies on exhaustive enumeration of all possible voltage vectors within the finite control set. For simpler implementation the computational burden is not a concern, but the limited control set can demand a very high sampling frequency to reach good steady state performance. Indirect Predictive Control or Modulated Predictive Control addresses this by using a modulation stage, but classic implementation depends on duty cycle calculation that can limit constraints and non linearities in the cost function. This paper introduces the Informed Binary Search, an intelligent search algorithm capable of handling any constraints within the cost function, similarly to Finite Control Set, but with the benefit of fixed switching frequency. Experimental results on a two-level VSI driving an induction motor is shown but the method can be applied to every setup of machine and inverter. The results are compared with PTC, PTC with DSVM and MPTC and the metrics show improvements in torque and flux ripple and improvements of current THD up to 30% over classic M2PC while having similar computational burden.

Grid codes require High Voltage Direct Current (HVDC) converter stations to have AC Fault Ride-Through (AC-FRT) capability, which prevents disconnection during AC grid faults and requires them to support the faulty grid. A major challenge during AC-FRT operation is the faulty converter's reduced power capacity, which can lead to over-energizing of the Multi-Terminal Direct Current (MTDC) grid if said converter was exporting power from the MTDC grid prior to the fault. This paper presents an innovative strategy for dealing with DC grid over-energization during AC-FRT operation by utilizing the existing Metal Oxide Varistor (MOV) banks within DC Circuit Breakers (DCCBs) through a slight and economical modification to the DCCB architecture. By adopting this strategy, it is possible to reduce the size or number of Dynamic Break System (DBS) used in the MTDC grid. The proposed strategy is able to prevent overvoltage by clamping the DC voltage throughout the MTDC grid. Furthermore, it is able to attain a uniform distribution of energy dissipation in DCCBs throughout the MTDC grid without the need for communication and coordination and achieves a smooth transition between normal and AC-FRT operation and vice versa.

This paper derives a set of bulldozer dynamics to test a three-dimensional (3D) blade controller. This work was motivated by a desire to mimic how a human operator would control a bulldozer blade to build an autonomous system control module and by the lack of existing work on 3D bulldozer blade control. To that end, three-dimensional surface-to-track and surface-to-blade dynamics were derived to test the proposed 3D bulldozer blade controller in simulation. The track dynamics included considerations for track slip, while the blade dynamics accounted for non-symmetric mounds in front of the blade. The 3D bulldozer blade control effort was implemented using fuzzy logic to mimic a human operator's nonlinear modes of blade control, thereby following a Human Decision Making Model (HDMM). The inputs to the main portion of the control effort were the yaw, pitch, and roll errors between the blade's current orientation and the desired orientation. Accordingly, the outputs were the blade's yaw, pitch, and roll velocities. The pitch error, in particular, was tied to a separate fuzzy controller that outputted the desired cutting depth depending on the observed soil type, against tying back to the HDMM.

In my previous paper, "Exploring the Potential of ESP8266: A Wireless Control Experiment" (can be accessed by: https://doi.org/10.36227/techrxiv.170793844.49329362/v1), I documented the development of a smart switch prototype on a breadboard inside a local Wi-Fi network. This initial exploration served as a foundation, inviting readers to consider advancing the project to a professional PCB for practical application. I'm pleased to report that our initial tests on the protoboard were successful, validating the feasibility of the concept. Encouraged by this challenge and driven by the pursuit of innovation, I embarked on transforming this prototype into a practical application product that can be integrated into any automation or IoT project, creating a marketready solution that is currently being tested in real-world applications. This paper details the transformation of the project from a protoboard to a PCB, outlining its practical daily applications in home automation. It aims to inspire students, engineers, and IoT enthusiasts by demonstrating the transition from concept to real-world application. This work particularly seeks to motivate those in the engineering field who have not yet engaged in hands-on projects, offering them the satisfaction of creating something tangible and applicable on their own.

The increasing complexity of linguistic tasks has driven a need for more adaptable and contextually aware language models, capable of producing accurate and coherent outputs even when faced with ambiguous or evolving input. Traditional token embedding methods, which rely on static representations, often fall short in capturing the complex relationships between tokens and their contextual dependencies. Introducing a novel approach, Dynamic Token Embedding Sampling (DTES), allows embeddings to be dynamically adjusted during inference based on the evolving input sequence, offering a significant improvement in predictive accuracy and reduction in hallucinations. Through seamless integration into transformer architectures, DTES enhances both the flexibility and efficiency of token processing, ensuring more precise and contextually relevant outputs across various natural language processing tasks. Experimental results demonstrate considerable improvements in key metrics, including accuracy, inference speed, and factual consistency, further highlighting the potential of DTES to optimize language models for real-world applications. The method's scalability and adaptability across diverse linguistic domains make it a compelling advancement in the ongoing effort to improve the performance and reliability of language models in increasingly complex environments.

Tracheotomy is commonly performed for patients needing prolonged intubation, airway obstruction, and neck injuries. Accurate placement of the incision and the tracheal window is paramount in order to avoid complications. Current surgical technique heavily relies on palpating cartilage landmarks on the neck to place the incision. In order to achieve the accelerated goals of the robot-assisted subtask in a tracheotomy, this paper proposes a novel autonomous palpation-based acquisition strategy-RASEC in the tracheal region, which can interactively determine the next best acquisition point in order to maximize the expected acquisition information and minimize the expected costs of palpation procedure. We employ a Gaussian Process (GP) to model the distribution of hardness and utilize anatomical information as a priori input to guide the point of palpation for medical robots. The dynamic tactile sensor based on the resonant frequency is introduced to measure tissue hardness in the tracheal region by millimeter-scale gentle contact to secure the interaction. To figure out the drawbacks that the existing kernel functions are not sufficiently optimal in this scenario, we investigate the kernel fusion method to blend the Squared Exponential (SE) kernel with Ornstein-Uhlenbeck (OU) kernel. Moreover, we further regularize the exploration and greed factors, and the tactile sensor's moving distance and the robotic base link's rotation angle during the incision localization process are considered new factors in the acquisition strategy. Simulation and physical phantom experiments are conducted for comparison with stateof-the-art GP-based exploration approaches. The results show that the sensor's moving distance was optimized to 53.1% and the rotation angle of the base was optimized to 75.2% of the previous values without sacrificing overall performance capabilities. The satisfying algorithmic index (average precision 0.932, average recall 0.973, average F1 score 0.952) with fewer central estimation distance errors (0.423 mm) and high resolution (1 mm) indicates the performance of the proposed RASEC in terms of exploration efficiency, cost awareness, and localization accuracy for incision localization and recommendation in real robot-assisted subtask in the tracheotomy procedure.

Stability of multi-vendor power electronics-dominated grids requires proper specifications for converters interoperability, e.g. tuning boundaries for the converters' control loops to be respected by all vendors. The confidentiality and variety of implementation of the converters' control system by different vendors and the nonlinear dynamics of power converters make this task arduous. This paper formulates this research question as a robust stability problem, embedding frequency-dependent uncertainties in the converter control systems and analytically quantifying the amount of uncertainty able to warranty the power grid stability. Grid following converters are considered and modeled with a nonlinear grey-box approach, in contrast with conventional small-signal black-box models which suffer from inaccuracy when their operating point deviates from the nominal one. This paper also demonstrates both analytically and with Hardware-In-the-Loop (HIL) experiments that different converters operating points (e.g. active power injections, ac bus voltages) radically affect their stability, empathizing the necessity of nonlinear converters models to realize accurate power grid stability analyses.

RoboAgent is an innovative integrated hardware-software system designed to simulate human-like behavior in interactions with short video platforms such as Instagram, TikTok, WeiXin, XiaoHongShu, and YouTube. By combining a three-axis writing robot, computer vision capabilities, and advanced language models, RoboAgent facilitates controlled, reproducible studies of user behavior and engagement on these platforms. This paper details RoboAgent's design, including its system architecture and components, and presents results from experiments demonstrating high interaction success rates, precise icon detection, and generation of contextually relevant comments. RoboAgent represents a significant step forward in automated testing and analysis of social media interactions, offering valuable insights into user dynamics and platform engagement.

Achieving high-quality radar forward-looking imaging (FLI) presents significant challenges due to the poor azimuth resolution, which is constrained by the physical antenna aperture size and specific imaging geometry. In this paper, an efficient super-resolution FLI method that integrates wavefront modulation with beam scanning is proposed. Without resorting to complex super-resolution algorithms to solve the imaging equation, this method generates a distinct echo structure and then enhances resolution by extrapolating the azimuth information of the targets. First, a novel chirp-modulated beam is developed, functioning as a chirp pulse propagating along the azimuth direction during scanning. By leveraging the flexibility of modulation, chirp beams with varying azimuthal chirp rates are produced and employed successively in the scanning process, resulting in a series of two-dimensional (2-D) echo matrices. Subsequently, these matrices are compiled into a threedimensional (3-D) echo cube, which provides three degrees of freedom derived from the transmitting signal's bandwidth, beam scanning, and modulation parameters. Moreover, a maximum a posteriori (MAP)-based signal extrapolator is applied along the modulation dimension to augment target's azimuth information and enlarge the separability between different targets. Finally, super-resolution azimuth profiles are obtained through applying the Fourier transform (FT) along the scanning dimension to the extrapolated signal. Simulation results verify the effectiveness of the proposed method, showing a twofold to fourfold enhancement in azimuth resolution. Furthermore, a case study on sea-surface target imaging confirms its practical utility.

Extended reality (XR) and 6G networks are set to transform mobile immersive experiences, with privacy and security being paramount in XR communications. Achieving secure and reliable XR experiences while meeting high-resolution and low-latency needs are challenging for wireless networks. A novel security-aware cross-layer communication management framework is proposed, employing zero-trust spatiotemporal physical layer level manipulations for moving-target defense (MTD). Driven by deep reinforcement learning (DRL) and realtime monitoring, the proposed framework adaptively reprograms the network configuration to maximize the user's quality of experience (QoE), reduce the overall latency, and minimize the attacker's intercept probability. The framework was evaluated in a simulated scenario featuring an indirect multi-hop communication setup. The results show that the proposed framework effectively and efficiently secures XR user communications while maintaining QoE outperforming conventional Q-learning algorithms, making it suitable for mission-critical XR applications.

A modification to the Gyrator Capacitor (GC) magnetic model is proposed to correct errors that may arise when running the GC model of magnetic structures on common circuit simulators. The proposed method adds a conduction path to the DC component of the magnetic flux, due to a DC bias current, a path that is absent in the original model. This helps to distribute correctly the MMFs across the capacitors. The conduction path is realized by placing resistors in parallel to the capacitors to help distribute correctly the DC component of the MMFs across the capacitors. Theoretical considerations backed by circuit simulation validate the proposed approach.

Measurement-oriented adversarial attacks, particularly ghost attacks, pose a significant challenge to the robustness of multi-object tracking (MOT) systems due to their covert nature. These attacks not only deceive the system and degrade tracking accuracy but also impose additional computational overhead proportional to the number of ghost targets. Under high-density attacks, the sensing node’s computational capacity can become fully saturated, potentially leading to a denial-of-service (DoS) scenario. To ensure the integrity and accessibility of multi-sensor, multi-object tracking systems, this work introduces the Average Likelihood for Attack Resilient Multi-object (ALARM) filtering method. This approach enhances efficiency and robustness by effectively rejecting false measurements during the Bayesian update. The proposed method mitigates the impact of ghost measurements at an early stage and demonstrates significantly improved computational performance compared to standard filters under high-density attacks. This makes ALARM a practical and efficient solution, particularly for large-scale networks susceptible to attacks.