Search Results (256)

Motorcyclists face a disproportionately high risk of severe injury or death compared to other road users, highlighting the need for intelligent rider assistance technologies. This paper presents an initial, modular, and interpretable AI pipeline that generates context-aware safety advice from first-person motorcycle videos with practical inference latency suitable for on-device deployment, framing large language models as interpretable cognitive support agents for motorcycle safety. The system integrates lightweight perception and reasoning components to emulate the function of an Advanced Rider Assistance System (ARAS). Video frames are processed at 1 FPS using Pixtral, a Mistral-based multimodal large language model (MLLM), to produce descriptive scene captions, while YOLOv8 identifies key objects such as vehicles, pedestrians, and road hazards. A Mistral-small language model then fuses this information to generate concise, imperative safety tips. Preliminary evaluations on publicly available motorcycle POV datasets demonstrate promising performance in terms of contextual accuracy, interpretability, and scalability, suggesting potential for real-world deployment in low-resource or embedded environments. The proposed framework offers interpretable, context-aware safety assistance that is particularly valuable for young and newly licensed riders during the transition from supervised training to independent riding, where real-time hazard interpretation support is most needed. Full article

Population ageing and chronic disease are increasing demand for rehabilitation, while resources remain limited. This review does not report an implemented end-to-end system; instead, it proposes a modular workflow framework for applying large AI foundation models across rehabilitation. Organised into four stages—assessment, prescription, execution, and monitoring—we summarise recent evidence and highlight techniques most suitable at each stage. In assessment, multimodal models can enable more continuous and objective functional measurement from heterogeneous sensor and imaging data. In prescription, large language models can support evidence-informed, personalised plan formulation by synthesising guidelines and patient context. In execution, vision–language–sensor models can provide real-time feedback for telerehabilitation and adherence support. In monitoring, longitudinal and cross-setting data integration can facilitate risk prediction and early warning for safety and long-term management. We also discuss practical adaptation options (e.g., parameter-efficient fine-tuning) and propose a clinimetric-oriented evaluation framework to assess validity, reliability, and generalisability. By mapping AI capabilities to concrete workflow tasks, the framework provides a theoretical foundation and roadmap for reproducible research and future translation toward a universal rehabilitation model. Full article

This study presents an AI-based, human-in-the-loop decision support system designed for large-scale institutional query routing and response generation. The proposed system combines semantic text classification with large language model-based response generation to assist administrative staff in handling high-volume natural language requests from various system users, while preserving human oversight. Using a dataset of 135,359 real student and staff interactions collected over 15 years, the system was designed, deployed, and evaluated in a live university information portal. The classification component achieved 95.88% accuracy in evaluation and 82.21% staff acceptance in practice, while 94.81% of AI-generated draft responses were adopted with minor edits. Operational evaluation showed a 30.8% reduction in resolution time, a 32.6% decrease in misrouting, and an increase in user satisfaction from 3.6 to 4.9 out of 5. The system is implemented as a modular RESTful API to ensure interoperability with existing Student Information Systems, with analysis code available upon request to support replication in similar resource-constrained environments. The results illustrate how human-in-the-loop AI systems can support improvements in service quality, efficiency, and institutional capacity in resource-constrained environments, providing a transferable applied AI framework for scalable decision support in complex administrative domains. Full article

Conventional G-banded karyotyping remains indispensable in prenatal diagnosis but continues to rely on labor-intensive, expertise-dependent visual examination. To address these challenges, we developed a modular artificial intelligence (AI) workflow that automates chromosome interpretation from amniotic fluid metaphase images. The system integrates image denoising, chromosome segmentation, overlap screening, and morphology-based classification, and was trained using 13,223 clinical cases comprising more than 50,000 manually annotated chromosomes. Across training, temporal validation, and independent testing cohorts, classification accuracy remained consistently high (97.45%, 96.95%, and 95.72%, respectively). The overlap-recognition module further reduced downstream errors by reliably identifying composite chromosome regions. When applied to unsorted metaphase images from a later clinical cohort, the workflow successfully generated draft karyotypes without manual sorting and maintained close concordance with expert review. These findings demonstrate that an AI-assisted pipeline can support cytogenetic laboratories by streamlining the most labor-intensive steps of karyotyping, potentially enhancing diagnostic efficiency while preserving interpretive reliability. Full article

Clinical diagnostic laboratories continue to face growing pressure from rising test volumes, increasingly complex testing menus, significant workforce shortages, and expectations for faster turnaround times at sustainable cost. Total laboratory automation (TLA) has become a central strategy for improving efficiency in high-volume laboratories, where integrated systems from Abbott, Roche, Siemens Healthineers, and Beckman Coulter have demonstrated substantial reductions in turnaround time, error rates, and labor requirements. Evidence across multiple health systems shows that TLA improves performance and stabilizes laboratory operations even during workload peaks. Despite these gains, large segments of pre-analytical and post-analytical workflows remain manual, especially tasks related to specimen transportation, bench-level manipulation, instrument tending, and troubleshooting. Recent progress in collaborative robotics (cobots), autonomous mobile robots (AMRs), and hospital service robots demonstrates that these technologies can complement TLA by addressing not only the logistical and dexterous tasks that fixed automation lines cannot reach but also enabling robots that can work safely right alongside humans in a shared space. Cobots have shown sub-millimeter precision in colony picking and other fine-motor tasks, though typically at lower throughputs than dedicated track modules, and AMRs have demonstrated reliable transport of pathology carts and medical supplies through large clinical environments. Meanwhile, humanoid-capable mobile manipulators, like Moxi from Diligent Robotics, deployed in hospitals are already completing hundreds of thousands of supply deliveries, indicating real-world significance. Here, we integrate technical, regulatory, operational, and business perspectives on TLA, collaborative robotics, and mobile platforms. We discuss real-world efficiency gains, regulatory expectations under the CLIA and United States FDA, and the emerging case for hybrid automation ecosystems that combine TLA islands, cobotic workcells, AMRs, and AI-enabled orchestration. We argue that the next decade of laboratory automation will move beyond monolithic tracks with robots toward flexible, modular robotic systems designed to operate safely together with humans and to augment the increasingly strained laboratory workforce. This not only allows clinical staff to dedicate more time to patient care but also ensures greater reliability and scalability for essential services throughout demanding hospital environments. Full article

Human pose classification, along with related tasks such as action recognition, is a crucial area in deep learning due to its wide range of applications in assisting human activities. Despite significant progress, it remains a challenging problem because of high inter-class similarity, dataset noise, and the large variability in human poses. In this paper, we propose a lightweight yet highly effective modular attention-based architecture for human pose classification, built upon a Swin Transformer backbone for robust multi-scale feature extraction. The proposed design integrates the Spatial Attention module, the Context-Aware Channel Attention Module, and a novel Dual Weighted Cross Attention module, enabling effective fusion of spatial and channel-wise cues. Additionally, explainable AI techniques are employed to improve the reliability and interpretability of the model. We train and evaluate our approach on two distinct datasets: Yoga-82 (in both main-class and subclass configurations) and Stanford 40 Actions. Experimental results show that our model outperforms state-of-the-art baselines across accuracy, precision, recall, F1-score, and mean average precision, while maintaining an extremely low parameter count of only 0.79 million. Specifically, our method achieves accuracies of 90.40% and 87.44% for the 6-class and 20-class Yoga-82 configurations, respectively, and 94.28% for the Stanford 40 Actions dataset. Full article

Effective clash management is critical to the success of complex construction projects, yet BIM coordinators face severe information overload when modern detection tools generate thousands or even millions of collision reports, making interdisciplinary coordination increasingly difficult. This article presents a conceptual framework for using AI for collision triage in a Building Information Modeling (BIM) environment. Previous approaches have focused mainly on collision detection itself and simple, rule-based prioritization, rarely exploiting the potential of Artificial Intelligence (AI) methods for post-processing of results, which constitutes the main innovation of this work. The proposed framework describes a modular system in which collision detection results and data from BIM models, schedules (4D), and cost estimates (5D) are processed by a set of AI components, offering adaptive, data-driven decision support unlike static rule-based methods. These include: a classifier that filters out irrelevant collisions (noise), algorithms that group recurring collisions into single design problems, a model that assesses the significance of collisions by determining a composite ‘AI Triage Score’ indicator, and a module that assigns responsibility to the appropriate trades and process participants. The framework leverages supervised machine learning methods (gradient boosting algorithms, selected for their effectiveness with tabular data) for noise filtering, density-based clustering (HDBSCAN, chosen for its ability to detect clusters of varying densities without predefined cluster count) for clash aggregation, and multi-criteria scoring models for priority assessment. The article also discusses a potential way to integrate the framework into the existing BIM workflow and possible scenarios for its validation based on case studies and expert evaluation. The proposed conceptual framework represents a step towards moving from manual, intuitive collision triage to a data- and AI-based approach, which can contribute to increased coordination efficiency, reduced risk of errors, and better use of design resources. As a conceptual study, the framework provides a foundation for future empirical validation and its limitations include dependency on historical training data availability and the need for calibration to project-specific contexts. Full article

The rapid aging of bridges has increased interest in real-time, data-driven monitoring for predictive maintenance and safety management; however, practical deployment on in-service bridges remains limited. This paper presents lessons learned from long-term field deployment of real-time bridge joint monitoring systems on three in-service highway bridges and demonstrates how these insights can support the transition toward Industry 5.0. A unified framework is introduced to integrate key enabling technologies, including Internet of Things (IoT), digital twins, and artificial intelligence (AI), into a practical, human-centric monitoring architecture. Best practices for achieving durable, site-compliant, and cost-effective system design are summarized, with emphasis on sensor selection, wireless communication strategies, modular system development, and maintaining seamless operation. The development of a Docker-based analytics and visualization platform illustrates how interactive dashboards enhance human–machine collaboration and support informed decision-making. The role of advanced analytical tools, including digital twins, AI, and statistical modeling, in providing reliable structural assessments is highlighted, along with guidance on balancing cloud and edge computing for energy-efficient performance under constraints such as limited power, weather exposure, and site accessibility. Overall, the findings support the development of scalable, resilient, and human-centric real-time monitoring systems that advance data-driven decision-making and directly contribute to the realization of Industry 5.0 objectives in bridge health management. Full article

Microfluidic chip technologies, also known as lab-on-a-chip systems, have profoundly transformed laboratory medicine by enabling the miniaturization, automation, and rapid processing of complex diagnostic assays using minimal sample volumes. Recent advances in chip design, fabrication methods—including 3D printing, modular and flexible substrates—and biosensor integration have significantly enhanced the performance, sensitivity, and clinical applicability of these devices. Integration of advanced biosensors allows for real-time detection of circulating tumor cells, nucleic acids, and exosomes, supporting innovative applications in cancer diagnostics, infectious disease detection, point-of-care testing (POCT), personalized medicine, and therapeutic monitoring. Notably, the convergence of microfluidics with artificial intelligence (AI) and machine learning has amplified device automation, reliability, and analytical power, resulting in “smart” diagnostic platforms capable of self-optimization, automated analysis, and clinical decision support. Emerging applications in fields such as neuroscience diagnostics and microbiome profiling further highlight the broad potential of microfluidic technology. Here, we present findings from a comprehensive review of recent innovations in microfluidic chip design and fabrication, advances in biosensor and AI integration, and their clinical applications in laboratory medicine. We also discuss current challenges in manufacturing, clinical validation, and system integration, as well as future directions for translating next-generation microfluidic technologies into routine clinical and public health practice. Full article

We developed an AI-assisted zero-trust control system at low capital expenditure to retrofit brownfield Ethernet environments without disruptive hardware upgrades or costly software-defined networking migration. Legacy network infrastructures in small and medium-sized enterprises (SMEs) lack the flexibility and programmability required by modern zero-trust architectures, creating a persistent security gap between static Layer-1 deployments and dynamic cyber threats. The developed system addresses this gap through a modular architecture that integrates genetic-algorithm-based virtual local area network (VLAN) optimization, large language model-guided firewall rule synthesis, threat-intelligence-driven policy automation, and telemetry-triggered adaptive isolation. Network assets are enumerated and evaluated through a risk-aware clustering model to enable micro-segmentation that aligns with the principle of least privilege. Optimized segmentation outputs are translated into pfSense firewall policies through structured prompt engineering and dual-stage validation, ensuring syntactic correctness and semantic consistency. A retrieval-augmented generation pipeline connects live telemetry with historical vulnerability intelligence, enabling rapid policy adjustments and automated containment responses. The system operates as an overlay on existing managed switches, orchestrating configuration changes through standards-compliant interfaces such as simple network management protocol and network configuration protocol. Experimental evaluation in a representative SME testbed demonstrates substantial improvements in segmentation granularity, refining seven flat subnets into thirty-four purpose-specific VLANs. Compliance scores improved significantly, with the International Organization for Standardization/International Electrotechnical Commission 27001 rising from 62.3 to 94.7% and the National Institute of Standards and Technology Cybersecurity Framework alignment increasing from 58.9 to 91.2%. All 851 automatically generated firewall rules passed dual-agent validation, ensuring reliable enforcement and enhanced auditability. The results indicate that the system developed provides an operationally feasible pathway for legacy networks to achieve zero-trust segmentation with minimal cost and disruption. Future extensions will explore adaptive learning mechanisms and hybrid cloud support to further enhance scalability and contextual responsiveness. Full article

As artificial intelligence (AI) continues to expand across clinical environments, healthcare is transitioning from static decision-support tools to dynamic, autonomous agents capable of reasoning, coordination, and continuous interaction. In the context of geriatric medicine, a field characterized by multimorbidity, cognitive decline, and the need for long-term personalized care, this evolution opens new frontiers for delivering adaptive, assistive, and trustworthy digital support. However, the autonomy and interconnectivity of these systems introduce heightened cybersecurity and ethical challenges. This paper presents a Secure Agentic AI Architecture (SAAA) tailored to the unique demands of geriatric healthcare. The architecture is designed around seven layers, grouped into five functional domains (cognitive, coordination, security, oversight, governance) to ensure modularity, interoperability, explainability, and robust protection of sensitive health data. A review of current AI agent implementations highlights limitations in security, transparency, and regulatory alignment, especially in multi-agent clinical settings. The proposed framework is illustrated through a practical use case involving home-based care for elderly patients with chronic conditions, where AI agents manage medication adherence, monitor vital signs, and support clinician communication. The architecture’s flexibility is further demonstrated through its application in perioperative care coordination, underscoring its potential across diverse clinical domains. By embedding trust, accountability, and security into the design of agentic systems, this approach aims to advance the safe and ethical integration of AI into aging-focused healthcare environments. Full article

This paper presents an artificial intelligence (AI)-assisted multi-objective topology optimization of a 48 V interior permanent magnet synchronous motor (PMSM) intended for mid-drive e-bike applications. The machine features a 48-slot, 8-pole stator–rotor combination with $\Delta$-shaped three buried magnets per pole, and is coupled to a multi-stage gearbox that adapts its high-speed, low-torque output to a human-scale crank speed. The design problem simultaneously maximizes average torque and efficiency while minimizing torque ripple by varying key stator slot dimensions and magnet geometries. A modular MATLAB–ANSYS Maxwell framework is developed in which finite element simulations are driven by a Bayesian optimization (BO) loop augmented by a large language model (LLM) with retrieval-augmented generation (RAG). The LLM acts as a memory-based agent that proposes candidates, shapes Gaussian Process priors, and incorporates natural language rules expressing qualitative design knowledge. Two AI-assisted trials are compared against a multi-objective Artificial Hummingbird Algorithm benchmark, RAG + BO with and without natural language input. All three methods converge to a similar Pareto region with average torque around 5.4–5.7 Nm, torque ripple of approximately 12.8–14.2%, and efficiency near 93.3–93.6%, suitable for geared e-bike drives. The LLM-guided trial achieves this performance with a 20.1% reduction in simulation expenses relative to the BO baseline and by about 48% compared to the Artificial Hummingbird Algorithm. The results demonstrate that integrating LLM guidance into Bayesian optimization improves sample efficiency while providing interpretable design trends for PMSM topologies tailored for light electric vehicles. Full article

Scaffold-aware artificial intelligence (AI) models enable systematic exploration of chemical space conditioned on protein-interacting ligands, yet the representational principles governing their behavior remain poorly understood. The computational representation of structurally complex kinase small molecules remains a formidable challenge due to the high conservation of ATP active site architecture across the kinome and the topological complexity of structural scaffolds in current generative AI frameworks. In this study, we present a diagnostic, modular and chemistry-first generative framework for design of targeted SRC kinase ligands by integrating ChemVAE-based latent space modeling, a chemically interpretable structural similarity metric (Kinase Likelihood Score), Bayesian optimization, and cluster-guided local neighborhood sampling. Using a comprehensive dataset of protein kinase ligands, we examine scaffold topology, latent-space geometry, and model-driven generative trajectories. We show that chemically distinct scaffolds can converge toward overlapping latent representations, revealing intrinsic degeneracy in scaffold encoding, while specific topological motifs function as organizing anchors that constrain generative diversification. The results demonstrate that kinase scaffolds spanning 37 protein kinase families spontaneously organize into a coherent, low-dimensional manifold in latent space, with SRC-like scaffolds acting as a structural “hub” that enables rational scaffold transformation. Our local sampling approach successfully converts scaffolds from other kinase families (notably LCK) into novel SRC-like chemotypes, with LCK-derived molecules accounting for ~40% of high-similarity outputs. However, both generative strategies reveal a critical limitation: SMILES-based representations systematically fail to recover multi-ring aromatic systems—a topological hallmark of kinase chemotypes—despite ring count being a top feature in our structural similarity metric. This “representation gap” demonstrates that no amount of scoring refinement can compensate for a generative engine that cannot access topologically constrained regions. By diagnosing these constraints within a transparent pipeline and reframing scaffold-aware ligand design as a problem of molecular representation our work provides a conceptual framework for interpreting generative model behavior and for guiding the incorporation of structural priors into future molecular AI architectures. Full article

The rapid advancement of connected and autonomous vehicles (CAVs) has the potential to revolutionize road transportation, promising significant improvements in safety, efficiency, and sustainability. However, traditional CAV architectures are predominantly modular and rule-based. They struggle with interaction, cooperation, and adaptability in complex mixed-traffic environments. Moreover, the substantial infrastructure investment required and the absence of compelling killer applications have limited large-scale deployment of CAVs and roadside units (RSUs), resulting in insufficient penetration to realize the full safety benefits of CAV applications and creating a deployment stalemate. To address the above challenges, this paper proposes an innovative connected autonomous vehicle system, termed AQ-CAV, which leverages recent advances in AI agents and QR codes. AI agents are employed to enable cooperative, self-adaptive, and intelligent vehicular behavior, while QR codes provide a cost-effective, accessible, robust, and scalable mechanism for supporting CAV deployment. We first analyze existing CAV systems and identify their fundamental limitations. We then present the architectural design of the AQ-CAV system, detailing the components and functionalities of vehicle-side and infrastructure-side agents, inter-agent communication and coordination mechanisms, and QR code-based authentication for AQ-CAV operations. Representative applications of the AQ-CAV system are investigated, including a case study on emergency response. Preliminary results demonstrate the feasibility and effectiveness of the proposed system, which achieves significant safety improvements at low system cost. Finally, we discuss the key challenges faced by AQ-CAV and outline future research directions that require exploration to fully realize its potential. Full article

Deep reinforcement learning (DRL) has been widely adopted to solve decision-making problems in complex environments, demonstrating high performance across various domains. However, DRL-based FPS agents are typically trained with a traditional, monolithic policy that integrates heterogeneous functionalities into a single network. This design hinders policy interpretability and severely limits structural flexibility, since even minor design changes in the action space often necessitate complete retraining of the entire network. These constraints are particularly problematic in game development, where behavioral characteristics are distinct and design updates are frequent. To address these issues, this study proposes a Modular Reinforcement Learning (MRL) framework. Unlike monolithic approaches, this framework decomposes complex agent behaviors into semantically distinct action modules, such as movement and attack, which are optimized in parallel with specialized reward structures. Each module learns a policy specialized for its own behavioral characteristics, and the final agent behavior is obtained by combining the outputs of these modules. This modular design enhances structural flexibility by allowing selective modification and retraining of specific functions, thereby reducing the inefficiency associated with retraining a monolithic policy. Experimental results on the 1-vs-1 training map show that the proposed modular agent achieves a maximum win rate of 83.4% against a traditional monolithic policy agent, demonstrating superior in-game performance. In addition, the retraining time required for modifying specific behaviors is reduced by up to 30%, confirming improved efficiency for development environments that require iterative behavioral updates. Full article