In many IIoT-based yard operations, accurately identifying the spatial position of containers is becoming increasingly relevant as operators try to automate stacking and retrieval processes by technologies like Container Spatial Localization (CSL). Despite this automation in IIoT, RTK-GPS-based container stacker positioning frequently lacks precision, which causes disruptions in stacking and reduces efficiency in space utilisation. Though it offers placement precision accurately up to 3 cm, this is still insufficient in high-volume Yard Management Systems (YMS). Consequently, this yields to variable container orientation, waste of usable space, increased man input is required in handling goods, and potential automated system failures. This research proposes a novel methodology that combines conventional RTK-GPS measurements with angular information captured from a BHI-260AP-based spreader sensor, allowing the system to correct container placement errors arising from orientation rather than only from positioning. In addition to the spatial positioning problem, we found that continuous IIoT operation raises concerns regarding energy use, particularly when micro-controllers remain active throughout the task cycle. As a solution, this integrates a dynamic task scheduling approach that puts the device in sleep modes whenever computation is not required. In our experiments, this strategy improved overall power efficiency by 34.44%, which makes long automated operation more practical.

Large AI models are reshaping intelligent manufacturing from isolated automation toward knowledge-intensive, model-assisted production systems. Yet their industrial value depends not on model scale alone, but on how language, vision, code, sensor data, engineering knowledge, and feedback mechanisms are integrated into deployable manufacturing workflows. This review examines recent progress in large AI models for intelligent manufacturing, covering model architectures, adaptation strategies, system integration, and applications across product development, production processes, equipment maintenance, and manufacturing services. A lifecycle-based framework is used to organize the literature and distinguish model capabilities from the data resources, retrieval mechanisms, simulation and optimization tools, digital twins, edge-cloud infrastructure, and human validation required for deployment. Current evidence suggests that large models show more reliable value in bounded, information-rich tasks, whereas safety-critical control and production-scale autonomy remain insufficiently validated. The review further summarizes challenges in data quality, domain adaptation, reliability, interpretability, latency, cybersecurity, cost, benchmarking, and responsibility allocation. By linking application scenarios, system-level enablers, and evidence maturity, this review provides a structured perspective for assessing the practical value of large AI models in manufacturing.

In recent years, researchers have leveraged single-agent reinforcement learning to boost educational outcomes and deliver personalized interventions; yet this paradigm provides no capacity for inter-agent interaction. Multi-agent reinforcement learning (MARL) overcomes this limitation by allowing several agents to learn simultaneously within a shared environment, each choosing actions that maximize its own or the group's rewards. By explicitly modeling and exploiting agent-to-agent dynamics, MARL can align those interactions with pedagogical goals such as peer tutoring, collaborative problem-solving, or gamified competition, thus opening richer avenues for adaptive and socially informed learning experiences. This survey investigates the impact of MARL on educational outcomes by examining evidence of its effectiveness in enhancing learner performance, engagement, equity, and reducing teacher workload compared to single agent or traditional approaches. It explores the educational domains and pedagogical problems addressed by MARL, identifies the algorithmic families used, and analyzes their influence on learning. The review also assesses experimental settings and evaluation metrics to determine ecological validity, and outlines current challenges and future research directions in applying MARL to education.

With the rapid development of the Industrial Internet of Things (IIoT), flexible manufacturing systems impose stringent requirements on network lifetime and communication reliability in industrial wireless networks. However, in complex industrial environments, the inherent energy constraints of sensor nodes and the highly dynamic nature of network topology significantly hinder the prolongation of network lifetime and the improvement of Quality of Service. To address these issues, this paper proposes a Non-Cooperative Game-based and Multi-Hop (NCG-MH) data transmission optimization strategy. Specifically, the cluster head (CH) selection process is formulated as a non-cooperative game, where nodes determine their probability of becoming CHs via utility maximization. A CH retention mechanism is introduced to reduce unnecessary role switching during the effective service period. In addition, a deterministic multi-hop data forwarding strategy is designed based on residual energy and transmission distance, enabling optimal path planning while effectively avoiding the high exploration cost inherent in reinforcement learning-based approaches. Simulation results demonstrate that the proposed strategy significantly enhances network lifetime, energy efficiency, and communication performance compared with existing schemes.

Infrared image generation is essential in scenarios with low illumination or complex environments, but scarcity of aligned visible-infrared data and the lack of physical realism in generated results remain challenges. However, existing generative models often overlook the thermodynamic principles underlying infrared imaging, resulting in synthetic images that are visually plausible yet physically inaccurate. In paper, we propose Infrared Physics-guided Latent Diffusion (IPLD), a novel framework that integrates physics guided modeling into a latent diffusion process for high-fidelity synthesis of infrared images. Central IPLD is the Temperature-Emissivity-Environmental Radiance (TeR) decomposition, which decomposes thermal signals into temperature, emissivity, and environmental radiance components, governed by the laws blackbody radiation. To enhance the environmental radiance modeling, we introduce Environmental Radiance Map Estimation (ERME), a hybrid local-global estimation mechanism that preserves both spatial detail thermal consistency. Furthermore, a novel Skip Connection Diffusion Transformer (SCDT) is proposed strengthen and balance semantic structure and fine-grained details during image reconstruction. Extensive experiments on public datasets demonstrate that IPLD outperforms state-of-the-art Generative Adversarial Network (GAN)-based and diffusion-based methods, achieving superior results in Structural Similarity Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), Learned Perceptual Image Patch Similarity (LPIPS), Fr & eacute;chet Inception Distance (FID) metrics. Ablation studies validate the complementary value of TeR, ERME, and SCDT in improving radiative realism. Our approach establishes a new paradigm for physically grounded image translation, offering enhanced generalization and reliability for downstream perception tasks such target detection.

In the Industrial Internet of Things (IIoT) environment, the deployment of the Generative Digital Twin (GDT) system faces the complexity of real-time synchronization and objective conflicts between end-devices and edge gateways. Therefore, we propose a Stackelberg-based generative task offloading (STA-GTO) algorithm to address the deployment challenges of the GDT system. Specifically, we model the different objectives of end-devices and edge gateways in the GDT system as two optimization problems, and develop the conditional temporal diffusion model (CTDM) to predict the future environmental disturbance states in the GDT system. Finally, the two optimization problems are transformed into a Stackelberg game problem, and we develop an offloading strategy generation algorithm based on dual-timescale Multi-Agent Proximal Policy Optimization (MAPPO), thereby obtaining a local approximate Stackelberg equilibrium solution. Experimental results demonstrate that the STA-GTO algorithm can effectively cope with environmental dynamics and outperforms the benchmark algorithms in both the total system utility and the synchronization latency of the GDT system.

Digital twins have emerged as a key enabling technology in Industry 4.0, enabling virtual representations of physical systems that evolve in real time through data synchronization. They are widely applied to industrial robotic systems such as manipulators, collaborative robots, and automated production cells for applications including process monitoring, predictive maintenance, and control optimization. Despite their growing adoption, many existing digital twin implementations implicitly assume that sensor measurements are accurate and reliable. In practice, measurements of key variables such as position, velocity, force, and system states are affected by uncertainty arising from noise, bias, drift, and environmental disturbances, which can significantly impact the reliability of digital twin predictions. This paper proposes a metrology-driven conceptual framework for uncertainty-aware digital twins (UA-DTs), where measurement uncertainty is explicitly modeled, quantified, and propagated through the digital twin. The proposed framework integrates metrological principles such as calibration, traceability, and uncertainty evaluation with digital twin architectures to enable probabilistic state estimation and reliability-aware decision-making. Although the framework is developed in the context of Industry 4.0, its applicability is demonstrated through a healthcare robotics use case, where robotic systems operate in safety-critical environments requiring strict reliability guarantees. Healthcare robotics is considered a specialized cyber–physical application domain of Industry 4.0 technologies, where uncertainty-aware monitoring is essential for safe human–robot interaction. Simulation-based experiments demonstrate how the proposed framework enables reliable state estimation, uncertainty propagation, anomaly detection, and safety monitoring. The results highlight that explicitly modeling measurement uncertainty improves the trustworthiness of digital twin predictions and supports risk-aware operation in both industrial and healthcare robotic systems.

Federated Learning (FL) enables collaborative model training across decentralized edge devices while preserving data privacy. However, real-world deployments are severely hindered by high uplink communication overhead and heterogeneous on-device computational constraints. To address these joint bottlenecks, We present Energy-Aware Adaptive Quantization and Freezing (EA-AQF), a unified framework that co-optimizes communication and computation. EA-AQF introduces a tiered mapping policy that assigns stage-wise activation bitwidths based on a dynamic client resource indicator. To further reduce local training costs, we implement a deterministic, precision-conditioned stage-freezing policy that disables gradient updates for computationally expensive network stages during backpropagation. Additionally, weight quantization is applied to client updates to minimize the communication footprint. We evaluate EA-AQF across diverse architectures, including CNN-7, CNN-17, and ResNet-20 on CIFAR-10, CIFAR-100, and MNIST datasets, and LSTM and MLP on IMDB and HAR datasets, under varying statistical distributions. Experimental results demonstrate that EA-AQF achieves a consistent communication reduction of 72.9 × , which is 1.6 × higher than prior state-of-the-art methods, with a marginal accuracy trade-off typically within 2-3 kgCO_2e for resource-intensive tasks. Notably, in highly heterogeneous tasks, EA-AQF maintains robust convergence positioning it as a highly efficient solution for sustainable edge AI.

Human activity prediction (HAP) is crucial for enabling intelligent smart home services; yet, it is often hindered by the scarcity of high-quality, multidimensional datasets. Existing datasets are typically fragmented, capturing either long-term activity sequences or short-term device interactions, but rarely both in a unified manner. Traditional data collection methods are costly and time-consuming, while conventional simulation techniques struggle to generate diverse and logically coherent behavior sequences. To address these limitations, we propose SmartLLM, a novel large language model (LLM)-based simulation framework for automated generation of multidimensional smart home datasets. SmartLLM simulates simulated agents with distinct profiles (e.g., old man, remote worker, and holiday maker) performing daily activities within configurable home environments, generating temporally aligned sequences across activity-device-sensor dimensions. We generate two months of simulated data for three user profiles and validated their plausibility through activity distribution visualization, statistical perplexity analysis, and case studies. Multidimensional feature validation experiments further demonstrate that our multidimensional data significantly enhances the accuracy of activity prediction models compared to using single-dimensional features. This work successfully addresses key bottlenecks in smart home data acquisition and provides a scalable, high-quality data foundation for advancing smart home algorithm research.

Generative world models facilitate robotic planning through future prediction, but existing navigation models (NWM) lack 3D geometric grounding, failing in complex 6DoF humanoid environments. We present Geo-NWM (GeometricNavigation World Model), a framework integrating conditional diffusion transformers with sparse visual SLAM priors to enforce geometric consistency. A consistency loss formalized on the SE(3) manifold penalizes physics-violating pixel motions, stabilizing diffusion gradients and mitigating trajectory hallucinations. The dual-stream architecture fuses a SLAM geometric anchor with a generative diffusion dreamer via a Geometric Gate layer. We introduce the Dream-to-Reality ATE (D-ATE) metric to quantify physical fidelity. Validation on the Unitree G1 in MuJoCo demonstrates an 80% reduction in trajectory drift, providing a robust foundation for safe human-robot interaction in shared environments.

According to Kings Research [1] the global Wearable Market is projected to reach US$ 192.14 billion by 2030, growing at a CAGR of 15.8% from 2023 to 2030, with an increasing growth caused by the COVID-19. The industry and public sector are then pushing for innovative WCS solutions with high levels of dependability and trustworthiness that can efficiently operate in increasingly complex scenarios. Great strides have been made to realize WCS for the 24/7 monitoring of single users based on 3-tier architectures involving wearables, edge, and cloud systems. However, new requirements targeting cooperative multiuser scenarios demand radically new approaches, as promoted by the community-oriented WCS (CO-WCS). The COMMON-WEARS project is developing novel models and architectures for next-generation CO-WCS, defining a rigorous engineering methodology, with associated formal verification and simulation tools, to drive the development lifecycle of CO-WCS and creating a pilot in a real instrumented environment to support activities of surgical teams in operating rooms. COMMON-WEARS strives to move the research front by developing CO-WCS featuring multi-user collectives of smart wearables and body sensor networks (BSN), with applications, e.g., in domestic, urban, manufacturing, emergency, and working environments. This is achieved by combining aggregate computing, collective opportunistic IoT, machine learning, and WCS/BSN architectures. Herein, we discuss lessons learned from prototyping new concepts in collaborative wearable computing, in the contest of challenging use cases, including healthcare, emergency response and pandemic management, which are strongly impacted by smart wearables.

Reliable fault detection in water distribution systems (WDS) is critical to maintaining service continuity and mitigating costly infrastructure failures. However, existing solutions often rely on supervised learning and dense sensor deployments, which are impractical in real-world scenarios with scarce labeled data and limited edge resources. To address these challenges, this paper proposes a lightweight U-Net autoencoder framework enhanced with semantic compression for unsupervised fault detection. Semantic compression transforms raw sensor streams into compact statistical and spectral descriptors, thereby reducing computational complexity and communication overhead. The autoencoder leverages separable convolutions in depth and skip connections to efficiently reconstruct normal operating patterns, while faults are identified by thresholding reconstruction errors with data-driven calibration strategies. Experiments on real-world WDS datasets with controlled fault injection (including drift, stuck-at, spike, and noise anomalies) demonstrate that the proposed method outperforms conventional autoencoders and baseline anomaly detectors. Specifically, it achieves higher event-level F1 Scores and recall rates while reducing model size and inference latency, enabling deployment on edge devices. This work presents a novel unsupervised fault-detection pipeline that balances accuracy and efficiency, providing a generalizable paradigm for anomaly detection in resource-constrained Internet of Things (IoT) environments.

Federated learning (FL) presents a promising method for preserving data privacy while enabling collaborative training of machine learning models on decentralized devices. Nevertheless, the development of effective model representations remains essential due to constraints in computational resources and communication bandwidth. This study examines the potential of integrating quantization techniques, such as Post-Training Quantization (PTQ), and Quantization-Aware Training (QAT), into the FL pipeline to address these challenges. While posttraining quantization (PTQ) reduces communication costs and accelerates inference, quantization-aware training (QAT) enables the model to maintain accuracy despite reduced precision. This analysis provides a quantitative assessment of energy consumption and associated $C O_{2}$ emissions. Experiments on the MNIST and Fashion-MNIST datasets indicate that both PTQ and QAT reduce energy consumption, communication costs, and memory usage, with only a marginal decrease in accuracy. These findings suggest a practical approach to enhancing the efficiency and sustainability of FL systems.

An industry 5.0 revolution is characterized by advanced automation and human-centric design resulting in an unprecedented growth in the electronics sector. This advancement comes at the cost of a surge in electronic waste (E-waste) generation. In the past, many researchers have reported on E-waste recycling and management; however, the efficient collection of domestic E-waste still remains a critical challenge. This research paper presents a novel approach to domestic E-waste management by developing a smart E-Bin equipped with an Electronic Waste Detection and Bin-Level Control System (EDBLCS), IoT setup, and a YOLOv11-powered (EW YOLO) computer vision system. This innovative solution selectively collects only E-waste, ensuring accurate identification and preventing contamination with other waste streams, with the mAP@0.50 score increased to 0.90074 by Epoch 50, while mAP@0.50-0.95 reached 0.73899 using YOLOv11. The primary contribution of this work is the integration of YOLOv11-based real-time detection with an IoT-enabled smart E-Bin framework to enable selective, edge-oriented domestic E-waste segregation.

Federated Learning (FL) has become a foundational paradigm for privacy-preserving distributed intelligence, yet its scalability remains fundamentally constrained by communication bottlenecks, device heterogeneity, and the challenges of training under statistically non-IID data. Quantization is one of the most effective mechanisms for mitigating these limitations, reducing both uplink/downlink payloads and on-device computation. This paper provides the first FL-centric systematic review of quantization, introducing a novel taxonomy organized around FL-specific dimensions, including client heterogeneity, aggregation consistency, communication-scheduling adaptation, non-IID robustness, privacy/security integration, and hardware/energy co-optimization. Beyond cataloging existing methods, we analyze how quantization interacts with core FL behaviors such as client drift, partial participation, convergence stability, secure aggregation, and differential privacy. We further identify cross-method insights, open research gaps, and design guidelines for practitioners deploying quantized FL on mobile, IoT, and edge platforms. This survey thus establishes quantization not merely as a compression technique, but as a fundamental systems component shaping the performance, robustness, and practicality of modern FL.

Smart-meter theft detection requires learning from fine-grained electricity consumption data, whose centralized processing poses significant privacy risks. Federated learning (FL) mitigates these risks by decentralizing training, but providing rigorous user-level differential privacy (DP) under non-IID data and heterogeneous client behavior remains challenging. Existing DP-FL approaches rely on fixed global clipping bounds for client updates, which substantially overestimate sensitivity when privacy loss is composed using R & eacute;nyi Differential Privacy (RDP), zero-Concentrated DP (zCDP), or Moments Accountant (MA) frameworks, leading to excessive noise and degraded utility. This work proposes an adaptive clipping-based RDP accountant that incorporates empirical, round-wise update magnitudes into privacy accounting by rescaling each round's RDP contribution according to the observed clipping ratio. The method is optimizer-agnostic and is evaluated with FedAvg, FedProx, and SCAFFOLD on the SGCC smart-meter theft dataset under IID and Dirichlet non-IID partitions. Experimental results show consistently tighter privacy bounds and improved model utility compared to classical DP accountants, demonstrating the effectiveness of sensitivity-aware privacy accounting for practical differentially private FL.

The rapid proliferation of Internet of Vehicles (IoV) technology has significantly enhanced traffic efficiency and driving safety, yet it has also introduced severe security challenges due to malware and cyberattacks. This paper proposes a novel Fractional-Order Attack-Defense Game model (FADG-IoV) to address dynamic malware propagation in IoV environments. By integrating fractional-order dynamics, the model accounts for communication delays, traffic density heterogeneity, and channel fading, capturing memory-dependent behaviors inherent in IoV systems. We introduce the Fractional-Order Attack-Defense Game Sparse Attention Multi-Agent Soft Actor-Critic (FADG-SMASAC) algorithm, a model-free reinforcement learning approach that leverages sparse attention mechanisms to achieve adaptive and robust control without requiring a known system model. Through multi-baseline experiments, we validate the FADG-IoV model and FADG-SMASAC algorithm, demonstrating superior convergence, scalability, and robustness compared to existing methods. Our findings highlight the effectiveness of fractional-order game-theoretic strategies in enhancing IoV security against dynamic malware threats, paving the way for future research in adaptive defense mechanisms. Note to Practitioners-The rapid advancement of Internet of Vehicles (IoV) technology has revolutionized traffic management and vehicle safety, but it also exposes IoV systems to significant security risks from malware and cyberattacks. These threats can disrupt communication networks, compromise vehicle coordination, and degrade system performance in real-world scenarios. To address these challenges, this work introduces a Fractional-Order Attack-Defense Game model (FADG-IoV) and the Fractional-Order Attack-Defense Game Sparse Attention Multi-Agent Soft Actor-Critic (FADG-SMASAC) algorithm. Our approach leverages fractional-order dynamics to model memory-dependent behaviors, such as communication delays and traffic density variations, while employing a model-free reinforcement learning framework with sparse attention mechanisms to achieve robust and adaptive malware defense. Practitioners can implement this approach to enhance IoV security without requiring precise system models, as the algorithm adapts dynamically to evolving threats. The inclusion of sparse attention ensures scalability, making it suitable for large-scale IoV networks. Our multi-baseline experiments demonstrate improved convergence and resilience compared to traditional methods, offering a practical solution for securing IoV systems against dynamic cyber threats. This framework also provides a foundation for developing advanced adaptive defense strategies in other networked systems.