ABSTRACT. E-moped micromobility growth continues globally, and its safety awaits critical attention as its low helmet use rate effectively exposes users, especially the unhelmeted ones, to high chances of head-related injuries. This research aims to explore helmet usage intentions of riders in shared e-mopeds in China via an extended theory of planned behavior (TPB) framework. The results, based on data from 461 respondents, were examined by employing partial least squares structural equation modeling (PLS-SEM). Findings reveals that perceived behavioral control (0.243) was the strongest predictor with subjective norms (0.176) and attitude (0.113). In extending the TPB model, perceived safety (β = 0.194), law enforcement (β = 0.102) were also found to have positive impacts, whereas perceived inconvenience (β = -0.168) was discovered to be an obstacle. The results show that adoption intention concerning helmets within the micromobility setting is determined by a combination of psychological, socio-contextual, and functional considerations. Therefore, successful interventions should be multi-facet. Recommendations were made to improve the perceived control of the riders, utilize social influence, support the safety advantages, make systems re-design to reduce the inconvenience, and work with law enforcement. The study equips policy-makers and micromobility operators with an evidence base to design specific approaches that enhance rider safety and foster sustainable urban travel.

ABSTRACT. The growth of urbanization and complexity of traffic environment requires more smart, efficient and flexible transportation systems. The digitization of transportation through technologies such as internet of things (IoT) and AI has dramatically reshaped the way that drivers interact with the surroundings and their vehicles. This paper propose a Context-Aware Driving Assistance System (CA-DAS) that utilizes sensor fusion, semantic context modelling, and machine learning-based to deliver tailored assistance across dynamic scenarios. Our model was developed and tested using the CARLA simulator across 150 scenarios, including urban, rural, and various driver conditions. Furthermore, a Likert-scale survey was conducted with 20 drivers to measure their satisfaction with the system. The results show that CA-DAS achieved 94.2% accuracy in risk detection, reducing the average time to warning by 28%, and decreasing lane deviation under driver condition(fatigue) by 32%. The driver satisfaction was averaged 4.6 out of 5, indicating strong user trust and acceptance. The findings demonstrated effectiveness of the system as a proactive solution for enhancing safety. Our study contributes to the field in developing a scalable and flexible framework for context-aware driving assistance system in semiautonomous and human-driven vehicles

ABSTRACT. Smart vehicles increasingly employ artificial intelligence to enhance driving safety and performance. Continuous monitoring of the driver is vital to prevent accidents and ensure safe operation. Vision-based sensing provides a non-invasive means to assess the driver behavior, alertness, and attention. Reliable and real time detection of drowsiness is essential to support higher levels of vehicle automation. Classical vision-based approaches that relied on eye landmarks and handcrafted ratios often struggle under challenging conditions such as poor illuminations, variations in head movements, and occlusions. Although convolutional neural networks (CNNs) have significantly improved detection accuracy, they still face limitations in modeling long-term dependencies and generalizing across diverse driving environments. Eye state (open vs. closed) serves as a fundamental visual cue for blink detection, attention monitoring, and higher-level drowsiness estimation. Most existing studies and publicly available datasets predominantly provide drowsy or alert labels at the frame or segment level based on full-face without direct supervision on the observable cue of eye state. In addition, in certain situations or for cultural reasons, full-face images may not be available due to religious or privacy considerations, leaving only the eye region visible. In such cases, accurate eye-state detection becomes particularly crucial. To address these challenges, we propose an ensemble framework for driver eye-state detection in smart vehicles. The architecture integrates three transformer encoders namely ViT, DeiT, and Swin-T. We preprocessed the original UTA-RLDD and NTHU-DDD video dataset and generated eye-state annotations for the extracted eye regions using Eye Aspect Ratio (EAR) computation. The model was evaluated on MRL-Eyes, UTA-RLDD-Eye, and NTHU-DDD-Eye state annotations. Experimental results demonstrate promising and consistent performance for the proposed ensemble model across all eye-state conditions and datasets.

ABSTRACT. Lip reading, the process of understanding spoken words through visual observation of lip movements, holds significant promise for improving accessibility in autonomous transportation systems. This research introduces a visual Arabic lip-reading model aimed at facilitating inclusive access to in-vehicle information for people with hearing challenges. As transportation technologies evolve toward greater autonomy and intelligence, the exchange between passengers and vehicles must go beyond sound-based communication. The proposed framework allows users to engage with the vehicle through lip gestures, enabling natural interaction for accessing real-time updates in their native language. This study addresses this gap by presenting an empirical investigation into Arabic visual lip reading, emphasizing the challenges unique to this language. To advance research in this area, we constructed a comprehensive dataset designed to capture the diversity of Arabic speakers, accounting for variations in pronunciation, dialect, and speaking styles. Furthermore, we explored the efficacy of combining 3D Convolutional Neural Networks (3DCNN) with Bidirectional Long Short-Term Memory (Bi-LSTM) networks to perform visual speech recognition tasks in Arabic. The results from our experiments reveal the complexity of processing Arabic speech, particularly as the number of target classes increases. These findings highlight the need for continued innovation in model architecture and dataset development to overcome the unique challenges posed by Arabic lip reading, paving the way for more inclusive and robust visual speech recognition systems.

ABSTRACT. In Saudi Arabia, camel-vehicle collisions (CVCs) contribute significantly to fatalities and injuries. Studies indicate that CVCs may account for up to 25% of traffic-related deaths and represent a substantial portion of spinal cord injuries in certain regions. These findings highlight the urgent need for an effective camel detection and driver alert system. Advanced detection and real-time driver alert technologies, such as dynamic variable message signs (VMS), can improve driver awareness and reaction times, offering a practical solution. This paper aims to enhance road safety in Saudi Arabia’s desert regions by developing and deploying a state of-the-art Camel Detection and Driver Alert System (CDDAS). The proposed system is designed to reduce fatalities and injuries while also limiting the economic losses resulting from CVCs. We evaluate the performance of three YOLO models—YOLOv8, YOLOv11, and YOLOv12—on a custom camel detection dataset. All models were trained under identical conditions, including the same dataset, training configurations, image sizes, and 100 training epochs, to ensure a fair comparison. The results show that YOLOv8 and YOLOv11 outperformed YOLOv12 in both detection accuracy and training efficiency, achieving the highest mean Average Precision (mAP) scores with relatively short training times. YOLOv12 demonstrated reduced accuracy without offering notable improvements in training performance.

ABSTRACT. Supply chain management is a key area of study that focuses on the effective coordination of material, information, and financial flows across production and distribution networks. SCM plays a vital role in ensuring sustainability, resilience, and competitiveness by optimizing processes. However, SCM inherently involves complex decision-making scenarios under uncertain and conflicting environments. To address these challenges, this research proposes a multi-attribute decision-making approaches combined with an advanced hybrid mathematical framework known as Complex Spherical Fuzzy Soft Sets (CSpFSS). First, the notion of CSpFSS is introduced and illustrated through an example. Subsequently, two distinct distance measures for any two CSpFSS are presented, forming the basis of the proposed MADM algorithmic approach. The developed algorithm is then applied to a case study concerning the selection of the best supplier from among multiple alternatives. To validate and check the effectiveness of the presented approach, a statistical analysis employing various statistical tools is carried out. Furthermore, to establish the superiority of the proposed model, a detailed comparative analysis with existing distance-based MADM models is provided. Finally, the research concludes with a synthesis of the key findings and outlines potential future research directions.

ABSTRACT. Deadheading in last-mile delivery significantly increases operational costs, fuel consumption, and carbon emissions which makes critical challenges to the logistics sector. To address this, a multi-objective optimization framework is developed and solved using five heuristic algorithms: Genetic Algorithm(GA), Particle Swarm Optimization(PSO), Aquila Optimization(AO), Grey Wolf Optimization (GWO), and Simulated Annealing (SA). Adaptive parameter tuning is applied to enhance algorithm performance, with each method evaluated based on cost, solving time, and delivery assignment efficiency. Among them, Simulated Annealing emerges as the most effective approach, offering valuable insights into the trade-offs between solution quality, computational efficiency, and operational feasibility.

ABSTRACT. This paper addresses a multi-product, multi-vehicle Inventory Routing Problem with Transshipments and Backordering (IRPTB), integrating replenishment, inventory sharing, and backordering decisions into a single optimization framework. The problem is formulated as a mixed-integer linear program (MILP) and solved using a matheuristic approach based on the Fix-and-Optimize (F&O) strategy to balance solution quality and computational efficiency. Computational experiments on benchmark datasets demonstrate that the proposed method consistently achieves near-optimal solutions with substantial computational speed-up up to 15 times faster than CPLEX; across different demand variability levels. Results indicate that allowing transshipments and backordering simultaneously can reduce total system cost by up to 10% while improving service levels through better inventory balance and reduced lost sales. The study highlights the effectiveness, robustness, and scalability of the proposed matheuristic and provides managerial insights for applying coordinated delivery and transshipment planning in complex, demand-uncertain distribution networks.

ABSTRACT. With growing environmental challenges and rising demands for sustainability, researchers are increasingly focusing on energyefficient scheduling methods. Time-of-Use (ToU) electricity tariffs play a crucial role in balancing supply and demand, particularly in energy-intensive industries. This paper addresses the Permutation Flow Shop Scheduling problem (PFSP) with the objective of minimizing the total energy cost (T EC) under ToU tariffs, while constraining the schedule’s makespan within a specified time horizon. A Q-learning-based Genetic Algorithm (QL-GA) is proposed, which leverages Q-learning (QL) for dynamic parameter control. The proposed algorithm is evaluated against a classical GA on 25 instances from the challenging VRF benchmark (Vallada et al., 2015), which include both small and large instances with up to 200 jobs and 40 machines. The results demonstrate that integrating Q-learning significantly reduces the T EC and consistently outperforms the classical GA. These outcomes underscore the significance of incorporating reinforcement learning techniques into metaheuristic frameworks. This work contributes to advancing sustainable manufacturing through intelligent scheduling optimization.

ABSTRACT. In light of growing environmental concerns and rising energy costs, the manufacturing industry is increasingly focusing on energyefficient scheduling. This paper addresses the Non-Permutation Flow Shop Scheduling (NPFS) problem under Time-of-Use (TOU) pricing, where job sequences may vary across machines. The objective is to minimize the Total Energy Cost (T EC) while maintaining a constraint on the makespan to remain within the time horizon, thereby balancing production efficiency and sustainability. To achieve this, VNS-QL, a hybrid algorithm, is proposed, which uses Q-learning for operator selection within the Variable Neighborhood Search framework. The algorithm is evaluated through experiments on 20 challenging instances adapted from Vallada et al., 2015 and extended to incorporate TEC under TOU pricing. Results show that VNS-QL consistently outperforms VNS in reducing TEC. These findings highlight the value of integrating reinforcement learning into metaheuristic frameworks to guide the search more efficiently, achieving notable energy savings

ABSTRACT. Mega satellite constellations provide unprecedented capabilities for high-resolution Earth observation, supporting applications such as environmental monitoring, urban mapping, and disaster management. However, the limited onboard power, computational resources, and intermittent connectivity of satellites pose significant challenges for real-time image analysis. This paper introduces SatelliteEdgeNet, a secure, efficient, and scalable framework designed for distributed satellite imagery analysis. The framework combines a lightweight, multi-scale Convolutional Neural Network (CNN) optimized for low-power space devices with a privacy-preserving Federated Learning (FL) scheme. Each satellite performs local model training on its high-resolution imagery, employing advanced preprocessing. Differential privacy is applied to both local gradient updates and global aggregation to ensure the confidentiality of sensitive spatial information. To optimize performance, SatelliteEdgeNet incorporates gradient clipping, structured pruning, and 8-bit quantization, reducing communication overhead and computational demand while maintaining high model accuracy. Experimental results on real-world satellite imagery demonstrate that the framework achieves up to 99.65% classification accuracy, robustly handles dynamic satellite availability, and efficiently balances computation, bandwidth, and privacy. SatelliteEdgeNet provides a practical, secure, and resource-aware solution for next-generation satellite constellations requiring real-time, collaborative, and privacy-preserving image analysis.

ABSTRACT. Low-Earth-Orbit (LEO) constellations can support resilient positioning, navigation, and timing (PNT). We propose a simple and interpretable LEO-only pipeline that learns link-wise pseudorange residuals using a gated recurrent unit (GRU) and then applies a standard least-squares (LS) estimator. Input features include LEO pseudoranges, satellite azimuth and elevation, and atmospheric corrections (including ionospheric and tropospheric delays). From held-out data, we form per-link user-equivalent range errors (UERE) to weight LS and apply a geometry–quality gate. We introduce Time-in-Spec (TiS), defined as the fraction of epochs that meet a UERE-weighted threshold. Only LS estimations within TiS are accepted. The proposed method is trained on three suburban trajectories and evaluated on a distinct fourth trajectory that differs in both route and time. Evaluation results indicate that 48.91% of the test trajectory falls within TiS, with the 3D (RMSE) reduced from 6.12 m to 2.17 m, corresponding to a 64.5% reduction. The maximum observed error is reduced to 3.03 m from 23.32 m, representing an 87.0% improvement. Results from twenty-five Monte Carlo (MC) simulations further confirm the consistency and stability of the method's performance. The findings indicate that integrating GRU-based range correction with UERE-weighted LS estimation provides accurate and reliable LEO-only positioning.

ABSTRACT. GNSS receivers in smart mobility ecosystems are increasingly exposed to intentional and unintentional jamming.We present a fast, interpretable pipeline that types common jammer families (continuous wave, linear chirp, pulsed, wide-band noise) using short I/Q windows. The method applies a continuous wavelet transform (CWT) with a Morse wavelet to obtain a scalogram, from which we derive an interpretable feature set: ridge slope (Hz/s) and goodness-of-fit R2 along the dominant time–frequency ridge, time-domain energy envelope statistics (duty cycle), scale/PSD entropies as spread measures, and a scale-bandwidth proxy. A lightweight rulebased classifier—expressed in a few thresholded relations—maps features to jammer types. This enables transparent tuning and operator diagnosis via compact plots (PSD, scalogram with ridge, time envelope). On synthetic I/Q sampled at 5 MHz using ∼54 ms windows, the rules cleanly separate the four families and remain robust after threshold retuning (e.g., CW versus. wideband) demonstrated through sensitivity tables and qualitative figures. The pipeline runs in real time per window and avoids heavy training. The approach is deployable for on-device monitoring, provides explainable decisions for operators, and forms a compact feature substrate for future ML back-ends when real RF datasets become available.

ABSTRACT. Electric Vehicles (EVs) are becoming the center of future transportation, which rely heavily on Brushless DC motors due to their high efficiency and reliability, their reliance on electronic control units and controller area network (CAN). However, EVs are vulnerable to cyberattacks such as Denial-of-Service (DoS) and False Data Injection. This research work presents an event-triggered control (ETC) method for an electric-vehicle traction drive (EVTD) subject to network-induced DoS interruptions. Using a BLDC motor model, we design a hybrid, attack-aware ETC scheme that guarantees finite-time input-to-state stability in nominal operation, preserves finite-time stability (FTS) under DoS intervals, and eliminates Zeno behavior through strictly positive inter-event times. A Lyapunov analysis provides explicit triggering conditions and DoS admissibility bounds. Simulation results demonstrate fast convergence of speed and position states with substantially fewer network transmissions than periodic sampling. Without DoS, the controller achieves finite-time regulation with only ~50 events over the horizon. under DoS, stability is retained with ~85 events, still far below a time-triggered baseline, while the hybrid trigger adapts transmission density around attack windows. The proposed scheme reduces bus utilization, limits performance degradation during DoS, and offers practical tuning rules that balance tracking quality and communication frugality for CAN-based EV traction systems.

ABSTRACT. This paper presents the finite-element model of a Line Start Permanent Magnet Synchronous Motor (LSPMSM) for simulating early stator Inter-Turn Short-Circuit (ITSC) faults. Developed in ANSYS Maxwell, the model captures the impacts of incipient ITSC, represented as partial turn shorting under different conditions. Signals for voltage, current, and rotor speed were recorded from both healthy and faulty conditions to generate a physically-based synthetic dataset. This dataset is meant for feature benchmarking and evaluating early fault detection algorithms. The model is made publicly accessible and designed for scalability, facilitating further analysis and future diagnostic studies. Spectral analysis revealed a strong correlation between the fundamental and lower-order harmonics and fault severity, confirming their utility as diagnostic features in ITSC fault detection.

ABSTRACT. This study presents a hybrid Visible Light Positioning (VLP) and Vision–Camera localization framework for industrial Au- tonomous Guided Vehicles (AGVs). The proposed system fuses photodiode-based VLP ranging, camera-based LED geometry, and vision anchor corrections using an Unscented Kalman Filter (UKF). A comprehensive modeling and simulation environment was developed under realistic Lambertian lighting conditions with ambient noise and field-of-view (FOV) constraints to evaluate accuracy and stability. The inclusion of a camera subsystem enhances the geometric consistency of VLP, providing continuous global localization even under degraded optical conditions. Experimental results show that the hybrid VLP + Camera + Vision configuration achieves a median localization error of 10.03 cm and a 90th-percentile error of 10.30 cm, representing a 77% im- provement over VLP-only and a 34% gain over VLP + Vision. The system also achieves the lowest smoothness index (J = 0.013), indicating highly stable and deterministic motion estimates. These results demonstrate that integrating vision geometry with VLP creates a scalable, RF-free, and ISO-compliant localization solution for smart-factory AGV navigation.

ABSTRACT. Drone Technology, especially swarm drones have been adopted widely across various sectors. Such as surveillance, intelligent monitoring, rescue operations and many more. Formation flight can extend the endurance of multirotor swarms, yet the optimal geometry for quadcopters remains under characterized. We present a parametric computational study of five quadcopter V formations that evaluates the impact of half V angle (30°, 45°, 60°) and inter drone spacing (20–35 cm) on aerodynamic drag and energy use. Using steady 2D RANS (k ω SST) in ANSYS Fluent with a 15 m/s uniform inlet and 5 % turbulence intensity, we first establish a single quadcopter baseline (drag ≈ 14.345 N), then quantify per position and formation average drag across configurations. The minimum formation average drag occurs at a 30° half V angle with 20 cm spacing in our 2D setup. Per position results reveal substantially higher drag for the two outer trailing quadcopters; for example, at the optimum we observe a representative drag vector of [9.51, 9.06, 8.62, 17.73, 17.72] N, motivating a position swapping algorithm that schedules time in each slot based on measured drag, equalizes battery depletion, and extends collective flight time. We also articulate the limits of 2D modeling which cannot capture upwash/downwash benefits. Overall, the study combines aerodynamic optimization with simple cooperative control to improve swarm endurance, with relevance to surveillance, search and rescue, and networked UAV missions.

ABSTRACT. Autonomous underwater vehicles (AUVs) are essential for marine applications. However, ensuring stable operation in dynamic and uncertain underwater environments remains a critical challenge due to complex hydrodynamic behavior and unpredictable external disturbances. This paper presents a Particle Swarm Optimization optimized linear quadratic regulator (LQR) controller deployed in an underactuated AUV for a depth-tracking problem. The AUV model considers the non-linearities inherent to the system and the external uncertainties, such as hydrodynamic and environmental uncertainties. A double-loop Adaptive Line of Sight (ALOS) LQR controller coupled with extended state observer is simulated and compared with the proposed controller. The proposed controller scheme yields faster settling time and lowers mean absolute error of the depth tracking error than the double loop strategies.

ABSTRACT. Mass gatherings such as Hajj and Umrah represent some of the most challenging crowd-management scenarios in the world. Within Al-Masjid Al-Haram, the Mataf area where pilgrims perform the circumambulation ritual (Tawaf) experiences extreme pedestrian densities that pose operational, safety, and design challenges. This study evaluates alternative Mataf layouts using a calibrated multi-agent pedestrian simulation. Field observations, video analytics, and spatial data from the busiest times of Ramadan and Hajj served as the basis for calibration. The existing architecture, a suggested symmetric layout, and Option A which shifts ingress points and enhances merge protection were the three configurations that were examined. Density, average speed, travel time, throughput, and effective capacity across concentric radial bands were among the performance indicators. Option A reduced local conflicts and improved egress flow balance while increasing overall capacity by about 47% when compared to the current arrangement. These results offer a replicable methodology for large-scale religious crowd modelling and quantifiable recommendations for safer, more effective Mataf activities.

ABSTRACT. Accurate learned dynamics are essential for tracking, feedback, and planning on industrial manipulators, yet single–step prediction error often fails to reflect how models behave when iterated in simulation or control. We study plant–level, torque–driven system identification on a 7-DoF KUKA LWR iiwa using real joint state and torque logs. Two linear baselines are compared: a memoryless linear state–space surrogate (LS–SS) and a history–linear ARX model trained with ridge regression. Beyond standard next–step metrics, we evaluate models with measured–input long-horizon rollouts, trajectory-level in/out-of-distribution (ID/OOD) splits, and 95% confidence intervals across seeds. On KUKA, ARX consistently achieves the lowest one–step RMSE and exhibits slower error growth over horizons up to 1000 steps than LS–SS, although both accumulate error with horizon. These results hold under an ID→OOD shift constructed from the most aggressive trajectory in the set. Our contributions are: (i) a reproducible plant–level SysID protocol for manipulators that complements single–step metrics with measured–input rollouts and residual tests; and (ii) a head-to-head analysis showing that short history in a linear ARX estimator materially improves iterative stability relative to an H=1 surrogate. We conclude that one–step accuracy alone is insufficient for models intended for iterative use and recommend long-horizon, measured-input evaluation as standard practice for manipulator SysID.

ABSTRACT. The present paper proposes a DenseNet121-based automated pavement distress detection model using a granular three-class prediction of asphalt, crack, and pothole images based on a Bangladesh-specific road surface damage dataset. Seven ImageNet pretrained convolutional neural networks (DenseNet121, EfficientNetB3, ResNet50, InceptionV3, Xception, VGG16, MobileNetV2) are trained in a common, leakage-safe pipeline with identical output heads, optimization parameters, and data splits. Models are evaluated on a held-out test set using accuracy, precision, recall, F1-score, and cross-entropy loss. DenseNet121 is the best-performing network, achieving 0.9470 test accuracy and 0.9469 weighted F1-score with balanced class-wise performance across asphalt, crack, and pothole categories, demonstrating its suitability for fine-grained pavement texture analysis. The framework offers a replicable benchmark for Bangladesh road networks and supports comparative architectural analysis to guide the selection of efficient backbones for practical pavement management systems.

ABSTRACT. Weak subgrade soils in transportation infrastructure often limit bearing capacity and long-term performance, especially under heavy traffic and harsh environmental conditions. Conventional stabilization methods improve strength but are associated with high carbon emissions. This study presents a sustainable and smart alternative using phosphogypsum (PG), an industrial by-product, combined with alkali-activated binders (AAB) made from fly ash (FA) and ground granulated blast furnace slag (GGBFS). The AAB system was activated with a sodium silicate–sodium hydroxide solution to enhance strength development and promote carbon utilization. Laboratory tests, including unconfined compressive strength (UCS) and Brazilian tensile strength (BTS), were performed to assess the mechanical performance of PG–AAB composites with varying FA and GGBFS contents for subgrade applications. The results showed notable gains in strength and durability, supported by microstructural evidence of cementitious phase formation responsible for improved compressive, shear, and tensile behavior after seven days of curing. The integration of industrial waste with AAB enhances subgrade resilience and supports the development of smart, low-carbon, and sustainable transportation systems.

ABSTRACT. The trapdoor phenomenon is the primary cause of surface settlements above buried structures, particularly common on highways with underground utilities. Soil arching develops due to a static, yielding zone that redistributes load towards stiffer adjacent zones, significantly affecting overburden pressure and surface settlement. Trapdoor tests, conducted in a controlled environment, enable the observation of load redistribution and the evaluation of the effects of cover height and geometry. This paper consolidates a comprehensive dataset of trapdoor simulations performed with the discrete element method, analyzing the problem using three dimensionless ratios:  = H/B,  = B'/B, and  = L/B. A concise predictive formula for the peak stress concentration factor peak is proposed, showing it increases with soil cover (H) and yielding width (B’), converging as the cover and H/B ratio grow, and incorporating a boundary multiplier to account for periodic versus wall boundaries. Using existing values across various geometries and heights, model parameters are estimated through non-linear least squares, then validated with cross-validation, parity plots, and residual diagnostics. Results are presented in design charts that map peak within the dimensionless parameter space, indicating the necessary cover to reach 90 to 95 percent of the asymptotic response. The study links particle-scale mechanisms with practical design guidance in a portable format for trapdoor analysis geometries.

ABSTRACT. Effective wastewater treatment is essential for achieving environmental sustainability, particularly given the growing global population and the increasing volume of wastewater produced. Accurate prediction of ammonia concentration is critical for optimizing treatment processes and ensuring regulatory compliance. This study presents an approach for integrating temporal deep learning models, LSTM and GRU, with explainable feature selection techniques to improve the accuracy and transparency of ammonia concentration prediction in wastewater systems. It also evaluates several machine and deep learning models on a novel dataset collected from North Water’s sewage facility in Delfzijl, Netherlands (2014–2017). In parallel, we apply and compare multiple feature selection techniques, including Backward Elimination, Recursive Feature Elimination, and Sequential Feature Selection, to identify the most informative predictors. These techniques are leveraged not only for model optimization but also as part of an Explainable AI framework to enhance model interpretability and transparency. To further validate the significance of the selected features, we employ SHapley Additive exPlanations to quantify each feature's contribution to model predictions. The experiments show that the LSTM model combined with Recursive Feature Elimination achieved the best performance (R$^2$ = 0.96, MAE = 1.044, RMSE = 1.49), effectively capturing temporal trends and highlighting key predictors such as influent nitrite and dissolved oxygen levels.

ABSTRACT. The global waste crisis is escalating, with solid waste generation expected to increase tremendously in the coming years. Traditional waste collection methods, particularly in remote or harsh environments like deserts, are labor-intensive, inefficient, and often hazardous. Recent advances in computer vision and deep learning have opened the door to automated waste detection systems, yet most research focuses on urban environments and recyclable materials, overlooking organic and hazardous waste and underexplored terrains such as deserts. In this work, we propose YOLO-SAT, an enhanced real-time object detection framework based on a pruned, lightweight version of YOLOv12 integrated with Self Adversarial Training (SAT) and specialized data augmentation strategies. Using the DroneTrashNet dataset, we demonstrate significant improvements in precision, recall, and mean average precision (mAP), while achieving low latency and compact model size suitable for deployment on resource-constrained aerial drones. Benchmarking YOLO-SAT against state-of-the-art lightweight YOLO variants further highlights its optimal balance of accuracy and efficiency. Our results validate the effectiveness of combining data-centric and model-centric enhancements for robust, real-time waste detection in desert environments.

ABSTRACT. Detecting objects in industrial context has proven to be a challenge because it requires a huge number of manual annotations, which restricts scalability, efficiency and flexibility. In this regard, we suggest ZELF (Zero-shot Ensemble Learning Framework), a modular method that is a combination of several zero-shot object detectors to produce high-quality pseudo-labels without human interaction. It is followed by training a supervised YOLOv8 model with these pseudo-labels, and it can be robustly trained on automatically annotated data. ZELF consists of three states of the art zero-shot detectors including YOLO-World, OWL-ViT and Grounding DINO and majority voting along with tailored post-processing strategies to generate high quality pseudo-labelled datasets. We use ZELF as a case study on the re-annotated CHV dataset, which would have multiple categories of PPE, though this paper targets helmet, vest, and gloves. It has been experimentally demonstrated that ZELF holds competitive and often better results than supervised baselines like YOLOv7, especially in structured image detection tasks like helmet detection, and competitive results with vests and near-agnostic results with gloves. The ensemble significantly outperforms single zero-shot models, which proves the superiority of integration. ZELF has a high potential of being an alternative to pipelines with heavy annotation usage, being applicable not only to PPE detection but also to the scope of retail, agriculture, and smart cities. This framework allows to automatize and keep safety of the logistics operations and facilitate intelligent infrastructure of intelligent cities and transport sustainability. ZELF achieves higher accuracy results than both YOLOv7 and YOLOv8 on the helmet class, attaining a class mAP@0.5 of 0.966, outperforming YOLOv8 (67.4%) and YOLOv7 (66.6%). More so, ZELF is used to aid human decision-making in mobility and logistics which is assisted by AI using flexible detection systems.

ABSTRACT. We built a simple tool that checks warehouse work posture from ordinary video and runs entirely in the web browser on the user’s device. The tool finds the person’s body joints, turns them into easy-to-read angles (back, neck, shoulders, knees, wrists), and then gives a risk score using well-known ergonomics checklists: REBA and RULA (standard methods for judging whole-body and upper-limb posture risk). For lifting tasks, the user can enter the box weight and how often it’s lifted; the tool then reports the NIOSH Lifting Index (a standard way to judge lifting difficulty). Because everything runs on-device, no video or keypoints are uploaded, which protects privacy and makes field use practical. The tool covers common warehouse tasks—lift/lower, carry, push/pull, and pick/place, and can mark lift cycles, repetition rate, and obvious twisting. We publish the exact rules and thresholds so results are transparent and reproducible. In tests across varied warehouse scenes (different shelf heights, box sizes, lighting, and PPE), we compare the tool’s output with expert ratings, check angle accuracy, measure how well it detects lift cycles, and report speed on typical laptops and phones. Results show quick, green/yellow/red feedback that workers and supervisors can understand and act on. We also note limits, such as hidden body parts, unknown loads unless provided, and camera angles that are too extreme.

ABSTRACT. In this paper, we focus on undertaking a systematic comparative study of four forecasting methods—ARIMA, Gradient Boosting (GB), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM)—for forecasting the weekly demand of products in the retail industry. The models are tested using real time-series data consisting of ten products and ten geographical areas under two scenarios: (i) forecasting demand across regions per product and (ii) forecasting product demand per region. Forecast accuracy was measured using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Results show that BiLSTM outperformed all other approaches, with RMSE and MAE reductions of up to 42.35% and 40.10%, respectively, compared with ARIMA. This superior performance is attributed to BiLSTM’s ability to capture complex temporal dependencies through its bidirectional architecture. LSTM and GB achieved modest improvements over ARIMA, which underperformed mainly due to its limitations in modeling nonlinearity and non-stationarity. These findings highlight the effectiveness of deep learning methods in retail demand forecasting and suggest promising future research directions involving hybrid architectures and the inclusion of exogenous variables.

ABSTRACT. Providing stable equipment availability is very essential for the overall efficiency of the supply chain. In this research, 11 different models of deep learning techniques, namely ANFIS, LSTM, GRU, Transformer, and DeepAR, are used for predicting the availability of equipment for the next day in the food processing industry. The research aims to improve the efficiency of the supply chain by integrating advanced predictive models and operation scheduling techniques. The proposed models are trained on three-year production data, considering Down Time, Idleness, and Production, and are used for predicting the value of Availability(t+1) using MAE and RMSE. The PyTorch LSTM model gives the best accuracy (MAE 0.0861, RMSE 0.1102) compared to other models, namely Darts LSTM and DeepAR. The research illustrates how artificial intelligence algorithms and predictive models can act as catalysts for optimized supply chain decision-making in Industry 4.0

ABSTRACT. Smart-city logistics increasingly depend on autonomous vehicles, unmanned aerial vehicles (UAVs), and roadside infrastructure that exchange safety beacons, sensor data, and control messages over vehicle-to-everything (V2X) links. These communications must satisfy strict latency and reliability constraints while remaining robust against evolving network behaviour and adaptive cyber-attacks. Conventional intrusion-detection or rule-based security mechanisms often degrade under concept drift and require frequent retraining. This paper proposes a lightweight, resource-aware cybersecurity framework for resilient logistics communication. A supervised CNN–BiLSTM encoder is trained offline on labeled traffic to learn temporal–spatial flow representations and then frozen to provide a stable 128-dimensional embedding space. On top of these embeddings, a contextual-bandit Deep Q-Network (DQN) acts as an adaptive policy head that classifies traffic efficiently using a cost-aware reward function. Warm-start imitation from the encoder’s soft outputs and an annealed auxiliary loss enable fast, stable convergence. Experiments on the UAVIDS-2025 dataset achieve a test accuracy of 99.14 %, precision of 99.33 %, recall of 99.58 %, and an AUC of 0.999. These results confirm that the proposed hybrid CNN–BiLSTM + DQN model delivers intelligent, resilient, and adaptive protection for secure autonomous-logistics communication in smart-city environments.

ABSTRACT. This study presents the design and development of Easy-Move, a prototype Mobility-as-a-Service (MaaS) platform tailored to the urban mobility challenges of Riyadh, Saudi Arabia. The city faces high dependence on private vehicles, limited multimodal connectivity, and underutilized public transport options. In response, Easy-Move proposes a unified digital platform that enables users to plan, book, and pay for entire multimodal trips—metro, bus, ride-hailing, and micro-mobility—through a single interface and a unified QR ticket. The system integrates real geospatial data from central and northern Riyadh, providing users with three personalized route options—fastest, cheapest, and most comfortable—based on their preferences. The prototype was developed using a Flask-based web application, incorporating session-based personalization and a simulation-driven backend for route planning and payment logic. Basic user testing and survey responses were used to assess usability and behavioral interest in multimodal travel. Compared to global platforms like Whim and Citymapper, Easy-Move offers a more localized, Vision 2030-aligned solution for the Saudi context. It contributes to Saudi national goals for reducing traffic congestion, increasing public transport use, and advancing smart city initiatives. The research underscores the relevance of user-centered smart mobility in emerging cities.

ABSTRACT. Maintenance planning and control refers to the coordinated activities for determining maintenance needs, scheduling maintenance work, and ensuring that maintenance execution aligns with operational requirements. In public transport sectors, maintenance planning and control is critical for efficient operations, yet their maintenance workshops still use traditional computerized maintenance management software and lack advanced decision support tools. In this regards, this paper proposes an artificial intelligence based maintenance planning and control (AI-MPC) platform that aims to intelligently optimize maintenance workflows in maintenance workshops. The proposed platform considers a holistic approach and integrates maintenance planning with maintenance control in a unified system. As a systematic methodology, focus group workshop with industry and academic experts is conducted to gather the requirements and an expert view on the development of the platform. The focus group was organized around four key dimensions i.e., practicality and usability, integration and adaptability, functional and structural feasibility, and framework completeness. The expert feedbacks were properly analyzed and based on that a comprehensive multi-layer platform architecture is developed that highlight the important aspects of integration, intelligence, and user interface, etc., as suggested by the experts. According to our knowledge, there is no such solution available and that, our proposed AI-MPC is a novel comprehensive maintenance platform that can enable real-time, data-driven maintenance planning and control, reducing unplanned downtime and improving resource utilization in maintenance facilities.

ABSTRACT. Sustainable mobility is central to the 2030 Agenda, but the extent to which research in this domain contributes to the Sustainable Development Goals (SDGs) remains unaddressed. Addressing this gap, this paper is the first to systematically analyze the sustainable mobility literature through the explicit lens of the SDGs, offering a panoramic view of how transport-related research has aligned with global development priorities over time. We analyzed a Scopus corpus of 9,513 publications (2010–2025; articles and reviews in English) and segmented it into three periods to capture pre-SDG uptake, early operationalization, and recent consolidation. Period-level SDG mapping was conducted using the JRC SDG Mapper applied to concatenated titles, abstracts, and author keywords. To contextualize these signals, we complemented the mapping with a bibliometric overview (production, venues, actors) and thematic analyses using VOS viewer (keyword co-occurrence) and Biblioshiny (trend topics, thematic maps, thematic evolution). Findings reveal a consistent focus on SDG 11 (specially targets 11.2/11.3), a sharp rise in SDG 13, and steady progress in SDGs 7 and 9. The field has shifted from planning-focused research to implementing energy- and climate-friendly solutions, with a surge in publications from 2021–2025, spanning transport, energy, and environmental journals, and involving global researchers. However, SDGs 5, 10, 14, and 16 remain underexplored, with SDG 3 often implied rather than explicit. This study provides policymakers, funders, and researchers with an evidence-based map of where sustainable mobility research already contributes to the SDGs and where targeted incentives are needed before 2030.

ABSTRACT. The shift towards a sustainable energy future has made green hydrogen one of the key elements of the long-term energy diversification and decarbonization policies in Saudi Arabia. Since the Kingdom has immense potential in terms of renewable energy sources and has a visionary national strategy of implementing the hydrogen economy, the development of effective and affordable transport infrastructure is essential to facilitate the use of large-scale hydrogen application. The proposed research paper constructs an optimization-based model to design domestic green hydrogen transportation network to reduce the total supply chain cost in Saudi Arabia. The suggested model takes into consideration eight cities of Saudi Arabia, for analysing numerous types of transportation: pipelines, compressed gas trailers, and liquid hydrogen tankers, which connect renewable production sites to industrial demand centres, refuelling stations, and storage facilities all over the Kingdom. A mixed-integer linear programming (MILP) model is developed to ensure the optimization of transport modes configuration, and infrastructure capacities under uncertain demand scenarios. The objective of the model is to minimize the total system cost which includes the production, Storage and transportation cost, subject to technical, capacity, and geographical limits. The study will help in advancing the hydrogen roadmap in Saudi Arabia by providing a quantitative decision support instrument in the infrastructure planning and investment prioritization. The model is a strategic guide to policymakers and industry stakeholders who want to see a cost-effective and resilient domestic hydrogen logistics network that will facilitate green transition of energy and vision 2030 goals of the Kingdom.

ABSTRACT. University campuses offer unique opportunities to enhance walkability due to their compact structure, controlled environments, and concentrated daily activities. However, many campuses remain car-dependent despite short inter-building distances. This paper suggests a computational, optimization-based framework for evaluating and improving walkability on the campus of King Fahd University of Petroleum and Minerals (KFUPM) in Dhahran, Saudi Arabia. Using a parametric workflow created in the Rhino/Grasshopper environment, an OpenStreetMap-derived mobility model was built with the Urbano plugin to simulate pedestrian accessibility based on Walk Score methodology. Analysis of the baseline walkability indicated a car-dependent environment, motivating targeted spatial interventions within the academic area. A generative design method was used to enhance walkability with the Galapagos evolutionary solver that determined optimal locations for newly introduced amenities. The model was optimized by minimizing cumulative network-based distances between new and existing amenities, enabling it to explore a wide variety of spatial scenarios and find the most accessible locations. The scenario-based evaluation approach evaluated two types of amenity and embedded the optimized solutions into the mobility model for the evaluation, which assesses pedestrian accessibility. Results indicate that even marginal, well-crafted interventions, if well placed with careful strategy, can substantially improve walkability, leading to a 15.33% improvement in Walk Score. Placed optimized amenities aligned with highly connected pedestrian corridors, underlining the influence of network geometry on accessibility. The research verifies the efficacy of combined generative design, network analysis, and optimization methods and suggests a scalable, data-driven framework to support pedestrian-oriented development in campus and similarly controlled urban environments.

ABSTRACT. Swarm unmanned surface vehicles (USVs) have become a key technology for future maritime defense due to their significant capability for cooperative and autonomous operations. Based on swarm coordination strategies, encirclement tasks make a significant contribution to target containment, interception, and area protection. This study investigates a multi-agent reinforcement learning (MARL) approach for the swarm USV encirclement task, comparing two algorithms: the basic Multi-Agent Proximal Policy Optimization (MAPPO) and its recurrent extension, MAPPO-LSTM. These algorithms were trained in Unity 3D simulation platform using ML-Agents toolkit. Three defenders cooperatively surrounded a target under adapted real-map water environment. The performance evaluation is conducted using four metrics (accumulative reward, angular coverage, maximum angular gap, and rotation number of encirclement). Experimental results show that MAPPO-LSTM reaches better cumulative rewards and improved temporal stability, while the basic MAPPO model produces broader spatial coverage and tighter angular formation. The use of LSTM improves motion smoothness and coordination through temporal memory, resulting in more consistent encirclement behavior. These findings highlight the trade-off between spatial completeness and temporal coherence in USV swarm encirclement and emphasize the potential of the MARL framework for smart maritime defense applications.

ABSTRACT. This paper presents a decentralized control strategy for mobile robot swarms navigating unknown environments. The approach is developed and tested using foot-bot robots in the ARGoS simulator. In contrast to many potential field (PF)–based methods that rely on prior map knowledge or partial coordination, our controller handles simultaneous trajectory tracking, obstacle avoidance, and formation keeping entirely through local sensing and neighbor interaction. Each robot independently computes attractive, repulsive, and formation related forces at runtime without centralized control or environmental maps. The framework demonstrates strong adaptability across various scenarios, including narrow passages, wide obstacles, and dynamic swarm reconfigurations. % The results the framework is scalable and robust, making it a practical PF-based strategy for decentralized smart mobility applications.

ABSTRACT. The deployment of unmanned aerial vehicle (UAV) swarms in urban mobility applications such as traffic monitoring, infrastructure inspection, and emergency response is challenged by dense obstacles, limited communication, and the need for scalable coordination strategies. This paper aims to develop a fully decentralized control framework that enables multiple UAVs to maintain formation, avoid obstacles, and track mission objectives in complex urban environments using only local information. To achieve this objective, a Laplacian-based consensus formation controller is integrated with an adaptive Artificial Potential Field (APF) method for obstacle avoidance, while a leader-based tracking strategy guides the swarm centroid toward user-defined waypoints. Stability of the overall system is established through Lyapunov-based analysis, demonstrating asymptotic convergence of formation and tracking errors. Python-based simulations validate the proposed approach in obstacle-rich scenarios, showing smooth formation convergence, collision-free navigation, and consistent preservation of safety distances (20 px inter-UAV and 30 px UAV–obstacle) under dynamic goal changes. The results indicate that the proposed decentralized and safety-aware control strategy is well suited for real-time urban UAV operations. From a policy and planning perspective, the framework supports scalable aerial cooperation with minimal communication requirements, providing a practical building block for integrating UAV swarms into future smart-city and urban mobility ecosystems.

ABSTRACT. Tuning MPC weight matrices remains a tedious process that demands considerable expertise. A method is presented that enables a reinforcement learning agent to adjust gains online while enforcing Lyapunov-based bounds guaranteeing that every proposed gain lies inside a provably stable region. The resulting adaptive controller maintains stability regardless of policy network outputs. Across four UAV platforms spanning a 200-fold mass range (27 g to 5.5 kg), tracking improvements of 22–27% are observed on an aggressive 3D figure-8 trajectory spanning ±4.0m on x and y axes. Position root mean square error (RMSE) is reduced from 0.45–0.55m to 0.33–0.43m across all platforms, with variance reductions of 28–33%. No stability violations occurred throughout 60 evaluation trials. A projection operator clips out-of-bounds gains before they reach the MPC solver, acting as a hard safety layer. Sequential transfer learning reduces per-platform training by 75%. These findings demonstrate that formal stability constraints and learning-based adaptation can coexist effectively.

ABSTRACT. Smart city mobility increasingly depends on fleets of autonomous and remotely piloted vehicles that support logistics, delivery, and emergency response. Their continuous connectivity through wireless control and sensing expands the attack surface and exposes every node to diverse, rapidly evolving cyber threats. A single detection model cannot remain effective across such heterogeneous and dynamic environments. At the same time, most aerial or ground vehicles operate under strict energy and computation limits, which prevent the use of heavy deep neural networks (DNNs). Intelligent decision-making is also required to coordinate local learning and global updates efficiently. To address these challenges, we design a hierarchical intrusion detection framework that separates heavy and lightweight learning roles across the network. The proposed system trains a global model at a ground control station (GCS) on aggregated traffic data, and each unmanned aerial vehicle (UAV) maintains a compact few-shot learner tuned to its own local traffic. A deep Q-network (DQN) controller on the ground decides when to update models and how to allocate retraining effort among clients. This architecture allows rapid local adaptation and global resilience without overloading airborne resources. Experiments on the UAVIDS-2025 dataset show that the few-shot UAV models reach about 99 % detection accuracy with minimal computation, while the heavier convolutional neural network–gated recurrent unit (CNN–GRU) global model achieves nearly 98 %. The coordinated DQN controller reduces redundant retraining and communication, proving the feasibility of federated few-shot learning for secure and resource-aware smart-city UAV operations.