Mohammad El-Yabroudi

Work zones are essential for infrastructure maintenance, yet remain among the most hazardous environments for both workers and motorists. Despite widespread use of Temporary Traffic Control (TTC), incident rates remain elevated, suggesting limitations in current practices and insufficient integration of Internal Temporary Traffic Control (ITTC). This study employs a mixed-methods survey design to evaluate practitioner perceptions of TTC and ITTC effectiveness, identify perceived incident contributing factors, examine implementation challenges, and assess mitigation measures. Data were collected from 60 experienced transportation professionals using structured Likert-scale questions and open-ended responses. Quantitative data were analyzed using repeated-measures ANOVA to compare perceived incident contributors and mitigation effectiveness, while qualitative responses were analyzed using thematic analysis with affinity-based clustering of practitioner responses. Results indicate that distracted driving (M = 9.2) and speeding (M = 8.3) were perceived as the dominant contributors to work-zone crashes, significantly exceeding improper TTC and ITTC implementation. Among mitigation measures, visible police presence was rated most effective (M = 8.0, p < 0.001), outperforming automated ticketing and static pre-construction planning. Qualitative thematic analysis identified seven improvement domains: work-zone design and management, enforcement and speed control, driver education, worker training and certification, policy development, technology solutions, and inter-agency coordination. The findings demonstrate that sustainable work-zone safety requires an integrated socio-technical approach combining behavioral enforcement, adaptive TTC and ITTC management, structured inspection practices, and organizational accountability. The study provides practitioner-informed guidance for agencies and contractors to institutionalize ITTC within specifications, prioritize high-impact enforcement strategies, and adopt safety management frameworks shaped by field-tested professional expertise to address recurrent safety challenges in active construction environments.

Robotic grasping based on monocular vision remains challenging due to perception uncertainty, pixel-to-world mapping errors, and a tight per-frame budget. This paper presents a lightweight vision-based multi-object robotic grasping framework that integrates object detection, polynomial pixel-to-world mapping, and rule-based grasp execution into a unified pipeline. Object detection is carried out with a You Only Look Once version 13 (YOLOv13) detector, into which lightweight structural modules are inserted for the tabletop grasping task. To enable accurate perception-to-action mapping, a polynomial calibration model is introduced to transform image-space detections into planar world-space coordinates, effectively correcting geometric distortions caused by preprocessing. Based on the mapped targets, a rule-based grasp planning strategy is developed to generate physically feasible grasp poses for a Universal Robots 5 (UR5) robotic arm without relying on depth sensors or learning-based grasp prediction. The proposed framework is implemented and validated in CoppeliaSim. Quantitative results show end-to-end grasp success rising from 86% with the center-point baseline to 94% with the proposed strategy on the same 50 paired placements, while mapping and planning together add about 1 ms per frame. The gain comes from enforcing execution feasibility on an already sufficient map, not from a more accurate mapping model.

The Internet of Medical Things (IoMT) plays a crucial role in smart healthcare by leveraging medical devices that generate vast amounts of data for monitoring and diagnosis. IoMT networks are prone to challenges like dynamic topology, computational limitations, and power constraints that make it difficult to process data. Offloading computationally intensive operations to high-capacity computational nodes is necessary. However, medical devices provide nonindependent and identically distributed (non-i.i.d.) data in different formats, making it more complex. Conventional rule-based offloading schemes are inefficient in enhancing network performance and hence have high latency and energy consumption. This article introduces a secure computational offloading framework using deep reinforcement learning. With deep deterministic policy gradient (DDPG) and federated learning as the basis, the framework: 1) observes the network; 2) learns offloading decisions; and 3) safeguards data privacy. It reduces energy and latency with a security guarantee. A~cloud-fog architecture is implemented with a DDPG-based local model, which is executed at the fog broker, and a global model that is executed at the cloud server. We use advance encryption standard (AES) + encryption for secure transmission of data (ECC). MATLAB simulations are conducted for encrypted and unencrypted offloading and exhibit up to 12% improvement in encrypted cases and 13% in unencrypted cases compared to existing solutions.

Autonomous vehicles (AVs) are expected to eventually replace traditional vehicles that require human drivers. In recent years, several AV manufacturers have begun on-road testing to validate the safety of these vehicles. California is one of the few states to permit such testing, regulating it through a permit system. To ensure transparency and public awareness, the state mandates that any licensed AV manufacturer conducting on-road tests report crashes involving AVs. This must be conducted using a standardized format known as OL316, a requirement that has been in place since late 2014. While previous research has explored AV crash data, most studies have focused on specific timeframes without covering the entire period since 2014. Moreover, converting the data from PDFs to machine-readable formats has often been a manual process, and the description text field in reports has rarely been fully analyzed. This article presents a comprehensive, machine-readable dataset of AV crashes from 2014 to September 2024, along with publicly available parsing code to streamline future data analysis. Additionally, we provide an updated statistical analysis of AV crashes during this period. Furthermore, we leverage Generative AI (GenAI) to analyze the description text field of the OL316 reports. This analysis identifies common crash scenarios, contributing factors, and additional insights into moderate and major incidents. The final dataset comprises 728 crash entries. Notably, only 2% of the crashes were categorized as major, while 14% were classified as moderate. Furthermore, 43% of the crashes occurred while the AV was stationary, whereas 55% took place while the AV was in motion. Our GenAI analysis indicates that, in many instances, human drivers of non-autonomous vehicles were at fault. Common causes include rear-end collisions due to insufficient following distances, traffic violations such as running red lights or stop signs, and reckless behaviors like lane boundary violations or speeding.

Electric vehicles (EVs) are becoming increasingly popular as an efficient transportation solution but they also present unique challenges for energy management. Bi-directional charging (BDC) is a solution that allows EVs to not only consume energy from the grid but also supply energy back to the grid. This facilitates vehicle-to-load (V2L) integration, where EVs can act as mobile power sources for homes, buildings, and the grid. V2L enables better energy management by utilizing EVs as a flexible resource to balance grid demand and supply in the proposed system. This is achieved through intelligent coordination between the EVs, charging stations, and the grid, using smart meters and communication networks. Integration of BDC and V2L also enables EVs to provide backup power during grid outages, reduce the need for costly grid infrastructure, and support renewable energy integration. BDC with V2L integration is a promising approach for optimal energy management in EVs and can play a significant role in the future of sustainable transportation and energy systems. The proposed model reached 95.13% charging efficiency, 95.03% energy management, 95.69% power rating, 96.28% voltage support and 87.99% temperature management.

Pixel-level depth information is crucial to many applications, such as autonomous driving, robotics navigation, 3D scene reconstruction, and augmented reality. However, depth information, which is usually acquired by sensors such as LiDAR, is sparse. Depth completion is a process that predicts missing pixels’ depth information from a set of sparse depth measurements. Most of the ongoing research applies deep neural networks on the entire sparse depth map and camera scene without utilizing any information about the available objects, which results in more complex and resource-demanding networks. In this work, we propose to use image instance segmentation to detect objects of interest with pixel-level locations, along with sparse depth data, to support depth completion. The framework utilizes a two-branch encoder–decoder deep neural network. It fuses information about scene available objects, such as objects’ type and pixel-level location, LiDAR, and RGB camera, to predict dense accurate depth maps. Experimental results on the KITTI dataset showed faster training and improved prediction accuracy. The proposed method reaches a convergence state faster and surpasses the baseline model in all evaluation metrics.

LiDAR point cloud clustering is an essential part of a wide range of applications such as object detection, object recognition, and localization. In this paper, we focus on Density Based Spatial Clustering of Applications with Noise (DBSCAN) as a very promising and efficient algorithm to cluster LiDAR point cloud. However, it requires two tuning parameters that are manually specified based on prior knowledge of the domain. Furthermore, the selected tuning parameters cannot be modified once the clustering process begins. Environments in domains like robotics and autonomous driving are very dynamic, and hence, automated and adaptive selection of these parameters is preferred. In this work, we propose to estimate DBSCAN tuning parameters automatically based on a field of view division scheme, and points’ distances and densities within the division scheme. Furthermore, we adaptively estimate the tuning parameters for every point cloud frame separately. We evaluated the proposed method using the KITTI dataset, and the preliminary results show that the proposed method can achieve comparable results to the original DBSCAN without the need for manually selecting the parameters.