Pub Date : 2025-11-28DOI: 10.1109/OJITS.2025.3638660
Anubhav Pandey;Swapnil Yadav;Tejasvi Alladi;Fei Richard Yu
Railway infrastructure plays a critical role in transportation systems, and its routine inspection is crucial for ensuring operational stability and safety. Traditional railway inspection methods often rely heavily on fixed sensors and human monitoring, which are expensive to set up and time-consuming, respectively. This paper presents an autonomous drone-based railway monitoring system to detect structural defects and obstructions on railway tracks in real time. The flight stack comprises modern robotic frameworks such as PX4-Autopilot and ROS2. The sensor stack consists of an RGB camera for object detection and a depth camera for altitude estimation. Two parallel object detection pipelines, regular and oriented bounding box (OBB) YOLOv11 models, are fine-tuned to enhance detection accuracy under challenging visual conditions. Simulation results demonstrate the system’s effectiveness in detecting anomalies like sleeper misalignments and railway track obstructions. The system performance is tested with varying model sizes. The YOLOv11n model achieved an F1-score of 0.92 and an average latency of 59 ms per frame, providing a strong balance between accuracy and speed. Controller evaluations across speeds up to 1 m/s showed lateral and yaw RMSEs of 0.30 m and 2.01 deg, respectively, confirming stable and precise navigation. These findings highlight the potential of autonomous aerial systems to supplement or replace traditional railway inspection methods.
{"title":"An Autonomous Drone-Based Framework for Real-Time Railway Monitoring Using YOLO-Based Defect Detection","authors":"Anubhav Pandey;Swapnil Yadav;Tejasvi Alladi;Fei Richard Yu","doi":"10.1109/OJITS.2025.3638660","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3638660","url":null,"abstract":"Railway infrastructure plays a critical role in transportation systems, and its routine inspection is crucial for ensuring operational stability and safety. Traditional railway inspection methods often rely heavily on fixed sensors and human monitoring, which are expensive to set up and time-consuming, respectively. This paper presents an autonomous drone-based railway monitoring system to detect structural defects and obstructions on railway tracks in real time. The flight stack comprises modern robotic frameworks such as PX4-Autopilot and ROS2. The sensor stack consists of an RGB camera for object detection and a depth camera for altitude estimation. Two parallel object detection pipelines, regular and oriented bounding box (OBB) YOLOv11 models, are fine-tuned to enhance detection accuracy under challenging visual conditions. Simulation results demonstrate the system’s effectiveness in detecting anomalies like sleeper misalignments and railway track obstructions. The system performance is tested with varying model sizes. The YOLOv11n model achieved an F1-score of 0.92 and an average latency of 59 ms per frame, providing a strong balance between accuracy and speed. Controller evaluations across speeds up to 1 m/s showed lateral and yaw RMSEs of 0.30 m and 2.01 deg, respectively, confirming stable and precise navigation. These findings highlight the potential of autonomous aerial systems to supplement or replace traditional railway inspection methods.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1655-1666"},"PeriodicalIF":5.3,"publicationDate":"2025-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11271341","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729455","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-28DOI: 10.1109/OJITS.2025.3637819
Yi Liang;Rongjun Fu;Lai Zhou
Motion cueing algorithms (MCAs) enable realistic motion cues on limited-workspace platforms for flight training. In upset prevention and recovery training, conventional MCAs often fail to deliver critical sensory cues, which reduces pilot training performance. This study proposes a model predictive control-based MCA optimized using a deep Q-network. Motion cueing weights for specific scenarios are learned under platform constraints. As a result, the optimized MCAs improved overall perceptual tracking by at least 11.6% and increased the subjective ratings by at least 7.8%. These findings demonstrate the effectiveness of the proposed method in enhancing motion realism and training efficiency under high-demand conditions.
{"title":"Deep Q-Network-Based Optimization of Model Predictive Control Motion Cueing Algorithm for Specific Scenario","authors":"Yi Liang;Rongjun Fu;Lai Zhou","doi":"10.1109/OJITS.2025.3637819","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3637819","url":null,"abstract":"Motion cueing algorithms (MCAs) enable realistic motion cues on limited-workspace platforms for flight training. In upset prevention and recovery training, conventional MCAs often fail to deliver critical sensory cues, which reduces pilot training performance. This study proposes a model predictive control-based MCA optimized using a deep Q-network. Motion cueing weights for specific scenarios are learned under platform constraints. As a result, the optimized MCAs improved overall perceptual tracking by at least 11.6% and increased the subjective ratings by at least 7.8%. These findings demonstrate the effectiveness of the proposed method in enhancing motion realism and training efficiency under high-demand conditions.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"29-40"},"PeriodicalIF":5.3,"publicationDate":"2025-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11271304","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830791","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-26DOI: 10.1109/OJITS.2025.3637700
Chenlei Han;Michael Frey;Frank Gauterin
With the increasing level of driving automation, localization and navigation are not only used to provide positioning and route guidance information for users, but are also important inputs for vehicle control.Odometry localization method is the most widely used localization method due to its good short-term accuracy and cost-effectiveness, despite its known limitations like drift and environment dependency. Optimizing odometry remains a valuable area of research. By using the UKF-based odometry localization methode for vehicles with increased maneuverability introduced in our previous work, this paper presents a simulation-based optimization method to improve the accuracy of the odometry. This proposed simulation-based optimization method aims to achieve the accuracy goal with low computation effort. The covariance matrices of the UKF-based odometry are optimized by the particle swarm algorithm. In order to make the in simulation optimized covariance also applicable in the real vehicle, sensor error models are built up to generate realistic sensor signals. To reduce the computation effort during optimization an efficient driving maneuver, which covers more vehicle states is generated and used instead of normal parking maneuvers. The use of the efficient driving maneuver has been shown to reduce the optimization effort by approximately 60% without sacrifice the optimization effect. The efficacy of the optimized covariance matrices in enhancing odometry accuracy has been validated in both simulated and real-driving tests. The optimized odometry can reach an average end position error of $11cm$ and average end orientation error of 0.4°. Furthermore, a sensitivity analysis of sensor accuracy and noise level on odometry has been performed in the simulation environment with the help of the proposed optimization methods. Odometry using sensors of various accuracy and noise levels are optimized to achieve its best performance. The simulation results indicate the importance of the IMU sensor in the odometry localization method. This conclusion is supported by the results of a real driving test that used two IMU sensors with different accuracy and noise levels. The results of the sensitivity analysis provides a basis for sensor selection in vehicle system design.
{"title":"A Simulation-Based Efficient Optimization Method of an Odometry Localization Filter for Vehicles With Increased Maneuverability","authors":"Chenlei Han;Michael Frey;Frank Gauterin","doi":"10.1109/OJITS.2025.3637700","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3637700","url":null,"abstract":"With the increasing level of driving automation, localization and navigation are not only used to provide positioning and route guidance information for users, but are also important inputs for vehicle control.Odometry localization method is the most widely used localization method due to its good short-term accuracy and cost-effectiveness, despite its known limitations like drift and environment dependency. Optimizing odometry remains a valuable area of research. By using the UKF-based odometry localization methode for vehicles with increased maneuverability introduced in our previous work, this paper presents a simulation-based optimization method to improve the accuracy of the odometry. This proposed simulation-based optimization method aims to achieve the accuracy goal with low computation effort. The covariance matrices of the UKF-based odometry are optimized by the particle swarm algorithm. In order to make the in simulation optimized covariance also applicable in the real vehicle, sensor error models are built up to generate realistic sensor signals. To reduce the computation effort during optimization an efficient driving maneuver, which covers more vehicle states is generated and used instead of normal parking maneuvers. The use of the efficient driving maneuver has been shown to reduce the optimization effort by approximately 60% without sacrifice the optimization effect. The efficacy of the optimized covariance matrices in enhancing odometry accuracy has been validated in both simulated and real-driving tests. The optimized odometry can reach an average end position error of <inline-formula> <tex-math>$11cm$ </tex-math></inline-formula> and average end orientation error of 0.4°. Furthermore, a sensitivity analysis of sensor accuracy and noise level on odometry has been performed in the simulation environment with the help of the proposed optimization methods. Odometry using sensors of various accuracy and noise levels are optimized to achieve its best performance. The simulation results indicate the importance of the IMU sensor in the odometry localization method. This conclusion is supported by the results of a real driving test that used two IMU sensors with different accuracy and noise levels. The results of the sensitivity analysis provides a basis for sensor selection in vehicle system design.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1580-1595"},"PeriodicalIF":5.3,"publicationDate":"2025-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11269360","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674661","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-26DOI: 10.1109/OJITS.2025.3637305
Iván Gómez;Sergio Ilarri
Efficient urban traffic management is a crucial challenge in modern smart cities, especially in densely populated areas with complex and dynamic traffic conditions. In this paper, we tackle the traffic prediction problem and present a lightweight architecture that combines sensor embeddings with dense layers, sustaining strong performance across both short- and long-term forecasting horizons while substantially reducing training time and enabling fast inference times. In comparative evaluations, our approach matches or surpasses the accuracy of more complex methods and consistently improves efficiency. To foster reproducibility, we release the code along with an enriched dataset that integrates traffic flows with contextual features such as weather conditions, temporal variables, and urban attributes. The richness and coverage of this dataset exceed those of existing public resources, enabling deeper and more comprehensive analyses of traffic dynamics. Overall, we demonstrate that a lightweight, well-designed architecture can achieve high performance and practical scalability for urban mobility management.
{"title":"Advanced Prediction of Traffic at Different Temporal Scales Using Heterogeneous Data Sources","authors":"Iván Gómez;Sergio Ilarri","doi":"10.1109/OJITS.2025.3637305","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3637305","url":null,"abstract":"Efficient urban traffic management is a crucial challenge in modern smart cities, especially in densely populated areas with complex and dynamic traffic conditions. In this paper, we tackle the traffic prediction problem and present a lightweight architecture that combines sensor embeddings with dense layers, sustaining strong performance across both short- and long-term forecasting horizons while substantially reducing training time and enabling fast inference times. In comparative evaluations, our approach matches or surpasses the accuracy of more complex methods and consistently improves efficiency. To foster reproducibility, we release the code along with an enriched dataset that integrates traffic flows with contextual features such as weather conditions, temporal variables, and urban attributes. The richness and coverage of this dataset exceed those of existing public resources, enabling deeper and more comprehensive analyses of traffic dynamics. Overall, we demonstrate that a lightweight, well-designed architecture can achieve high performance and practical scalability for urban mobility management.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1539-1550"},"PeriodicalIF":5.3,"publicationDate":"2025-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11269361","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674735","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-26DOI: 10.1109/OJITS.2025.3637341
Manfredi Napolitano;Alessandra Somma;Alessandra De Benedictis;Nicola Mazzocca
Urban mobility systems are growing increasingly complex due to rapid urbanization, evolving transportation demands, and rising congestion levels, which lead to travel delays and increased environmental impact. These challenges underscore the importance of developing advanced modeling, monitoring, and management approaches that can capture the dynamics of urban mobility systems and support more efficient, sustainable and adaptive transportation solutions. Mobility Digital Twin (MoDT) has emerged as a promising paradigm for monitoring, managing, and predicting urban mobility by maintaining a synchronized digital replica of the transportation system and enabling feedback to the physical infrastructure. However, most existing MoDT implementations remain fragmented, and tightly bound to specific use cases, limiting scalability, reusability, and broader applicability. This paper presents FlowTwin, a comprehensive methodology and software framework for developing a MoDT for traffic flow monitoring. The proposed approach introduces a dual-phase methodology: an offline phase for modeling and calibration using macroscopic and microscopic traffic flow models to capture both aggregate and fine-grained driver dynamics; and an online phase that enables simulation, monitoring, and feedback generation. The presented architecture is aligned with the capabilities proposed by Digital Twin Consortium, ensuring structured design and extensibility. The feasibility and effectiveness of FlowTwin are validated through its instantiation in the Italian city of Bologna, resulting in BoMoDT. BoMoDT operates on traffic flow streams generated by a city-scale emulator built from real-world mobility data, enabling continuous monitoring, simulation, and feedback within a realistic but controlled environment. Quantitative evaluations confirm BoMoDT’s capability to support accurate traffic simulation and responsive monitoring under diverse traffic scenarios.
{"title":"FlowTwin: A Digital Twin for Traffic Flow Monitoring","authors":"Manfredi Napolitano;Alessandra Somma;Alessandra De Benedictis;Nicola Mazzocca","doi":"10.1109/OJITS.2025.3637341","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3637341","url":null,"abstract":"Urban mobility systems are growing increasingly complex due to rapid urbanization, evolving transportation demands, and rising congestion levels, which lead to travel delays and increased environmental impact. These challenges underscore the importance of developing advanced modeling, monitoring, and management approaches that can capture the dynamics of urban mobility systems and support more efficient, sustainable and adaptive transportation solutions. Mobility Digital Twin (MoDT) has emerged as a promising paradigm for monitoring, managing, and predicting urban mobility by maintaining a synchronized digital replica of the transportation system and enabling feedback to the physical infrastructure. However, most existing MoDT implementations remain fragmented, and tightly bound to specific use cases, limiting scalability, reusability, and broader applicability. This paper presents FlowTwin, a comprehensive methodology and software framework for developing a MoDT for traffic flow monitoring. The proposed approach introduces a dual-phase methodology: an offline phase for modeling and calibration using macroscopic and microscopic traffic flow models to capture both aggregate and fine-grained driver dynamics; and an online phase that enables simulation, monitoring, and feedback generation. The presented architecture is aligned with the capabilities proposed by Digital Twin Consortium, ensuring structured design and extensibility. The feasibility and effectiveness of FlowTwin are validated through its instantiation in the Italian city of Bologna, resulting in BoMoDT. BoMoDT operates on traffic flow streams generated by a city-scale emulator built from real-world mobility data, enabling continuous monitoring, simulation, and feedback within a realistic but controlled environment. Quantitative evaluations confirm BoMoDT’s capability to support accurate traffic simulation and responsive monitoring under diverse traffic scenarios.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1551-1568"},"PeriodicalIF":5.3,"publicationDate":"2025-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11269355","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674734","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Autonomous driving stands at the forefront of next-generation mobility, driven by advances in Artificial Intelligence (AI), deep learning, and real-time sensor data processing. While Foundation Models (FMs)–large, pre-trained neural networks capable of generalizing across tasks–have revolutionized fields such as natural language processing and computer vision, their integration into autonomous driving remains limited and fragmented. This paper addresses this critical gap by systematically reviewing the application of FMs across the autonomous driving pipeline, from perception and scene understanding to reasoning, planning, and synthetic dataset generation. We classify over 70 models by architecture, modality, task type, and input/output structure, and provide a unified framework for understanding their role in intelligent vehicle systems. Key contributions include: (i) a taxonomy of FMs deployed in perception, reasoning, and control; (ii) identification of current limitations in real-time deployment, interpretability, and safety assurance; and (iii) emerging trends such as prompt-based fine-tuning, multimodal grounding, and generative scenario synthesis. Our findings highlight both the opportunities and challenges of incorporating FMs into safety-critical autonomous systems and outline promising directions for future research in edge-efficient, robust, and explainable FM-based driving models.
{"title":"Foundation Models in Autonomous Driving: A Review of Current Tasks and Applications","authors":"Artemis Stefanidou;Elena Politi;George Dimitrakopoulos","doi":"10.1109/OJITS.2025.3633871","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3633871","url":null,"abstract":"Autonomous driving stands at the forefront of next-generation mobility, driven by advances in Artificial Intelligence (AI), deep learning, and real-time sensor data processing. While Foundation Models (FMs)–large, pre-trained neural networks capable of generalizing across tasks–have revolutionized fields such as natural language processing and computer vision, their integration into autonomous driving remains limited and fragmented. This paper addresses this critical gap by systematically reviewing the application of FMs across the autonomous driving pipeline, from perception and scene understanding to reasoning, planning, and synthetic dataset generation. We classify over 70 models by architecture, modality, task type, and input/output structure, and provide a unified framework for understanding their role in intelligent vehicle systems. Key contributions include: (i) a taxonomy of FMs deployed in perception, reasoning, and control; (ii) identification of current limitations in real-time deployment, interpretability, and safety assurance; and (iii) emerging trends such as prompt-based fine-tuning, multimodal grounding, and generative scenario synthesis. Our findings highlight both the opportunities and challenges of incorporating FMs into safety-critical autonomous systems and outline promising directions for future research in edge-efficient, robust, and explainable FM-based driving models.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1522-1538"},"PeriodicalIF":5.3,"publicationDate":"2025-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11251045","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145612098","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Tugboat–barge coordination in inland waterway transportation presents critical multi-objective optimization challenges due to interdependent constraints including fleet capacity, operational costs, dynamic tidal conditions, and temporal accessibility windows. Traditional approaches fail to effectively address these complex interdependencies in constrained inland waterway environments. This paper proposes Multi-Objective Generative Adversarial Learning and Search for Intelligent Transportation Systems (MGALS-ITS), integrating reinforcement learning-based construction, generative adversarial network-driven local search, and adaptive optimization specifically for tugboat–barge scheduling in tidal inland waterways. The Reinforcement Learning (RL) component learns from constraint patterns to generate feasible, cost-efficient coordination schedules for tugboat–barge operations. A conditional Wasserstein Generative Adversarial Network (GAN) refines solutions through learned neighborhood exploration, while adaptive strategies and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) enable real-time cost-makespan trade-offs. Experimental validation on comprehensive inland waterway scenarios involving 103 tugboats, 80 barges, and 48 customer destinations demonstrates superior performance over conventional scheduling methods. MGALS-ITS achieves lowest operational costs and shortest completion times, surpassing Long Short-Term Memory and Random Forest (LSTM+RF) baselines while generating 15.4% more diverse solutions and 31% more feasible configurations than existing systems, with 20–35% greater resilience against operational disruptions. This research positions MGALS-ITS as an adaptive decision support framework for tugboat–barge operations in inland waterway networks, offering significant performance improvements for tidal waterway logistics optimization.
{"title":"Intelligent Multi-Objective Tugboat–Barge Scheduling for Inland Waterway Operations Using Generative Adversarial Learning and Reinforcement-Based Optimization","authors":"Rapeepan Pitakaso;Kanchana Sethanan;Thanatkij Srichok;Kongkidakhon Worasan;Kuo-Jui Wu","doi":"10.1109/OJITS.2025.3631650","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3631650","url":null,"abstract":"Tugboat–barge coordination in inland waterway transportation presents critical multi-objective optimization challenges due to interdependent constraints including fleet capacity, operational costs, dynamic tidal conditions, and temporal accessibility windows. Traditional approaches fail to effectively address these complex interdependencies in constrained inland waterway environments. This paper proposes Multi-Objective Generative Adversarial Learning and Search for Intelligent Transportation Systems (MGALS-ITS), integrating reinforcement learning-based construction, generative adversarial network-driven local search, and adaptive optimization specifically for tugboat–barge scheduling in tidal inland waterways. The Reinforcement Learning (RL) component learns from constraint patterns to generate feasible, cost-efficient coordination schedules for tugboat–barge operations. A conditional Wasserstein Generative Adversarial Network (GAN) refines solutions through learned neighborhood exploration, while adaptive strategies and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) enable real-time cost-makespan trade-offs. Experimental validation on comprehensive inland waterway scenarios involving 103 tugboats, 80 barges, and 48 customer destinations demonstrates superior performance over conventional scheduling methods. MGALS-ITS achieves lowest operational costs and shortest completion times, surpassing Long Short-Term Memory and Random Forest (LSTM+RF) baselines while generating 15.4% more diverse solutions and 31% more feasible configurations than existing systems, with 20–35% greater resilience against operational disruptions. This research positions MGALS-ITS as an adaptive decision support framework for tugboat–barge operations in inland waterway networks, offering significant performance improvements for tidal waterway logistics optimization.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1500-1521"},"PeriodicalIF":5.3,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11240121","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145612113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-11DOI: 10.1109/OJITS.2025.3631661
Zongyuan Wu;Ben Waterson;Craig B. Rafter;Bani Anvari;Yadan Yan
This paper develops an Estimation of Unequipped Vehicle with Occupancy (EUVO) algorithm to predict departure times and occupancy levels for vehicles across different approaching lanes in mixed traffic environments containing both Connected Vehicles (CVs) and Unequipped Vehicles (UVs). The algorithm integrates multi-source data from CVs, loop detectors and roadside cameras. After processing by EUVO algorithm, a Person-based Deep Deterministic Policy Gradient (PB-DDPG) algorithm is proposed to improve the performance of person-based traffic control under low CV penetration rates and reduce computational complexity using Deep Reinforcement Learning (DRL). The integration of EUVO and PB-DDPG algorithms reconstructs the states of both CVs and UVs, combining vehicle occupancy levels and excess waiting time as input data. Through a trial-and-error training process, it derives optimal signal timing solutions with flexible actions and person-based rewards. The method remains effective even at low CV penetration rates, ranging from 0% to 20%. The algorithm is evaluated across two study sites in Hull and Birmingham, U.K., under various traffic scenarios. Results show that compared with the vehicle-based DQTSC-M model, PB-DDPG reduces average person delay and the number of person stops by approximately 18.3% and 19.6%, respectively. It also exhibits faster convergence and more stable performance than the Double Deep Q Network (DDQN) model. In addition, the EUVO algorithm significantly improves the performance of PB-DDPG in reducing average person delay and stops under the following conditions: CV penetration rates below 90%, UV position estimation errors within 6 meters, loop detection errors below 50%, loop detection latency within 2 seconds, and camera occupancy detection errors below 30%.
本文提出了一种带占用率的无装备车辆估计（EUVO）算法，用于预测混合交通环境中车辆在不同接近车道上的出发时间和占用水平，该混合交通环境包含了联网车辆（cv）和无装备车辆（UVs）。该算法集成了来自cv、环路检测器和路边摄像头的多源数据。在EUVO算法处理后，提出了一种基于人的深度确定性策略梯度（PB-DDPG）算法，以提高低CV渗透率下基于人的交通控制性能，并利用深度强化学习（DRL）降低计算复杂度。结合EUVO和PB-DDPG算法，将车辆占用率和超额等待时间作为输入数据，重构cv和uv的状态。通过一个反复试验的训练过程，它可以通过灵活的行动和基于人的奖励来获得最佳的信号定时解决方案。即使在CV渗透率较低的情况下（从0%到20%），该方法仍然有效。该算法在英国赫尔和伯明翰的两个研究地点进行了各种交通场景的评估。结果表明，与基于车辆的DQTSC-M模型相比，PB-DDPG分别减少了约18.3%和19.6%的平均人员延误和人员停车次数。与双深度Q网络（Double Deep Q Network， DDQN）模型相比，它具有更快的收敛速度和更稳定的性能。此外，EUVO算法显著提高了PB-DDPG在减少平均人员延迟方面的性能，并在CV渗透率低于90%，UV位置估计误差在6米以内，环路检测误差低于50%，环路检测延迟在2秒以内，摄像机占用检测误差低于30%的条件下停止。
{"title":"An Unequipped Vehicle State Estimation Algorithm to Augment the Person-Based Control in Low Connected Vehicle Penetration Rates via Deep Reinforcement Learning","authors":"Zongyuan Wu;Ben Waterson;Craig B. Rafter;Bani Anvari;Yadan Yan","doi":"10.1109/OJITS.2025.3631661","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3631661","url":null,"abstract":"This paper develops an Estimation of Unequipped Vehicle with Occupancy (EUVO) algorithm to predict departure times and occupancy levels for vehicles across different approaching lanes in mixed traffic environments containing both Connected Vehicles (CVs) and Unequipped Vehicles (UVs). The algorithm integrates multi-source data from CVs, loop detectors and roadside cameras. After processing by EUVO algorithm, a Person-based Deep Deterministic Policy Gradient (PB-DDPG) algorithm is proposed to improve the performance of person-based traffic control under low CV penetration rates and reduce computational complexity using Deep Reinforcement Learning (DRL). The integration of EUVO and PB-DDPG algorithms reconstructs the states of both CVs and UVs, combining vehicle occupancy levels and excess waiting time as input data. Through a trial-and-error training process, it derives optimal signal timing solutions with flexible actions and person-based rewards. The method remains effective even at low CV penetration rates, ranging from 0% to 20%. The algorithm is evaluated across two study sites in Hull and Birmingham, U.K., under various traffic scenarios. Results show that compared with the vehicle-based DQTSC-M model, PB-DDPG reduces average person delay and the number of person stops by approximately 18.3% and 19.6%, respectively. It also exhibits faster convergence and more stable performance than the Double Deep Q Network (DDQN) model. In addition, the EUVO algorithm significantly improves the performance of PB-DDPG in reducing average person delay and stops under the following conditions: CV penetration rates below 90%, UV position estimation errors within 6 meters, loop detection errors below 50%, loop detection latency within 2 seconds, and camera occupancy detection errors below 30%.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1476-1499"},"PeriodicalIF":5.3,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11240116","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-11DOI: 10.1109/OJITS.2025.3631716
Weiguo Yi;Longteng Wang
In autonomous driving under complex weather conditions, issues such as low detection accuracy and poor model generalization persist due to unstable illumination, target blurring, image occlusion, and enhanced background noise. To address these challenges, this paper proposes an improved multi-scale object detection algorithm for autonomous driving based on the YOLO11s model. Initially, a small-object detection layer is added to the neck structure to enhance detection accuracy for minor targets and reduce the misdetection rate of distant vehicles during autonomous driving. Second, the LightSDI lightweight spatial-dimensional interaction module is introduced to optimize the original feature fusion layers, improving the model’s detection accuracy in harsh environments and strengthening its semantic perception capability. Third, DK_SCDown replaces the conventional downsampling convolution (Conv) to leverage multi-scale feature extraction and dynamic weighted fusion, thereby reinforcing receptive field coverage and dynamic feature selection while achieving parameter lightweighting through depthwise separable convolution. Finally, the Wise-IoU v2 loss function, more suitable for object detection tasks in complex environments, is adopted to mitigate issues like target boundary ambiguity and severe occlusion; it normalizes geometric distances between targets via center-point normalization to guide more rational regression. On the public SODA10M dataset, mAP@0.5 improves by 7.0% compared to the baseline model, while the parameter count decreases by 19%. Moreover, excellent performance on additional public datasets, including BDD100K, KITTI, and DAWN, further demonstrates the generalization capability of the improved model.
{"title":"Robust Object Detection for Autonomous Driving in Adverse Weather Conditions With Multi-Scale Feature Enhancement","authors":"Weiguo Yi;Longteng Wang","doi":"10.1109/OJITS.2025.3631716","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3631716","url":null,"abstract":"In autonomous driving under complex weather conditions, issues such as low detection accuracy and poor model generalization persist due to unstable illumination, target blurring, image occlusion, and enhanced background noise. To address these challenges, this paper proposes an improved multi-scale object detection algorithm for autonomous driving based on the YOLO11s model. Initially, a small-object detection layer is added to the neck structure to enhance detection accuracy for minor targets and reduce the misdetection rate of distant vehicles during autonomous driving. Second, the LightSDI lightweight spatial-dimensional interaction module is introduced to optimize the original feature fusion layers, improving the model’s detection accuracy in harsh environments and strengthening its semantic perception capability. Third, DK_SCDown replaces the conventional downsampling convolution (Conv) to leverage multi-scale feature extraction and dynamic weighted fusion, thereby reinforcing receptive field coverage and dynamic feature selection while achieving parameter lightweighting through depthwise separable convolution. Finally, the Wise-IoU v2 loss function, more suitable for object detection tasks in complex environments, is adopted to mitigate issues like target boundary ambiguity and severe occlusion; it normalizes geometric distances between targets via center-point normalization to guide more rational regression. On the public SODA10M dataset, mAP@0.5 improves by 7.0% compared to the baseline model, while the parameter count decreases by 19%. Moreover, excellent performance on additional public datasets, including BDD100K, KITTI, and DAWN, further demonstrates the generalization capability of the improved model.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1459-1475"},"PeriodicalIF":5.3,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11240119","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560748","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-30DOI: 10.1109/OJITS.2025.3626948
Debora Russo;Franca Rocco Di Torrepadula;Luigi Libero Lucio Starace;Sergio Di Martino;Nicola Mazzocca
The development of advanced data-driven Intelligent Transportation Systems (ITS) strongly relies on the availability of representative mobility datasets. While several datasets are publicly available, practically none explicitly represent anomalous mobility scenarios such as strikes, road closures, or sudden spikes in mobility demand due to special events, also due to the lack of standardized annotations for anomalies. Moreover, existing datasets often do not include fine-grained mobility traces due to privacy concerns, and generally do not fully capture the actual variability of real-world conditions. This poses a significant challenge for ITS researchers and practitioners, requiring accurate, annotated data to model, simulate, and analyze the effects of disruptive events on urban mobility. To address these gaps, in this paper, we present a solution for automatically generating synthetic urban mobility datasets including various anomalous scenarios. Built on top of the well-known SUMO framework, our solution is designed to apply to any urban road network, as it leverages open data sources to create detailed, scenario-specific datasets. The tool features a Graphical User Interface, empowering users, including non-technical staff such as urban planners and decision-makers, to easily generate realistic datasets, including fully customizable anomalous scenarios. We show the effectiveness of our proposal by conducting a case study based on the city of Genoa, Italy, leveraging publicly available data provided by the city’s Municipality. In the case study, we show how the solution can be employed to easily generate detailed mobility datasets involving different anomalous scenarios, and how the resulting datasets can be used to perform different fine-grained mobility analyses. Additionally, we assess the realism and consistency of the generated data by validating the internal plausibility and coherence of the synthetic mobility flows, including verification that spatio-temporal patterns align with widely accepted urban mobility principles. By democratizing access to high-quality, annotated mobility data for anomalous conditions, we envision that our tool could significantly contribute to the field of urban mobility research and practice.
{"title":"A Framework for Generating Synthetic Urban Mobility Datasets With Customizable Anomalous Scenarios","authors":"Debora Russo;Franca Rocco Di Torrepadula;Luigi Libero Lucio Starace;Sergio Di Martino;Nicola Mazzocca","doi":"10.1109/OJITS.2025.3626948","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3626948","url":null,"abstract":"The development of advanced data-driven Intelligent Transportation Systems (ITS) strongly relies on the availability of representative mobility datasets. While several datasets are publicly available, practically none explicitly represent anomalous mobility scenarios such as strikes, road closures, or sudden spikes in mobility demand due to special events, also due to the lack of standardized annotations for anomalies. Moreover, existing datasets often do not include fine-grained mobility traces due to privacy concerns, and generally do not fully capture the actual variability of real-world conditions. This poses a significant challenge for ITS researchers and practitioners, requiring accurate, annotated data to model, simulate, and analyze the effects of disruptive events on urban mobility. To address these gaps, in this paper, we present a solution for automatically generating synthetic urban mobility datasets including various anomalous scenarios. Built on top of the well-known SUMO framework, our solution is designed to apply to any urban road network, as it leverages open data sources to create detailed, scenario-specific datasets. The tool features a Graphical User Interface, empowering users, including non-technical staff such as urban planners and decision-makers, to easily generate realistic datasets, including fully customizable anomalous scenarios. We show the effectiveness of our proposal by conducting a case study based on the city of Genoa, Italy, leveraging publicly available data provided by the city’s Municipality. In the case study, we show how the solution can be employed to easily generate detailed mobility datasets involving different anomalous scenarios, and how the resulting datasets can be used to perform different fine-grained mobility analyses. Additionally, we assess the realism and consistency of the generated data by validating the internal plausibility and coherence of the synthetic mobility flows, including verification that spatio-temporal patterns align with widely accepted urban mobility principles. By democratizing access to high-quality, annotated mobility data for anomalous conditions, we envision that our tool could significantly contribute to the field of urban mobility research and practice.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"6 ","pages":"1439-1458"},"PeriodicalIF":5.3,"publicationDate":"2025-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11222764","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145510221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：