Pub Date : 2025-12-31DOI: 10.1109/OJITS.2025.3650057
M. Pobar;G. Paulin;M. Ivasic-Kos;A. Vorkapic
This work presents an automated, real-time approach for localized sea state estimation using a single camera mounted on a ship’s bridge. A labeled dataset of sea surface images, collected during regular operation of an overseas liner, is built and used to train deep neural networks to estimate sea state on the Beaufort scale from 1 to 8. To improve robustness and better capture operational variability, a substantially enlarged test set is constructed relative to previous work, enabling a more comprehensive evaluation under diverse navigation and environmental conditions. Additionally, to mitigate the scarcity of rarely occurring sea states in real-world operation, a synthetic training dataset is generated that simulates a wide range of sea and weather conditions while preserving key physical relationships to increase variability in illumination and wave appearance without degrading realism. We evaluated state-of-the-art convolutional and transformer-based architectures, including Resnet-101d, DeiT III, Swin transformer, XciT and CoAtNet. The impact of different synthetic-to-real training data ratios in both RGB and grayscale domains is systematically examined, yielding a 6% improvement in test accuracy and mean F1 score and a reduction of the maximum error from 7 to 3 Beaufort. Finally, a temporal voting framework that aggregates predictions over several consecutive frames further reduces the maximum error to 2 Beaufort and achieves 96% intra-1-class accuracy and an F1 score of 62%, substantially outperforming a baseline trained only on real data without temporal voting. Impact Statement—The maritime industry is increasingly adopting artificial intelligence to enhance operational efficiency, improve navigational safety and achieve sustainability goals. Ensuring navigational safety, as well as the environmental and economic efficiency of maritime operations, requires an accurate assessment of sea state. Traditionally, sea state is visually assessed using the Beaufort scale, which relates wind speed to sea surface state and classifies it into 13 classes (0-12). Since this method relies on visual observation, it is subjective and prone to human error, so automating sea state assessment using computer vision methods can provide an effective monitoring alternative. The models proposed in this paper achieved intra-1-class accuracy of 96% on a varied test set, demonstrating the effectiveness of this approach for robust sea state estimation.
{"title":"Automated Real-Time Localized Sea State Estimation During Navigation Based on the Beaufort Scale","authors":"M. Pobar;G. Paulin;M. Ivasic-Kos;A. Vorkapic","doi":"10.1109/OJITS.2025.3650057","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3650057","url":null,"abstract":"This work presents an automated, real-time approach for localized sea state estimation using a single camera mounted on a ship’s bridge. A labeled dataset of sea surface images, collected during regular operation of an overseas liner, is built and used to train deep neural networks to estimate sea state on the Beaufort scale from 1 to 8. To improve robustness and better capture operational variability, a substantially enlarged test set is constructed relative to previous work, enabling a more comprehensive evaluation under diverse navigation and environmental conditions. Additionally, to mitigate the scarcity of rarely occurring sea states in real-world operation, a synthetic training dataset is generated that simulates a wide range of sea and weather conditions while preserving key physical relationships to increase variability in illumination and wave appearance without degrading realism. We evaluated state-of-the-art convolutional and transformer-based architectures, including Resnet-101d, DeiT III, Swin transformer, XciT and CoAtNet. The impact of different synthetic-to-real training data ratios in both RGB and grayscale domains is systematically examined, yielding a 6% improvement in test accuracy and mean F1 score and a reduction of the maximum error from 7 to 3 Beaufort. Finally, a temporal voting framework that aggregates predictions over several consecutive frames further reduces the maximum error to 2 Beaufort and achieves 96% intra-1-class accuracy and an F1 score of 62%, substantially outperforming a baseline trained only on real data without temporal voting. Impact Statement—The maritime industry is increasingly adopting artificial intelligence to enhance operational efficiency, improve navigational safety and achieve sustainability goals. Ensuring navigational safety, as well as the environmental and economic efficiency of maritime operations, requires an accurate assessment of sea state. Traditionally, sea state is visually assessed using the Beaufort scale, which relates wind speed to sea surface state and classifies it into 13 classes (0-12). Since this method relies on visual observation, it is subjective and prone to human error, so automating sea state assessment using computer vision methods can provide an effective monitoring alternative. The models proposed in this paper achieved intra-1-class accuracy of 96% on a varied test set, demonstrating the effectiveness of this approach for robust sea state estimation.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"136-154"},"PeriodicalIF":5.3,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11320450","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929561","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 : 2025-12-30DOI: 10.1109/OJITS.2025.3649601
Chandranil Chakraborttii;Lin Cheng
With the growing number of vehicles in urban areas, traffic jams and parking inefficiencies are becoming an increasing challenge. To address this, Intelligent Parking Systems have been proposed for predicting parking availability and improving traffic flow, thereby improving driving experience. However, deploying such a system across various environments remains challenging because parking patterns vary between neighborhoods, streets, and cities. Traditional machine learning models require extensive retraining for new environments, which makes them impractical for large-scale, real-world use. In this paper, we introduce TransPark, a new transfer learning framework where a base model is pre-trained on a pool of similar datasets. This process helps the model learn generalized parking patterns. Next, we adapt the model to new target regions with minimal fine-tuning, which significantly reduces computational costs. TransPark is suitable where parking patterns differ in residential, commercial, and mixed-use areas. This approach reduces the resources needed to build separate models for each environment. We examine datasets from various urban neighborhoods in San Francisco and evaluate parking availability predictions in time intervals ranging from 1 to 60 minutes. We show TransPark effectively models complex spatio-temporal parking patterns by using a hybrid attention mechanism that decouples spatial learning (between streets) and temporal learning (between time-steps). By leveraging VPCL-driven pretraining and the hybrid attention mechanism, TransPark achieves 25-70% accuracy improvement over prior methods based on statistical, deep learning, and transformer baselines while reducing the computational cost of retraining by over 55%. This promises real-world deployability in diverse parking environments.
{"title":"TransPark: Leveraging Transfer Learning With Transformers for Intelligent Parking Systems","authors":"Chandranil Chakraborttii;Lin Cheng","doi":"10.1109/OJITS.2025.3649601","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3649601","url":null,"abstract":"With the growing number of vehicles in urban areas, traffic jams and parking inefficiencies are becoming an increasing challenge. To address this, Intelligent Parking Systems have been proposed for predicting parking availability and improving traffic flow, thereby improving driving experience. However, deploying such a system across various environments remains challenging because parking patterns vary between neighborhoods, streets, and cities. Traditional machine learning models require extensive retraining for new environments, which makes them impractical for large-scale, real-world use. In this paper, we introduce TransPark, a new transfer learning framework where a base model is pre-trained on a pool of similar datasets. This process helps the model learn generalized parking patterns. Next, we adapt the model to new target regions with minimal fine-tuning, which significantly reduces computational costs. TransPark is suitable where parking patterns differ in residential, commercial, and mixed-use areas. This approach reduces the resources needed to build separate models for each environment. We examine datasets from various urban neighborhoods in San Francisco and evaluate parking availability predictions in time intervals ranging from 1 to 60 minutes. We show TransPark effectively models complex spatio-temporal parking patterns by using a hybrid attention mechanism that decouples spatial learning (between streets) and temporal learning (between time-steps). By leveraging VPCL-driven pretraining and the hybrid attention mechanism, TransPark achieves 25-70% accuracy improvement over prior methods based on statistical, deep learning, and transformer baselines while reducing the computational cost of retraining by over 55%. This promises real-world deployability in diverse parking environments.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"166-189"},"PeriodicalIF":5.3,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11318639","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929667","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 : 2025-12-29DOI: 10.1109/OJITS.2025.3649213
Zijian Huang;Xiuhong Wang;Xuefeng Yuan;Xinyao Guo;Ziyou Su
For port air-pollution monitoring, which is characterized by high mobility and strong temporal variability, automatic path planning is critical for enabling shipborne multi-UAV cooperation. This paper focuses on path optimization under the simultaneous motion of a “carrier ship–multi-UAV–multiple target vessels” system and develops a time-dependent cooperative planning model. With the objective of minimizing the total monitoring completion time, the model explicitly represents the full cycle of launch–rendezvous–sampling–recovery, incorporating dynamic rendezvous, sampling time windows, and battery-closure constraints, and replaces static Euclidean distances with a dynamic distance matrix. A deep learning–dynamic ant colony optimization fusion algorithm (DL-DACO) is proposed, in which a neural network adaptively tunes key ACO parameters, while the pheromone update rule integrates elite reinforcement, annealed heavy-tailed perturbations, and time-varying evaporation rates to enhance global search capability and reduce the risk of premature convergence. Simulation results show that, compared with a fixed-base monitoring strategy, the proposed dynamic mobile-base strategy reduces the total monitoring time by 17.87%. Furthermore, compared with existing algorithms, the proposed fusion algorithm shortens the total monitoring time by 3.36% and 4.25%, respectively, demonstrating that the developed model and algorithm can substantially improve the efficiency of UAV-based pollutant monitoring and the path-planning capability of shipborne multi-UAV cooperative environmental monitoring.
{"title":"Synergistic Path Planning of Shipborne Multi-UAV Systems for Port Atmospheric Pollution Monitoring","authors":"Zijian Huang;Xiuhong Wang;Xuefeng Yuan;Xinyao Guo;Ziyou Su","doi":"10.1109/OJITS.2025.3649213","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3649213","url":null,"abstract":"For port air-pollution monitoring, which is characterized by high mobility and strong temporal variability, automatic path planning is critical for enabling shipborne multi-UAV cooperation. This paper focuses on path optimization under the simultaneous motion of a “carrier ship–multi-UAV–multiple target vessels” system and develops a time-dependent cooperative planning model. With the objective of minimizing the total monitoring completion time, the model explicitly represents the full cycle of launch–rendezvous–sampling–recovery, incorporating dynamic rendezvous, sampling time windows, and battery-closure constraints, and replaces static Euclidean distances with a dynamic distance matrix. A deep learning–dynamic ant colony optimization fusion algorithm (DL-DACO) is proposed, in which a neural network adaptively tunes key ACO parameters, while the pheromone update rule integrates elite reinforcement, annealed heavy-tailed perturbations, and time-varying evaporation rates to enhance global search capability and reduce the risk of premature convergence. Simulation results show that, compared with a fixed-base monitoring strategy, the proposed dynamic mobile-base strategy reduces the total monitoring time by 17.87%. Furthermore, compared with existing algorithms, the proposed fusion algorithm shortens the total monitoring time by 3.36% and 4.25%, respectively, demonstrating that the developed model and algorithm can substantially improve the efficiency of UAV-based pollutant monitoring and the path-planning capability of shipborne multi-UAV cooperative environmental monitoring.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"110-125"},"PeriodicalIF":5.3,"publicationDate":"2025-12-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11318011","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929647","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 : 2025-12-24DOI: 10.1109/OJITS.2025.3647830
Shuyi Tan;Chao Huang;Yi Zhang;Yang Wang
Accurate trajectory prediction is essential for autonomous driving systems to make safe and efficient decisions. Traditional global message-passing methods, though effective at capturing mutual interactions, suffer from an $O(N^{2})$ parameter complexity, which limits their scalability in high-density traffic environments. To address this, we propose a message-passing approach based on local neighborhoods, which reduces the complexity to $O(N cdot K_{max })$ by restricting each node’s interactions to its most relevant neighbors. On the Argoverse 1 motion forecasting benchmark, our model achieves a minADE6 of 0.739 and a minFDE6 of 1.133 with only 1.56M parameters, improving both metrics over a global message-passing baseline. On Argoverse 2, it attains a $mathrm {minFDE}_{6}$ of 1.196 and an MR6 of 12.2. These results demonstrate that local neighborhood message passing can simultaneously enhance prediction accuracy and computational efficiency, offering a scalable and practical solution for motion prediction in autonomous driving systems.
{"title":"Scalable Trajectory Prediction via Local Neighborhood Interactions","authors":"Shuyi Tan;Chao Huang;Yi Zhang;Yang Wang","doi":"10.1109/OJITS.2025.3647830","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3647830","url":null,"abstract":"Accurate trajectory prediction is essential for autonomous driving systems to make safe and efficient decisions. Traditional global message-passing methods, though effective at capturing mutual interactions, suffer from an <inline-formula> <tex-math>$O(N^{2})$ </tex-math></inline-formula> parameter complexity, which limits their scalability in high-density traffic environments. To address this, we propose a message-passing approach based on local neighborhoods, which reduces the complexity to <inline-formula> <tex-math>$O(N cdot K_{max })$ </tex-math></inline-formula> by restricting each node’s interactions to its most relevant neighbors. On the Argoverse 1 motion forecasting benchmark, our model achieves a minADE6 of 0.739 and a minFDE6 of 1.133 with only 1.56M parameters, improving both metrics over a global message-passing baseline. On Argoverse 2, it attains a <inline-formula> <tex-math>$mathrm {minFDE}_{6}$ </tex-math></inline-formula> of 1.196 and an MR6 of 12.2. These results demonstrate that local neighborhood message passing can simultaneously enhance prediction accuracy and computational efficiency, offering a scalable and practical solution for motion prediction in autonomous driving systems.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"126-135"},"PeriodicalIF":5.3,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11313861","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929646","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}
Accurate driving risk prediction is essential for preventing traffic accidents, particularly in complex mountain tunnel environments where conventional assessment methods often fall short. This study presents a novel approach for quantifying and predicting driving risk under multi-vehicle interactions scenarios. A weighted comprehensive risk matrix is constructed by integrating Time-to-Collision (TTC) and Interaction Strength (IS), taking into account the behavior of surrounding vehicles. A risk representation framework centered on the ego vehicle and a multi-level risk classification scheme are proposed. To capture the spatial and temporal dynamics of driving risk, a hybrid deep learning model is proposed, combining Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Attention mechanisms. The model was validated using real-world trajectory data from six mountain tunnels in Chongqing, China. The model achieved prediction accuracies of 83% in car-following and 76% in lane-changing scenarios, outperforming traditional methods. The proposed model significantly enhances the identification and prediction of abnormal driving behaviors under highly interaction conditions, offering a valuable tool to improve intelligent driving safety in mountain tunnels.
{"title":"A Risk Identification and Prediction Model for Intelligent Driving Under Multi-Vehicle Interactions in Mountain Tunnel Environments","authors":"Xiaoyu Cai;Minghua Zhang;Cailin Lei;Ling Jin;Bo Peng","doi":"10.1109/OJITS.2025.3647953","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3647953","url":null,"abstract":"Accurate driving risk prediction is essential for preventing traffic accidents, particularly in complex mountain tunnel environments where conventional assessment methods often fall short. This study presents a novel approach for quantifying and predicting driving risk under multi-vehicle interactions scenarios. A weighted comprehensive risk matrix is constructed by integrating Time-to-Collision (TTC) and Interaction Strength (IS), taking into account the behavior of surrounding vehicles. A risk representation framework centered on the ego vehicle and a multi-level risk classification scheme are proposed. To capture the spatial and temporal dynamics of driving risk, a hybrid deep learning model is proposed, combining Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Attention mechanisms. The model was validated using real-world trajectory data from six mountain tunnels in Chongqing, China. The model achieved prediction accuracies of 83% in car-following and 76% in lane-changing scenarios, outperforming traditional methods. The proposed model significantly enhances the identification and prediction of abnormal driving behaviors under highly interaction conditions, offering a valuable tool to improve intelligent driving safety in mountain tunnels.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"155-165"},"PeriodicalIF":5.3,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11313849","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929519","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 : 2025-12-11DOI: 10.1109/OJITS.2025.3643797
Atharva Ghate;Olamide Akinyele;Qilun Zhu;Robert Prucka;Miriam A. Figueroa-Santos;Morgan J. Barron;Matthew P. Castanier
Autonomous off-road navigation requires coping with unstructured terrain, intermittent obstacles, and tight real-time computational constraints, challenges that often exceed the capabilities of conventional motion-planning and control pipelines. This paper proposes the Cascaded Reinforcement Learning and Model Predictive Path Integral (CRM) framework, which integrates a curriculum-trained Reinforcement Learning (RL) critic for global planning with a fallback-enabled Model Predictive Path Integral (MPPI) controller for local refinement. Unlike prior RL-MPPI methods, the proposed approach incrementally teaches the RL critic obstacle avoidance, rollover prevention, and traction constraints, thereby improving the accuracy of terminal cost estimates. To safeguard against unconverged RL outputs in new or out-of-distribution states, we embed a logic-based fallback that reverts MPPI to baseline costs whenever the RL-driven terminal value is judged unreliable. In simulations on representative of off-road environments, CRM achieves success rates higher by 70%, lowers sample requirements up to 90% compared to MPPI alone, and avoids collisions more effectively than standalone RL methods. These results underscore the necessity of curriculum-informed critics and robust fallback strategies for safe and efficient off-road autonomy.
{"title":"Cascaded RL-MPPI Framework for Off-Road Vehicles: Integrating Global Maps and SLAM","authors":"Atharva Ghate;Olamide Akinyele;Qilun Zhu;Robert Prucka;Miriam A. Figueroa-Santos;Morgan J. Barron;Matthew P. Castanier","doi":"10.1109/OJITS.2025.3643797","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3643797","url":null,"abstract":"Autonomous off-road navigation requires coping with unstructured terrain, intermittent obstacles, and tight real-time computational constraints, challenges that often exceed the capabilities of conventional motion-planning and control pipelines. This paper proposes the Cascaded Reinforcement Learning and Model Predictive Path Integral (CRM) framework, which integrates a curriculum-trained Reinforcement Learning (RL) critic for global planning with a fallback-enabled Model Predictive Path Integral (MPPI) controller for local refinement. Unlike prior RL-MPPI methods, the proposed approach incrementally teaches the RL critic obstacle avoidance, rollover prevention, and traction constraints, thereby improving the accuracy of terminal cost estimates. To safeguard against unconverged RL outputs in new or out-of-distribution states, we embed a logic-based fallback that reverts MPPI to baseline costs whenever the RL-driven terminal value is judged unreliable. In simulations on representative of off-road environments, CRM achieves success rates higher by 70%, lowers sample requirements up to 90% compared to MPPI alone, and avoids collisions more effectively than standalone RL methods. These results underscore the necessity of curriculum-informed critics and robust fallback strategies for safe and efficient off-road autonomy.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"41-60"},"PeriodicalIF":5.3,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11298481","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830788","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 : 2025-12-09DOI: 10.1109/OJITS.2025.3642136
Yongjia Nian;Hao Liu;Renwen Chen;Xintong Hou;Aocheng He
This paper investigates the challenge of topology optimization for UAV swarms in dynamic environments and proposes a reinforcement learning窶電riven distributed framework. Under the centralized training and decentralized execution (CTDE) paradigm, a MADDPG-based topology reconfiguration algorithm is developed that integrates partial observability with a bi-directional interest game, enabling nodes to achieve distributed Nash equilibrium decisions under local information constraints. At the communication layer, a channel model, topology maintenance scheme, and CSDMA-based distributed slot allocation process are introduced to ensure reliable connectivity in the presence of interference and dynamic node access. Simulation results show that the proposed method attains faster convergence, greater robustness, lower communication latency, and higher path efficiency than benchmark approaches such as MST and PSO, with reconfiguration completed within milliseconds. These results highlight both the effectiveness and scalability of the framework for large-scale swarm networking. Beyond its theoretical contributions, the approach holds practical promise for deployment in critical scenarios such as emergency communications, disaster relief, and mission-critical operations, offering a viable pathway toward intelligent UAV swarm networks.
{"title":"Distributed Real-Time Topology Reconfiguration for UAV Swarms via MADDPG","authors":"Yongjia Nian;Hao Liu;Renwen Chen;Xintong Hou;Aocheng He","doi":"10.1109/OJITS.2025.3642136","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3642136","url":null,"abstract":"This paper investigates the challenge of topology optimization for UAV swarms in dynamic environments and proposes a reinforcement learning窶電riven distributed framework. Under the centralized training and decentralized execution (CTDE) paradigm, a MADDPG-based topology reconfiguration algorithm is developed that integrates partial observability with a bi-directional interest game, enabling nodes to achieve distributed Nash equilibrium decisions under local information constraints. At the communication layer, a channel model, topology maintenance scheme, and CSDMA-based distributed slot allocation process are introduced to ensure reliable connectivity in the presence of interference and dynamic node access. Simulation results show that the proposed method attains faster convergence, greater robustness, lower communication latency, and higher path efficiency than benchmark approaches such as MST and PSO, with reconfiguration completed within milliseconds. These results highlight both the effectiveness and scalability of the framework for large-scale swarm networking. Beyond its theoretical contributions, the approach holds practical promise for deployment in critical scenarios such as emergency communications, disaster relief, and mission-critical operations, offering a viable pathway toward intelligent UAV swarm networks.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"74-92"},"PeriodicalIF":5.3,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11288063","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886652","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}
Timely ambulance allocation is essential for Emergency Medical Services (EMS) to deliver life-saving care effectively. Conventional methods often struggle to adapt to the unpredictable nature and locations of emergencies. Within EMS, efficient resource management is crucial for ensuring rapid and effective responses. While much emphasis has been placed on optimizing the deployment of ambulances from fixed stations, managing specialized critical care response units—known as Charlie vehicles in Qatar EMS—presents a distinct challenge. These rapid response cars are vital for providing advanced care in challenging situations, and their dynamic deployment requires a more flexible management strategy. Effectively relocating Charlie vehicles to areas with high anticipated demand after they have responded to an emergency introduces unique challenges that differ from traditional ambulance redeployment approaches. This paper proposes a novel dynamic redeployment system specifically for optimizing the allocation of critical care response vehicles, including those involved in patient transfers. Utilizing a Deep Reinforcement Learning (DRL) framework, we create a deep scoring network that prioritizes and navigates various dynamic factors at each station. Experiments using real-world data from Qatar EMS demonstrate that our system significantly outperforms existing methods. For instance, our approach achieves faster average response times and improved critical response rates compared to the leading baseline method. Notably, we observe a substantial 21.55% reduction in average response time (AveRT) and an 18.34% increase in relative response time (RelaRT) in comparison to actual operational metrics. Our approach effectively shortens the time needed to reach patients, thereby increasing the likelihood of timely treatment and improving overall patient care outcomes.
{"title":"A Dynamic Redeployment System for Critical Care Paramedic Units in Qatar Utilizing Deep Reinforcement Learning","authors":"Reem Tluli;Ahmed Badawy;Saeed Salem;Muhammad Hardan;Sailesh Chauhan;Guillaume Alinier","doi":"10.1109/OJITS.2025.3642001","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3642001","url":null,"abstract":"Timely ambulance allocation is essential for Emergency Medical Services (EMS) to deliver life-saving care effectively. Conventional methods often struggle to adapt to the unpredictable nature and locations of emergencies. Within EMS, efficient resource management is crucial for ensuring rapid and effective responses. While much emphasis has been placed on optimizing the deployment of ambulances from fixed stations, managing specialized critical care response units—known as Charlie vehicles in Qatar EMS—presents a distinct challenge. These rapid response cars are vital for providing advanced care in challenging situations, and their dynamic deployment requires a more flexible management strategy. Effectively relocating Charlie vehicles to areas with high anticipated demand after they have responded to an emergency introduces unique challenges that differ from traditional ambulance redeployment approaches. This paper proposes a novel dynamic redeployment system specifically for optimizing the allocation of critical care response vehicles, including those involved in patient transfers. Utilizing a Deep Reinforcement Learning (DRL) framework, we create a deep scoring network that prioritizes and navigates various dynamic factors at each station. Experiments using real-world data from Qatar EMS demonstrate that our system significantly outperforms existing methods. For instance, our approach achieves faster average response times and improved critical response rates compared to the leading baseline method. Notably, we observe a substantial 21.55% reduction in average response time (AveRT) and an 18.34% increase in relative response time (RelaRT) in comparison to actual operational metrics. Our approach effectively shortens the time needed to reach patients, thereby increasing the likelihood of timely treatment and improving overall patient care outcomes.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"93-109"},"PeriodicalIF":5.3,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11288020","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886653","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}
Ensuring that longitudinal control in autonomous driving is accurate, robust, and smooth is key to enhance vehicle autonomy and reduce driver intervention, improving user acceptance of autonomous vehicles. Vehicles have complex dynamics that make accurately following the speed reference in various driving situations a challenging task. Model-Free Control (MFC) has shown its performance and robustness in systems which are difficult to model or with time-varying dynamics, making it relevant for this application. In this paper, a cascade control architecture based on MFC is proposed. This strategy keeps the MFC principle of simplicity in control while, due to the cascade structure, using all the information generated by the motion planner and the measured speed and acceleration, which are easy to obtain. Regulators with this structure have been systematically designed to keep the tracking quality, safety and passenger comfort in a wide variety of driving situations.These regulators have been evaluated both in simulation and real-world scenarios, showing improvements in robustness and performance when compared with the baseline.
{"title":"Model-Free Speed Tracking Control for Automated Cars","authors":"Marcos Moreno-Gonzalez;Antonio Artuñedo;Jorge Villagra","doi":"10.1109/OJITS.2025.3640943","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3640943","url":null,"abstract":"Ensuring that longitudinal control in autonomous driving is accurate, robust, and smooth is key to enhance vehicle autonomy and reduce driver intervention, improving user acceptance of autonomous vehicles. Vehicles have complex dynamics that make accurately following the speed reference in various driving situations a challenging task. Model-Free Control (MFC) has shown its performance and robustness in systems which are difficult to model or with time-varying dynamics, making it relevant for this application. In this paper, a cascade control architecture based on MFC is proposed. This strategy keeps the MFC principle of simplicity in control while, due to the cascade structure, using all the information generated by the motion planner and the measured speed and acceleration, which are easy to obtain. Regulators with this structure have been systematically designed to keep the tracking quality, safety and passenger comfort in a wide variety of driving situations.These regulators have been evaluated both in simulation and real-world scenarios, showing improvements in robustness and performance when compared with the baseline.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"1-15"},"PeriodicalIF":5.3,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11278736","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886651","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}
With the rapid advancement of autonomous driving technology, the achievement of complex trajectory prediction for human-like driving behaviors has become a critical research focus. Traditional data-driven models exhibit substantial limitations in replicating human driving logic and cognitive processes, constraining their adaptability and robustness across diverse driving scenarios. This study proposes and validates a novel Data-knowledge-driven Human-like logic Trajectory Prediction model (DHTP) using a bidirectional hybrid modeling approach. It incorporates an attention mechanism, memory reasoning, and autonomous evolution modules. The performance is assessed using multiple quantitative metrics and experimentally validated in real-world driving scenarios, including the urban expressway and highway weaving areas. The experimental results show that the DHTP model significantly outperforms the baseline model, showcasing enhanced accuracy and robustness across diverse driving conditions. Additionally, it rapidly converges to the global optimal solution, particularly in highly dynamic environments. The results indicate that optimizing the attention mechanism and autonomous evolution module allows the DHTP model to successfully simulate human driving logic and behavioral patterns. This study can help to facilitate AV-HV interaction and supports cognitive module advancement toward autonomy.
{"title":"Incomprehensible But Intelligible Human Logics: Toward a Data-Knowledge-Driven Trajectory Prediction Model","authors":"Jiming Xie;Jianhua Li;Yaqin Qin;Jiachen Ren;Hongjian Liang;Liang Chen;Yulan Xia","doi":"10.1109/OJITS.2025.3640704","DOIUrl":"https://doi.org/10.1109/OJITS.2025.3640704","url":null,"abstract":"With the rapid advancement of autonomous driving technology, the achievement of complex trajectory prediction for human-like driving behaviors has become a critical research focus. Traditional data-driven models exhibit substantial limitations in replicating human driving logic and cognitive processes, constraining their adaptability and robustness across diverse driving scenarios. This study proposes and validates a novel Data-knowledge-driven Human-like logic Trajectory Prediction model (DHTP) using a bidirectional hybrid modeling approach. It incorporates an attention mechanism, memory reasoning, and autonomous evolution modules. The performance is assessed using multiple quantitative metrics and experimentally validated in real-world driving scenarios, including the urban expressway and highway weaving areas. The experimental results show that the DHTP model significantly outperforms the baseline model, showcasing enhanced accuracy and robustness across diverse driving conditions. Additionally, it rapidly converges to the global optimal solution, particularly in highly dynamic environments. The results indicate that optimizing the attention mechanism and autonomous evolution module allows the DHTP model to successfully simulate human driving logic and behavioral patterns. This study can help to facilitate AV-HV interaction and supports cognitive module advancement toward autonomy.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"7 ","pages":"412-433"},"PeriodicalIF":5.3,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11278737","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175851","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：