Pub Date : 2026-02-06DOI: 10.1109/tase.2026.3662192
Ying Jing, Hong Zheng, Yuchuan Ji
{"title":"Prior-Guided and Gaussian Mixture-Refined Network for Industrial Anomaly Detection and Localization","authors":"Ying Jing, Hong Zheng, Yuchuan Ji","doi":"10.1109/tase.2026.3662192","DOIUrl":"https://doi.org/10.1109/tase.2026.3662192","url":null,"abstract":"","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"27 1","pages":""},"PeriodicalIF":5.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134368","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/tase.2026.3662003
Ming Sun, Yu Wang, Bo Yang, Li He, Hong Zhang
{"title":"Accurate and Robust UWB Localization with Incomplete Measurements based on Multi-Modal Diffusion Model","authors":"Ming Sun, Yu Wang, Bo Yang, Li He, Hong Zhang","doi":"10.1109/tase.2026.3662003","DOIUrl":"https://doi.org/10.1109/tase.2026.3662003","url":null,"abstract":"","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"1 1","pages":""},"PeriodicalIF":5.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134377","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TASE.2026.3661872
Rui Liu;Cong Wu;Yifan Zhang;Haiying Song;Fei Yuan;Wei Dai;Wen Jung Li;Jun Liu
Accurate and robust cell detection in microscopic images is a fundamental yet challenging task due to diverse imaging conditions, dense cell distributions, and morphological variability. In this study, we propose FACellDet, a novel dual-branch hierarchical encoder-decoder framework that integrates adaptive feature alignment and a new supervisory signal to enhance cell detection. Specifically, a Coordinate-Attention-based Feature Alignment (CAFA) module is introduced to address spatial misalignment during multi-scale feature fusion, substantially improving cell detection precision. Furthermore, we design a Focal Attenuated Distance (FAD) map as an intermediate representation, providing highly discriminative and spatially informative cues, particularly in crowded regions. FACellDet features a dual-branch architecture, with the main branch predicting FAD maps for cell detection, while the auxiliary branch generates density maps to estimate cell counts for suppressing false detections. Extensive experiments on diverse cell types and imaging modalities from multiple public and in-house datasets demonstrate that our approach outperforms state-of-the-art methods in detection accuracy, while maintaining strong adaptability and robustness across challenging biomedical imaging scenarios. These results underscore the potential of FACellDet as an accurate and generalizable solution for automated cell detection in heterogeneous microscopic cell images, thereby facilitating reliable cell analysis to accelerate biomedical research and clinical workflows. Note to Practitioners—This work addresses the need for accurate and efficient cell detection and counting in biomedical images, where manual methods are time-consuming and error-prone, and existing automated approaches often struggle with dense or diverse cells. FACellDet offers a practical deep learning solution adaptable to various cell types and imaging conditions, improving both detection accuracy and robustness through innovative feature alignment and enhanced supervisory signals. This system can streamline laboratory workflows and support high-throughput research and clinical diagnostics. While FACellDet demonstrates strong performance across challenging datasets, its current deployment requires adequate computational resources. Future development could focus on creating lightweight versions and integrating the framework with automated imaging systems, further broadening its accessibility and impact in routine biomedical practice.
{"title":"Feature-Aligned Cell Detection for Heterogeneous Microscopic Images With Focal Attenuated Distance Transform","authors":"Rui Liu;Cong Wu;Yifan Zhang;Haiying Song;Fei Yuan;Wei Dai;Wen Jung Li;Jun Liu","doi":"10.1109/TASE.2026.3661872","DOIUrl":"10.1109/TASE.2026.3661872","url":null,"abstract":"Accurate and robust cell detection in microscopic images is a fundamental yet challenging task due to diverse imaging conditions, dense cell distributions, and morphological variability. In this study, we propose FACellDet, a novel dual-branch hierarchical encoder-decoder framework that integrates adaptive feature alignment and a new supervisory signal to enhance cell detection. Specifically, a Coordinate-Attention-based Feature Alignment (CAFA) module is introduced to address spatial misalignment during multi-scale feature fusion, substantially improving cell detection precision. Furthermore, we design a Focal Attenuated Distance (FAD) map as an intermediate representation, providing highly discriminative and spatially informative cues, particularly in crowded regions. FACellDet features a dual-branch architecture, with the main branch predicting FAD maps for cell detection, while the auxiliary branch generates density maps to estimate cell counts for suppressing false detections. Extensive experiments on diverse cell types and imaging modalities from multiple public and in-house datasets demonstrate that our approach outperforms state-of-the-art methods in detection accuracy, while maintaining strong adaptability and robustness across challenging biomedical imaging scenarios. These results underscore the potential of FACellDet as an accurate and generalizable solution for automated cell detection in heterogeneous microscopic cell images, thereby facilitating reliable cell analysis to accelerate biomedical research and clinical workflows. Note to Practitioners—This work addresses the need for accurate and efficient cell detection and counting in biomedical images, where manual methods are time-consuming and error-prone, and existing automated approaches often struggle with dense or diverse cells. FACellDet offers a practical deep learning solution adaptable to various cell types and imaging conditions, improving both detection accuracy and robustness through innovative feature alignment and enhanced supervisory signals. This system can streamline laboratory workflows and support high-throughput research and clinical diagnostics. While FACellDet demonstrates strong performance across challenging datasets, its current deployment requires adequate computational resources. Future development could focus on creating lightweight versions and integrating the framework with automated imaging systems, further broadening its accessibility and impact in routine biomedical practice.","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"23 ","pages":"4375-4387"},"PeriodicalIF":6.4,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134375","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TASE.2026.3661751
Yajuan Liu;Haoran Ma;Ziqiu Song
It is difficult to obtain an accurate mechanism model of a floating wind turbine due to the large coupling disturbance of wind and wave at sea, and the control accuracy is difficult to guarantee, so maintaining stable power output is a challenge. Therefore, an event-triggered model-free adaptive control (ET-MFAC) collective pitch angle strategy is proposed for the NREL 5MW floating offshore wind turbine. The proposed method combines input and output Model-free adaptive control (IO-MFAC) with an improved Event-triggered mechanism (ETM). The improved ETM is based on the intensified ETM, which evaluates the weighted historical state error, and innovatively introduces an adaptive adjustment factor to realize real-time adjustment of the trigger frequency according to the system operating state, and reduces the computational burden of IO-MFAC. Meanwhile, the stability of IO-MFAC is proved based on the strict contraction mapping theory, which guarantees the stability of Bounded Input Bounded output (BIBO) and the monotonic convergence of tracking error. Experimental results on the OpenFAST/Simulink simulation platform show that the ET-MFAC strategy is superior to the traditional method in terms of rated power tracking, computational load reduction and robustness, especially in the extreme coupling conditions of strong turbulence and strong sea state. Note to Practitioners—The motivation for this study stems from the practical challenges faced in controlling offshore floating wind turbines, which operate in extremely complex and uncertain environments. Traditional pitch control methods often rely on accurate system models, which are difficult to obtain and computationally expensive, or cannot dynamically adapt to the changing ocean environment. The proposed ET-MFAC strategy provides a practical alternative that does not require accurate turbine modeling and can significantly reduce the computational burden of the controller. ET-MFAC combines an event-triggered mechanism that considers multiple historical trigger signals with data-driven control, and adaptively adjusts the trigger frequency according to real-time output errors, initiating pitch angle adjustment only when necessary. A high-fidelity NREL 5MW wind turbine model is used for simulation, and the results show that the ET-MFAC has more stable control performance than the traditional variable pitch controller under the condition of wind and wave coupling. This strategy provides a promising avenue to achieve more reliable and efficient operation of floating wind turbines and reduce maintenance and operation costs.
{"title":"Blade Pitch Control for Floating Wind Turbines via Event-Triggered Model-Free Adaptive Control Strategy","authors":"Yajuan Liu;Haoran Ma;Ziqiu Song","doi":"10.1109/TASE.2026.3661751","DOIUrl":"10.1109/TASE.2026.3661751","url":null,"abstract":"It is difficult to obtain an accurate mechanism model of a floating wind turbine due to the large coupling disturbance of wind and wave at sea, and the control accuracy is difficult to guarantee, so maintaining stable power output is a challenge. Therefore, an event-triggered model-free adaptive control (ET-MFAC) collective pitch angle strategy is proposed for the NREL 5MW floating offshore wind turbine. The proposed method combines input and output Model-free adaptive control (IO-MFAC) with an improved Event-triggered mechanism (ETM). The improved ETM is based on the intensified ETM, which evaluates the weighted historical state error, and innovatively introduces an adaptive adjustment factor to realize real-time adjustment of the trigger frequency according to the system operating state, and reduces the computational burden of IO-MFAC. Meanwhile, the stability of IO-MFAC is proved based on the strict contraction mapping theory, which guarantees the stability of Bounded Input Bounded output (BIBO) and the monotonic convergence of tracking error. Experimental results on the OpenFAST/Simulink simulation platform show that the ET-MFAC strategy is superior to the traditional method in terms of rated power tracking, computational load reduction and robustness, especially in the extreme coupling conditions of strong turbulence and strong sea state. Note to Practitioners—The motivation for this study stems from the practical challenges faced in controlling offshore floating wind turbines, which operate in extremely complex and uncertain environments. Traditional pitch control methods often rely on accurate system models, which are difficult to obtain and computationally expensive, or cannot dynamically adapt to the changing ocean environment. The proposed ET-MFAC strategy provides a practical alternative that does not require accurate turbine modeling and can significantly reduce the computational burden of the controller. ET-MFAC combines an event-triggered mechanism that considers multiple historical trigger signals with data-driven control, and adaptively adjusts the trigger frequency according to real-time output errors, initiating pitch angle adjustment only when necessary. A high-fidelity NREL 5MW wind turbine model is used for simulation, and the results show that the ET-MFAC has more stable control performance than the traditional variable pitch controller under the condition of wind and wave coupling. This strategy provides a promising avenue to achieve more reliable and efficient operation of floating wind turbines and reduce maintenance and operation costs.","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"23 ","pages":"4365-4374"},"PeriodicalIF":6.4,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134369","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TASE.2026.3661086
Yazhou Tian;Yuangong Sun;Bing Liu
Significant achievements have been made in the analysis of linear coupled differential-difference systems (CDDSs). However, the study of nonlinear CDDSs, particularly those with time-varying characteristics and exogenous inputs, presents substantial challenges. This paper proposes, for the first time, a reachable set estimation method for time-varying homogeneous nonlinear coupled differential-difference systems (HNCDDSs) with exogenous inputs. By introducing a novel state transformation and a method developed in positive systems, we establish an explicit sufficient condition that ensures all system states converge exponentially to a specified ball when the homogeneity degree of the system is less than or equal to one. Building upon this analytical framework, for the homogeneity degree of the system greater than one, we further derive a criterion guaranteeing the states converge polynomially within a bounded region. These theoretical findings not only extend but also improve existing results in the literature, which are effectively supported by two specific numerical examples. Note to Practitioners—Coupled differential-difference systems play a key role to characterize the behaviors of the dynamic systems in control field, such as electrical engineering, fluid dynamics, and multi-agent systems. It is significant to explore the reachable set estimation of nonlinear CDDSs with exogenous inputs. Moreover, since most physical systems are inherently time-varying, the reachable set estimation of time-varying HNCDDSs has become a critical issue that urgently needs to be addressed. Traditional approaches, such as the Lyapunov-Krasovskii functional method, often prove ineffective for time-varying systems, as they typically lead to either unsolvable Riccati differential equations or indefinite linear matrix inequalities (LMIs). To overcome these challenges, this study proposes a novel state transformation combined with a method developed in positive systems to estimate the reachable set of time-varying HNCDDSs with exogenous inputs, and derives more general results compared with existing conclusions.
{"title":"Reachable Set Estimation for Time-Varying Homogeneous Coupled Differential-Difference Systems With Exogenous Inputs","authors":"Yazhou Tian;Yuangong Sun;Bing Liu","doi":"10.1109/TASE.2026.3661086","DOIUrl":"10.1109/TASE.2026.3661086","url":null,"abstract":"Significant achievements have been made in the analysis of linear coupled differential-difference systems (CDDSs). However, the study of nonlinear CDDSs, particularly those with time-varying characteristics and exogenous inputs, presents substantial challenges. This paper proposes, for the first time, a reachable set estimation method for time-varying homogeneous nonlinear coupled differential-difference systems (HNCDDSs) with exogenous inputs. By introducing a novel state transformation and a method developed in positive systems, we establish an explicit sufficient condition that ensures all system states converge exponentially to a specified ball when the homogeneity degree of the system is less than or equal to one. Building upon this analytical framework, for the homogeneity degree of the system greater than one, we further derive a criterion guaranteeing the states converge polynomially within a bounded region. These theoretical findings not only extend but also improve existing results in the literature, which are effectively supported by two specific numerical examples. Note to Practitioners—Coupled differential-difference systems play a key role to characterize the behaviors of the dynamic systems in control field, such as electrical engineering, fluid dynamics, and multi-agent systems. It is significant to explore the reachable set estimation of nonlinear CDDSs with exogenous inputs. Moreover, since most physical systems are inherently time-varying, the reachable set estimation of time-varying HNCDDSs has become a critical issue that urgently needs to be addressed. Traditional approaches, such as the Lyapunov-Krasovskii functional method, often prove ineffective for time-varying systems, as they typically lead to either unsolvable Riccati differential equations or indefinite linear matrix inequalities (LMIs). To overcome these challenges, this study proposes a novel state transformation combined with a method developed in positive systems to estimate the reachable set of time-varying HNCDDSs with exogenous inputs, and derives more general results compared with existing conclusions.","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"23 ","pages":"4341-4349"},"PeriodicalIF":6.4,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134370","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TASE.2026.3661491
Yu Chen;Yuanli Cai;Yifan Deng;Jiaqi Liu
This paper proposes a robust event-triggered filtering algorithm for stable tracking of complex dynamic systems under nonstationary heavy-tailed noise, packet dropouts, and communication congestion. First, the one-step prediction and likelihood probability density functions are modeled as Gaussian–Student’s t mixtures, with the unknown degrees of freedom characterized by a Gamma distribution, thereby constructing a hierarchical state-space model that adapts to dynamic noise variations. Second, by integrating a stochastic event-triggered mechanism with a Bernoulli process to establish a hybrid packet dropout model and introducing a compensation mechanism for missing data reconstruction, a variational Bayesian inference framework is employed to derive a novel filtering algorithm that can simultaneously handle nonstationary heavy-tailed noise and packet dropouts. Furthermore, the boundedness of the prediction error covariance is analyzed, and sufficient conditions are derived to ensure mean-square exponential stability. Finally, tracking simulations validate the effectiveness of the proposed algorithm, demonstrating its superior estimation accuracy and robustness under low communication overhead, even in the presence of packet dropouts and nonstationary heavy-tailed noise. Note to Practitioners—State estimation and tracking in complex dynamic systems are often challenged by nonstationary heavy-tailed noise, packet dropouts, and communication congestion. These issues can significantly degrade estimation accuracy and system stability. Traditional filtering methods typically struggle to handle both the robustness requirements under non-Gaussian noise and the information loss caused by packet dropouts. Moreover, although frequent data transmission helps keep filters up to date, it also leads to high communication load and increased energy consumption, limiting system reliability and practical deployment. To address these challenges, this paper proposes a robust filtering algorithm based on a stochastic event-triggered mechanism. It dynamically models noise using a hybrid Gaussian–Student’s t distribution combined with a Gamma distribution, and incorporates a Bernoulli process to model packet dropout. A compensation mechanism is introduced to jointly handle nonstationary heavy-tailed noise and missing data. The proposed algorithm guarantees mean-square exponential stability, significantly reduces communication costs, and achieves high estimation accuracy and robustness. It offers an effective solution for tracking systems in complex environments by jointly addressing estimation performance and communication efficiency.
{"title":"Stochastic Event-Triggered Robust Tracking Algorithm Under Nonstationary Heavy-Tailed Noise and Packet Dropouts","authors":"Yu Chen;Yuanli Cai;Yifan Deng;Jiaqi Liu","doi":"10.1109/TASE.2026.3661491","DOIUrl":"10.1109/TASE.2026.3661491","url":null,"abstract":"This paper proposes a robust event-triggered filtering algorithm for stable tracking of complex dynamic systems under nonstationary heavy-tailed noise, packet dropouts, and communication congestion. First, the one-step prediction and likelihood probability density functions are modeled as Gaussian–Student’s t mixtures, with the unknown degrees of freedom characterized by a Gamma distribution, thereby constructing a hierarchical state-space model that adapts to dynamic noise variations. Second, by integrating a stochastic event-triggered mechanism with a Bernoulli process to establish a hybrid packet dropout model and introducing a compensation mechanism for missing data reconstruction, a variational Bayesian inference framework is employed to derive a novel filtering algorithm that can simultaneously handle nonstationary heavy-tailed noise and packet dropouts. Furthermore, the boundedness of the prediction error covariance is analyzed, and sufficient conditions are derived to ensure mean-square exponential stability. Finally, tracking simulations validate the effectiveness of the proposed algorithm, demonstrating its superior estimation accuracy and robustness under low communication overhead, even in the presence of packet dropouts and nonstationary heavy-tailed noise. Note to Practitioners—State estimation and tracking in complex dynamic systems are often challenged by nonstationary heavy-tailed noise, packet dropouts, and communication congestion. These issues can significantly degrade estimation accuracy and system stability. Traditional filtering methods typically struggle to handle both the robustness requirements under non-Gaussian noise and the information loss caused by packet dropouts. Moreover, although frequent data transmission helps keep filters up to date, it also leads to high communication load and increased energy consumption, limiting system reliability and practical deployment. To address these challenges, this paper proposes a robust filtering algorithm based on a stochastic event-triggered mechanism. It dynamically models noise using a hybrid Gaussian–Student’s t distribution combined with a Gamma distribution, and incorporates a Bernoulli process to model packet dropout. A compensation mechanism is introduced to jointly handle nonstationary heavy-tailed noise and missing data. The proposed algorithm guarantees mean-square exponential stability, significantly reduces communication costs, and achieves high estimation accuracy and robustness. It offers an effective solution for tracking systems in complex environments by jointly addressing estimation performance and communication efficiency.","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"23 ","pages":"4428-4441"},"PeriodicalIF":6.4,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134371","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/tase.2026.3661969
Wenjun Xu, Liang Yang, Guanyu Lai, Yong Chen
{"title":"Finite-Time Adaptive Visual Tracking Control of Manipulators with Parameter Uncertainties","authors":"Wenjun Xu, Liang Yang, Guanyu Lai, Yong Chen","doi":"10.1109/tase.2026.3661969","DOIUrl":"https://doi.org/10.1109/tase.2026.3661969","url":null,"abstract":"","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"91 1","pages":""},"PeriodicalIF":5.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134376","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/tase.2026.3660212
Marios-Nektarios Stamatopoulos, Shridhar Velhal, Avijit Banerjee, George Nikolakopoulos
{"title":"Optimal Safety-Aware Scheduling for Multi-Agent Aerial 3D Printing with Utility Maximization under Dependency Constraints","authors":"Marios-Nektarios Stamatopoulos, Shridhar Velhal, Avijit Banerjee, George Nikolakopoulos","doi":"10.1109/tase.2026.3660212","DOIUrl":"https://doi.org/10.1109/tase.2026.3660212","url":null,"abstract":"","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"89 1","pages":""},"PeriodicalIF":5.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134380","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TASE.2026.3662294
Ciyun Lin;Yujia Wang;Bowen Gong;Hongchao Liu
A consistent and lane-level vehicle trajectory offers substantial spatial-temporal insights to understand driving behavior, which is essential to improve safety and efficiency in merging zones. Therefore, an innovative framework was devised that merges a graph convolutional network (GCN) with the Transformer model for vehicle trajectory and intention prediction using roadside LiDAR data. First, an unsupervised intention extraction algorithm was developed to establish a driving intention dataset via high-resolution and lane-level vehicle trajectories extracted from roadside LiDAR in merging zones. Then, a coupled vehicle model was proposed to model the interaction between merging and mainline vehicles. Finally, a multitask framework was designed to capture the spatial dependencies via GCN and temporal patterns via Transformer for predicting vehicle trajectory and intention. Experimental results demonstrated that the proposed framework outperforms the state-of-the-art algorithms in vehicle trajectory prediction, securing MAE and RMSE values under 0.62 m and 1.10m, respectively. The intention prediction achieves an average precision rate of 92.76%. Ablation studies highlighted the ability of GCN and the coupled vehicle model to refine vehicle trajectory and intention prediction accuracy, reducing trajectory prediction errors by approximately 0.3 m and 0.4 m for MAE and RMSE, respectively, and enhancing intention prediction accuracy by 16%. Note to Practitioners—The practical problem addressed in this paper is the need to improve safety and efficiency in merging zones, where vehicle interactions often lead to congestion and accidents. Accurate prediction of vehicle trajectories and intentions is critical for enabling advanced driver-assistance systems (ADAS) and advanced traffic management systems (ATMS). This paper introduces an innovative framework that combines a graph convolutional network (GCN) with a Transformer model to predict vehicle trajectories and intentions using roadside LiDAR data. The framework is designed to operate in real-world scenarios, leveraging high-resolution, lane-level trajectory data to model interactions between merging and mainline vehicles.Our solution significantly improves prediction accuracy, achieving mean absolute error (MAE) and root mean square error (RMSE) values below 0.62 m and 1.10 m, respectively, for trajectory prediction, and a 92.76% precision rate for intention prediction. These results outperform existing state-of-the-art methods, demonstrating the effectiveness of the GCN in capturing spatial dependencies and the Transformer in modeling temporal patterns. The coupled vehicle interaction model further refines predictions, reducing trajectory errors by approximately 0.3 m (MAE) and 0.4 m (RMSE) and improving intention prediction accuracy by 16%. These improvements support real-time risk assessment and proactive decision-making for connected and autonomous vehicles (CAVs), while also enabling roadside inf
一致的车道水平车辆轨迹为了解驾驶行为提供了大量的时空洞察，这对于提高合并区的安全性和效率至关重要。因此，设计了一个创新的框架，将图形卷积网络（GCN）与Transformer模型结合起来，利用路边激光雷达数据进行车辆轨迹和意图预测。首先，开发了一种无监督意图提取算法，通过从合并区域的路边激光雷达提取高分辨率车道级车辆轨迹来建立驾驶意图数据集。然后，提出了一种耦合车辆模型来模拟合并车辆与干线车辆之间的相互作用。最后，设计了一个多任务框架，通过GCN捕获空间依赖关系，通过Transformer捕获时间模式，用于预测车辆轨迹和意图。实验结果表明，该框架在车辆轨迹预测方面优于现有算法，MAE和RMSE值分别在0.62 m和1.10m以下。意向预测的平均准确率为92.76%。烧蚀研究强调了GCN和耦合车辆模型改进车辆轨迹和意图预测精度的能力，将MAE和RMSE的轨迹预测误差分别降低了约0.3 m和0.4 m，并将意图预测精度提高了16%。从业人员注意事项：本文讨论的实际问题是需要提高合并区的安全性和效率，在合并区，车辆的相互作用经常导致拥堵和事故。准确预测车辆轨迹和意图对于启用先进的驾驶员辅助系统（ADAS）和先进的交通管理系统（ATMS）至关重要。本文介绍了一种创新的框架，该框架将图形卷积网络（GCN）与Transformer模型相结合，利用路边激光雷达数据预测车辆轨迹和意图。该框架旨在在现实场景中运行，利用高分辨率车道级轨迹数据来模拟合并车辆和干线车辆之间的相互作用。我们的方案显著提高了预测精度，轨迹预测的平均绝对误差（MAE）和均方根误差（RMSE）分别低于0.62 m和1.10 m，意图预测的准确率为92.76%。这些结果优于现有的最先进的方法，证明了GCN在捕获空间依赖关系和Transformer在建模时间模式方面的有效性。耦合车辆相互作用模型进一步改进了预测，将轨迹误差减少了约0.3 m （MAE）和0.4 m (RMSE)，并将意图预测精度提高了16%。这些改进支持联网和自动驾驶汽车（cav）的实时风险评估和主动决策，同时也使路边基础设施能够在复杂的合并场景中发出早期预警，并为人类驾驶员提供指导。然而，该框架依赖于高质量的激光雷达数据，这可能会限制其在传感器基础设施不足的地区的适用性。未来的工作可以探索整合其他数据源，如摄像头或车联网（V2X）通信，以增强鲁棒性。除了合并区，这种方法还可以适用于其他交通场景，如十字路口或高速公路出口，以进一步提高交通安全和效率。
{"title":"A Driving Trajectory and Intention Prediction Framework for Vehicle in Merging Zones Using Roadside LiDAR","authors":"Ciyun Lin;Yujia Wang;Bowen Gong;Hongchao Liu","doi":"10.1109/TASE.2026.3662294","DOIUrl":"10.1109/TASE.2026.3662294","url":null,"abstract":"A consistent and lane-level vehicle trajectory offers substantial spatial-temporal insights to understand driving behavior, which is essential to improve safety and efficiency in merging zones. Therefore, an innovative framework was devised that merges a graph convolutional network (GCN) with the Transformer model for vehicle trajectory and intention prediction using roadside LiDAR data. First, an unsupervised intention extraction algorithm was developed to establish a driving intention dataset via high-resolution and lane-level vehicle trajectories extracted from roadside LiDAR in merging zones. Then, a coupled vehicle model was proposed to model the interaction between merging and mainline vehicles. Finally, a multitask framework was designed to capture the spatial dependencies via GCN and temporal patterns via Transformer for predicting vehicle trajectory and intention. Experimental results demonstrated that the proposed framework outperforms the state-of-the-art algorithms in vehicle trajectory prediction, securing MAE and RMSE values under 0.62 m and 1.10m, respectively. The intention prediction achieves an average precision rate of 92.76%. Ablation studies highlighted the ability of GCN and the coupled vehicle model to refine vehicle trajectory and intention prediction accuracy, reducing trajectory prediction errors by approximately 0.3 m and 0.4 m for MAE and RMSE, respectively, and enhancing intention prediction accuracy by 16%. Note to Practitioners—The practical problem addressed in this paper is the need to improve safety and efficiency in merging zones, where vehicle interactions often lead to congestion and accidents. Accurate prediction of vehicle trajectories and intentions is critical for enabling advanced driver-assistance systems (ADAS) and advanced traffic management systems (ATMS). This paper introduces an innovative framework that combines a graph convolutional network (GCN) with a Transformer model to predict vehicle trajectories and intentions using roadside LiDAR data. The framework is designed to operate in real-world scenarios, leveraging high-resolution, lane-level trajectory data to model interactions between merging and mainline vehicles.Our solution significantly improves prediction accuracy, achieving mean absolute error (MAE) and root mean square error (RMSE) values below 0.62 m and 1.10 m, respectively, for trajectory prediction, and a 92.76% precision rate for intention prediction. These results outperform existing state-of-the-art methods, demonstrating the effectiveness of the GCN in capturing spatial dependencies and the Transformer in modeling temporal patterns. The coupled vehicle interaction model further refines predictions, reducing trajectory errors by approximately 0.3 m (MAE) and 0.4 m (RMSE) and improving intention prediction accuracy by 16%. These improvements support real-time risk assessment and proactive decision-making for connected and autonomous vehicles (CAVs), while also enabling roadside inf","PeriodicalId":51060,"journal":{"name":"IEEE Transactions on Automation Science and Engineering","volume":"23 ","pages":"4442-4453"},"PeriodicalIF":6.4,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134379","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","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：