STYLE-Net:风格感知与可靠性评估驱动的车辆轨迹预测模型

STYLE-Net:a Style-aware and Reliability-driven Vehicle Trajectory Prediction Model

摘要

轨迹预测的精度和稳定性关系到自动驾驶的安全.现有方法多聚焦几何轨迹和场景交互建模,忽略了驾驶风格的内在一致性,限制了预测性能提升.为此,本文提出了一种基于风格感知和可靠性评估的轨迹预测方法STYLE-Net.针对风格离散分类导致个体风格细粒度变化表征难的问题,设计了自监督连续风格模块.该模块基于双向LSTM和自注意力机制提取全局风格特征,再通过自监督学习得到精炼的连续风格特征和预测的可靠性分数,实现个性化风格的高效捕捉;针对突发场景下风格可能不稳定的问题,提出了可靠性驱动的动态融合策略,以可靠性分数调节连续风格特征与时空特征的融合权重,提升预测的鲁棒性.最后,构建了风格一致性驱动的联合优化框架,实现轨迹精度和风格一致性的协同提升.实验结果表明,STYLE-Net预测性能优异,实现了高效稳定且个性化的轨迹预测.

Abstract

The safety of autonomous driving is directly impacted by the accuracy and stability of trajectory prediction.Existing methods primarily focus on modeling of geometric trajectories and scene interactionsbut often overlook the inherent temporal consistency of driving styles.This limitation restricts improvements in prediction performance.To address these issues,this paper proposes a trajectory prediction model(named STYLE-Net) based on style awareness and reliability evaluation.A Self-Supervised Continuous Style Module is designed to overcome the challenges of capturing the fine-grained,continuous changes of individual styles,which are caused by discretized style representations in existing methods.This module leverages a Bi-LSTM and a self-attention mechanism to extract global style features.Through self-supervised learning,it produces refined continuous style features and a predicted reliability score,achieving efficient capture of personalized driving styles.A Reliability-Driven Dynamic Fusion Strategy is proposed to handle the potential style instability in abrupt scenes.This strategy uses the reliability score to modulate the fusion weight between the continuous style features and the spatio-temporal features,thus enhancing prediction robustness.Finally,a Style-Consistency-Driven Joint Optimization Frameworkis developed to achieve the synergistic improvement of trajectory accuracy and style consistency.Experiments demonstrate that STYLE-Net achieves superior performance,enabling efficient,stable,and personalized trajectory forecasting.

关键词

车辆轨迹预测 / 驾驶风格建模 / 动态融合 / 联合优化 / 深度学习

Key words

vehicle trajectory prediction / driving style modeling / dynamic fusion / joint optimization / deep learning

[1] Argo A I.Argoverse motion forecasting competition[EB/OL].https://eval.ai/web/challenges/challenge-page/454/leaderboard/1279,2025-10-09.
[2] Ju C,Wang Z,Long C,et al.Interaction-aware kalman neural networks for trajectory prediction[C]//Proceedings ofIEEE Intelligent Vehicles Symposium(IV),2020:1793-1800.
[3] Ye N,Zhang Y Y,Wang R C,et al.Vehicle trajectory prediction based on hidden Markov model[J].KSII Transactions on Internet and Information Systems(VTC-Fall),2016,10(7):3150-3170.
[4] YAN J H,LIU Z Y,WANG Z.Multi-scale convolutional vehicle trajectory prediction integrating spatiotemporal attention mechanism[J].Computer Engineering,2025,51(8):406-414.
[5] Zuo Z Q,Wang X Y,Guo S L,et al.Trajectory prediction network of autonomous vehicles with fusion of historical interactive features[J].IEEE Transactions on Intelligent Vehicles,2024,9(1):2171-2183.
[6] Zhang S M,Bai R,He R,et al.Research on vehicle trajectory prediction methods in urban main road scenarios[J].IEEE Transactions on Intelligent Transportation Systems,2024,25(11):1-17.
[7] Wu K S,Zhou Y,Shi H T,et al.Graph-based interaction-aware multimodal 2d vehicle trajectory prediction using diffusion graph convolutional networks[J].IEEE Transactions on Tntelligent Vehicles,2024,9(2):3630-3643.
[8] Zhou Z K,Ye L Y,Wang J P,et al.HiVT:hierarchical vector transformer for multi-agent motion prediction[C]//Proceedings ofIEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR),2022:8813-8823.
[9] Sheng Z H,Xu Y W,Xue S B,et al.Graph-based spatial-temporal convolutional network for vehicle trajectory prediction in autonomous driving[J].IEEE Transactions on Intelligent Transportation Systems,2022,23(10):17654-17665.
[10] Xu D,Lyu N C,Yang L.A proactive risk prediction framework for cut-in maneuvers incorporating inherent driving styles[J].IEEE Transactions on Intelligent Transportation Systems,2025,doi:10.1109/TITS.2025.3607860.
[11] Yang J Y,Lee J J,Antoniou C.Trajectory prediction for multiple agents in dynamic environments:factoring in traffic states and driving styles[J].IEEE Transactions on Intelligent Transportation Systems,2025,26(11):19281-19295.
[12] Li X,Yang B,Zeng X M,et al.StyleFormer:multi-agent joint trajectory prediction and planning in urban environments with driving style awareness[J].IEEE Transactions on Intelligent Transportation Systems,2025,26(11):20317-20333.
[13] Wang Y F.Personalized trajectory inference framework integrating driving behavior recognition and temporal dependency learning[C]//Proceedings of 5th International Conference on Artificial Intelligence and Industrial Technology Applications(AIITA),2025:1991-1997.
[14] Zhang Q X,Xing Y,Wang J X,et al.Interaction-aware and driving style-aware trajectory prediction for heterogeneous vehicles in mixed traffic environment[J].IEEE Transactions on Intelligent Transportation Systems,2025,26(7):10710-10724.
[15] Wu S W,Sun F,Zhang W T,et al.Graph neural networks in recommender systems:asurvey[J].ACM Computing Surveys,2023,55(5):1-37.
[16] Zhou C,Wang K.ST3GAN:a Chinese character font generation model combining structural-stroke consistency loss and style perception loss[C]//Proceedings of4th International Symposium on Computer Applications and Information Technology(ISCAIT),2025:883-891.
[17] Chen L,Bian Y,Zeng J B,et al.Style consistency unsupervised domain adaptation medical image segmentation[J].IEEE Transactions on Image Processing,2024,33(1):4882-4895.
[18] ZHANG K,LIANG W,CAO P,et al.Semi-supervised multi-label medical image classification model guided by label correlation[J].Journal of Chinese Computer Systems,2025,46(9):2160-2167.
[19] Wang X,Zhan Y,Zhao Y,et al.Semi-supervised crowd counting with spatial temporal consistency and pseudo-label filter[J].IEEE Transactions on Circuits and Systems for Video Technology,2023,33(8):4190-4203.
[20] Liang M,Yang B,Hu R,et al.Learning lane graph representations for motion forecasting[C]//Proceedings of European Conference on Computer Vision(ECCV),2020:541-556.
[21] Qi R C,Su H,Mo K C,et al.PointNet:deep learning on point sets for 3D classification and segmentation[C]//Proceedings ofIEEE Conference on Computer Vision and Pattern Recognition(CVPR),2017:77-85.
[22] Tang X L,Kan M N,Shan S G,et al.HPNet:dynamic trajectory forecasting with historical prediction attention[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR),2024:15261-15270.
[23] Wang M K,Ren X G,Jin R C,et al.FutureNet-LOF:joint trajectory prediction and lane occupancy field prediction with future context encoding[J].arXiv preprintarXiv:2406.14422,2024.
[24] Fan B,Yuan H F,Dong Y K,et al.Bidirectional agent-map interaction feature learning leveraged by map-related tasks for trajectory prediction in autonomous driving[J].IEEE Transactions on Automation Science and Engineering,2025,22(1):10801-10813.
[25] Zhou Z K,Wang J P,Li Y H,et al.Query-centric trajectory prediction[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR),2023:17863-17873.
[26] Lan Z Q,Jiang Y X,Mu Y,et al.SEPT:towards efficient scene representation learning for motion prediction[J].arXiv preprintarXiv:2309.15289,2023.
[27] Pei M L L,Shi S S,Chen X S,et al.Foresight in motion:reinforcing trajectory prediction with reward heuristics[J].arXiv preprintarXiv:2507.12083,2025.
[28] Jia X S,Wu P H,Chen L,et al.HDGT:heterogeneous driving graph transformer for multi-agent trajectory prediction via scene encoding[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(11):1-15.
[29] Wang M K,Zhu X G,Yu C Q,et al.GANet:goal area network for motion forecasting[J].IEEE International Conference on Robotics and Automation,2023:1609-1615,doi:10.1109/icra48891.2023.10160468.
附中文参考文献:
[4] 闫建红,刘芝妍,王 震.融合时空注意力机制的多尺度卷积车辆轨迹预测[J]. 计算机工程 ,2025,51(8):406-414.
[18] 张 凯,梁 伟,曹 鹏,等.标签关联性指导的半监督多标签医疗图像分类模型[J].小型微型计算机系统,2025,46(9):2160-2167.

基金资助

国家自然科学基金项目(62373251,62203306)资助.