端到端的多任务车辆自动驾驶行为决策模型

打开文本图片集

中图分类号：U461.6 文献标识码：A DOI：10.3969/j.issn.1674-8484.2025.04.011

Abstract：Toaddress the chalenges of spatiotemporal feature processingand inter-task dependencies in autonomous driving decision-making，this paper proposedanend-to-end drivingdecision model basedona 3D windowself-attntionmechanism.Byapplying windowself-attentionto computethe spatiotemporal featuresof theinputsequence，and combining multi-task learning with loss weightallcation，the model effctively extracts features fromdriving videosand predicts vehicle speed and steering angle.The results demonstrate that the proposed model achieves predictionaccuracies of 86.32% forsteering angleand 85.36% for vehicle speed， outperforming modelssuchasFMNet，Swin-Transformer，and MobileT-DSM.Moreover，itrequires only 57.48 GFLOPsofcomputational cost，exhibitingsuperiorspatiotemporal feature extractionaswellasa better trade-off between performance and efficiency.

Keywords：autonomousdriving;decision-making and control; deep learning;multi-task;atentionmechanism

道路交通安全是全球范围内一项重大的公共健康问题。（剩余15351字）