汽车工程

汽车工程 ›› 2018, Vol. 40 ›› Issue (12): 1494-1499.doi: 10.19562/j.chinasae.qcgc.2018.012.017

• • 上一篇    

基于端到端学习机制的高速公路行驶轨迹曲率预测*

焦新宇, 杨殿阁, 江昆, 曹重, 谢诗超, 王思佳   

  1. 汽车安全与节能国家重点实验室,智能新能源汽车协同创新中心,清华大学汽车工程系,北京 100084
  • 收稿日期:2017-09-04 出版日期:2018-12-25 发布日期:2018-12-25
  • 通讯作者: 杨殿阁,教授,E-mail:ydg@mail.tsinghua.edu.cn
  • 基金资助:
    基于中美合作的电动汽车前沿技术与应用联合研究项目(2016YFE0102200)、国家自然科学基金(61773234)和北京市科技计划课题(D171100005117002)资助。

Driving Trajectory Curvature Prediction of Vehicle on Highway Based on End-to-end Learning Mechanism

Jiao Xinyu, Yang Diange, Jiang Kun, Cao Zhong, Xie Shichao, Wang Sijia   

  1. State Key Laboratory of Automotive Safety & Energy, Collaborative Innovation Center of Intelligent New Energy Vehicle,The Department of Automotive Engineering of Tsinghua University, Beijing 100084
  • Received:2017-09-04 Online:2018-12-25 Published:2018-12-25

PDF (PC)

431

赞

摘要： 本文中基于端到端学习机制进行了高速公路场景下的车辆行驶轨迹预测。首先,为量化表达行驶轨迹,并对预测结果进行合理评价,建立了行车轨迹曲率预测模型与评价体系。然后,针对端到端的行驶轨迹曲率预测训练集中驾驶员行为决策的不确定性导致性能不佳的问题,采用场景切分和特征预提取的方法进行优化和实车试验验证。结果表明,该方法提高了高速公路行驶轨迹预测的准确性和可靠性。

关键词: 智能汽车, 端到端学习机制, 轨迹预测, 场景切分, 特征预提取

Abstract: The driving trajectory of vehicle on highway is predicted based on end-to-end learning mechanism in this paper. Firstly for quantitatively express the driving trajectory of vehicle and reasonably evaluating prediction results, a model along with an evaluation system for driving trajectory curvature prediction are established. Then in view of the problem of poor performance of end-to-end-based driving trajectory curvature prediction caused by the uncertainty of driver's behavior decision in training set, an optimization is conducted by using scene segmentation and feature pre-extraction techniques,which is verified by real vehicle test. The results show that the method adopted enhances the accuracy and reliability of driving trajectory prediction on highway

Key words: intelligent vehicle, end-to-end learning mechanishm, trajectory prediction, scene segmentation, feature pre-extraction