汽车工程 ›› 2022, Vol. 44 ›› Issue (6): 909-918.doi: 10.19562/j.chinasae.qcgc.2022.06.013

所属专题: 底盘&动力学&整车性能专题2022年

• • 上一篇    下一篇

基于车辆动力学响应特征的越野地面分类方法

赵健1,李雅欣1,佟静1,朱冰1(),武维祥2,孙博华1,韩嘉懿1   

  1. 1.吉林大学,汽车仿真与控制国家重点实验室,长春  130022
    2.上海捷能汽车技术有限公司,上海  201804
  • 收稿日期:2022-01-14 修回日期:2022-02-21 出版日期:2022-06-25 发布日期:2022-06-28
  • 通讯作者: 朱冰 E-mail:zhubing@jlu.edu.cn
  • 基金资助:
    吉林省自然科学基金(20210101057JC);国家自然科学基金(52172386);长沙市“揭榜挂帅”重大科技项目(kq2102008);青年科学基金项目(52102457)

Cross-Country Road Classification Method Based on Vehicle Dynamic Response Characteristics

Jian Zhao1,Yaxin Li1,Jing Tong1,Bing Zhu1(),Weixiang Wu2,Bohua Sun1,Jiayi Han1   

  1. 1.Jilin University,State Key Laboratory of Automotive Simulation and Control,Changchun  130022
    2.Shanghai E-Propulsion Auto Technology Co. ,Ltd. ,Shanghai  201804
  • Received:2022-01-14 Revised:2022-02-21 Online:2022-06-25 Published:2022-06-28
  • Contact: Bing Zhu E-mail:zhubing@jlu.edu.cn

摘要:

利用车辆动力学响应进行地面分类是越野智能汽车的关键技术之一。本文中提出了结合地面不平整度特征和力学特征进行越野地面分类的方法,对沙地、土路、水泥路和雪地进行分类。本方法中选取等效地面轮廓和车身垂向加速度作为地面不平整度特征,选取行驶阻力和轮速波动作为力学特征,设计了基于LSTM模型的越野地面分类器,对自行采集的车辆越野行驶数据集进行训练与测试,结果表明,分类正确率达到95.5%;最后,使用HMM模型实现了分类后处理,解决了分类结果在连续数据上跳变的问题,使该算法在连续越野数据上的地面分类正确率从88.44%提高到90.13%。

关键词: 越野地面, 地面分类, 分类后处理, 车辆响应, 机器学习

Abstract:

Terrain classification based on vehicle dynamic response is one of the key technologies of off-road intelligent vehicles. In this paper, a method of off-road terrain classification, combining the terrain roughness features and mechanical characteristics, is proposed, with the sand, dirt, cement and snow roads classified. In this method, the equivalent terrain profile and the vertical acceleration of vehicle body are selected as the characteristics of terrain roughness, with the driving resistance and wheel speed fluctuation as the mechanical characteristics, the cross-country road classifier is designed based on LSTM model, and the training and testing are conducted on the cross-country driving dataset of vehicle. The results show that the correct rate of terrain classification reaches 95.5%. Finally, the post-processing of classification is fulfilled by using HMM model to solve the issue of the abrupt change of continuous data, enabling the correct rate of the algorithm in terrain classification with off-road continuous data rises from 88.44% to 90.13%.

Key words: off-road terrain, terrain classification, classification post-processing, vehicle response, machine learning