摘要
人机共驾是现阶段汽车智能化发展的必经之路,在人机共驾中为了避免人机冲突,对驾驶人的人机共驾状态进行识别是实现和谐人机共驾的基础。然而现有研究较少考虑了该状态,同时相关识别方法多基于驾驶人生理信息,导致应用繁琐,不具备实用性。为此,设计了一种能够测量握力分布的智能方向盘系统硬件架构,并在此基础上开发了基于握力分布的驾驶人人机共驾状态识别方法。首先搭建了能够测量驾驶人双手握力分布的智能方向盘系统,在此基础上利用驾驶人在环试验台采集了15名驾驶人在不同人机共驾状态下的试验数据;然后根据试验数据通过递推最小二乘法对驾驶人的上肢肌肉阻抗特性参数进行了辨识,分析了不同状态下的驾驶人上肢肌肉特性;最后基于门控循环单元(Gated Recurrent Unit,GRU)构建了Hybrid-GRU(H-GRU)模型,将回归任务与分类任务混合,利用辨识得到的肌肉阻抗特性结果对模型中的回归部分进行预先训练,使模型具备了一定的先验知识,实现了从驾驶人握力分布到人机共驾状态的映射,并将H-GRU模型与常规GRU模型和支持向量机模型进行对比测试。结果表明:所建立的模型总体分类准确率达到97.59%,相比常规GRU模型和支持向量机模型分别提升6.97%和33.02%。所提出的基于方向盘握力分布的人机共驾状态识别方法不仅能够准确辨识驾驶人人机共驾状态,还能够输出驾驶人肌肉阻抗特性参数,可为驾驶人建模或人机共驾策略开发等提供帮助。
Currently,human-machine shared driving is the only way to develop intelligent vehicles.The recognition of the driver state is the basis for avoiding human-machine conflict in human-machine shared driving.However,existing research has seldom considered this state feature,and related recognition methods are primarily based on the driver’s physiological information,which is cumbersome to apply in practice.Therefore,in this study,a smart steering wheel system that can measure the distribution of grip force was designed,and on this basis,a recognition method for the human-machine shared-driving state was developed.First,a smart steering wheel system was built to measure the grip force distribution.The experimental data of 15 drivers in different shared driving states were collected via a driver-in-the-loop experiment.Then,based on the experimental data,the impedance characteristics of the driver’s muscles were identified using the recursive least squares method,and the characteristics under different states were analyzed.Finally,based on the gated recurrent unit(GRU),a hybrid-GRU(H-GRU)model that mixed the classification and regression tasks was constructed,in which the regression part was trained in advance using the identified impedance characteristic results to train the model with prior knowledge,and the mapping from the driver grip force distribution to the shared driving state was realized.Moreover,the H-GRU model was compared with the conventional GRU and SVM models.The results show that the classification accuracy of the H-GRU model reached 97.59%,which is 6.97%and 33.02%higher than that of the conventional GRU and SVM models,respectively.The proposed driver shared driving state recognition method based on the grip force distribution accurately identifies the state and also outputs the driver’s muscle impedance characteristic parameters,which is useful to model the driver and shared driving strategy development.
作者
韩嘉懿
朱冰
赵健
马驰
HAN Jia-yi;ZHU Bing;ZHAO Jian;MA Chi(State Key Laboratory of Automotive Simulation and Control,Jilin University,Changchun 130022,Jilin,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2022年第3期166-176,共11页
China Journal of Highway and Transport
基金
国家自然科学基金项目(51775235).
关键词
汽车工程
人机共驾
GRU神经网络
驾驶人状态
握力分布
驾驶人肌肉阻抗
递推最小二乘法
automotive engineering
human-machine shared driving
GRU
driver state
grip force distribution
driver muscle impedance
recursive least-squares method
作者简介
韩嘉懿(1992-),男,吉林长春人,工学博士研究生,E-mail:jiayi.han@qq.com;通讯作者:朱冰(1982-),男,吉林双辽人,教授,博士研究生导师,工学博士,E-mail:zhubing@jlu.edu.cn。
