汽车工程 ›› 2023, Vol. 45 ›› Issue (12): 2330-2337.doi: 10.19562/j.chinasae.qcgc.2023.12.015

所属专题: 智能网联汽车技术专题-感知&HMI&测评2023年

• • 上一篇    下一篇

不同认知负荷非驾驶任务下高度自动化驾驶接管绩效预测

马艳丽(),卢俊,朱洁玉(),韩笑雪   

  1. 哈尔滨工业大学交通科学与工程学院,哈尔滨 150090
  • 收稿日期:2023-04-09 修回日期:2023-06-17 出版日期:2023-12-25 发布日期:2023-12-21
  • 通讯作者: 马艳丽,朱洁玉 E-mail:mayanli@hit.edu.cn;zhujieyu9322@163.com
  • 基金资助:
    国家自然科学基金(52372325);黑龙江省自然科学基金(LH2020E056)

Take-over Performance Prediction Under Different Cognitive Loads of Non-driving Tasks in Highly Automated Driving

Yanli Ma(),Jun Lu,Jieyu Zhu(),Xiaoxue Han   

  1. School of Transportation Science and Engineering,Harbin Institute of Technology,Harbin  150090
  • Received:2023-04-09 Revised:2023-06-17 Online:2023-12-25 Published:2023-12-21
  • Contact: Yanli Ma,Jieyu Zhu E-mail:mayanli@hit.edu.cn;zhujieyu9322@163.com

摘要:

在高级自动化驾驶中,准确预测驾驶接管绩效对于提高自动驾驶接管安全水平意义重大。文中设计了不同认知负荷非驾驶任务下驾驶接管场景,分析不同认知负荷非驾驶任务下驾驶接管绩效指标与脑电指标的显著性,采用驾驶人eSense数值和脑电波数据作为输入特征,构建了基于随机森林的驾驶接管绩效预测模型,分析了模型在3、5、7、9 s 4种时间窗下的预测性能,并开展了模型的有效性验证。结果表明,不同认知负荷非驾驶任务下的接管时间、最大横向加速度、最小TTC及驾驶人eSense等存在显著性差异;随机森林在9 s时间窗下的预测性能最佳,其准确率达到了0.94;随机森林的预测准确率和micro-AUC面积高于支持向量机、朴素贝叶斯和逻辑回归。研究方法可有效预测驾驶员自动驾驶接管绩效,并为驾驶员与自动驾驶车辆之间的交互设计提供理论依据。

关键词: 交通工程, 接管绩效预测, 随机森林, 认知非驾驶任务, 高度自动化驾驶

Abstract:

In highly automated driving, accurate prediction of takeover performance is of great significance to improve the safety of automated driving takeover. Based on the design of driving take-over scenarios under different cognitive load non driving-related tasks (NDRT), the significance of takeover performance indicators and EEG indicators under different cognitive load NDRT of automated driving is analyzed. Using eSense value and brainwave data of the driver as input, a prediction model of take-over performance based on random forest is constructed to analyze the prediction effect of the model within time windows of 3, 5, 7 and 9 s, and the validity of the model is verified. The results show that there are significant differences in take-over time, maximum lateral acceleration, minimum TTC and driver’s eSense under different loads of NDRT. It is found that the random forest has the best prediction performance within 9 s time window, with the accuracy of 0.94. The prediction accuracy and micro-AUC area of random forest are higher than the results of support vector machine, naive Bayes and logistic regression. The proposed method can effectively predict the take-over performance and provide a theoretical basis for the interaction design between the driver and the autonomous vehicle.

Key words: traffic engineering, takeover performance prediction, random forest, cognitive DNRT, highly automated driving