自动驾驶接管绩效的影响因素、模型与评价方法综述

Review of Take-over Performance of Automated Driving:Influencing Factors,Models,and Evaluation Methods

有条件自动驾驶系统无法应对所有驾驶场景,因而需要驾驶人在必要情况下进行接管。驾驶人的接管绩效对于有条件自动驾驶车辆的安全性、驾乘体验与接受度具有重要意义。近年来,大量文献从不同视角对驾驶人接管绩效这一主题进行了广泛研究,但仍然存在一些问题亟待解决。从驾驶人接管绩效影响因素、接管绩效模型与接管绩效评价方法3个方面综述了驾驶人接管绩效的研究现状。首先,从驾驶人因素、交通环境因素和自动驾驶系统因素3个维度对驾驶人接管绩效影响因素的相关研究进行综述。其次,对现有驾驶人接管绩效模型,包括经典统计模型、机器学习模型与结构方程模型进行总结。最后,总结了现有原始接管绩效评价指标以及接管绩效综合评价方法。分析发现:现有接管绩效影响因素的量化指标仍不够全面,现有接管绩效模型的可解释性和预测精度难以兼顾,现有接管绩效评价方法尚需进一步完善。有鉴于此,未来研究首先需要基于驾驶人群体主观评价提出接管绩效的全面评价方法,然后以此为基础从人-机-环境维度全方面探索接管绩效影响因素的量化指标,最后考虑多种影响因素的复杂关联关系,建立高精度接管绩效预测模型,从而为提升驾驶人接管绩效提供理论支持,促进有条件自动驾驶的进一步发展。

Conditionally automated driving systems,though advanced,are not universally adept at managing all driving scenarios and require driver intervention when necessary.The efficacy of driver take-over is paramount for the safety,user experience,and broader acceptance of such automated vehicles.A plethora of recent studies rigorously examined driver take-over performance,but certain challenges persist.This study presented a systematic review of extant literature concerning driver take-over performance,encapsulating the influencing factors,the models,and the various evaluation methodologies employed.The determinants influencing take-over performance span driver-specific factors,traffic environment parameters,and features of the automated driving systems.Concerning the modeling of take-over performance,distinctions were drawn between classical statistical models,machine learning approaches,and structural equation models.The study further encapsulated prevailing evaluation indices specific to take-over performance,alongside holistic evaluation methodologies.Findings from the review pinpoint that current indicators for influencing factors lack comprehensiveness.Additionally,a discernible imbalance between interpretability and predictive accuracy is observed in the existing models.Furthermore,the present evaluation methods for take-over performance necessitate refinement.As a roadmap for future inquiries,this study advocates for the initiation of comprehensive measures of take-over performance based on subjective evaluation of human drivers.Then,there is an imperative to develop quantitative indicators of the influencing factors of take-over performance from human-machine-environment aspects.Conclusively,calls are made for crafting high-precision predictive models for take-over performance that duly recognize the intricate interdependencies of myriad influencing factors.Pursuing such avenues of research is vital to provide theoretical support for elevating driver take-over performance,thus propelling the evolution of conditionally automated driving.