自动驾驶环境下驾驶人接管行为结构方程模型

为提取自动驾驶环境下驾驶人接管行为的关键影响因素，使用驾驶模拟器和眼动仪进行自动驾驶环境下驾驶人接管试验；采集了11个受试者对5种接管情境的反应数据，包括车辆运行数据和眼部运动数据，并调查了受试者的个人属性；基于实测数据定性分析和情境差异定量分析的结果，利用AMOS软件建立了描述驾驶人接管行为的结构方程模型；假设纵向接管行为、横向接管行为和眼部运动行为是3个潜在变量，找到可以表征这3个潜在变量的9个观测变量；根据修正指数多次修正得到最终的结构方程模型，由此获得表征驾驶人接管行为的各变量间的关系及对应的参数。研究结果表明:驾驶人接管自动驾驶车辆的全过程可分为5个阶段，即感知反应、减速避让、加速回升、稳定恢复以及稳定运行；当左前方车辆汇入当前车道，此时驾驶人接管风险较高；横向驾驶行为与纵向驾驶行为、眼部运动行为均显著负相关，相关系数分别为-0.226和-0.223，纵向驾驶行为与眼部运动行为正相关，相关系数为0.152；平均速度、总体横摆角均值、一秒内扫视时间可分别高度解释驾驶人接管自动驾驶车辆时纵向、横向及眼部的潜在行为。可见，此模型能有效揭示驾驶人接管自动驾驶车辆的整体行为与局部行为，有助于改进人机交互模式与自动驾驶接管请求提示。 

Structural equation model of drivers'takeover behaviors in autonomous driving environment

Tests were conducted to explore the key factors that influence drivers' takeover behaviors in an autonomous driving environment using a driving simulator and an eye movement instrument. Data were collected from 11 participants who responded to 5 takeover scenarios, including vehicle and eye movement data, and the participants' personal attributes were investigated. According to the results of measured data processed by qualitative analysis and situational difference processed by quantitative analysis, a structural equation model was established using AMOS to describe drivers' takeover behaviors. The longitudinal takeover behavior, lateral takeover behavior, and eye movement behavior were the three potential variables. Nine observed variables were identified to represent the three potential variables. Based on the modification indices, the final structural equation model was obtained using multiple amendments. Thus, the relationships between all the variables and the corresponding parameters were obtained to describe the drivers' takeover behaviors. Research results show that the entire process in which a driver takes over an autonomous driving vehicle can be divided into 5 stages, including perception and reaction, deceleration and avoidance, acceleration and ascending, stable recovery, and stable movement. The drivers' takeover risk is higher when a left-front vehicle merges into the current lane. The lateral driving behavior is negatively correlated with the longitudinal driving or eye movement behavior, with correlation coefficients of -0.226 and -0.223, respectively. The longitudinal driving behavior is positively correlated with the eye movement behavior, with a correlation coefficient of 0.152. Average speed, mean of the overall yaw angle, and saccade time in a second can interpret the potential longitudinal, lateral, and eye behaviors, respectively, when drivers takeover autonomous driving vehicles. Therefore, the research can reveal drivers' overall and local behaviors when they takeover autonomous driving vehicles, and can help improve the human-computer interaction mode and takeover request hints in autonomous driving. 10 tabs, 7 figs, 30 refs.