Research On The Mechanism Of Drivers’ Physiological Responses And Construction Of Their Trust Model In Conditionally Automated Driving

Nowadays,automated vehicle technology is one of the forefront and hot spot research orientations in the field of transportation.It uses a variety of sensors such as millimeter wave radar,Li DAR,and GPS to sense their surroundings,handling all aspects of the dynamic driving tasks,and has the benefit of increasing the efficiency of traffic and improving traffic safety.In conditional automated driving,drivers can be fully engaged in non-driving-related tasks(NDRTs)and only needs to takeover control of the vehicle when encountering the system’s limitation situation.Since drivers are out of the control loop,they always have difficulty in negotiating the takeover transitions safely and may have physiological and psychological(e.g.,trust)response to the takeover request.However,in Society of Automotive Engineers(SAE)level 3 automated driving,existing research on the mechanisms of drivers’ physiological responses and trust modeling is still not comprehensive and in-depth.For this purpose,to address the above issues,we will investigate two aspects of drivers’ physiological response mechanism and trust modeling in conditional automated driving,with the main research content as follows:First,the relationships and effects between takeover criticality and situational factors(i.e.,takeover time budget,takeover frequencies,and scenario type)and drivers’ physiological responses(e.g.,skin electrical,heart rate,and pupil diameter)were examined.Based on the subjective scales and physiological parameters collected from the experiment of takeover criticality,a quantitative assessment model of takeover criticality was developed by a generalized linear mixed model to analyze the relationships and effects between the above variables and the takeover criticality.It was indicated that the takeover situations of the higher criticality have larger driver’s maximum HR,mean pupil size,and maximum change in the SC.Besides,the results showed that drivers reported lower takeover criticality ratings when they experienced the longer takeover time budget or at fifth to sixth takeovers.We also observed that the motion attribute of obstacles affects the criticality rating of the takeover situation.The takeover situation of dynamic obstacle had a higher takeover criticality rating than the static obstacle when drivers were in the same time to collision.Second,the mechanisms of human-automation interaction experiences(i.e.,takeover time budget,system usage time),trust propensity,and dynamic trust on physiological responses(e.g.,skin conductance,heart rate,and monitoring behavior)were explored.Based on subjective ratings(e.g.,trust propensity,dynamic trust)and physiological parameters collected from the experiment of driver trust,a multiple moderation mediating effect model(mediated by dynamic trust,moderated by trust propensity and human-automation interaction experiences)were constructed to elucidate the mechanisms underlying complex effects of these variables on physiological responses.The results showed that dynamic trust mediated the effects of human-automation interaction experience on skin conductance and monitoring ratio during non-takeover periods after experiencing takeover.Additionallty,high trust propensity mitigated the effects of takeover time budget on dynamic trust,while system usage time positively moderated the effects of dynamic trust on skin conductance and heart rate.Moreover,shorter takeover time budget weakened the relationship between dynamic trust and monitoring ratio.Third,a hybrid method of combining the subjective rating with the objective behavior indicator was adopted to investigate the mechanism of trust change during the takeover transition and its modeling.To more accurately measure the trust change direction during the takeover transition,firstly,a combination method of subjective and objective trust measurement was proposed.And then,the relationships and effects between drivers’ physiological responses,takeover-related factors,and trust-change directions during takeover transitions were explored by the statistical analyses.It was found that combining the change values of subjective trust scores and monitoring behaviors before and after the takeover could more accurately measure the direction of trust change during the takeover transition.In addition,the results of statistical analysis showed that skin conductance and heart rate during the takeover transition were negatively correlated with the direction of trust change.When drivers experienced more takeovers,longer takeover time budgets,and stationary vehicle scenarios,they were more likely to increase trust during the takeover.Finally,different typical machine learning methods(e.g.,random forest,support vector machine,and decision tree)were applied to establish the recognition models of the trust change direction during takeover transitions.The results indicated that the random forest model had the best recognition performance and can identify the change direction in trust during the takeover transition relatively accurately.Fourth,a driver trust model based on the deep learning method was constructed,and it was applied to the allocation of control authority in human-automation cooperative control.To identify driver trust more accurately,the combination of subjective and objective method proposed in Chapter 5 was used to measure trust in automated vehicles.And an identification model of driver trust in conditional automated driving was developed by label-smoothing-based convolutional neural and long short-term memory networks(CNN-LSTM-LS)fusing multimodal physiological signals and interaction experiences.The proposed model was also compared with four commonly used classification algorithms,such as support vector machines,convolutional neural networks,and long short-term memory networks.The comparison results indicated that the performance of the proposed model was better than the other models.In addition,a human-automation shared control framework integrating trust was constructed and an application study was carried out on the proposed model.The result indicated that the controller can adjust its control authority according to the levels of driver trust and achieve good obstacle avoidance safety.This application example provides theoretical guidance for continuously improving human-automation cooperative control methods in the future.These findings can provide some theoretical basis and technical support for the design of adaptive alarm system,in-vehicle environmental soothing technology,trust monitoring system and human-automation cooperative driving system,which have greater application prospects.