Worldwide, over a million fatal road accidents occur each year. The majority of these crashes are caused by inadequate longitudinal control performance such as late braking and close following. Despite the fact that car driving is primarily a visual task, there is yet no clear understanding of how drivers control their vehicle in safety-critical conditions as a function of visual information. Understanding the visual information needs of drivers in different safety-critical conditions is a prerequisite for designing and validating interventions (e.g., support systems and training programs) that reduce the number and severity of collisions. Longitudinal and lateral control tasks of driving an automobile have been extensively studied as tracking control problems. However, studies of drivers’ behaviour in collisions have been scarce in general and particularly in terms of drivers’ control performance. Most of the driving behaviour research considers the brake reaction time as the main factor in assessing drivers’ behaviour in safety-critical conditions. Comparatively little is known about the performance of drivers after the brake onset, including the dosing and duration of the brake pedal input. In addition, previous studies did not clearly investigate the effect of visual/physical conditions such as the gap and the relative velocity between drivers and other road objects, and the visibility condition of the road on the performance and risk assessment of drivers. In current traffic systems, drivers are responsible for navigating the vehicle safely. A variety of technological interventions have been developed to assist drivers in collision prone conditions. These technological systems often use absolute visual information (e.g., distance, time headway) to control the vehicle without taking into account driver perception. Therefore, there may be a mismatch between what such systems do and what drivers expect from such systems to do. Little human factors knowledge is available about how to design driver support systems that improve longitudinal control performance of drivers. The first objective of this thesis is to understand how the availability and quality of visual information in safety-critical driving conditions shapes drivers’ longitudinal control performance. Following this, the second objective of this thesis is to design and investigate the effectiveness of a technological solution for improving longitudinal control performance. A total of four driving simulator experiments were conducted that assessed the effects of degradedvision and augmented-vision conditions in safety-critical stopping and car following tasks. The first experiment (Chapter 2) examined the effect of visual information on braking performance of drivers faced with a decelerating lead car. Four lead-car braking conditions were created by varying the deceleration of the lead car (1.7 vs. 6.5 m/s2) and the distance between the participant’s car and the lead car (13.4 vs. 33.4 m). Three visual conditions were tested: lead-car brake lights, no lead-car brake lights, and visual occlusion at the onset of lead-car deceleration. The braking behaviour of drivers has been analysed by relating the braking input of the driver to the visual information of the driving condition. The results showed that an occlusion as short as 0.4 s (about the duration of a glance on the speedometer) can dramatically increase crash risk. This implies that if following at a 0.5 s time headway (a short but not unrealistic headway), any off-road glance should be avoided. Brake lights were found to reduce brake reaction times when the lead-car deceleration was small (1.7 m/s2) but had little added value when the lead car engaged in an emergency stop (6.5 m/s2). In summary, the results of the first experiment indicate that an off-road glance when the most critical driving condition (short headway, high deceleration of the lead vehicle) occurs significantly increases the number of crashes. Even alert drivers require continuous visual information to be able to avoid collisions in very critical conditions. The second experiment (Chapter 3) investigated the braking performance of drivers when stopping at a stationary target as a function of the temporal demand of the braking event and the presence versus absence of visual information during braking. The access to visual information was manipulated by occluding the screen at the start of half of the braking trials, and the temporal demand was manipulated by changing the time-to-arrival (TTA) at the onset of braking. Contrary to expectations, it was found that the lack of visual information after the brake onset reduced the maximum brake force applied by drivers, especially in braking events with short TTAs (? 4 s). For the larger TTA values (? 6 s), participants in the occlusion condition stopped too early and at variable positions on the road as compared to the non-occluded condition. In the occlusion condition, participants were likely to apply an intermediate brake pedal depression, whereas in the non-occluded condition participants applied low or high pedal depressions. Overall, the findings indicate that without vision, drivers underestimate the required brake input for optimum performance in safety-critical conditions. The availability of visual information during a stopping task improves performance, even when the stopping task is urgent. This is in line with the findings in Chapter 2 which showed that drivers need (continuous) visual information even when an ‘open loop’ braking action would in theory suffice. The third experiment (Chapter 4) investigated the underlying causes of the paradoxical phenomenon that drivers adopt short distance headways in fog compared to clear visibility conditions. The effects of visual information (fog vs. clear weather) and automation (adaptive cruise control vs. manual driving) on the subjective feeling of risk (measured during driving using a touch screen) and steering activity at different distance headways were examined. The results show that participants’ feeling of risk was lower in clear weather than in fog, especially when the headway was large. It is concluded that participants in fog try to keep the lead car just in sight, and that the lead car provides a guide resulting in reduced lateral control activity. In line with the findings of Chapters 2 and 3, a lack of visual information of the lead car was found to be detrimental for the performance of drivers. It is concluded that, having access to continuous visual information is so critical that drivers reduce their headway to improve the availability and quality of visual information. The results suggest that except for extremely short headways, keeping the vehicle at the edge of the visibility threshold reduces the perceived risk. The results also showed that extremely short headways induce elevated feelings of risk, even when the driving task is automated. It is argued that adaptive cruise control systems should either avoid extremely short headways or include a driver information system to reduce the level of risk that drivers perceive in very close following distances. In the final experiment (Chapter 5), a head-up display (dubbed Rear Window Notification Display, or RWND) was developed to improve the driver’s manual car-following performance by continuously visualizing the lead car’s acceleration and time headway on the rear window of the lead car. The effect of the system (RWND off vs. on) on the car following performance was determined when following a lead car driving with variable speed. The results showed that the RWND reduced both the mean and standard deviation of time headway, but did not increase the occurrence of potentially unsafe headways of less than 1 s. Using a linear car-following model, it was shown that when assisted by the display, participants improved their performance by adopting higher control gains with respect to inter-vehicle distance, relative speed, and acceleration compared to when they were not assisted. In Chapter 6 a short literature review is provided on human factors issues of automated driving. It is shown that automation is no panacea and may actually lead to new types of risk compared to manual driving, such as overreliance, loss of skills, and behavioural adaptation. Several design solutions are proposed that inform and involve the human driver about the situation ahead and the automation status. Moreover, several design requirements are proposed for a cooperative adaptive cruise control (CACC) system that allows for platooning with short headways. The results of this chapter reinforce statements made in the earlier chapters that drivers need to be properly informed about the environment and automation status. In Chapter 7, the results are summarized and the findings of Chapters 2 to 5 are interpreted by means of perceptual control models. A comparison between the experimental results and the reviewed theoretical models suggests that the perceptual sensitivity of drivers improves when the distance headway decreases, which in turn improves the accuracy of drivers’ longitudinal control performance. The control models also support the performance results of the RWND system, where direct operational information about the acceleration and deceleration of the lead car provided bypasses the perceptual sensitivity threshold of drivers. Driving simulators have been considered as suitable tools with relative validity to test the effects of the availability and quality of visual information on longitudinal control of the vehicle in collision-prone conditions. Driving simulator provides an environment free of physical risk even when a driver fails to avoid a collision. The visual and kinematics conditions of simulated driving scenes are also controllable to a great extent. This chapter also further justifies the ecological validity of the tasks, and the kinematics and the frequency of the events tested in the previous experiments, and suggests several future research directions on related road safety issues. When a collision is imminent and there is a need for a rapid manoeuvre within a very small time frame (less than a few seconds), drivers who are not fully occupied by the driving task do not have their full attention resource available to intervene. Having an understanding of the limitations of drivers in these safety-critical conditions is a prerequisite for designing technologies that aim to enhance the performance of drivers in such situations. This thesis generated knowledge on how drivers visually control their vehicle in safety-critical conditions by showing the critical role of visual feedback in such situations and how disturbances in visual information during these conditions affect longitudinal control performance. The thesis also showed how drivers reduce their following distance as an adaptation mechanism to cope with the performance decline when the quality of visual information is degraded. Such knowledge led to the development of a RWND that keep drivers ‘in the loop’ while benefiting from technological advances. The findings of this work highlight the deficiencies that exist in drivers’ control of the vehicle in safety-critical situations and demonstrated the viability of cooperation between the human driver technologies, such as the RWND, to support drivers’ intervention in situations prone to longitudinal crashes. The results suggest that the RWND can be used along with CACC to increase network capacity without degrading safety. Mental workload and distraction effects should be evaluated in further experiments, including on-road testing in a naturalistic environment and with a more diverse population. (Author/publisher)
Samenvatting