Study on Accident-Avoidance Mechanism in Driver-Vehicle System When Activating Level-2 ADAS

Abstract

Advanced Driver Assistance Systems (ADAS) in passenger cars, such as Adaptive Cruise Control (ACC) and Lane Keeping Assist (LKA), have recently been widely deployed. However, these systems are not designed for high-risk situations, yet some drivers over trust the system and engage in secondary tasks. These behaviours may lead to serious accidents. On the other hand, it has been reported that the use of ACC reduced approximately half of the collision rate on highways based on public road data. This study aims to clarify the mechanism how ADAS affects driver behaviour and enhances accident prevention performance. Specifically, the focus is on the forward collision risk and the driver’s brake operating behaviour. Through the analysis using a driving simulator, this study compared driver behaviour in high collision risk situations in the case with and without ACC. The analysis of subject drivers’ internal parameters confirmed the driver behaviour that avoided a collision by using ADAS to take over control and brake earlier.

1 Introduction

Advanced Driver Assistance Systems (ADAS), such as Adaptive Cruise Control (ACC) and Lane Keeping Assist (LKA), have recently become widespread. These systems are designed to reduce the driver’s workload. Consequently, these systems cannot handle high accident risk situations. However, some drivers over trust ADAS and this results in distracted driving. Several serious accidents have occurred when using ADAS due to driver overconfidently trusting the system and sometimes lead to fatal accidents [1, 2]. On the other hand, Otsuki et al. reported that the ratio of collisions on highways is reduced to less than half with the use of ACC [3]. These facts indicate that although ADAS alone do not have high accident-avoidance performance, ADAS can avoid accidents by influencing on driver before the accident risk increases. The accident reduction mechanism has not been quantitatively clarified yet.

This research aims to quantitatively clarify the mechanism by which ADAS influences driver behavior and enhances accident-avoidance performance before accident risk increases. To clarify driver behavior when using ADAS, experiments with subject drivers were conducted using a driving simulator. Two experiments, gaze movement during distracted driving and driver collision avoidance behavior, were examined respectively under two conditions, with and without ADAS. This analysis was based on the hypothesis that active deceleration intervention and warning of ADAS hasten the driver's perception of forward risk even though ADAS induce forward inattention. This paper focuses on acceleration and deceleration support in the front and rear directions and verifies the hypothesis using the situation where the preceding vehicle suddenly brakes in distracted driving.

2 Driving Simulator Experiment

2.1 Experiment Overview

Two experiments were conducted for monitoring eye movement and driver risk-avoidance behavior respectively. A motion-based driving simulator with four electric actuators, which conduct motion cueing in the pitch and roll directions, was used. The main task of the subject driver during the experiment was to follow a preceding vehicle traveling at varying speeds of 80–100 km/h on the highway. In addition, as a subtask to replicate distracted driving, the participants were instructed to answer simple questions on a fixed smartphone within designated sections. This task was designed to imitate operating a car navigation system. The subtask sections appeared several times on the driving course. The driving simulator was also equipped with ACC and LKA to mimic the commercial Level-2 ADAS. Before starting the test, the experiment participants agreed to give informed consent and practiced using the driving simulator to familiarize themselves with the simulator environment. These experiments were approved by the Ethical Review Committee of the Tokyo University of Agriculture and Technology.

2.2 Gaze Movement

In order to quantify the forward inattention state during doing the subtask, all participants put on glasses-type eye tracker, Tobii pro Glasses 2. The setting of the experiment is shown in Fig. 1. These test runs were conducted for each subject under two conditions: ACC ON and ACC OFF, with LKA always activated to reduce the driving burden. The order of testing was changed for each half to account for order effects. 24 subjects who hold driving licenses participated in the experiment. Their average age was 26.1 years (range: 21–54). To guarantee uniformity in the risk perception for the preceding vehicle even without the activation of ACC, zones indicating the recommended maximum and minimum headway distances were displayed in the driving sections without secondary task execution.

Fig. 1.

The setting of driving simulator experiment using eye tracker.

2.3 Driver Risk-Avoidance Behavior

In order to evaluate the collision risk and quantify the driver’s brake stepping behavior, the preceding vehicle suddenly braked at the end of the final subtask section. This emergency scenario could not be avoided solely by relying on ACC and the drivers were not informed about it before the test. These test runs were conducted for each subject under two conditions: ADAS ON (ACC and LKA were activated) and ADAS OFF (manual driving). 7 subjects who hold driving licenses participated in the experiment. Their average age was 29.7 years (range:21–49). Considering the physical workload experienced by the participants and the impact of familiarity to the driving scenario, every trial was carried out with a gap of more than one week between each trial.

3 Driver Behavior Analysis

3.1 Gaze Movement Analysis

To ascertain whether the driver's visual focus was on the preceding vehicle or not, a bounding box was assigned to the preceding vehicle on the front display, followed by the computation of the distance to the viewpoint coordinates. If the calculated distance was smaller than the threshold value, the viewpoint was considered to be close to the preceding vehicle and the driver was considered to be gazing at the preceding vehicle. Although there are various definitions for the size of the visible range, an effective field of view (within approx. 15 deg to the left and right, 8 deg above and 12 deg below) [4] was defined as the range within which the behavior of the preceding vehicle was visible in this experiment, where information could be gazed at using only eye movements and information was instantly visible. If this range is applied to the simulator environment, it corresponds to a range of approximately 11 m to the left and right of the preceding vehicle, and a size of 300 pixels of the front display. Thus, the threshold of the distance between the preceding vehicle and the gaze point was defined as 300 pixels, which defined the state of gazing at the preceding vehicle in this experiment. An example of the range of the preceding vehicle's viewable area on the front display is shown in Fig. 2. The preceding vehicle's bounding box is a yellow frame, and if the gaze coordinates are within a blue range of 300 pix from its boundary, it was judged that the driver is in the state of gazing at the preceding vehicle.

Fig. 2.

Example of possible surroundings for preceding vehicle on front display.

As the set Time Headway(T_hw) of ACC was 1.8 s, each forward gaze duration (eye point is within the blue area) and looking away duration (eye point is out of the blue area) were calculated for the subtask section where the average T_hw was within 1.8 ± 0.3 s, respectively. When using ACC, the average forward gaze duration is a little shorter than when not using ACC as shown in Fig. 3-a. Moreover, when using ACC, the average looking away duration is quite longer than when not using ACC and there is a significant difference(*p < 0.05) as shown in Fig. 3-b. These results indicate that drivers who are using ACC tend to overtrust the system and fall into an inattentive state when doing distracted driving.

Fig. 3.

Average duration time of each subject compared between ACC ON and ACC OFF.

3.2 Collision Risk

The collision risk in the scene of the preceding vehicle’s sudden braking was quantified using iTTC [s⁻¹], which is defined as the inverse of the Time-To-Collision (TTC). This iTTC indicates a higher collision risk for larger values and a lower collision risk for smaller values. Figure 4 shows an example of a collision risk reduction result when using ADAS. In the case of ADAS OFF, this driver did not step on the brake pedal in time for a collision, while the risk of collision increased. On the other hand, in the case of ADAS ON, the increased risk of collision was reduced, thus avoiding a collision.

In summary, in the experiment, 4 out of 7 participants had a smaller maximum iTTC in ADAS ON than in ADAS OFF, reducing the collision risk as shown in this example.

Fig. 4.

Time history of inverted Time-To-Collision iTTC from the time instant that the preceding vehicle started sudden braking.

3.3 Brake Pedal Operation Behavior

To find out why the collision risk became lower when using ADAS, the driver brake pedal operation behavior was modeled by considering the driver reaction delay from the sudden braking initiation of the preceding vehicle as the dead time, and the driver brake pedal stroke as the first-order delay system. Figure 5 shows the block diagram of brake stepping behavior. In the case of ADAS ON, ACC deceleration control was activated, and the driver can easily perceive forward risk through the posture change by deceleration and the alert. Three parameters, Reaction delay τ [s], Time constant T_b [s], and Gain H_b [-], after the sudden braking of the preceding vehicle were identified for each participant and compared depending on with or without ADAS.

Fig. 5.

Block diagram of driver-vehicle braking behavior model.

Figure 6 shows the relationship between the amount of brake pedal stroke and each parameter. The vertical axis shows the amount of driver’s brake pedal stroke, which was calculated by defining the time from the start of emergency braking by the preceding vehicle until the driver started to brake as the reaction dead time τ, the maximum amount of brake operation by the driver as the gain H_b, and the time from the start of driver braking until the value of 63.2% (= 1–e⁻¹) of gain H_b was reached as the time constant T_b was defined and calculated.

Fig. 6.

Relationships of braking behavior parameters

The result was that all 4 participants, who reduced the collision risk in ADAS ON compared to ADAS OFF, had a shorter reaction delay when using ADAS. Furthermore, these 4 participants braked after ACC deceleration had started but before the collision alert sounded. Thus, when the preceding vehicle conducted sudden braking, it was confirmed that the use of ADAS with ACC deceleration enabled the driver to avoid the risk of collision by bringing the driver back to primary driving task more quickly than without ADAS, even though ADAS alone could not avoid the risk.

4 Conclusion

This paper conducted quantitative analyses of the change in driver’s eye movement and driver’s behavior in the high collision risk situation, depending on whether ADAS was used or not. The following two findings were identified.

• A) When using ADAS, drivers tend to overconfidently trust the system and lose focus on the risk ahead when they are distracted while driving.

• B) The use of ADAS with active deceleration intervention enabled drivers, who tend to be more distracted, to bring the drivers back to primary driving task more quickly and improved their performance in avoiding critical accidents.

The subsequent stage involves the clarification of how drivers perceive ACC deceleration and its impact on collision avoidance actions. The findings from experimental study will contribute to the design of next-generation ADAS to enhance safety for unsafe drivers who tend to drive while distracted.