Keywords

1 Introduction

Advanced driver assistance systems (ADAS) such as adaptive cruise control (ACC) and lane keeping assist (LKA) have recently been developed. Such systems help older drivers in particular to drive their car. To use the ADAS or other in-car devices (e.g., car navigation system, air conditioning), the drivers have to operate buttons and levers in their car to enable them and to change settings while controlling their car. However, older drivers have difficulties engaging in such dual task. In this study, we investigate how cognitive aging affects the ability to enable or set the ADAS while driving a car using cognitive modeling.

1.1 Driving and Aging

Many previous studies have investigated the effect of aging on driving performance [1,2,3]. The researchers in most studies measured older drivers’ abilities to control the car, such as the response time while braking and behavior at an intersection. However, a dual task driving situation requires other cognitive abilities beyond those required when just driving a car.

For older drivers, the dual task situation is more difficult than for younger drivers because the cognitive abilities required for performing the dual task are affected by aging [4]. Should they encounter a dangerous situation once, it is possible that older drivers will give up on using the useful support systems. It is, therefore, important to specify what type of cognitive aging would result in what type of danger. This research focuses on the impact of cognitive aging in time perception.

1.2 Dual Task and Time Perception

When people conduct a dual task, they process each task alternately by switching their goal. This means that they process only one goal-focused task at a time.

The situation investigated in this study involves the primary task of driving the car and the secondary task of changing the system state. Namely, the driving goal is the main goal and the enabling goal is the secondary goal. When drivers pursue the driving goal, they control the car; similarly, when they pursue the enabling goal, they operate the in-car device to change the state of the system.

Kujala and Salvucci (2015) investigated drivers’ cognitive processes in a similar situation [5]. The drivers switched their goal from the primary one (i.e., driving) to the secondary one (a task other than driving) when the main task was stable. They switched back to the main task at two points of time: (1) following the achievement of the sub-goal of the secondary task and (2) when a certain period of time had lapsed.

Previous studies demonstrated that older people have difficulties performing dual tasks [6, 7]. One possible cause of this impaired performance is a decline in time perception. The internal clock of older people tends to be slower than that of younger people [8, 9]. This decline in time perception affects when the older drivers switch back to the driving goal, especially their judgment whether a certain period of time is passed. Based on their internal clock, older people may not estimate how long they engage in the enabling goal correctly (i.e., they estimate shorter), and as a result, they may not to switch back to the driving goal with sufficient rapidity. In this study, we investigate the effects of declining time perception on the performance of dual tasks in the driving situation.

2 Task

Our target task in this study was to change the state of the ACC from an off to a “set” state. The ACC is one of the ADASs that automatically adjusts the car’s speed and helps to maintain a certain distance from the vehicle ahead.

The virtual car driven by our model was based upon the Levorg by Subaru. Figure 1 shows the devices in the Levorg related to the ACC. The state of the ACC is changed by using three of the six buttons on the steering wheel, two of which were used for the task in this study. The ACC monitor in the instrumental panel displays the current state of the ACC.

Fig. 1.
figure 1

Buttons for controlling the ACC and the ACC monitor.

Full size image

To change the state of the ACC from off to “set”, the driver must push two buttons in order. First, they push a “cruise” button, which takes the ACC into the standby state. Then, by pushing a “set” button, the ACC switches to the set state and starts adjusting the speed of the car.

3 User Model

3.1 Cognitive Modeling

A cognitive modeling approach was used to investigate the effect of aging on task performance. A cognitive modeling approach uses a computer model of human cognitive processes, which can perform a task by simulating human action. We constructed a user’s cognitive model that could enable the ACC and drive the car. Then, the value of the parameters related to the time perception were modified, and the changes in performance observed. The purpose of this study was not to evaluate the effectiveness of the model or parameter settings, but to observe what happened when the computer model aged virtually.

3.2 Adaptive Control of Thought–Rational

We used the Adaptive Control of Thought–Rational 7.0 (ACT–R) to construct the user model. ACT–R is a cognitive architecture consisting of multiple modules as shown in Fig. 2, which can replicate human cognitive processes, including both internal and external processes [10].

Fig. 2.
figure 2

Structure of ACT–R.

Full size image

The ACT–R model receives perceptual stimuli from the environment through perceptual modules and changes the environment using motor modules. We added steering and pedal modules to control the car, and a device-operation module to push the buttons in the car.

The ACT–R has two memory modules: declarative and procedural modules. The procedural module includes the production rules (i.e., “IF \(\cdots \), THEN \(\cdots \)”). The declarative module stores declarative knowledge. The ACT–R selects the subsequent action based on the current goal, knowledge retrieved from the declarative module, and information from the environment. A selected production rule is activated, and the ACT–R model adds the operations to the environment.

3.3 Declarative Knowledge

We provided the user model with two types of declarative knowledge needed to achieve the task. One was the knowledge of instructions shown in Table 1, which were used as the sub-goals of the enabling goal. Each knowledge of the instruction entailed an operation to be conducted and the object to be used. In addition, the knowledge of the function of the buttons was used to identify whether a button was the target one.

Table 1. Sub-goals for enabling ACC.
Full size table

3.4 Model’s Behavior

We constructed our user model based on an ACT–R model by Kujala and Salvucci (2015) [5]. Their model simulated a process of searching for a target music title from an in-car display while driving a car. Their model switched goals from the primary driving goal to the search goal when the driving was stable. The stability of driving was evaluated by assessing the vehicle’s lateral position in the lane and its lateral velocity [11]. It switched back to the primary driving goal when the sub-goals of the search goal (i.e., finding the target or finishing searching a current page) were achieved, or when the model was aware of the passage of time.

An overview of our user model is shown in Fig. 3, which was designed to alternately activate a driving goal and an enabling goal. We prepared four sub-goals for the enabling goal as shown in Table 1. Our user model switched between the two goals in a similar manner to that described in the study by Kujala and Salvucci (2015) [5].

Fig. 3.
figure 3

Schematic overview of the user model.

Full size image

The user model enabled the ACC through the following procedures. First, it retrieved an instruction and set it as the current goal. When the sub-goal was the retrieved instruction which value of the operation was “push,” the model moved its attention to one of the buttons on the steering wheel. The function of the attended button was retrieved. If the function was the target one (i.e., a value of the object), the model pushed the button; if not, the model moved its attention to the next button. When the sub-goal was the retrieved instruction which value of the operation was “check,” the model moved its attention to the ACC monitor and read it, to check whether the state of the ACC was the state required to meet its goal.

3.5 Manipulation of Time Perception

The ACT–R model was able to perceive the duration of time by counting internal ticks [12]. Based on the human cognitive mechanism, as the ticks counted increased, the tick interval became longer. An interval for the nth tick (\(t_n\)) was calculated using the following equations:

$$\begin{aligned} t_0&= start + \varepsilon _1\\ t_n&= a*t_{n-1} + \varepsilon _2 \end{aligned}$$

where start was the initial tick interval, and \(\varepsilon _1\) and \(\varepsilon _2\) were noise; parameter a decided how slow the ticks became as time passed.

To simulate the decline in time perception of an aged person, the value of this parameter a (hereinafter, referred to as the “interval parameter”) was increased. As the value of the interval parameter increased, the estimated time for a certain period of time became shorter (i.e., decline in time perception). The default value of the interval parameter was 1.10, which increased by 0.08 till 1.26; namely, 1.10, 1.18, and 1.26 were used as the interval parameter values. The interval parameter of 1.10 was the default model not affected by cognitive aging. The other two longer interval parameters simulated the progressive effects of aging on time perception.

We decided the time limit for one operation of the in-car device based on the model by Kujala and Salvucci (2015) [5]. When the model counted a certain ticks, it switched back to the driving goal. The initial number of ticks was 17 ticks. When the model returned to the driving task under stable driving conditions, the model increased the time limit by one tick; if not, the model reset the limit to its initial value. The initial limit corresponded to 0.5 s, 1.0 s, and 2.5 s respectively, with each value of the interval parameter as 1.10, 1.18, and 1.26. For all other parameters, we used the default values of ACT–R.

4 Simulation

4.1 Procedure of One Trial

The driving environment of our model was a straight three-lane highway. The model drove its car in the middle lane without changing lanes. During the first 10 s, the model only controlled the vehicle to ensure its driving was stable. Each model began the dual task to enable the ACC after 10 s had passed. The model performed the dual task following the leading vehicle, driving at about 100 km/h. Each trial was terminated 10 s after the model changed the ACC state to “set”. The time duration for the model to conduct the dual task was analyzed. Additionally, to investigate vehicle control performance we added angles to the steering angle based on a synthetic sine wave.

4.2 Simulation Results

We ran each model 100 times. Figure 4 shows the time duration needed to enable the ACC as well as the lateral deviations. The value of the interval parameter had no effect on time duration.

Fig. 4.
figure 4

Mean duration time and lateral deviations.

Full size image
Fig. 5.
figure 5

Mean duration time and lateral deviations in the modified environment.

Full size image

The lateral deviation was an index of driving performance while setting the ACC state, which assessed the deviation from the mean lateral position. The lateral deviation was found to increase as the parameter value increased. When the interval parameter was increased from 1.18 to 1.26 (0.107) the lateral deviation changed larger than that when it increased from 1.10 to 1.18 (0.157).

4.3 Modification of Environment

As we described above, three of the buttons had no relation to the ACC, one of which activated a lane keep assist system, with the other two used to change driving modes. We ran our user model in an environment where those three buttons were removed. Figure 5 shows the duration times and lateral deviations in the modified environment. Both values were found to reduce dramatically, and the aging effect disappeared, meaning that in each case the sub-goal was achieved before the model focused too long on the enabling goal.

5 Discussion

The model developed investigated how a decline in the ability to perceive time affects driving performance in a dual task scenario. The simulation results showed that while cognitive aging affected the control of the car position in the lane, it had no effect on the time taken to change the state of ACC from off to “set”.

This relationship between time perception and lateral deviation was explained based on the goal switching behavior. The user model had to switch back to the driving goal frequently, or their car strayed over the line and crashed into other cars. When the model attended to the in-car device, they could not acquire information from the driving environment.

Older people, as well as the older models in this study, have difficulty estimating how much time has passed due to the decline in time perception that results from cognitive aging [4, 8]. Therefore, they may continue the operation of the in-car device and neglect to control the car due to a lack of awareness of how much time has passed. The mean time duration from switching to the enable goal to switching back to the driving goal was 0.503, 0.737, and 0.941 s, for each value of the interval parameter, 1.10, 1.18, and 1.26.

Driving is not the only situation where the effects of time perception can be observed. Older people neglect their main tasks for a longer period of time in other dual task situations.

5.1 Time Perception and Time for Enabling

A decline in time perception did not affect the duration of time needed to complete enabling the ACC. This is because of the trade-off relationship between enabling the ACC and controlling the car. The model’s goal switching had the features shown in Fig. 6. This diagram was constructed based on the models’ process logs.

Fig. 6.
figure 6

Conceptual goal switching diagram. The white bars show the duration while the model engaged in the driving goal, and the gray bars show that while it engaged in the enabling goal. The total gray bar areas are identical in the two models.

Full size image

The amount of time needed to achieve all sub-goals (the sum of the gray bars) did not differ since the interval parameter did not affect the ability to operate the ACC (e.g., find the button and check the ACC state). As each engagement by the default value model for the enabling goal was relatively short, each adjustment of the car’s position (i.e., driving goal) was completed in a short time. In contrast, the larger value (older) model engaged in the enabling goal over a longer time interval, during which the car deviated from the center of the lane; consequently, each adjustment (i.e., driving goal) required a longer period of time. For these reasons, only the lateral deviation was affected by the interval parameter setting.

5.2 Advantages of Cognitive Modeling

As demonstrated in this study, cognitive modeling can be a useful method to investigate the human cognitive process during complex activities. It is impossible to explore the effects of only one factor, such as time perception, based upon naturalistic observations of complex human activities such as driving. However, by using cognitive modeling, we can investigate how a certain factor affects a target activity. Note that there are many other factors to be considered, time perception is not the only one that affects driving performance.

Cognitive modeling also can verify the functionality of new interfaces based on the user’s cognitive processes. In this study, we tested the situation with fewer buttons, where the lateral deviation was small and constant, despite the decline in time perception. However, it is difficult to merely reduce the number of buttons, because recent vehicles have numerous functions that the driver must control. Our study suggests that further consideration of the positioning and grouping of buttons may help older people to use the driving support functions more effectively.

6 Conclusion

The simulation results suggested that the older people tended to concentrate for too long on the non-driving task; as a result, the lateral deviation increased. An additional simulation demonstrated that by removing the unrelated buttons, the effect of aging was eliminated. Car designers need to develop a user interface that makes it easier to switch back to the primary driving goal.