Research Report
Changing Lanes with Conditional Driving Automation: The Case of Overtaking Slower Traffic on Highways
2024 Volume 2024 Issue 2 Article ID: JRJ20240202
Details
2024 Volume 2024 Issue 2 Article ID: JRJ20240202
本研究の目的は，条件付き自動運転車のドライバーに高速道路上の渋滞をどのように知らせるかという疑問に答えることである．我々は，被験者内で4つのシステムデザインを比較するために，40人の参加者を用いてドライビングシミュレータ実験を行った．ベースライン条件として，システムはドライバーに知らせずに渋滞の後方をゆっくり走り続ける．2番目のシステムは，渋滞を手動で追い越した方が良いことをドライバーに知らせるように設計された．第3のシステムは，ボタンを押すことによって，システムが自動的に渋滞を追い越すことをドライバーに要求するものである．一方，第4のシステムは，6秒後に自律的な追い越しが開始されることをドライバーに知らせ，ドライバーはボタンを押すことでそのプロセスを取り消すことができる．その結果，ドライバーは3番目と4番目のデザインよりも1番目と2番目のデザインを好んだ．自動追い越しに関しては，ドライバーは第3のシステムを第4のシステムよりも信頼していた．セーフティ・クリティカルな条件がなかったため，追い越しステアリングの挙動は，人間のドライバーとシステムとの間の大きな違いを反映していない．これらの観察結果は，人間中心の自動化というコンセプトに従って設計された自動運転システムは，ユーザーにより受け入れられることを示している．
Abstract
This study investigates the impact of different designs of Automated Driving Systems (ADSs) on driver behavior, trust, and acceptance when approaching a traffic jam in conditionally automated vehicles. A driving simulation experiment was conducted to compare four ADS designs. While ADS-1 (baseline) can only decelerate to continue automated driving behind the traffic jam, ADS-2 can also guide the driver to avoid slow traffic. ADS-3 can overtake the traffic jam automatically if the driver approves the maneuver by pushing a “decision button”. In contrast, ADS-4 can autonomously overtake the traffic jam without driver approval unless the driver aborts the maneuver by pushing a “decision button” within 6 seconds. In all designs, driver takeover of the vehicle was optional. Results indicate that drivers preferred ADS-1 and ADS-2 more than ADS-3 and ADS-4 because they did not have to make a critical lane-change decision. However, the drivers trusted and accepted ADS-3 more than ADS-4, because they were able to confirm safety on the adjacent lane before letting ADS-3 proceed with the overtaking maneuver. These observations indicate that automated driving systems can be more accepted by the user when designed following the concept of human-centered automation.
1. Introduction
Automated driving has been primarily introduced for reasons of safety and also for driver workload reduction, congestion relief, and power consumption 1, 2). For example, these systems relieve a driver from tedious control tasks by employing lane-keeping assistance and adaptive cruise control to regulate lateral and longitudinal vehicle motions 3-5) . In addition, these contribute to road safety by employing active safety systems such as autonomous emergency braking and steering systems to prevent drivers from colliding with vehicles, pedestrians, and other obstacles 6-8) .
Although Automated Driving Systems (ADSs) have the potential to advance vehicle mobility and save lives, they still have limitations and imperfections and may fail when they are most needed 9). On the one hand, inappropriate human trust in these systems can affect drivers’ behavior toward the system, affecting their capability to detect inadequate system performances 10). On the other hand, the excessive use of ADS may yield unintended negative consequences, such as skills and morale degradation and loss of situation awareness 11-13). It may also cause automation-induced complacency when human operators become overly satisfied with their capabilities and thereby create preventable hazards 14). In addition, drivers assisted by an ADS are more likely to be engaged in non-driving related activities, which would impair their capability to respond to critical situations 15). Furthermore, drivers are likely to become overly dependent on ADSs, which may not continuously function as intended 16).
The introduction of partial driving automation systems in certain commercial vehicles enables drivers to occasionally experience hands- and feet-free automated driving in limited operational design domains such as highways 17). However, the drivers must monitor the traffic environment and supervise the system status 18). One challenge is the maintenance of drivers’ engagement and vigilance, both of which can be performed effectively only when an adequate level of driver workload exists 4, 19, 20). Another challenge is to train the driver to supervise the system because the supervision of automated vehicles is a more demanding task as it requires a higher level of human skills when compared to that of manual driving 2, 21).
However, conditional driving automation systems that can control a vehicle for an extended time without driver engagement and intervention require less supervision than partial driving systems 22). Therefore, drivers of conditional driving automation may be out of the control and monitoring loop. As a result, it is more difficult to gain control over the vehicle when required than it is with partial driving automation 23). For example, while a driver is reading a book or using a smartphone, the system may encounter a situation that it is not designed to manage. Hence, it would request the driver to resume manual control within a short time (e.g., within 10 s) 24). In critical situations, a major challenge is to appropriately transform the driver from an out-of-the-loop state to an in-the-loop state to resume driving their vehicle safely 25).
To address the abovementioned challenges, we propose a cooperative human-machine interface (HMI) design to support drivers’ decision-making and control for the safe and practical use of partial and conditional driving automation. Cooperative HMI can enable humans to share decision-making and action implementations via automation to simultaneously achieve a function or task 26). The sharing of decision-making and control between the user and the system may improve the human’s comprehension of the automated process and thereby facilitate them in developing an appropriate level of trust toward the system. Following the concepts of human-centered systems design 27, 28), we designed an HMI, in which the human can be more involved in automated processes to improve the driver’s attention and comprehension of the traffic situation and to reduce the requirement for control transition during non-critical automated driving conditions. Specifically, we focused on partial and conditional driving automation, in which the interaction between the driver and system is a primary concern, and the driver must be able to regain full or partial control of the vehicle when necessary or when requested by the system.
A human-centered system may support human perception and decision-making but may not implement an automatic action without human directives. According to the ten levels of automation (LOA) proposed by 26) , a human-center system design can be implemented under LOA-6 or less, as shown in Fig. 1. In this study, we combined the concept of human-center automation and LOA to design a cooperative HMI for partial and automated driving systems. We proposed an ADS that issues a request to engage (RTE) in response to non-critical changes in the driving environment. The proposed RTE may suggest an appropriate action to be carried out by the driver (LOA-4), execute that suggestion if the driver approves (LOA-5), or allow the driver a restricted time to veto before automatic execution (LOA-6). It differs from a request to intervene (RTI), which is typically issued during hazardous conditions, such as the attainment of an operational or functional limitation by the system or imminent collision situations 29) . An RTE does not necessarily imply that the human driver must resume manual control. It is a communication between the system and the driver to maintain an appropriate level of driver engagement and responsibility by monitoring the driving environment and approving or disapproving the system’s decisions. For example, on the one hand, the safety of the driver and other road users can be higher when an ADS detects and performs tactical tasks (e.g., lane changing and passing) under human directives compared to when the maneuver is performed by the system without human monitoring. On the other hand, driver resumption of vehicle control can also be improved when the system guides the driver toward a safer next course of action. These can address issues related to human disengagement, overreliance, and out-of-the-loop performance associated with the use of partial and conditional ADSs.
Fig. 1 Human perceptual cycle and levels of automation. LOA-6 represents the borderline between human-center system design in which the human has the final authority over the system and automation-center system design in which the automation has the final authority over the system.
The proposed designs of conditional driving automation with cooperative HMI were examined using a driving simulator experiment in which the drivers were exposed to smooth and congested traffic during automated driving on a two-lane highway. Four designs were compared. When the ADS detects traffic congestion on its main lane while the adjacent lane is available: ADS-1 continues automated driving behind the traffic jam without providing an RTE; ADS-2 issues an RTE to encourage the driver to resume manual control and overtake the congestion; ADS-3 issues an RTE to request the driver's permission to overtake the congestion automatically; ADS-4 issues and RTE to overtake the congestion automatically unless the driver disapproves it within a restricted time. The hypothesis is that driver engagement and attention depend on the extent to which the driver is required to interact with the system during automated driving. Furthermore, it is hypothesized that drivers’ acceptance of an H may decrease when they are required to be more involved in the automated process. It is likely that an increase in driver involvement in the automated process may enhance driver comprehension and driving performance while reducing the driver’s acceptance of and trust in the system. The study outcomes can provide a basis for accelerating the implementation of automated driving in the real world.
2. Method
2. 1 Participants and Apparatus
This study conducted a driving simulation experiment with forty human participants (20 males and 20 females) aged between 22 and 69 years old. This experiment was designed and conducted according to the Code of Ethics and Conduct published by the Japanese Psychological Association ( https://psych.or.jp/) and under the approval of the Ethical Committee of the Faculty of Engineering, Information, and Systems of the University of Tsukuba, Japan.
The experiment was carried out on a motion-based driving simulator, as shown in Fig. 2. The simulator comprised a cockpit with an adjustable driver chair, a steering wheel, brake and gas pedals, and an automatic transmission system. The driver’s front field-of-view was projected onto a 120° curved screen (85 × 30 in). Two small screens (5 × 4 in each) were provided to display the side fields-of-view, and a central screen (5 × 2.3 in) was provided to display the rear field-of-view. The simulator was connected to an external computer, in which the experimental scenarios and ADS were designed. The drivers can activate and deactivate the system by shifting the gear stick between D for manual driving and D3 for automated driving modes (see D/D3 ADS activation in Fig. 2). The automated driving mode can also be deactivated by the direct intervention of drivers using the steering wheel or gas/brake pedals.
Fig. 2 Driving simulator at the Laboratory for Cognitive Systems Science, University of Tsukuba. The top-left picture illustrates the driver’s scene (side and rearview mirrors, HMI display, and driver monitoring). The top right picture depicts the simulator's interior components (steering wheel, automatic transmission, and dashboard). The bottom left picture visualizes the main driving lane of the automated vehicle (AV) and the passing lane. The bottom middle picture represents the driving simulator cockpit. The bottom right picture shows the participant sitting hands- and feet-free inside the cockpit during automated driving.
2. 2 Experimental Scenarios
The training and testing experimental scenarios were implemented on a straight two-lane highway with a light and smooth traffic in the passing lane. Only daylight, a dry road, and clear weather conditions were simulated. Each testing scenario was divided into four sequential scenes, as illustrated in Fig. 3. In Scene-1 (~1 min), the participants were required to start driving the vehicle manually and attain the speed limit of 60 km/h. Scene-2 (~5 min) began when the participants activated the ADS and practiced conditional automated driving at a speed of 60 km/h with a smooth traffic flow on both lanes. During Scene-3 (~3 min), the vehicle approached slow traffic (20 km/h) while under automated driving such that the system decelerated the vehicle to synchronize with the congested traffic speed. However, the right-hand lane was available with cars passing at 60 km/h and 10 s time headway between them. In this part of the driving course, the drivers could freely decide to keep or change lanes manually or automatically based on their perception of the situation and ADS design and capabilities. Finally, in Scene-4 (~1 min), the participants were asked to resume control and stop the vehicle.
Fig. 3 Test scenario. Scene-1: driver must drive the vehicle manually and attain 60 km/h speed. Scene-2: driver switches to automated driving mode. Scene-3: ADS decreases vehicle speed from 60 to 20 km/h in response to slow traffic ahead (traffic congestion on the AV's main lane), whereas traffic flows smoothly on the adjacent lane. Based on the ADS design and capabilities, the drivers can decide whether to resume manual control and change/keep lanes or let the system decide the next course of the vehicle (keep or change lanes). Scene-4: drivers have to resume manual control and stop the vehicle.
2. 3 HMI Design
For the purpose of this experiment and based on the experimental scenarios, four main system states were used, as shown in Table 1. While HMI-1 was displayed when the automated driving mode was deactivated and the driver managed the entire driving task manually, HMI-2 was displayed when the automated driving mode was activated and the system performed the entire driving task during smooth traffic conditions. However, HMI-3 was displayed when the automated driving mode was turned on during traffic congestion to draw the driver's attention toward changes in the surrounding traffic. HMI-4 represents the RTE issued by the system to the driver during abnormal but non-critical traffic conditions (e.g., traffic congestion) while the automated driving mode was turned on in the presence of a better driving course. An LCD screen (7 × 4 in) installed in the dashboard to the left of the driver was used as a visual HMI to display the system status. An acoustic alert accompanied the variations in visual HMI messages to inform the drivers regarding the changes in the HMI and system states.
Table 1 HMI display of automated driving system states.
2. 4 ADS Design
The proposed ADS is a low-speed automated lane-keeping assistance system (i.e., a highway traffic jam pilot) that masters lateral and longitudinal vehicle motions for an extended period without driver intervention 17, 30, 31) . Based on LOAs and the concept of human-center systems design, four algorithms are proposed herein to manage Scene-3. These algorithms involve a method to inform the driver regarding traffic congestion and necessary lane-change maneuvers, as detailed in Table 2. Although all the systems are equivalent to conditional driving automation systems, these differ in terms of the capability to detect and perform lane-change maneuvers, as follows:
Table 2. Algorithm of ADS and driver behaviors during Scene-3.
| ADS | Level of automation | ADS Algorithm |
|---|---|---|
| 1 | LOA-3 |
if slower traffic is detected: synchronize speed, and display HMI-3 if the driver resumes manual control: deactivate the ADS, and display HMI-1 else if the driver does not respond: maintain the speed according to traffic congestion, and continue automated driving |
| 2 | LOA-4 |
if slower traffic is detected: synchronize speed, and display HMI-3 for 4 s if the driver resumes manual control within 4 s: deactivate the ADS, and display HMI-1 else if the driver does not respond: display HMI-4 (RTE) for 6 s if the driver resumes manual control within 6 s: deactivate the ADS, and display HMI-1 else if the driver does not respond within 6 s: maintain the speed according to traffic congestion, continue automated driving, and display HMI-3 |
| 3 | LOA-5 |
if slower traffic is detected: synchronize speed, and display HMI-3 for 4 s if the driver resumes manual control within 4 s: deactivate the ADS and display HMI-1 else if the driver does not respond within 4 s: display HMI-4 (RTE) for 6 s if the driver presses the “approve” button within 6 s: change lanes automatically, accelerate, and display HMI-2 else if the driver resumes manual control within 6 s: deactivate the ADS, and display HMI-1 else if the driver does not respond within 6 s: maintain the speed according to traffic congestion, continue automated driving, and display HMI-3 |
| 4 | LOA-6 |
if slower traffic is detected: synchronize speed, and display HMI-3 for 4 s if the driver resumes manual control within 4 s: deactivate the ADS, and display HMI-1 else if the driver does not respond within 4 s: display HMI-4 (RTE) for 6 s if the driver presses the “disapprove” button within 6 s: maintain the speed according to traffic congestion, continue automated driving, and display HMI-3 else if the driver resumes manual control within 6 s: deactivate the ADS, and display HMI-1 else if the driver does not respond within 6 s: change lanes automatically, accelerate, and display HMI-2 |
1) ADS-1: The RTE is implemented under LOA-3, in which the automation supports human perception and situation recognition. The capability of ADS-1 in overtaking traffic congestion is limited. Therefore, the system continues automated driving at a slow speed on the same lane. The system displays HMI-3 to attract the driver’s attention toward the traffic congestion. Based on their observation and perception of the situation, the drivers must decide whether to override the system and change lanes manually to circumvent the traffic congestion or continue with automated driving on the slower lane.
2) ADS-2: The RTE is implemented under LOA-4, in which the automation supports human decision-making. First, ADS-2 informs the driver regarding the situation (i.e., HMI-3). Subsequently, the system issues an RTE (HMI-4) that is designed under LOA-4, in which the automation recommends an alternative to support human decision-making. The RTE advises the driver to resume control and manually shift toward a faster lane. The drivers can disregard the RTE or resume manual driving to bypass the traffic congestion. If the driver disregards the RTE for more than 6 s, the system status is returned to HMI-3, and automated driving is continued on the slow lane.
3) ADS-3: The RTE is implemented under LOA-5, in which the automation supports human decision-making and action under the human directive. After informing the driver regarding the traffic congestion (HMI-3), ADS-3 issues an RTE (HMI-4) that is designed under LOA-5, in which the automation recommends an alternative option and executes it when the driver approves it. The RTE recommends a lane-change maneuver to circumvent congested traffic and requests the driver’s permission to perform the task automatically. If the driver approves the automatic lane-change maneuver by pushing the decision button shown in Fig. 2, the system automatically bypasses the traffic congestion and continues automated driving on the faster lane. If the driver disregards the RTE for more than 6 s, the system status is returned to HMI-3, and automated driving is continued on the slow lane.
4) ADS-4: The RTE is implemented under LOA-6, in which the automation gives the human a limited time to react or object before executing necessary actions automatically. When ADS-4 detects traffic congestion, the system issues HMI-3 to inform the driver regarding the congested traffic. It then issues an RTE (HMI-4) that is designed under LOA-6, in which automation provides the human a limited time to disapprove the automatic actions. The RTE informs the driver that an automatic lane-change maneuver would begin in 6 s. If the driver pushes the decision button (approve/disapprove button in Fig. 2), the system status is returned to HMI-3, and automated driving is continued on the slower lane. Otherwise, the system automatically bypasses the traffic congestion and continues automated driving on the faster lane.
2. 5 Tasks and Procedure
First, all the participants were briefed regarding the purpose of the study, experiment design, and ethical rights. Upon obtaining their consent for participation in the experiment, the participants were instructed to complete a short demographic survey. The critical instructions were to maintain safe driving performance and pay attention to the HMI display. Each participant was explained in detail using PowerPoint slides regarding the operation of the driving system. Then, they performed two familiarization and training drives (5 min each): 1) manual drive and 2) automated drive under light and smooth traffic conditions.
The testing phase comprised four drives (9–10 min each). In each test drive, the participant was presented with one of the four ADS designs such that each design was tested once during the testing phase. The participants were explained about the operation of each system design before they started driving with the system. The sequence in which the ADS designs were experienced was counterbalanced among the participants using the Latin-square method. The participants were allowed a short break (10 min) between test drives, including the brief for the subsequent trial. Finally, they completed post-experiment questionnaires regarding their trust, understanding, acceptance, and controllability with regard to each ADS design. For each participant, the entire experiment was completed within two hours in one day. The entire experiment was completed in two weeks period.
2. 6 Experimental Design and Data Analysis
This experiment followed a within-subject design such that each driver was presented with the four ADS designs as an independent variable. For all analyses, the statistical significance was set to α=0.05 and was determined by applying the Chi-square test, repeated measures univariate analyses of variance (ANOVA), post-hoc, and t-test using IBM SPSS Statistics. Sphericity violations, normality, and homogeneity of all dependent variables were checked to ensure the applicability of the selected test 32). All dependent variables were extracted from scene-3, as illustrated in Fig. 4.
To understand how drivers reacted and used the different ADS designs, we provided descriptive statistics of driver reactions (e.g., resumed manual control, responded to the RTE, pushed the decision button, and lane-keep or change) under each system. Further, we used the time headway (TH) between the subject and adjacent vehicles as a safety indicator to assess the possibility of drivers' involvement in dangerous lane-change maneuvers. TH was calculated as the difference between the time (in seconds) when the front of the subject vehicle arrives at a point on the highway and the time the front of the adjacent vehicle arrives at the same point, as indicated in Fig. 4.
Steering wheel and pedal indicators were used to determine the lane change response time and maximum steering wheel angle as dependent variables to evaluate the lateral and longitudinal driving behaviors during the automated and manual driving modes. The lane-change response time was calculated as the time in which the HMI status changed from HMI-2 to HMI-3. That is, it is the time between the instant when the slow traffic ahead is detected and the instant when the vehicle starts steering away toward the adjacent lane, as shown in Fig. 4. It was measured to evaluate the driver’s perception of the situation with or without an RTE (i.e., HMI-4) and to assess the effect of cooperative ADS design on driving behavior and preference during optional lane-change situations. The maximum steering wheel angle was determined as the maximum steering wheel input by the driver or the system during lane-changing maneuvers between the point where the vehicle steers away to circumvent traffic congestion and the point where the vehicle is driving forward in the passing lane, as shown in Fig. 4. The aim was to compare the vehicle lateral control performance between the case wherein the driver resumed manual driving and changed lanes and that wherein the ADS changed lanes autonomously.
For all the systems, the drivers’ engagement was measured based on their attention and response to the HMI and surrounding traffic changes 13) and the manner in which the control was transferred from the system to the driver. The driver off-road glance was determined to evaluate the drivers’ attention and engagement level as well as RTE effectiveness. The significant effect of ADS design was investigated using one-way repeated measures, and the results were further compared between systems using Tukey's HSD. The acceleration pedal and throttle control percentages were evaluated and compared using the t-test to investigate the longitudinal driving behavior and energy-saving efficiency.
Fig. 4 Scene-3: the driver had to decide the next driving course based on each system design. Dependent variables were used to express time intervals for which the time headway, lane-change response time, and maximum steering angle.
3. Results
3. 1 Driver Reaction, Lane-change Safety, and System Usage
Table 3 presents descriptive statistics of driver reaction to changes in traffic conditions and HMI during automated driving and the mean and standard deviation of the total system activation time for each ADS design. Although ADS-1 did not request driver intervention when approaching traffic jam, 36 participants (90%) decided to take over the vehicle control, and four participants (10%) did not perform any action and let the system drives their vehicle slowly behind the traffic jam. Out of the 36 participants who resumed manual driving, 34 participants changed lanes to bypass the traffic jam, while two participants continued slow driving in the same lane. In ADS-2, 37 participants responded to the system's recommendation to take over the control and change lanes, while only three participants did not perform any action. For ADS-3 driving condition, 36 participants (90%) pushed the decision button to let the system changes lanes and bypasses traffic jam automatically. However, two participants decided to take over the vehicle and change lanes manually, and two participants did not respond to the system request and continued the slow automated driving. For ADS-4 driving condition, 15 participants (37.5%) pushed the decision button to disapprove the automatic lane change by the system.
In contrast, 25 participants (62.5%) did not push the button and let the system proceed with the lane change to bypass the traffic jam. Under this condition, of the 15 participants who disapproved the automatic lane change, three participants decided to continue with the slow automated driving behind the traffic jam, and 12 participants overtook the vehicle control right after pushing the button. Out of those 12 participants, nine drivers changed lanes automatically, and three drivers continued manual driving behind the traffic jam.
When the automatic lane change is not available, a Chi-square comparison of the number of manual lane changes between ADS-1 and ADS-2 conditions indicates no significant effect of using the RTE under the ADS-2 condition (χ2 (1) = 10.1, p = 0.61). However, when the automatic lane change is available, a comparison of the numbers of manual and automatic lane changes under ADS-3 (χ2 (1) = 41.1, p < 0.001) and ADS-4 (χ2 (1) = 29.4, p < 0.01) conditions highlights the effectiveness of using RTE. While all 34 participants who resumed the manual control and changed lanes under ADS-1 pushed the decision button to request the automatic lane change under ADS-2, only 25 of them let ADS-4 to perform the automatic lane change. It is noteworthy that the participants who did not push the decision button under ADS-4 have pushed the decision button under ADS-3, allowing both systems to proceed with the automatic lane change. It is also interesting to know that the six lane-keep cases under ADS-1 and ADS-4 were committed by the same participants, while the three and two lane-keep cases under ADS-2 and ADS-3, respectively, were committed by different participants.
Table 3 Comparisons of how drivers responded to HMI and traffic changes, the accumulative time of automated driving, and the percentage of ADS usage.
| ADS | Driver reaction in scene-3 | Time headway (s) | Automated driving time (s) | ADS usage | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| resume manual driving | Push decision button | Lane change (N) | Lane keep | M | SD | M | SD | |||
| Manual | Automatic | |||||||||
| 1 | 36/40 | NA | 34/40 | NA | 6/40 | 1.12 | 0.56 | 308.21 | 38.15 | 77% |
| 2 | 37/40 | NA | 37/40 | NA | 3/40 | 0.83 | 0.80 | 278.12 | 20.74 | 67.5% |
| 3 | 2/40 | 36/40 | 2/40 | 36/40 | 2/40 | 0.47 | 0.63 | 380.60 | 35.91 | 91% |
| 4 | 12/40 | 15/40 | 9/40 | 25/40 | 6/40 | 0.55 | 0.39 | 353.90 | 37.89 | 81.5% |
| N: number; M: mean; SD: standard deviation | ||||||||||
The lane-change maneuver safety was assessed using the time headway (TH) indicator, as shown in Table 3. Although the largest TH mean value was under ADS-1 condition (1.12 s), which is consistent with what was found in previous studies 33), it is significantly shorter than the recommended 3 s TH value 34). The TH values under other conditions were lower than 1 s, which can be critical and may result in a lane-change crash or near-crash given the variability in driver alertness, skills, and capabilities. However, there was no crash recorded in this experiment.
To compare the system usage among the ADS types, the automated driving time was calculated as the cumulative time of ADS operation, during which the system completely controlled the vehicle, as presented in Table 3. A one-way repeated measures ANOVA revealed a significant effect of the ADS type on the automated driving time (F(3, 156) = 72.83, p < 0.01). These results provide preliminary information on the implications of system and HMI designs for the manner in which end users would use the system. Post-hoc tests with Tukey’s HSD indicated significant differences in automated driving time between systems (p < 0.001). The highest mean level was recorded under ADS-3 (M = 380.60), wherein the system performed lane change maneuvers under human approval. The smallest mean level was recorded for ADS-2 (M = 278.12).
The ADS usage was calculated as the mean percentage of the automated driving time periods and the frequency of using the system functionalities. In general, a significant difference is observed between the various system settings. This difference indicates the effect of using RTE and the extent to which the driver is engaged in decision-making and action implementation during automated driving. Although the use of RTE increased the system usage compared with ADS-1, the different ADS capabilities resulted in more differences among ADS-2, ADS-3, and ADS-4. Furthermore, although both ADS-3 and ADS-4 can perform automatic lane change maneuvers, the difference in HMI design strategies between the two systems affected the drivers’ response to the RTE. Therefore, it can be concluded that even for systems with similar capabilities (e.g., ADS-3 and ADS-4), the manner in which a human manages different system settings based on the information conveyed to them can vary. This, in turn, can cause differences in automation usage and efficacy.
3. 2 Lane-change Response Time and
Fig. 5 compares the lane-change response time between the four system designs. One-way repeated measures ANOVA revealed the significant effect of the ADS type on the response time (F(3, 128) = 11.35, p < 0.001). Multiple comparisons using Tukey’s HSD indicated that the lane-change response time under ADS-1 and ADS-4 was significantly longer than that under ADS-2 and ADS-3 (p < 0.001). Although there was no significant difference between ADS-1 and ADS-4, the results of standard deviation highlight the wide variations among the drivers in terms of perception of lane-change situations under ADS-1 (SD = 8.362) compared with the system’s perception of the same situations under ADS-4 (SD = 1.170). The results showed that the drivers responded more actively and promptly under ADS-2 and ADS-3 compared with ADS-1. Given that the response times of ADS-2 and ADS-3 were comparable and differed from that of ADS-4, the RTE methods of ADS-2 and ADS-3 reduced the response time more compared with that of ADS-4.
Fig. 5 Lane change response time to avoid traffic disturbance during Scene-3. The mean and standard deviation (error bar) were calculated from data of 132 lane-change maneuvers (there are 34 and 37 manual lane changes for ADS-1 and ADS-2, respectively; and 36 and 25 automatic lane changes for ADS-3 and ADS-4, respectively).
3. 3 Maximum Steering Wheel Angle
Fig. 6 compares the mean and standard deviation of the maximum steering wheel angle among the four ADS designs. A one-way repeated measures ANOVA revealed significant effects of the ADS design (F(3, 128) = 9.73, p < 0.01). The largest and smallest mean level of steering-wheel angle were recorded under ADS-1 (M = 0.562) and ADS-3 (M = 0.393), respectively. Multiple comparisons with Tukey’s HSD indicated that the maximum steering angles under ADS-1 and ADS-2 were significantly larger than those under ADS-3 and ADS-4 (p < 0.01). However, there were no significant differences between ADS-1 and ADS-2 (p = 0.977) and between ADS-3 and ADS-4 (p = 0.890). Although these results highlight the differences between manual lane-change maneuvers (i.e., ADS-1 and ADS-2) and automatic lane-change maneuvers (i.e., ADS-3 and ADS-4), the steering behavior was relatively smooth and stable under all driving conditions. A reason for this was that the situations encountered by the drivers were not critical, and they could decide and act without haste. Another reason was that the drivers were not compelled to resume manual control.
Fig. 6 Maximum steering angle during lane-change maneuvers with each ADS design.
3. 4 Off-road Glance
The driver’s face behavior was recorded using a video camera and processed using eye-tracking software to extract the driver’s eye movement. The off-road glance was calculated as the proportion of time spent by the driver gazing away from the roadway in front as well as the side and rearview mirrors during the automated driving mode. Off-road glances exceeding 3 s were considered accumulatively during each drive 35, 36). The objective was to assess the driver’s attention to the road and surrounding traffic during the automated driving mode.
Fig. 7 presents the percentage of off-road glances for each ADS design. According to a one-way repeated measures ANOVA, the system design was observed to have a significant effect on the drivers’ attention and glancing behavior (F (3, 156) = 55.25, p < 0.001). Multiple comparisons with Tukey’s HSD revealed significant differences between systems (p < 0.001), except between ADS-3 and ADS-4. The off-road glance under ADS-2 was significantly larger than those under ADS-3 and ADS-4, although the RTE was available in the three systems. These differences can be attributed to the HMI designs and system capabilities in performing lane-change maneuvers, which also affected the driver reaction and the total time of automated driving, as presented in Table 3. Such explanation may also be supported by the comparable off-road glance results under ADS-3 and ADS-4, both of which can perform lane-change maneuvers automatically. Meanwhile, the significant difference in off-road glance values between ADS-1 and AD-2 indicates that the use of RTE is effective in improving the driver’s attention.
Fig. 7 Proportion of drivers’ off-road glances for each ADS design during the entire drive. Error bars denote standard deviations.
Further analysis of drivers’ eye-glance locations revealed that all of the participants who initiated manual lane changes checked the right mirror for at least 2 s before steering toward the adjacent lane. However, 11 participants pressed the decision button without checking the mirror under ADS-3, and six participants failed to check their right mirror prior to initiating automatic lane change by ADS-4. These results suggest that, although the drivers' off-road glances were decreased during ADS-3 and ADS-4, some drivers failed to scan the adjacent lane prior to changing lanes by the system. More specifically, some drivers were focusing on the front window and not on the adjacent lane when the system initiated the lane-change maneuver.
3. 5 Energy Consumption
The acceleration pedal position was determined as the percentage of the maximum acceleration pedal input by the driver while resuming manual control after the vehicle speed was reduced to 20 km/h during Scene-3. It was compared with the throttle control percentage value, which was calculated as the percentage of the maximum acceleration input by the system to overtake the traffic jam during Scene-3 automatically. The acceleration pedal position and throttle control percentage represent the energy consumption by the human driver and ADS, respectively, when they attempted to circumvent the congestion and recover the vehicle’s initial speed (from 20 to 60 km/h). One objective was to compare the manual and automated longitudinal vehicle motion control when traffic disturbances were encountered. Another objective was to compare manual and automated driving in terms of energy consumption and thereby assess the potential environmental impact of introducing automated driving technology.
Fig. 8 shows a comparison of the cumulative percentage of energy consumption between manual and automated driving. A total of 89 manual lane-change maneuvers were compared with 61 automatic lane-change maneuvers. A dependent samples t-test indicated that the acceleration pedal position was significantly larger than the throttle control percentage (t = 7.91, df = 148, p < 0.001). Automated driving resulted in better energy consumption compared with manual driving for an identical task under similar circumstances. These results indicate that the extension of ADS capabilities would not only increase system usage but also conserve energy.
Fig. 8 Comparison of energy consumption between manual lane-change maneuvers (N = 89) and automated lane-change maneuvers (N = 61) for all the systems. Error bars denote standard deviations. APP: Acceleration Pedal Position as pressed by the driver during manual control. TCP: Throttle Control Percentage as controlled by the system during the automated driving mode.
3. 6 Subjective Assessment
The subjective driving experiences of the participants under each ADS design were accumulated and analyzed, as shown in Fig. 9. The drivers were asked to answer five questions regarding their perceptions of understanding, trust, acceptance, controllability, and attention to a 10 cm line by assigning a score between 0 (not at all) and 10 (absolutely). After completing all driving trials and before administrating the questionnaires, we have provided a comprehensive explanation of the four systems to remind the participants one more time about the characteristics of each system. According to two-way repeated-measures ANOVA, the ADS design affected the drivers’ responses to all the questions except for that on drivers’ attention (understanding: F(3, 159) = 2.74, p < 0.05; trust: F(3, 159) = 6.18, p < 0.01; acceptance: F(3, 159) = 5.22, p < 0.01; controllability: F(3, 159) = 2.69, p < 0.05; attention: F(3, 159) = 0.11, p = 0.95). In general, drivers preferred ADS-1 over other systems. However, ADS-2 presented patterns similar to those of ADS-1, whereas ADS-4 received the lowest ratings. The results of conducting multiple comparisons with Tukey’s HSD between systems are as follows:
1) Understanding (To what extent do you think you could understand the system?): The comparisons did not reveal significant differences between systems. However, the participants could understand ADS-1 and ADS-2 slightly more easily than ADS-3 and ADS-4.
2) Trust (To what extent do you think the system is trustworthy?): The comparisons revealed significant differences between ADS-1 and ADS-4 (p < 0.05) and between ADS-2 and ADS-4 (p < 0.01). Although the participants’ rating of all the systems exceeded the mid-value of the scale, ADS-4 was rated lower than the other systems. When the participants were questioned about the reason, they reported that it was difficult to trust a system that provided a very short time period (i.e., 6 s) to decide whether they should cancel its action in the presence of other vehicles passing at a higher speed in the adjacent lane.
3) Acceptance (To what extent do you think you prefer to use the system in the real world?): The drivers rated ADS-1 and ADS-2 significantly higher than ADS-3 and ADS-4 (p < 0.05). Meanwhile, no significant differences were observed between ADS-1 and ADS-2 and between ADS-3 and ADS-4.
4) Controllability (To what extent do you think you were in control of the vehicle during system activation?): Although the analysis did not reveal a significant difference in drivers’ ratings among the four ADS designs, the drivers rated ADS-4 lower than the other systems, which were rated almost equally.
5) Attention (To what extent do you think you can focus on traffic during a lane change?): The drivers’ rating of their attention to the roadway and surrounding traffic during the operation of the ADS did not differ among the four systems. In general, the participants perceived that they could focus on the roadway adequately during automated driving. However, the off-road glance results presented in Fig. 7 reveal a different outcome, indicating that the drivers were not completely aware of their behavior during automated driving. This observation affects the implementation demands of ADSs, which still require partial or complete attention of the driver on the road to ensure safety. For example, drivers may rely excessively on partial driving automation systems and engage in non-driving related tasks, which would result in higher trust in the system. In contrast, they may unnecessarily interrupt the operation of conditional or high driving automation systems, thereby affecting system usage and causing system disuse 37).
Fig. 9 Subjective assessment of each ADS corresponding to five post-driving-experiment questions. Error bars denote standard deviations.
4. Discussions
ADSs are designed to safely control motor vehicles for an extended period without human monitoring and intervention. However, the capabilities of these systems in managing various traffic conditions are still limited, which necessitates occasional human monitoring and intervention to achieve adequate system performances. This study set out to design human-centered HMI to support drivers’ attention to the road and comprehension of the automated process to reduce the requirement for human intervention during conditional automated driving. The human-centered HMI strategies were proposed to maintain sufficient driver engagement during automated driving for a better driver–ADS interaction without affecting system effectiveness and acceptance. Herein, humans can share certain decisions and control with ADSs to regulate the vehicle without requiring control transition. A driving simulation experiment investigated the impact of the proposed cooperative HMIs on drivers’ interaction with ADSs.
The study observations indicated that sharing decision-making and control between human drivers and ADS resulted in significant improvements in driver engagement and attention to the surrounding traffic during automated driving. Although the drivers preferred the baseline system (ADS-1), the objective assessment showed that the best vehicle lateral control and driver reaction time results were achieved with ADS-3. In this system, the RTE (HMI-4) provided better system capabilities for performing tactical tasks under the human directive. The results of lane-change response time under ADS-2 (wherein the RTE supported only driver’s decision-making) were comparable to that under ADS-3 and better than those under ADS-1 and ADS-4. However, drivers’ off-road glances indicated that the drivers were more attentive under ADS-3 and ADS-4 than under ADS-1 and ADS-2. These observations highlighted the importance of informing drivers regarding traffic disturbances and the method to address such situations.
Another important observation was that the best level of driver engagement, longest automated driving time, and least control transition was achieved under ADS-3, indicating the effectiveness of using the RTE. The results of system usage supported this, wherein the percentage of ADS-3 usage was higher than those for the other systems. These observations revealed the effects of the HMI strategy and control algorithm (see Table 2) on the drivers' behavior toward the system. For example, ADS-3 and ADS-4 were similar in their capability to perform automatic lane-change maneuvers but different in terms of the control algorithm. This similarity and difference resulted in significant statistical differences between the two systems in terms of control input and system usage (see Table 3).
One unanticipated finding was that the time headway between the subject vehicle and vehicles in the passing lane was critical under the ADS-3 and ADS-4 conditions. The time headway under the ADS-2 condition was less critical compared to ADS-3 and ADS-4 but lower than the recommended value (i.e., 3 s). Although the time to change lanes was the largest under the ADS-1 condition, the drivers maintained safer time headway during the lateral maneuver. In real-world driving, critical time headway can significantly affect the behavior of other road users, resulting in incidents 38). These experimental findings suggest that human-centered system designs, while they can improve the interaction between the driver and ADS, can also reduce safety and introduce discomfort to other users who will interact with ADS.
The results of manual and automated driving performance of lane-change tasks show that the ADS performed better than the human driver. However, the results of the steering wheel behavior were not significantly different between ADS-1 and ADS-2 and between ADS-3 and ADS-4. These results are attributed to two factors. One factor is that the traffic congestion scenario is not a time- or safety-critical condition. The second factor is related to the road geometry, i.e., all the scenarios were evaluated on straight road sections with curvatures less than 0.0002 1/m to reduce the number of variables that are likely to affect vehicle behavior and control.
It is noteworthy that the energy consumption during automatic overtaking was more stable and significantly less than that during manual overtaking under similar traffic conditions. Although preliminary, these results indicate that automated driving may reduce energy consumption, benefit the environment and human health, and contribute toward addressing particular challenges in achieving carbon-neutral goals. Such results can also be used to highlight the effect of ADSs on reducing non-exhaust emissions released from the engine and tire frictions during the vehicle acceleration, deceleration, and steering maneuvers 39).
Subjectively, the overall evaluation showed that the drivers’ ratings of all the systems were above-average for all the questions. The results indicated that the drivers understood, preferred, and trusted ADS-1 and ADS-2 (both of which were incapable of automatic lane changing) more than ADS-3 and ADS-4. However, the driver’s evaluation of ASD-3 was marginally higher than that of ADS-4 for all the questions. In this regard, the drivers reported that the control algorithm of ADS-3 was more reasonable and convenient to understand compared with that of ADS-4. With ADS-3, the drivers could perceive the traffic situation and select the most appropriate time to push the button to start the automatic lane-change maneuver. In contrast, with ADS-4, the drivers had a limited time (6 s) to perceive the traffic situation and decide whether to allow the system to proceed with lane changing. This indication was supported by the drivers’ response to the controllability question, wherein the drivers rated ADS-1, ADS-2, and ADS-3 relatively closely and marginally higher than ADS-4. From the human-centered designs perspective, these results were not very encouraging. For example, the main goal of proposing ADS-3 was to enhance the driver's understanding of the automated process. However, drivers' comprehension of the baseline, in which the system provided less communication with the driver, was better than other conditions.
Although the observations of the current study were obtained under non-time-critical or complex traffic conditions, these may be effective for safety-critical and take-over situations during automated driving. The sharing of decision-making and control between human drivers and systems during non-critical automated driving can improve the time required to take over and the smoothness of control transition, both of which are essential for safe control transition during critical situations. Furthermore, it should be noted that all scenarios were implemented during the daylight under clear weather and good roadway conditions with light traffic in the passing lane. These factors can significantly affect driver response and appreciation of the ADS design. More specifically, drivers' interaction with and acceptance of ADS-3 and ADS-4 can be improved if they were tested in bad weather and road conditions. Therefore, this combination of findings has important implications for developing cooperative HMIs for automated vehicles, but there are still many unanswered questions about the effectiveness of these HMIs under different driving conditions. Additional experimental, observational, and field studies considering various environmental and traffic conditions will be needed to develop a robust model of driver interaction with cooperative systems.
The differences between ADS designs and drivers’ preferences can be influenced by certain limitations of the current study. One limitation is related to the experimental design; wherein each participant encountered four system settings intensively in a day. Although the sequence in which the system settings were experienced was counterbalanced using the Latin-square method, learning effects can influence the driver behavior during the subsequent drives. Although drivers’ behavioral adaptation to the RTE messages and different system settings was rapid for all the participants, the inclusion of drivers with prior experience in automated driving, as indicated in the demographic survey, may limit the comparability between systems. Another limitation is related to the time of test scenarios and situation criticality. Although four subjects had fallen asleep during the automated driving time, which affected their perception and reaction time during traffic disturbance, an extension of the time of automated driving may result in more unusual driving behaviors. Furthermore, the introduction of drivers to certain critical situations, such as extreme weather conditions, which have been excluded in this study, may broaden our understanding of the effectiveness of using the RTE in real-world driving.
5. Conclusions
This study used the levels of automation compatible with the concept of human-centered systems design to propose four HMI designs for conditionally automated vehicles during lane change situations. Results of this driving simulation experiment indicate that the support to drivers’ decision-making and control significantly reduced the perception-response time to variations in the system status and traffic disturbance. This, in turn, resulted in significant improvements in the driver’s performance of the subsequent action under cooperative systems compared with the baseline condition (ADS-1). The results of lateral and longitudinal control indicated that the best driving performance was achieved by ADS-3, with no significant differences were observed with ADS-4 because the automatic lane changing feature was also available. However, it was identified that the ADS with higher levels of automation (ADS-3 and ADS-4) while maintaining drivers in the monitoring loop, increased the usage of the system, and reduced the requirement for control transition, can affect the overall safety and result in a dangerous condition for other road users. These findings are consistent with a part of the first hypothesis, indicating that cooperative HMIs can improve driver interaction with the system, but violate the part stating that using RTE can result in safety benefits for other road users who will interact with automated vehicles.
The subjective evaluation revealed that the drivers preferred and trusted less intrusive systems (ADS-1 and ADS-2) over higher-capability systems (ADS-3 and ADS-4) that required more driver engagement during automated driving. Although it was expected that the increase in driver involvement in automated driving could enhance drivers' understanding of the system, the results indicated that drivers’ understanding of ADS-3 and AD-4 decreased compared to ADS-1 and ADS-2 in which the drivers were involved in the automated driving into a lesser extent. These findings are also not fully consistent with the second hypothesis of this study, indicating that increasing driver's engagement in automated driving while reducing the driver’s acceptance of and trust in the system, can also improve driver's comprehension of the system. Therefore, efforts must be continued to balance system capabilities and user preference while focusing on the potential implications on user behavior and acceptance.
The outcomes of this study contribute to the rapidly expanding field of automobile automation by providing a deeper insight into the significant role of cooperative HMI in improving driver–automated vehicle interaction and as a potentially reasonable approach for introducing safe automated driving. However, it is difficult to predict the manner in which drivers would adapt to ADSs with different capabilities in terms of long-term use. In particular, it is unclear how drivers’ understanding of the automated process and attention to surrounding traffic variations during hands-, feet-, and eyes-free automated driving can be maintained effectively while they engage in non-driving-related tasks. Cooperative HMIs that enable drivers to be more involved in the automated process may improve drivers’ capabilities during abrupt or unpredictable control transitions or when the system cannot manage the situation encountered. Hence, this is a critical area necessitating further investigations.
