The Effects Of System Functional Limitations On Driver Performance

Transcript

2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC) Banff Center, Banff, Canada, October 5-8, 2017 The Effects of System Functional Limitations on Driver Performance and Safety when Sharing the Steering Control during Lane-change Husam Muslim Makoto Itoh Risk Engineering Major, Systems and Information Department, University of Tsukuba Tsukuba, Japan [email protected] Faculty of Engineering, Systems and Information, University of Tsukuba Tsukuba, Japan [email protected] acceptable ways as possible. All these developments aimed at either guiding the driving task performance, such as the lane keeping assistance system that guides the LAC function process, or performing a task or subtask autonomously, such as the adaptive course control system (ACC) that is capable of automatic headway and speed maintenance [4, 5]. In short, the authors found several control strategies in the literature in which LAC and LOC functions were differently allocated between both interacting agents. The control distribution strategy in which the final control action of a specific task is determined by a human and automation simultaneously is called Shared Control [4, 6]. When the final control action of a specific task is determined entirely by either agent, human alone or automation alone, the system is then called Traded Control [4]. The lane-changing maneuver is a potentially hazardous task in which drivers might encounter various hazards, such as objects in blind spots and fast approaching cars, as is illustrated in Fig. 1 [7]. Changing lanes without checking the surrounding environment may increase the likelihood of accidents [8]. The benefit of automation assistance in such critical situations has been demonstrated sufficiently. Little is known, however, about how driver performance, overall situation awareness, and safety might be affected, when automation assistance develops functional limitations due to rapid changes in the driving environment. Abstract—Even highly sophisticated and reliable driver assistance systems might encounter critical situations that are outside system design capacity. This can lead to a conflict in viewpoints between the driver and automation. To evaluate the effects of such conflicts on human-machine interaction and cooperation, overall performance, and safety, an experiment was designed to determine driver responses to different imminent hazards during lane change while receiving automation assistance. Using a driving simulator, two types of lane change collision avoidance systems were provided: a haptic shared control that consists of auditory and haptic force warnings and an automatic shared control that includes auditory warnings and autonomous action. While the results of both systems were encouraging in terms of accident reduction when the hazards were within system design capacity, accidents were significantly increased when the hazards encountered were outside system design parameters. The acceptance by drivers was considerably influenced by the hazard type, and their feeling of control was affected by the type of assistance systems. Keywords—Safety; System Design; Control; automation interactions; Human-machine cooperation I. Human- INTRODUCTION Automation has long been employed in the aviation domain. Aircraft control was mainly divided into two essential functions: lateral and vertical navigations [1]. The control of either function can be allocated between human pilots and automation in either a task- or a situation- dependent manner: e.g., taking-off, climbing, cruising or landing tasks, and risky or routine situations. Similar to aviation, car driving tasks can also be generally divided into two functions: (1) lateral control (LAC), which can be controlled with the steering wheel, and (2) longitudinal control (LOC), which can be controlled with gas and brake pedals. In conventional manual driving, human drivers assume full time control of both LAC and LOC functions under all circumstances [2]. However, erroneous behavior by drivers, particularly in tactical maneuvers like changing lanes that require high cognition and skills, necessitated automation assistance [3]. Several methodologies have been developed to make driving tasks easier and safer in as many enjoyable and 978-1-5386-1644-4/17/$31.00 ©2017 IEEE Fig. 1. Hazardous lane changing manoeuvre To address the abovementioned issue, the authors set up a driving experiment using a motion-based driving simulator in which two types of driver assistance systems for avoiding collisions during a lane change were installed. Driver performance, human-machine cooperation, and overall safety were evaluated when supported by haptic feedback guidance 135 through the steering wheel or by automating the steering function. Whilst the driver had the final authority on the steering task when sharing the control with the haptic system, he or she could not steer the vehicle when the control was traded with the automatic system. A recent study by Itoh and Inagaki (2014) examined and compared both systems under hazardous conditions where system design capacity and driver expectations were aligned [9]. The present study investigated more human-automation interactions under various hazardous situations when the system function developed limitations. It was hypothesized that human-automation interaction problems arise when the human and automation understanding of a situation does not match. The expected outcome of this study was that human drivers appreciate the shared control (sharing the LAC function) more than the autonomous cooperative control (Trading the LAC function) for an automation design. It was also expected that the functional limitations of the haptic system would affect driver performance and safety to a lesser extent. II. 1) Type of Assistance Systems: The experiment employed two types of assistance systems for avoiding collisions with vehicles at blind spots. The participants were randomly and equally divided between assistance systems. The systems were engaged when drivers input a steering angle equal or larger than 0.033 rad to the direction of objects in blind spots. The design specifications were as follows: a) Haptic Lane Change Collision Avoidance System (HLCAS): This system resisted unsafe lane change by stiffening the steering wheel to warn the driver. When it was activated, the steering friction torque given by the system was increased from 1 Nms to 9.6 Nms. The torque value was determined so that the drivers would readily feel the change from the normal torque, while allowing them to override the additional torque. An auditory signal was given to the driver just before the steering wheel was made heavier. The system restored the original value of the steering wheel torque either immediately after detecting that there was no vehicle in the blind spot or as soon as the driver overrode the system. EXPERIMENT b) Automatic Lane Change Collision Avoidance System (A-LCAS): This system cancelled the steering input by the driver and autonomously controlled the lateral position of the vehicle within the current host lane to prevent lane change collisions. When the system was active, the driver was no longer able to control both vehicle directions by moving the steering wheel. An auditory signal was given to the driver just before the system was activated. Depending on the context, the driver could change the velocity to increase the distance from the vehicle in the blind spot. When the risk was no longer within the system boundary, the driver was provided with a second but different auditory signal to inform him/her that the system was about to be disengaged. Three seconds later, another signal similar to the second one was given, and the system was immediately deactivated. A. Participants There were 48 participants (24 males and 24 females), aged between 20 and 50 years (m=30.0; S.D.=9.2) in this driving experiment. Their driving experience ranged between 30,000 and 200,000 kilometers. B. Aparatus The study was conducted on a single seat cockpit driving simulator (HONDA’s DA-1105), as shown in Fig. 2. The driving environment was simulated by one 120° front screen and three small LCD screens to show front, side, and rear views, respectively. There were no additional screens to show blind spots. Thus, drivers had to either rely on automation assistance or reduce their speed to check for blind spots when they were about to change lanes. The driving course was a twolane highway ring, which was 6 km long. The course consisted of both curved and straight-line sections. 2) Type of Critical Events: In this experiment, each participant performed 24 lane-changing maneuvers consisting of 6 hazardous and 18 non-hazardous lane changes. In each of the maneuvers, there was a slow leading vehicle traveling at 70 km/h, which forced the host vehicle to initiate an overtaking lane-changing to keep the speed at 80 km/h as instructed. The hazardous lane changes were divided according to type of encountered hazard as a function of system design capacity. a) Within System Design (WSD): Each driver encountered four hazards represented by the presence of a vehicle in the blind spot. These hazards were within the system design capacity. Two of the events were encountered during the training phase and the other two were during the testing phase. Fig. 2. Driving Simulator b) Outside System Design (OSD): During the testing phase, the drivers encountered two hazards that were outside system design capacity. One of the two hazards was a fast approaching vehicle (100 km/h) in the cruising lane, which could be seen through the side mirror. However, the system was not designed to handle such event. The other hazard was divided into two parts: first, the blind spot vehicle, which could be detected by the system; and second, a suddenly stopped front vehicle. The C. Experimental Design To investigate system effectiveness, driver acceptance, and the impact of system design capacity on driver performance and safety, the experiment was designed using a 2 × 2 mixed factorial design with types of assistance systems and hazards. 136 drivers had to avoid the rear-end collision, while they were assisted by the system to avoid the side collision. suggested that the ability of the drivers to regain control can be highly important when hazards encountered are outside system design capacity. D. Dependent Variables: Data on the number of collisions, steering wheel reversal rate, and braking reaction time were collected, analyzed, and compared between and within groups. For the subjective evaluation, a Likert scale questionnaire was used. The drivers had to answer the following questionnaire items after experiencing each of the hazardous situations by ranking their feeling from “0: Not at all” to “7: Absolutely”. B. Driver Performance 1) Braking Reaction Time (BRT): All lane changing manoeuvres performed in this experiment were to avoid a front hazard, i.e., a slow leading vehicle. It was therefore necessary to analyze the braking reaction time of the drivers, as shown in Fig. 3. BRT was measured as the time elapsed from the first auditory signal indicating a critical event, until the first driver pressing the brake pedal. Two-way repeated measures ANOVA showed that there was no significant interaction between the assistance systems and types of hazards (F(1, 46)= 10.8, p>0.05). However, it reported a significant difference between the systems (F(1, 46)= 88.6, P<0.01) and between the type of hazards (F(1, 46)= 91.3, P<0.01). In contrast, Bonferroni pairwise comparisons showed a significant effect of assistance type on driver BRT (P<0.01). BRT under the haptic system was markedly less than that under the automatic system. Once the haptic system was activated, drivers readily felt the difference in the steering wheel torque, while the torque remained unchanged under the automatic system. This significantly highlighted the necessity for receiving a continuous feedback about the automation assistance status. 1) Acceptance: “To what extent do you think you would like to use the system in real world situations?” 2) Control: To what extent do you think you had the situation under control? 3) Effectiveness: To what extent do you think driving with the system improved safety? 4) Interference: To what extent do you think you were able to detect conflict with the system activity? E. Tasks and Procedures At the beginning, all participants were briefed on the ethical rights and experiment design, requirements and tasks. The participants were required to drive safely in the left lane of the two-lane experimental course and maintain a vehicle speed of 80 km/h. However, they could initiate lane-changing maneuver to avoid threats in their forward path. After performing familiarization and training drives, the participants had to perform two testing drives under WSD condition and two testing drives under OSD condition. The critical events sequence, triggering time and location were random and counterbalanced among drives to minimize the learning effect. III. RESULTS A. Number of collisions In total, 196 hazardous lane changes were analyzed and discussed in this experiment (48 participants* 2 WSD hazards* 2 OSD hazards). Table 1 lists the number of accidents for each assistance system by the types of accidents and encountered hazards. Multiple comparisons with Ryan’s procedure revealed statistically significant differences in the number of collisions between the types of encountered hazards for both systems (p<0.01). From this data, the significant effect of system design capacity was evident. The result supported the design guidelines suggested by Abbink et al. 2012 regarding the boundary and functionality of automation assistance [10]. TABLE I. Hazard type WSD OSD Fig. 3. Braking reaction time for critical events within and outside system design capacity for each system 2) Steering Wheel Reversal Rate (SWRR): The SWRR was used to evaluate how much the automation assistance affected the reactive steering behavior of drivers, such as steering accuracy and fluency. The SWRR was computed as SWRR = N/T, where N is number of changes of direction in steering wheel rotation during time T, and T is the time required to avoid a lane change collision in seconds. Fig. 4 represents the mean and standard deviation of the SWRR for each system and the type of hazards. According to the two-way repeated measures ANOVA, there were significant interactions and differences between assistance systems ((F(1, 46)= 855.3, P<0.01) and (F(1, 46)= 52.2, P<0.05) respectively). The data indicated that the SWRR during the activation of H-LCAS was notably lower than ALCAS. Between the types of hazards, ANOVA showed a significant difference for both systems (F(1, 46)= 101.4, P<0.01) and a significant interaction among the types of assistance systems and hazards (F(1, 46)= 34.9, P<0.01). NUMBER OF COLLISIONS BASED ON SYSTEM DESIGN CAPACITY Side collisions H-LCAS A-LCAS 6/48 0/48 23/48 20/48 Rear-end collisions H-LCAS A-LCAS 1/48 3/48 9/48 20/48 Between systems, the comparisons showed a significant difference only in the number of rear-end collisions when the encountered hazards were outside system design capacity. This 137 system design capacity, where the system was not activated; hence, drivers felt in charge of the steering control more than during hazards within system design capacity. Fig. 4. Steering wheel reversal rate for critical events within and outside system design capacity for each system C. Subjective Evaluation 1) Drivers acceptance of automation assistance: Fig. 5 depicts the willingness of the drivers to use the system in real world driving. The Wilcoxon rank sum test revealed a significant difference between hazardous events for each assistance system (Z= −5.5 with H-LCAS and Z= −8.1 with ALCAS, p<0.01). While both systems were highly accepted when the hazards encountered were within system design capacity, acceptance was notably reduced when hazards outside system design capacity were encountered. However, HLCAS was less affected than A-LCAS. Between assistance systems, there was a significant difference only when the hazards were outside system design capacity (Z= −5.3, p<0.01). The reason could be that the drivers remained in charge of steering under all circumstances. Fig. 6. Subjective rating by drivers on their ability to regain control based on system design capacity 3) Driver evaluation of system effectiveness: To better understand the expectations of the drivers from the system and how they might rely on its assistance, it was necessary to subjectively evaluate system effectiveness to improve safety. As shown in Fig. 7, the drivers thought that both systems were efficient in improving safety during lane change. However, experiencing hazard outside system design capacity significantly affected driver ratings (Z= -6.3, p<0.01). Although the drivers rated the automatic system slightly higher than the haptic system when the encountered hazards were within system design capacity, their rating of the haptic system was affected less when they encountered hazards outside system design capacity. The significant difference between assistance systems when experiencing hazards outside system design capacity (Z= -5.5, p<0.01) suggested that the effectiveness of the system may depend on the ability of the driver to regain control and assist the system when necessary. Fig. 5. Subjective rating by drivers on acceptence of the automation assistance based on system design capacity 2) Driver feeling of having the situation under control: Fig. 6 illustrates the mean and standard deviation of the subjective ratings by drivers on their feeling of control. There were significant differences between assistance systems in all hazardous events (Z= −6.6 in WSD and Z= −3.8 in OSD, p<0.01). Automation authority could be considered a significant cause of the difference between assistance systems. When driving with H-LCAS, the test showed no significant differences between hazard types. This could be attributed to the fact that drivers were able to control the steering during system activation. However, the statistically significant difference between hazard types was only with A-LCAS (Z= −4.7, p<0.01). The results for A-LCAS seemed surprising because driver experienced some hazardous events outside Fig. 7. Ssubjective rating by drivers on their ability to regain control based on system design capacity 2) Driver ability to detect interference with automation action: Fig. 8 depicts the subjective assessment by drivers of their ability to detect interferences with system activity. It was assumed that if the driver was not able to detect interference, comparing his/her analysis to the one in the system, he/she could be surprised by the automation action. The rank-sum test showed a significant effect of the types of hazardous situations on both the assistance systems (Z= -2.3 with H-LCAS and Z= 2.8 with A-LCAS, p<0.01), but there was no statistically 138 significant difference between systems (Z= -0.9 with H-LCAS and Z= -0.6 with A-LCAS, p>0.05). When facing within-system-design imminent hazards, the drivers were able to detect and perceive decisions and recognize actions of either assistance type. This could be an indication of the understanding of the automation system by the drivers. When the hazards were outside system design capacity, the results revealed that the drivers were surprised by the limitation of the assistance system in such risky situations. This could be an indication of the significant effect of the functional limitation of automation on driver behavior in addition to traffic safety. study extended these findings by illustrating how humans and automation could interact when encountering unpredictable hazards that were outside system design capacity. When the encountered hazard was outside system design capacity, drivers were likely to overestimate the capabilities of the automation system. Whilst the drivers were able to recognize the fast approaching vehicle in the cruising lane, it was difficult to point out the exact human factors involved in such hazardous situations and why accidents dramatically increased. Perhaps, some drivers changed lanes expecting that the system would be activated if the situation is risky, i.e., overreliance. Others might fail to estimate the approximate speed of the fast approaching vehicle, i.e., human reliability. To avoid such consequences, the driver should be aware of what type of hazards the system is designed to handle. Considering the highly complex and rapidly dynamic driving environment, automation should be developed with clear boundaries and functionalities. The study found that driver performance was significantly (negatively) affected by failure on the part of the machine to cope with sudden traffic changes. Furthermore, the overall driver situation awareness was critically degraded leading to greater hazards. Analysis of the subjective ratings by drivers showed that machine failure significantly influenced driver response to all factors mentioned in the questionnaire (Trust, Acceptance, Safety, and Control). In summary, the results illustrated how driver performance and assessment of automation and overall safety were significantly affected by control authority in addition to system performance. Although drivers were able to control the longitudinal position of the vehicle, their feeling of control was considerably reduced when the system dominated the execution of the lateral maneuver. It can, therefore, be assumed that the effect of trading lateral or longitudinal control between humans and automation on driver performance and the feeling of control varied contexts dependently. Although both steering and pedals tasks were involved, these results are only applicable to automated steering task and this work should be extended to include the automated pedals tasks. Fig. 8. Subjective rating by drivers on their ability to regain control based on system design capacity IV. DISCUSSIONS The present experiment investigated how different levels of automation authority and the types of encountered hazards affected the interaction between a human driver and an assistance system when the control of the steering system was shared or traded between the human driver and automation assistance. The experiment was carried out in a driving environment where the drivers encountered hazards within and outside system design capacity. First, the willingness of the drivers to cooperate with the assistance system was significantly affected by the agent, human or automation, which had the final authority to control the steering task. In the case of the haptic system, there was no conflict of authority because the human driver was always the final decision maker in the process. The conflict occurred when driving with the automatic system, particularly when encountering hazards outside system design capacity. In short, human drivers cooperated better with automation when they had the final word in the decision process. These results were in agreement with those previously obtained in the automobile and aviation domains [10, 11]. When the system design capacity and expectations of the drivers were aligned, the haptic system was, to an extent, efficient at maintaining safety. However, a more powerful intervention, i.e., the automatic system, was needed to achieve further collision reduction. Results of the subjective assessment suggested that both systems were trustworthy, improved safety, and were accepted by the drivers. So far, these results are in line with those previously found in the literature [9, 12]. While previous studies drew their findings based on fully reliable systems and hazards within system design capacity, the present V. CONCLUSIONS For the examined assistance systems and studied experimental conditions, the following conclusions could be drawn:  Drivers comparably accepted and appreciated both assistance systems when the encountered hazard could be handled by the system. However, the system with a low level of authority was particularly accepted when the drivers encountered sudden traffic hazards while they were assisted by the system.  A proper balance of dynamic control allocation in shared control and cooperative systems should be context dependent, by taking into account how drivers react and regain control when the system develops an unpredictable functional limitation.  It was easier for the drivers to build their model of the system when they were the final authority. 139  The overall driver performance was improved for the situation when driving with the haptic system. The effect of the negative interference of the haptic guidance system on the driver during intended lane changing was also less than that of the automatic system. ACKNOWLEDGMENT The authors would like to thank Dr. Pacaux-Lemoine for her significant contribution in the design of the previous pilot experiment. We are also indebted to the SMC2016 Shared Control committee and members for their insightful comments and feedback. VI. LIMITATIONS AND FUTURE PERSPECTIVES REFERENCES In this study, the activation duration of the automatic lane change collision avoidance system depended on driver reaction. To be more specific, the system was activated when the drivers initiated a lane changing maneuver while there was a vehicle in the blind spot. In this case, the automatic system continued engagement until the blind spot vehicle came out of the boundaries of the system. Accordingly, the duration if system activation might be longer than expected, even though the hazard was avoided. This can potentially increase humanmachine interaction problems. It would be more efficient and be accepted more readily by drivers if the duration of the automatic system activation could be adjusted depending on the situation. Alternatively, it would be useful to investigate the most effective time duration for system activation with minimum side effects. The present study makes several noteworthy contributions to the development of automotive shared control systems for collision avoidance. It also provides additional evidence with respect to the role of humans and automation in safety-critical situations and on how to determine the final authority in making in a decision. This study has demonstrated, for the first time that machine reliability is not the only essential factor for an efficient assistance system design. The way in which the control authority is managed between the human driver and automation system plays a significant role in realizing a cooperative work environment between them. Although the current study is based on a relatively large sample of participants, further studies to assess the long-term effects of shared control systems on driver performance and safety are necessary. In real world driving, these types of support systems might not be activated very often; therefore, once the system is activated, it may lead to some surprises. [1] T. Inagaki, “Adaptive automation: sharing and trading of control”, In: E. Hollnagel, ed. Handbook of cognitive task design, Mahwah, NJ: Lawrence Erlbaum Associates, pp. 147–169, 2003. [2] S. Huang, W. Ren, and S.c. Chan, “Design and Performance Evaluation of Mixed Manual and Automated Control Traffic”, IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, Vol. 30, No. 6, pp. 661–673, 2000. [3] D.W. Van Der Wiela, J.M.M. Van Paassena, M. Mulderb, M. Muldera, and D.A. Abbinkb, “Driver Adaptation to Driving Speed and Road Width: Exploring Parameters for Designing Adaptive Haptic Shared Control”, IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2015. [4] T. B. Sheridan, “Telerobotics, automation, and human supervisory control”, Cambridge, MA: MIT Press, 1992. [5] F. Mars, M. Deroo, and J.-M. Hoc, “Analysis of Human-Machine Cooperation When Driving with Different Degrees of Haptic Shared Control”, IEEE Trans. Haptics, Vol. 7, No. 3, pp. 324–333, 2014. [6] D.A. Abbink and M. Mulder, “Neuromuscular analysis as a guideline in designing shared control”, Adv Haptics Vol. 109, pp. 499–516, 2010. [7] G. M. Fitch, S. E. Lee, S. Klauer, J. Hankey, J. Sudweeks, and T. Dingus, “Analysis of Lane-Change Crashes and Near-Crashes”, NHTSA Report, DOT HS 811 147, June 2009. [8] L. Barr and W.G. Najm, “Crash problem characteristics for the intelligent vehicle initiative”, TRB 80th Annual Meeting, Washington, DC, No. 01-2471, 2001. [9] M. Itoh, and T. Inagaki, “Design and evaluation of steering protection for avoiding collisions during a lane change”, Ergonomics, Vol. 57, No. 3, pp. 361–373, 2014. [10] A.D. Abbink, M. Mulder and E.R. Boer, “Haptic shared control: smoothly shifting control authority?”, Cogn Tech Work, Vol. 14, pp. 19–28, 2012. [11] M.S. Young, N.A. Stanton, and D. Harris, “Driving Automation: Learning from Aviation about Design Philosophies,” Int J. Vehicle Des., vol. 45, pp. 323-338, 2007. [12] N. Cramer, and M. Zadeh, “Blind-Spot Detection and Avoidance Utilizing In-Vehicle Haptic Force Feedback”, SAE Technical Paper 2011-01-0556. doi:10.4271/2011-01-0556, 2011. 140