Driving Performance After Self-Regulated Control Transitions in Highly Automated Vehicles

Introduction

Automated vehicles show promise in reducing road accidents (Eriksson, Banks, & Stanton, 2017; World Health Organisation, 2015) but in their current form are no panacea for road safety (Eriksson & Stanton, 2017b; Gold, Damböck, Lorenz, & Bengler, 2013; Gold, Korber, Lechner, & Bengler, 2016; Kalra & Paddock, 2016). As drivers engage a contemporary automated driving system, they are decoupled from the operational and tactical levels of control, leaving only high-level strategic goals to be dealt with by the driver (Bye, Hollnagel, & Brendeford, 1999; Michon, 1985) while still being expected to resume control when the vehicle reaches the limits of its operational design domain (ODD; the ODD may include geographic, roadway, environmental, traffic, speed, and/or temporal limitations of automated driving availability; SAE J3016, 2016) or when a system failure or sudden, unexpected event forces a transition back to manual control (SAE J3016, 2016). This intermediate form of automation has been deemed hazardous as drivers are required to monitor and be able to regain control at all times (Casner & Schooler, 2015; Seppelt & Victor, 2016; Stanton, 2015). This is a form of “driver initiated automation,” where the driver is in control regardless of whether the system is engaged or disengaged (Banks & Stanton, 2015, 2016; Lu, Happee, Cabrall, Kyriakidis, & de Winter, 2016; SAE J3016, 2016) contrary to “system initiated automation” where the controlling agent, be it driver or the automated driving system is in control, is determined by artificial intelligence (Gordon, Kozloski, Kundu, Malkin, & Pickover, 2017) in a MABA-MABA fashion (Dekker & Woods, 2002). The intermittent transitions of control and the sharing of task-relevant information between the driver and driving automation can be described in terms of distributed cognition (DCOG) (Hollan, Hutchins, & Kirch, 2000; Hutchins, 1995). DCOG offers a paradigm shift in describing how a human interacts with other humans, artifacts, and artificial agents, describing it as a “system” where cognition, knowledge, and mental models are distributed between agents in the joint cognitive system, henceforth referred to as system (Hollnagel & Woods, 2005). In such a system, coordination and communication between system entities are of the utmost importance (Christoffersen & Woods, 2002; Eriksson & Stanton, in press; Hutchins, 1995; Stanton, 2014). The functioning of a cognitive system has been described by Hollnagel and Woods (2005) in the contextual control model (COCOM) as a cybernetics inspired tracking loop. The COCOM model describes how the control of a system can be lost and gained and how different levels of control influence performance. As time progresses in a task such as driving, dynamic shifting between four different control modes can be made, dependent on the time horizon, and the predictability of the situation.

Hollnagel and Woods (2005) describes four control modes:

OPEN IN VIEWER

Figure 1.

The relationship between Hollnagel and Woods (2005) control levels and available time and predictability of a situation.

Driver assistance systems on Level 1/2 (SAE J3016, 2016) have shown detrimental effects on driver behavior when drivers were asked to resume control compared to manual driving. Young and Stanton (2007) observed an increase in brake reaction time of approximately 1 second when a leading vehicle suddenly braked requiring driver intervention when using adaptive cruise control, which is the approximate time it takes a driver to respond to a sudden braking event while engaged in fully manual driving (Summala, 2000). An additional 0.3-second increase was found when adaptive steering was added (SAE J3016, 2016, Level 2). It is evident that the introduction of automated driving systems on Levels 1 and 2 have detrimental effects on driver readiness and driving performance. In a review of the literature of transitions of control in Level 3 automated driving systems, Eriksson and Stanton (2017b) found that it takes 1.14 to 15 seconds to respond to a request to intervene in an externally paced transition (system or event paced). In a previous analysis of the data from the experiment presented in this article, it was also found that drivers took between 1.97 and 25.75 seconds to respond to a request to resume control when they were able to pace the transitions (Eriksson & Stanton, 2017b). The increase in response times may be attributed to the fact that the control activities involved in driving are normally “automatic” activities that require little or no conscious effort to be executed (Norman, 1976, p. 70). When these activities are disrupted, by for example automation requesting manual control, conscious control is required to the detriment of “manual-driving” performance.

Russell et al. (2016) showed that drivers who are unaware of changes to vehicle driving characteristics (the steering torque profile) show declined steering performance, which may lead to over- or undershoot, indicating scrambled control. Merat, Jamson, Lai, Daly, and Carsten (2014) showed that it takes approximately 40 seconds for the driver to regain control stability after a control transition. Notably, the control transition used in the experiment of Merat et al. (2014) was system initiated and lacked a takeover request (TOR), which is featured in other recent research into control transitions in automated driving (e.g., Damböck, Bengler, Farid, & Tönert, 2012; Eriksson & Stanton, 2017b). The lack of human-machine interface (HMI) to convey the need to resume control may have contributed to scrambled control behavior in the first 40 seconds after resuming control. Similarly, Desmond, Hancock, and Monette (1998) found larger heading errors and poorer lateral control in the first 20 seconds after resuming control from automated driving following a failure compared to compensating for a wind gust in manual driving, hinting at scrambled control. Moreover, Gold et al. (2013) showed that drivers that subjected to short lead times (5 vs. 7 seconds) for the TOR elicited shorter response times, but performance post takeover was characterized by harsh braking, rapid lane changing, and unnecessary full stops, indicating that drivers were at the scrambled level of control. Damböck et al. (2012) found that an 8-second lead time for TORs produced driving performance at the same level as manual driving, indicating that drivers experienced a higher, operational or tactical level of control. Evidently, there may be a relationship between the available time to resume control from the automated vehicle and the resulting level of control performance when control has been handed back to the driver.

Eriksson and Stanton (2017b) and Eriksson et al. (2017) argue that “driver-paced” transitions will be a commonly occurring type of control transition in SAE J3016 (2016) Level 3 and 4 systems, where there is enough foresight when it comes to identifying system limitations (e.g., through the fleet learning feature of Tesla Autopilot version 8; Tesla Motors, 2016), which in turn will increase the lead time between TOR and a transition to manual control (Eriksson & Stanton, 2017b). The increase in the time between a TOR and a transition to manual control extends the time horizon, which in turn could enable drivers to attain a higher tactical level of control compared to the externally paced transitions reported in the literature (Damböck et al., 2012; Desmond et al., 1998; Gold et. al., 2013; Merat et al., 2014; Russell et al., 2016). Eriksson et al. (in press) showed that when the reason for a TOR was highlighted through an augmented reality overlay, drivers exhibited opportunistic control by braking to buy time. This was not observed when the augmented reality display showed higher levels of semantics, such as arrows indicating safe paths, indicating a higher level of control. Hollnagel (1993) emphasized that the “essence of control is planning” (p. 168), which means that a sudden, forced transition to manual control likely has detrimental effects on driving performance unless appropriate support is given (Cranor, 2008; Eriksson & Stanton, 2017b, in press; Stanton, 2015; Stanton & Young, 1998).

In light of this, this study aims to explore whether there are any differences in driving aftereffects following a transition from automated to manual control in two conditions, one with and one without a secondary task, compared to baseline manual driving. This research aims to provide knowledge on whether the resulting level of control (in terms of driving performance) is affected by control transitions when the transition is driver initiated (Banks & Stanton, 2015, 2016; Lu et al., 2016) and whether the additional time available for self-regulation (Cooper, Vladisavljevic, Medeiros-Ward, Martin, & Strayer, 2009; Eriksson, Lindström, Seward, Seward, & Kircher, 2014; Kircher, Eriksson, Forsman, Vadeby, & Ahlstrom, 2017; Wandtner, Schumacher, & Schmidt, 2016) of the transition process has a positive effect on driving performance.

Method

Equipment

The experiment was carried out in a fixed based simulator at the University of Southampton. The simulator consisted of a Jaguar XJ 350 vehicle with pedal and steering sensors provided by Systems Technology Inc. as part of the STISIM Drive M500W system (http://www.stisimdrive.com/m500w). The driving environment is powered by the STISIM Drive Version 3 software engine providing a projected 140° field of view. Rearview and side mirrors were provided through additional projectors and video cameras. The Jaguar XJ instrument cluster was replaced with a 10.5” LCD panel to display computer-generated graphics components, in the case of this experiment, takeover requests.

When a takeover request was issued, the engine speed dial was hidden, and the request was shown in its place. The symbol asking for control resumption is shown in Figure 2a, and the symbol used to prompt the driver to reengage the automation is shown in Figure 2b.

OPEN IN VIEWER

Figure 2.

(a) On the left-hand side is a control transitions request to automated vehicle control presented in the instrument cluster, coupled with a computer-generated voice message stating “automation available.” (b) Right-hand side, the graphic is shown in the instrument cluster of a takeover request. The visual TOR was coupled with a computer-generated voice message stating “please resume control.”

The mode switching human-machine interface was located on a Windows tablet in the center cluster, consisting of two buttons, either to engage or disengage the automation. To enable dynamic dis- and reengaging of the automation, something STISIM currently does not support, bespoke algorithms for automated longitudinal and lateral control developed in Visual Basic 6 for the STISIM Open Module API platform were used (Eriksson, Spychala, De Winter, & Stanton, 2016; Eriksson & Stanton, 2017a).

Experiment Design

The experiment had a repeated-measures, within-subject design with three counterbalanced conditions: highly automated driving while passively monitoring the system, highly automated driving while engaged in a secondary task, and a manual baseline drive. In both automated conditions, the participants drove for approximately 20 minutes, at 70 mph on a 3-km, six-lane motorway, three lanes either side of a barrier separating the direction of travel, with mild curves and moderate traffic conditions. The route was mirrored between the two automated conditions to reduce familiarity effects while keeping the roadway layout consistent (Figure 3). In the manual driving conditions, drivers drove a shortened version (10 minutes) of the route in Figure 3 under identical traffic and road layout conditions as in the automated conditions. In the secondary task condition, drivers were asked to read an issue of National Geographic while the automated driving system was engaged to effectively remove them from the driving task. The drivers started in manual driving mode at the start of each automated driving condition, and the timer to trigger the prompts to transition between control modes was triggered after 2 minutes of manual driving. Consequentially, drivers spent between 2 minutes and 30 seconds and 2 minutes and 45 seconds in manual mode before being asked to engage the automated driving system (Figure 4).

OPEN IN VIEWER

Figure 3.

A bird’s eye view of the road layout in the passive monitoring condition (route was mirrored for the secondary task condition).

OPEN IN VIEWER

Figure 4.

Takeover procedure during an experimental drive. Each manual and automated driving section lasted between 30 and 45 seconds.

Following initial engagement of the automated driving system, the drivers were prompted to either resume control from or relinquish control to the automated driving system at a randomized interval ranging from 30 to 45 seconds, thus allowing for approximately 24 control transitions, of which half were to manual control.

Dependent Variables

The following metrics were calculated for each condition and participant for the duration of 20 seconds post transition to manual control in the automated condition. For the manual driving comparison, a section of road was randomly selected, and 20 seconds of driving performance data were extracted for comparison between the automated and manual conditions.

Standard deviation of steering wheel angle (degrees): This metric is related to driver workload. In normal driving conditions, drivers tend to make continuous small steering corrections to adjust their lane position as driving conditions change. When workload increases, these corrections decrease in frequency, resulting in the need for larger steering inputs to correct the lane positioning (Liu, Schreiner, & Dingus, 1999). This metric is defined in the following equation (cf. Knappe, Keinath, Bengler, & Meinecke, 2007, p. 2).

Mean absolute lateral position (centimeters): This metric describes lane keeping accuracy and is calculated in the following way: M L P = | ∑ i = 1 n d i n | , where d_i is the distance measured from the center of the vehicle to the lane center.

While a root mean square (RMS) of the lateral position may provide insights into large movements in lane position, mean absolute lateral position is used in combination with plotting the standard deviation as this not only provides the mean lateral position but also the RMS through the shaded standard deviation area shown in Figures 5 through 7. Furthermore, the standard deviation of the SDSWA metric is not included in Figures 8 through 10 as this effectively is a calculated standard deviation of a standard deviation metric and thus does not provide any additional information.

OPEN IN VIEWER

Figure 5.

A time-series analysis of the difference in the mean absolute lane position after resuming control from an automated driving system after engaging in passive monitoring compared to manual driving. The black dashed line in the middle graph indicates an alpha of 0.01, and the dashed-dotted line indicates a Bonferroni corrected alpha of 0.00083.

OPEN IN VIEWER

Figure 6.

A time-series analysis of the difference in the mean absolute lane position after resuming control from an automated driving system after engaging in a secondary task compared to manual driving. The black dashed line in the middle graph indicates an alpha of 0.01, and the dashed-dotted line indicates a Bonferroni corrected alpha of 0.00083.

OPEN IN VIEWER

Figure 7.

A time-series analysis of the difference in absolute mean lane position after resuming control with a secondary task and without a secondary task. The black dashed line in the middle graph indicates an alpha of 0.01, and the dashed-dotted line indicates a Bonferroni corrected alpha of 0.00083.

OPEN IN VIEWER

Figure 8.

A time-series analysis of the difference in the standard deviation of steering angle after resuming control from an automated driving system after engaging in passive monitoring compared to manual driving. The black dashed line in the middle graph indicates an alpha of 0.01, and the dashed-dotted line indicates a Bonferroni corrected alpha of 0.00083.

OPEN IN VIEWER

Figure 9.

A time-series analysis of the difference in the standard deviation of steering angle after resuming control from an automated driving system after engaging in a secondary task compared to manual driving. The black dashed line in the middle graph indicates an alpha of 0.01, and the dashed-dotted line indicates a Bonferroni corrected alpha of 0.00083.

OPEN IN VIEWER

Figure 10.

A time-series analysis of the difference in the standard deviation of steering angle after resuming control from an automated driving system after engaging in a secondary task compared to passive monitoring. The black dashed line in the middle graph indicates an alpha of 0.01, and the dashed-dotted line indicates a Bonferroni corrected alpha of 0.00083.

The authors argue that these metrics could be a good indicator of the level of control exerted by the driver. For example, higher values of SDSA while showing large variability in mean absolute lane position would indicate scrambled control, whereas large SDSA while maintaining low variability in absolute lane position would indicate opportunistic control, and low variability in both metrics would indicate a tactical level of control.

To determine the aftereffects of automated driving with and without a secondary task before a takeover request was issued, we compared the mean absolute lateral position and standard deviation of absolute steering wheel angle between the two conditions and a baseline drive. The performance metrics were sampled at 75 Hz and were aggregated into 60 bins (approximately 333.33 milliseconds of data/bin).

Analysis

The performance metrics were tested in a time-series, assessing the difference between conditions every 333.33 milliseconds for the full 20-second testing period through the use of paired t tests. The tests were corrected using the Bonferroni method, resulting in an alpha of 0.00083, which is represented in the graphs as a horizontal dashed-dotted line in the middle graph of each figure. An uncorrected alpha level of 0.01 is shown as a horizontal black line in the same graph.

The unconventional use of paired t tests used in this study is motivated by the increase in temporal resolution compared to using larger bin sizes (which inflates standard deviation, which has an influence on the resulting test results) and fewer t tests. It must be noted that despite conservative corrections of the significance level, the results should only be seen as an indication of an effect rather than direct evidence. However, should the occurrence of a Type 1 error have been present, it should have shown as a sudden spike in the graph. The authors argue that the reliability of the analysis has been increased through the addition of an effect size metric (Cohen’s d) showing the magnitude of the difference between the conditions.

The level of significance is represented through the negative base-10 logarithm of p, where large values represent small p values in a similar fashion to the Manhattan plot (previously used in Gibson, 2010; Petermeijer, Cieler, & de Winter, 2017; Tanikawa et al., 2012). Cohen’s d was calculated to assess effect size between conditions in the following way:

C o h e n ′ s D = M 2 − M 1 S D p o o l e d

S D p o o l e d = ( ( n 1 − 1 ) S D 1 2 + ( n 2 − 1 ) S D 2 2 ) n 1 + n 2 − 2 .

Results

Discussion

The aim of this article was to assess whether drivers exhibit any of the aftereffects in terms of driver control as described by Hollnagel and Woods (2005) through steering behavior and lane positioning. This was motivated by research suggesting that drivers adapt their behavior proactively depending on the situation and time pressure (Cooper et al., 2009; Eriksson et al., 2014; Eriksson, Marcos, Kircher, Västfjäll, & Stanton, 2015; Kircher et al., 2017; Kircher, Larsson, & Hultgren, 2014; Wandtner et al., 2016). If drivers are allowed to self-pace the transition from automated to manual driving, we hypothesize that the aftereffects and the seemingly scrambled levels of control, such as harsh braking, sudden lane changes, and poor lateral control with large heading errors (e.g. Desmond et al., 1998; Gold et al., 2013, 2016; Naujoks, Mai, & Nekum, 2014), observed in contemporary research into system-paced transitions of control would be reduced. Indeed, according to Hollnagel and Woods, levels of control in a system are dependent on the available time and predictability of a situation, where larger temporal resolution is rewarded with higher, safer levels of control and vice versa. Hollnagel (1993) emphasizes that the “essence of control is planning” (p. 168), implying that forced transitions to manual control likely have a detrimental effect on driving performance.

The results of this study indicate that when drivers were able to moderate their transition to manual control by means of the time they took to resume control (between 1.9 and 25.7 seconds as reported in Eriksson & Stanton, 2017b), they maintained lateral positioning after the transition to a level comparable to when driving manually in both the passive monitoring and distracted driving condition (Figures 5, 6), indicating a higher level of control than what is reported in the literature utilizing shorter lead times of 5 to 7 seconds (Gold et al., 2013, 2016), or an immediate failure resulting in a system paced transition to manual (Desmond et al., 1998). However, it was also found that drivers exerted more effort maintaining their position in the lane after control was resumed as shown by the significant increase in standard deviation of steering angle in the two automated conditions compared to the manual driving condition (Figures 8, 9). The stability of vehicle position while exerting more effort maintaining control performance could indicate that drivers were at the opportunistic level of control, approaching the tactical level. Research by Russell et al. (2016) found that drivers could not maintain the same level of steering control as during manual driving. An alternate explanation of the increase in steering control inputs could be that drivers are carrying out an opportunistic recalibration by exploring the vehicle dynamics and whether there has been a change to the lateral control dynamics since manual control was ceded. This is somewhat supported by Russell et al., who hypothesized that drivers would rely on an inaccurate mental model before reaching their previous manual driving performance while adapting to the manual driving task to the detriment of normal “internally automated” performance (Norman, 1976, p. 70). Thus, the increase in standard deviation of steering angle may be attributed to the fact that drivers had to resume a closed loop activity after being removed from the task for some time and therefore showed larger steering inputs while maintaining lane positioning comparable to manual driving (Russell et al., 2016).

When compared to externally paced transitions in the literature, the results obtained in this study show similar effects as reported by Merat et al. (2014), who found that drivers who were asked to resume control in a predictable manner performed better in terms of lateral positioning and steering controllability than drivers who were not expecting a transition to occur, much like in driver-paced versus externally paced transitions of control. The results of this study do indicate lesser lane deviation post-TOR for the two automated driving conditions compared to what is reported in Merat et al. This discrepancy in magnitude between Merat et al. and the results presented in this article could potentially be partially explained by the different bin sizes as a larger bin size will inherently inflate the standard deviation and may skew the mean due to changes over time. Contrary to Desmond et al. (1998), little effect on lateral positioning was found, indicating that drivers were able to attain an opportunistic or tactical level of control rather than the seemingly scrambled level of control in Desmond et al. Similarly, Gold et al. (2013) found that drivers exhibited indications of scrambled control by performing sudden lane changes or unnecessary use of the hard shoulder when a short lead time was given. Evidently, there may be a relationship between the available time to resume manual control and the resulting level of control, where higher levels of control are attained when drivers have more time to resume manual control and vice versa. This relationship is further strengthened by the results of Gold et al. (2016), who found that drivers performed worse and had more crashes when asked to resume control when traffic density increases while subject to a fixed 7-second lead time before a crash would occur. The findings by Gold et al. (2016) serve as an example of a decay to scrambled control as traffic density increased. This is in line with Eriksson et al. (2015), who showed that drivers need more time to assess a situation when complexity is high.

When comparing the aftereffects of a driver-paced transition on lateral positioning between the two automated driving conditions, no significant differences could be identified in lane positioning (Figure 7) and standard deviation of steering angle (Figure 10) despite the significant increase in TOR response time of 1.5 seconds (range, 1.9–25.7 seconds, median = 4.5 seconds in the passive monitoring condition; and range, 3.1–20.9 seconds, median = 6.06 seconds in the secondary task condition) in the distracted driving condition reported in Eriksson and Stanton (2017b), a previous analysis of the experiment reported in this article focusing on the response time distributions and effects of secondary tasks on response time. This lack of difference could be attributed to the drivers being able to self-moderate the transition (Cooper et al., 2009; Eriksson et al., 2014, 2015; Kircher et al., 2014, 2017; Wandtner et al., 2016), thus extending the temporal horizon, which results in maintaining higher levels of control. Eriksson and Stanton argue that part of the additional time it took drivers to resume control in the secondary task condition could be attributed to the repositioning required by virtue of holding the magazine. Indeed, Petermeijer, Abbink, and de Winter (2015) argue that the repositioning phase should be accounted for as part of the control resumption process.

Given that drivers were completely removed from the visual scanning of the road environment by virtue of the secondary task, in addition to the automation of lateral and longitudinal vehicle control, we argue that the additional 1.5 seconds before resuming control could be partially explained by drivers extending the time horizon to ensure a higher level of control. Thus, the additional time allotted to assess the situation may have been crucial in maintaining the same level of opportunistic or tactical control performance as was observed in the passive monitoring for lane positioning and steering behavior as shown in Figure 7 and Figure 10. These findings indicate that drivers who are allowed enough time between a takeover request and the transition to manual control are able to self-pace the transition process, thereby attaining a higher level of control and reducing the aftereffects on control performance observed in contemporary research utilizing externally paced transitions with shorter lead times.

Conclusions

Contemporary research into externally paced transitions to manual control from automated driving have shown indications of reduced driver control levels through detrimental effects on driving performance when manual control was resumed. In light of this, this article assessed the influence of driver-paced, nonurgent transitions to manual control on driving performance post transfer of control as this is deemed to be the most common type of transition in SAE Level 3 and 4 systems (Eriksson et al., 2017; Eriksson & Stanton, 2017b).

It was found that drivers who were able to pace the transition process back to manual control, irrespective of whether they were engaged in a secondary task or not, exhibited significantly more steering corrections than manual driving conditions while maintaining comparable positioning in the lane. These findings contradict some of the contemporary literature showing that drivers exhibit harsh braking, sudden lane changes, and poor lateral control with large heading errors (Desmond et al., 1998; Gold et al., 2013, 2016). This indicates that drivers attained a higher level of control when resuming control in a self-paced setting compared to an externally paced transfer of control. When the control transition aftereffects were assessed between the passive monitoring and the distracted driving condition, no significant differences in either lateral positioning or corrective steering behavior could be found. However, as was reported in a previous analysis of the response time data in this study in Eriksson and Stanton (2017b), it was found that when drivers resumed control in the distracted driving condition, they took an additional 1.5 seconds to do so. This indicates that drivers took more time to complete the transition to extend the time horizon, thus ensuring that a sufficient control level was maintained.

In conclusion, these results indicate that drivers who are able to self-moderate the time required to transfer back to manual control exhibit less of the detrimental effects observed in externally paced conditions (e.g., those found in Desmond et al., 1998; Gold et al., 2013, 2016). This is promising as a higher level of vehicle control could reduce the risk of accidents in situations where lead times between a request to resume manual control and where manual control is required, such as near the edge of the operational design domain (SAE J3016, 2016) of the automated driving system. Such anticipatory behavior is in line with research by Kircher et al. (2014), who found that users of adaptive cruise control anticipated situations where the system would perform poorly and disengaged the system before such a situation could occur, thus strategically pacing their interaction with the system. The authors would like to argue that this type of behavior is needed from automated vehicles in higher levels of automation (SAE Level 4), where the automation recognizes that it is approaching its ODD and triggers a takeover procedure well in advance to avoid an emergency. Examples of such behavior could be to request manual control when approaching a motorway exit, approaching an area where lane markings are faded, or even when the system detects that it is approaching an area with poor visibility, such as fog.

It must be noted that the experimental design in this article was designed to compress experience and normalize the experience of the transition process to avoid novelty effects and thus is not completely representative with regard to the frequency of requests to transition between automated and manual modes that drivers might experience in normal usage of such a system and that more research must be carried out in more naturalistic conditions to further shed light on the control transition process in SAE Level 4 systems. However, the results support the argument of Eriksson & Stanton (2017b) that the 5th to 95th percentiles of the range of driver takeover response time variability must be accounted for in designing the nonurgent takeover process for SAE Level 3 and 4 systems as this would ensure safer transitions to manual vehicle control.

Key Points