subscribe: Posts | Comments | Email

EXPERIMENT 5: The effects of deceleration on braking reactions as a function of preferred time-headway

0 comments
EXPERIMENT 5: The effects of deceleration on braking reactions as a function of preferred time-headway

 

This is chapter 8 from the thesis “From adaptive control to adaptive traffic behaviour” about traffic psychology and behavioural adaptation of drivers, by Wim van Winsum. The thesis is from 1996. It describes a number of behavioural experiments into car driving that were performed in a research driving simulator.

Other chapters of this thesis can be found here:

 

Abstract 

The manoeuvre of braking for a decelerating lead vehicle was separated into three sequential processes that were manipulated independently. The initial time-headway to the lead vehicle at the moment it started to decelerate affected reaction time. Primary deceleration of the lead vehicle manipulated the duration of the open-loop phase. From the moment the driver touched the brake pedal, the deceleration of the lead vehicle was changed. This secondary deceleration was assumed to affect the closed-loop phase of braking. The hypothesis was that drivers who prefer a small time-headway during car-following (short followers) differ from drivers who prefer to follow at a large time-headway (long followers) in both the open- and closed-loop phases. In that case an interaction is expected between following group (short vs. long follo­wers) and primary deceleration on the duration of the open-loop phase and between following group and secondary deceleration on the duration of the closed-loop phase, the maximum brake force exerted and the number of movement corrections. In general terms, these predictions could not be confirmed. The lack of confirmation of the hypothesis is explained in terms of task characteristics that resulted in startle reactions and vigilance effects.

 

8.1 Introduction

 

A number of studies of car driving have shown that behaviour on the tactical level such as speed choice and choice of time-headway in car-following is sensitive to a variety of factors that affect operational perfor­mance. For example, marijuana affects operational car driving performance while it also results in driving with a lower speed and in choosing a larger time-headway (THW) during car-following (Smiley, et al., 1981; Smiley et al., 1985; Smiley et al., 1987). Time-on-task results in the choice of a larger time-headway during car-follo­wing, accompanied by verbal reports of performance decrements, drowsiness and exhaustion (Fuller, 1981). Brookhuis et al. (1991) reported an increase in THW when using a car telephone while driving. These fin­dings suggest that behaviour on the tactical level is used by the driver to compensate for effects on operational performance. For the case of car-following this means that any factor that may lead to performance decrements in braking for lead vehicles may result in a compensatory increase of time-headway. From the same perspective it was studied whether individual differences in preferred THW are related to individual differences in braking performance in a number of experiments (Van Winsum, 1996; Van Winsum and Heino, 1996; Van Winsum and Brouwer, 1996). In Van Winsum and Heino (1996) and Van Winsum and Brouwer (1996) preferred THW proved to be consistent within the driver. This means that drivers are consistently short or long followers and, thus, that individual differences in THW are consis­tent.

In Van Winsum (1996) it was found that, under the instruction to brake as fast as possible as soon as a deceleration of the lead vehicle was detected, reaction time and motor-response times are not different between short and long followers. Also, no evidence was found for differences in the informa­tion processing stages of stimulus encoding of a braking lead vehicle and motor adjustment. However, when the driver did not know the deceleration of the lead vehicle in advance, short followers generated a faster motor response, i.e. they moved the foot faster from the accelerator pedal to the maximum brake position when the lead vehicle decelerated. This suggests that short followers differ from drivers with a larger preferred THW in the transformation of visual feedback to a required motor response.

In Van Winsum and Heino (1996) it was found that the initiation and control of braking are both affected by time-to-collision (TTC) at the moment the lead vehicle starts to brake. This suggests that TTC information is used for judging the moment to start braking and in the control of braking. No evidence was found for differences between short followers and long followers in the ability to accurately perceive TTC. However, short followers were better able to program the intensity of braking to required levels and tuned the control of braking better to the development of criticality in time during the braking process. It was concluded that short followers may differ from long followers in programming and execution of the braking response as a function of TTC information.

Van Winsum and Brouwer (1996) analyzed the braking response in terms of three sequential phases in the braking process. The first phase consists of the interval between the moment the lead vehicle starts braking and the moment the driver releases the accelerator pedal. This is measured by the reaction time (RT). The second phase consists of the open-loop ballistic motor response. It is measured as the interval between the moment the accelerator pedal is released and the moment the brake pedal is touched and referred to as Brake Initiation Movement Time (BIMT). The third phase is a closed-loop motor response during which visual feedback is used to control the braking response while braking. The duration of the open-loop phase was strongly determined by the TTC at the moment the accelerator pedal was released, while the duration of the closed-loop phase was strongly determined by the number of movement corrections in the brake pedal signal. It was found that short followers exhibited a faster open-loop motor response which was not caused by a smaller TTC at detection time. Also, short followers generated a faster closed-loop response which was caused by fewer movement corrections. These results again supported the hypothesis that short followers differ from long followers in the efficiency of programming and execution of the braking response.

According to Van Winsum and Brouwer (1996), the duration of the three phases is affected by different factors. The RT interval is assumed to be affected by the THW at the moment the lead vehicle starts braking. This means that RT is expected to be faster if the momentary time-headway at the moment the lead vehicle starts to decelerate is smaller. The open-loop interval (BIMT) is assumed to be affected by the primary deceleration of the lead vehicle. If the lead vehicle decelerates more strongly, TTC at detection time is smaller resulting in a faster open-loop response. The closed-loop interval (BCMT) is affected by the number of movement correcti­ons. If the deceleration of the lead vehicle changes after the subject’s braking response has started, the speed and intensity of the braking response has to be changed, based on visual feedback. This means that a change in the level of deceleration after the braking response has started (secondary deceleration) is assumed to result in more movement corrections and thus affects the duration of the closed-loop phase. If short followers differ from long followers in both the open- and closed loop phases, then an interaction is expected between following group (short vs. long follo­wers) and primary deceleration on BIMT and between following group and secondary deceleration on BCMT, the maximum brake force exerted and the number of movement corrections. These hypotheses are tested in the present experiment.

 

8.2 Method

 

Subjects. Twenty-two subjects participated in the experiment. The subjects were selected from the TRC database by the following procedure. First a preselection was made on the basis of age and driving experience. Only subjects between 25 and 40 years of age with a minimum driving experience of 10000 km that were known not to be susceptible to simulator sickness were preselected from the database, resulting in 150 persons. These were send a small photo-preference test that measures preferred THW. This test consisted of 6 numbered photographs with scenes of a lead vehicle on a highway at different distances in front of the car. The preselected subjects were required to fill out on a form the number of the photograph that best matched the THW chosen by the subject while driving with a speed of 110 km/h on a highway and to return the form if they were intere­sted in participating in the experiment. This test procedure has been shown to result in a reliable estimate of preferred time-headway during car-following on the road (Heino et al., 1992). From the returned forms, 11 subjects with a small preferred THW and 11 subjects with a larger preferred THW were invited for participation in the experiment. Subjects with a preferred photo number of less than or equal to 3 were assigned to the group of ‘short followers’. Subjects who preferred photo number 5 or 6 were assigned to the group of ‘long followers’, see table 1. These two groups are referred to as ‘THWpref groups’.

 

Table 1. Relation between photo number and headway on the photo preference test and number of subjects. DHW=distance-headway in meters, THW=time-headway in seconds.

 

 

Photo number      DHW        THW         number of subjects

 

1                            6                0.20          0

2                            11              0.36          2

3                            25              0.81          9

4                            33              1.08          0

5                            45              1.47          4

6                            65              2.13          7

 

 

Apparatus. The experiment was performed in the driving simulator of the Traffic Research Centre (TRC). This fixed-based simulator consists of two inte­grated subsys­tems. The first subsystem is a conventional simulator composed of a car (a BMW 518) with a steering wheel, clutch, gear, accele­ra­tor, brake and indica­tors connected to a Silicon Graphics Skywriter 340VGXT compu­ter. A car model converts driver con­trol actions into a displacement in space. On a projection screen, placed in front, to the left and to the right of the subject, an image of the outside world from the perspective of the driver with a horizontal angle of 150 degrees is projected by three graphi­cal videopro­jectors, controlled by the graphics software of the simulator. Images are presented with a rate of 15 to 20 frames per second, resulting in a sug­gestion of smooth move­ment. The visual objects are buil­dings, roads, traffic signs, traffic lights and other vehicles. The sound of the engine, wind and tires is presented by means of a digital soundsam­pler recei­ving input from the simulator computer.

The second subsystem consists of a dynamic traffic simu­la­tion with interacting artificially intelligent cars. For experimental purpo­ses different traffic situa­tions can be simulated. The simula­tor is described in more detail elsewhere (Van Wolffe­laar & Van Winsum, 1992 and Van Winsum & Van Wolf­felaar, 1993).

 

Procedure. The circuit was made of two-lane roads with a lane-width of 3 m., and straight road sections alternated with left-turning curved road sections. All roads had delineation with broken center lines and closed edge lines. Before the experiment started subjects practiced braking several times by approaching a traffic light that turned on red after a certain time-to-intersection was exceeded. This required the subjects to come to a full stop. After this, preferred time-headway was measured by the following procedure. The subject was instructed to drive with a fixed speed of 100 km/h, while continuously being overtaken by other cars. One of these cars merged in front of the subject and adopted a time-headway of 2 s. The subject was asked to rate on a scale from 1 to 10 how well the present THW resembled the THW normally adopted by the subject in similar situations on the road. If THW was too small it was increased with 0.5 s. If THW was too large it was decreased by 0.5 s. After this the subject was again asked for a rating. This continued until a definite peak was found in the subject’s rated THW, i.e. until the preferred THW was found. After this the braking trials started. The subject was instructed to drive with a constant speed of 100 km/h, to stay in the right lane and to avoid a collision with a lead vehicle in case it braked. While driving, the subject was overtaken by another vehicle every 5 seconds on average. The lead vehicle merged in front of the lead vehicle and started to drive at a fixed THW of either 0.8 or 1.2 seconds (THW condition). After a stable THW was reached it either pulled up again or it braked from 100 to 60 km/h. Braking occurred on average once in every 5 minutes. The lead vehicle applied either a deceleration of 3 or 6 m/s² (initial deceleration). After the subject touched the brake in response to the lead vehicle, the deceleration of the lead vehicle changed either to 3 or 6 m/s² (secondary deceleration), resulting in the following deceleration patterns for both THW conditions (0.8 vs. 1.2 s.): 3 to 3 m/s², 3 to 6 m/s², 6 to 3 m/s² and 6 to 6 m/s². Thus, the driver was subjected to a total of 8 braking trails, that were counterbalan­ced.

 

Data collection and analysis. During the braking trials the following data were sampled with a frequency of 50 Hz.: speed of the simulator car and the lead vehicle in m/s, accelera­tor pedal position, brake pedal position and force excerted in Nm, accele­ration in m/s², time-to-collision (TTC) and bumper to bumper distance from the lead vehicle. At t0 the lead vehicle started to brake.

 

Figure 1. Time history of braking and dependent variables.

The moment the accelerator pedal position was less then 4% after t0 was registered as tacc, and RT was computed as tacc-t0. The moment after tacc at which the brake pedal force was more than 3 Nm, was registered as tbr (the moment the brake pedal was touched). BIMT (Brake Initiation Movement Time, or the open-loop ballistic response) was computed as tbr-tacc. The maximum brake force was detected on-line and the moment this was reached was registered as tmaxbr. BCMT (Brake Control Movement Time, or the closed-loop braking response) was computed as tmaxbr-tbr. The maximum brake force excerted, MAXBRFO, was stored as well. During the closed-loop phase a number of decelerations typically occur in the brake pedal signal. These decelera­tions reflect movement velocity correc­tions. The number of decelerati­ons in the brake pedal signal (NRCOR) was analyzed, together with the maximum deceleration (DECmax) of the simulator car, the minimum TTC (TTCmin) and the minimum distance to the lead vehicle (DISmin) during the braking maneuver. In previous studies, the TTC on the moment the driver initiates braking, that is, TTC on tacc (TTCtacc) has proven to be an important variable controlling subsequent phases in the braking process. For this reason TTCtacc was analyzed as well. The time-history of braking can be seen in figure 1.

The dependent variables were analyzed with repeated measures analysis of variance with THW, initial deceleration and secondary deceleration as within-subjects factors, and THW-group as a between-subjects factor.

 

8.3 Results

Characteristics of THWpref groups. Table 2 presents the results of analysis of variance and the averages of THW as measured in the simulator, age, number of years licensed and kilometrage per year as a function of THWpref groups.

Table 2. Statistical effects and averages of THW as measured in the simulator, age, number of years licensed and kilometrage per year as a function of THWpref groups.

Dependent variable                F(22,1)     short                  long

 

THW simulator                      34.24**      1.5             2.9

Age                                            0.03        31.6          31.3

Years licensed                         0.65        12.8          11.3

Annual kilometrage               5.30*       28084       13227

 

** = p<0.01; * = p< 0.05.

 

There was a strongly significant difference between THWpref groups on the preferred THW as measured in the simulator. This supports the validity of the simulator for measuring car-following behaviour. There were no signifi­cant differences in age or number of years licensed to drive a car between short and long followers, but short followers drove significantly more kilometers per year.

 

Effects of manipulations. Table 3 lists the main effects of the manipulations on the dependent variables. The averages of RT, BIMT, BCMT, MAXBRFO and NRCOR as a function of the manipulations are shown in table 4. RT was significantly affected by the factor THW (0.8 vs. 1.2 s): a smaller THW at which the lead vehicle started to brake resulted in a smaller RT. BIMT was both affected by THW and by initial decele­ration: a smaller THW and a larger initial deceleration both resulted in a smaller BIMT. These effects match the significant effects of THW and initial deceleration on TTCtacc.

 

Table 3. Main effects of manipulations on dependent variables. ** = p<0.01; * = p< 0.05,

dec-1 represents primary deceleration and dec-2 secondary deceleration.

 

Dependent                     Independent       F(21,2)

 

RT                                  THW                  8.44 **

dec-1                 0.57

dec-2                 3.63

BIMT                             THW                  20.06 **

dec-1                 17.68 **

dec-2                 0.19

BCMT                            THW                  0.11

dec-1                 19.23 **

dec-2                 0.73

MAXBRFO                   THW                  26.83 **

dec-1                 41.68 **

dec-2                 71.64 **

NRCOR                         THW                  0.04

dec-1                 7.65 *

dec-2                 7.65 *

TTCmin                            THW                  14.89 **

dec-1                 77.23 **

dec-2                 35.13 **

DECmax                          THW                  17.43 **

dec-1                 65.31 **

dec-2                 57.20 **

DISmin                            THW                  81.40 **

dec-1                 80.23 **

dec-2                 9.02 **

TTCtacc                           THW                  15.69 **

dec-1                 160.42 **

dec-2                 0.86

 

Thus, if criticality, measured by TTCtacc, is higher the open-loop ballistic motor response is faster. The duration of the closed-loop phase, BCMT, was only significantly affected by initial deceleration, but not by THW or secondary decelera­tion: a larger initial deceleration resulted in a smaller BCMT. The maximum force, MAXBRFO, excerted on the brake pedal was signifi­cantly affected by all independent factors. Thus, a higher secondary deceleration, after the brake pedal was touched, resulted in a higher maximum brake force instead of a faster BCMT. The number of decelerations in the brake pedal signal (NRCOR) was both affected by initial and secondary deceleration. A larger initial and secondary deceleration both resulted in fewer decelerations in the brake pedal signal.

 

Table 4.  Averages as a function of the manipulated factors time-headway on which lead vehicle starts to decelerate, initial deceleration and secondary deceleration. RT, BIMT and BCMT in seconds, MAXBRFO in Nm. dec-1 represents primary deceleration and dec-2 secondary deceleration.

 

THW                                             0.8                                                                   1.2                           

dec_1                          3                                  6                                  3                                  6             

dec_2                 3                6                3                6                3                6                3               6      

 

RT                      0.73          0.79          0.74          0.88          0.91          1.05          0.91          0.83

BIMT                 0.63          0.73          0.49          0.50          0.98          0.89          0.66          0.58

BMCT                1.64          1.50          1.13          1.16          1.43          1.69          1.09          1.30

MAXBRFO       51.68        113.22      103.91      218.22      48.26        87.21        80.24        98.21

NRCOR             3.68          2.77          3.09          2.32          3.73          3.05          2.73          2.55

 

 

This indicates that this variable basically measures the necessity to move the pedal straight to the maximum without hesitation. Finally, a smaller initial THW resulted in a smaller minimum TTC, a larger deceleration and a smaller minimum distance to the lead vehicle. Similar effects were found for a larger initial deceleration by the lead vehicle and a larger secondary deceleration by the lead vehicle.

 

Effects of THWpref group. Table 5 lists the effects of THWpref group and interactions between THWpref group and the independent factors on the dependent variables.

 

Table 5. Main effects of THWpref and interactions on depen­dent variables. ** = p<0.01;

* = p< 0.05, dec-1 represents primary decele­ration and dec-2 secondary deceleration.

 

Dependent            Independent                F(21,2)

 

RT                         THWpref                     0.10

THWprefxTHW           0.08

THWprefxdec-1          0.60

THWprefxdec-2          0.25

BIMT                    THWpref                     0.36

THWprefxTHW           0.09

THWprefxdec-1          0.11

THWprefxdec-2          0.01

BCMT                   THWpref                     2.04

THWprefxTHW           0.13

THWprefxdec-1          0.30

THWprefxdec-2          0.67

MAXBRFO          THWpref                     0.76

THWprefxTHW           1.15

THWprefxdec-1          0.02

THWprefxdec-2          0.10

NRCOR                THWpref                     4.47 *

THWprefxTHW           0.45

THWprefxdec-1          8.18 **

THWprefxdec-2          4.03 *

 

 

There were no significant main effects of THWpref group on RT, BIMT, BCMT and MAXBR­FO. Also none of the interactions of THWpref with the independent factors reached significan­ce on any of these dependent variables. This means that these results do not support the hypotheses mentioned in the introduction. However, NRCOR, the number of movement corrections during the closed-loop phase, was significantly affected by THWpref group and revealed significant interactions of THWpref group with initial and secondary deceleration, see figure 2. The effects on NRCOR were as follows: only for the group of short followers was NRCOR affected by dec-1 (F(10,1)=20.65, p<0.001) and by dec-2 (F(10,1)­=12.08, <0.006). For the group of long followers, the effects of both dec-1 and dec-2 on NRCOR were not significant (F(10,1)=0.05, p<0.822 for dec-1 and F(10,1)=0.46, p<0.512 for dec-2).

 

These effects strongly indicate that long followers moved their foot directly to the maximum brake position, irrespecti­ve of the development of criticality in time, while short followers were more sensitive to the manipulations of initial and secondary deceleration on this measure.

Figure 2. NRCOR as a function of THWpref group, initial deceleration (dec-1)

and secondary deceleration (dec-2).

 

 

8.4 Discussion and conclusions

 

Based on a number of previous experiments it was tested whether short followers differ from long followers in both the open-loop and closed-loop phases of the braking process. This was tested by manipulating these phases. The open-loop phase was manipulated with two levels of initial deceleration of the lead vehicle. After the brake was touched by the subject, the deceleration of the lead vehicle changed. This secondary deceleration manipulated the closed-loop phase of the braking response. The hypotheses were:

1) there is an interaction between following group and primary deceleration on the duration of the open-loop phase (BIMT), defined as the interval between the moment the foot is released from the accelerator pedal and the moment the foot touches the brake pedal. This would support the idea that short followers differ from long followers in the open-loop phase.

2) there is an interaction between following group and secondary decelera­tion on the maxim­um brake force excerted, the number of movement corrections during the closed-loop phase and the duration of the closed-loop phase (BCMT), defined as the interval between the moment the foot touches the brake pedal and the moment the maximum brake force is excerted. This would support the idea that short followers differ from long followers in the closed-loop phase.

In general, these hypotheses were not supported. There was no signifi­cant interaction between following group and any of the independent factors, initial THW, primary and secondary deceleration, on RT, BIMT, BCMT and the maximum brake force. However, the interaction between following group and initial deceleration on the number of movement corrections was significant as was the interaction between following group and secondary deceleration on this variable. The number of movement corrections (NRCOR) during the closed-loop phase were conceived as an expression of uncertainty induced by a change in deceleration after the braking response was initia­ted. Although NRCOR was affected by secondary deceleration, it was also affected by primary deceleration. The pattern of effects suggests that NRCOR expresses the necessity to move the pedal straight to the maximum without hesitation. The results showed that only the group of short followers was sensitive to the effects of initial and secondary decelera­tion on NRCOR, while the long followers moved their foot to the maximum with the same number of movement corrections independent of primary and secondary deceleration. In a previous study it was found that NRCOR strongly determines the duration of the closed-loop phase. From this perspective, it would be expected that NRCOR and BCMT are affected by following group and the independent factors in a similar way. However, as was already apparent, there were no significant effects of following group on BCMT. Closer inspection of the data revealed that only in the trials where the secondary deceleration was high, the correlations between NRCOR and BCMT were significant, see table 6.

 

Table 6. Correlation between NRCOR and BCMT, depending on THWpref

group and secondary deceleration (dec-2).

 

dec-2

  3              6         

short                      0.30          0.68 **

long                       0.30                   0.59 **

 

 

This suggests that a causal relation between NRCOR and BCMT only exists if criticality is high enough.

 

The lack of support for the hypotheses may have been caused by specific task related factors. The subjects generally described the task as boring, mainly because of the long task duration and the low event-rate. There were only 8 braking trials over a duration of 45 minutes. This may have resulted in a vigilance task with two separate effects. On the one hand, the braking trials may have generated startle reactions, resulting in fast responses irrespective of the manipulations. On the other hand some responses may have been slow because of state-related factors. This would have resulted in a high variance in the data that was not caused by the manipulations of THW, initial and secondary deceleration. Figure 4 illustrates the distribution of RT as a function of the THW manipulation. It can be seen that the distributions are skewed on the right side, especially for the THW=1.2 condition, sugge­sting low-vigilance effects, although there are two distinctive peaks in the distributions.

 

Figure 4. Distributions of RT as a function of the THW manipulation.

 

Figure 5 and 6 show the distributions of BIMT as a function of initial deceleration for THW=0.8 and THW=1.2 respectively. These figures show that the distributions of BIMT are skewed on the right side and that the effects of initial deceleration are mainly caused by ‘outliers’ on the right side of the distributi­on. Especially for the THW=0.8 condition, the primary peak of the low deceleration condition occurs before the primary peak of the high deceleration condition, which obviously is not in the expected direction and opposite to the effects of the analyses of variance, which are based on the means. The distributions of BIMT in the THW=0.8 condition are visualized separately for the short followers and long followers in figure 7.

 

It can be seen that for the short followers there are two distinctive peaks as a function of initial deceleration in the expected direction, while for the long followers the primary peaks overlap. Moreover, it can be seen, that the primary peak in the BIMT of long followers occurs before the peaks of the short followers. This suggests that BIMT of long followers has suffered more from startle reactions resulting in BIMTs that were fast and not tuned to differences in initial deceleration, while the BIMTs of short followers were more sensitive to initial deceleration.

Thus, the distributions of the data and task-induced startle responses and low-vigilance effects may have given results that failed to support the hypotheses. This will be tested in the next experiment, with multiple measurements per manipulation, a higher event-rate and shorter task duration in order to prevent undesirable state-related effects.

Figure 5. Distribution of BIMT as a function of initial deceleration for the THW=0.8 conditi­on.

Figure 6. Distribution of BIMT as a function of initial deceleration for the THW=1.2 conditi­on.

Figure 7. Distribution of BIMT as a function of initial deceleration for the

THW=0.8 conditi­on, for short and long followers.


EXPERIMENT 4: Time-headway in car-following and operational performance during unexpected braking

0 comments
EXPERIMENT 4: Time-headway in car-following and operational performance during unexpected braking

This is chapter 7 from the thesis “From adaptive control to adaptive traffic behaviour” about traffic psychology and behavioural adaptation of drivers, by Wim van Winsum. The thesis is from 1996. It describes a number of behavioural experiments into car driving that were performed in a research driving simulator.

Other chapters of this thesis can be found here:

 

The relation between choice of time-headway during car-follo­wing and the quality of braking skills was studied in a driving simulator. The theoreti­cal perspective was that individual differen­ces in behaviour on the tactical level may be related to skills on the operational level of the driving task via a process of adaptation. In a sample of 16 young and middle-aged experienced drivers independent asses­sments were made of preferred time-headway and braking skill. Starting from modern theories of visual-motor learning, braking skill was analyzed in terms of a reaction time component, an open-loop visual-motor component, and a closed-loop visual-motor component involving the precise adjustment of braking (timing and force) to the situation. The efficiency of the visual-motor component of braking was a strong and significant predictor of time-headway in such a way that more efficient braking indicated a shorter preferred time-headway. This result appears to support the adaptation theory on an individual level.

 

7.1 Introduction

For many years it has been realized that individuals who have very good driving skills in the sense of great fluency and agility in performing the basic driving tasks of visual orientation and vehicular control, are not necessarily safe drivers (Williams and O’Neill, 1974; Evans, 1991). Traf­fic safety de­pends on what the driver will do in a given situa­tion and not on the maximum level of performance (Nääta­nen and Summala, 197­6) or, as Evans (1991) puts it, what is cruci­al is not how the driver can drive but how the driver does drive. The failure of driver skill models in explaining accident involvement has been attribu­ted to various adaptive mechanisms. For example, drivers with poor skills might compensate by driving slower, or, the other way around, very skilled drivers might tend to drive very fast. Because traditional skill models do not incorporate such compensatory mechanisms they are not suitable for assessing and understanding individual differences in the safety of traffic behaviour.

A solution to this problem may be the application of the hierarchical framework discussed by Michon (1985) to driving behaviour (Ranney, 1994). In this framework driving is viewed as a hierarchically organized set of tasks on the strategic, tactical and operational level. On the strategic level trip planning and the selection of trip goals and route occur. The tactical level includes, for in­stance, choice of speed on straight roads and in curves and choice of headway in car-follo­wing. Steering and control­ling the timing and intensity of braking are activities on the operational level. Traditionally the study of driving skill was aimed as the efficiency of performance on the operational level. However, in this article the importance of the interrelation between behaviour on the operational and the tactical level is stressed.  In this framework the adaptation problem may be understood as a compensation on the tactical and strategic levels of the driving task for individual differences in skills on the operational level (Brouwer et al., 1988). This adaptation theory has been used as an explanati­on for the relatively safe driving records of functionally impaired drivers. The link with driver safety now becomes clear. Driver safety may be defined in terms of the relationship between operational level skills and choices and preferences on the tactical level.

Recently, Van Winsum and Heino (1996) found some evidence with regard to the relationship between individual differences in operational level skills and tactical behaviour which appears to fit an adaptation theory on the individual level. In a study on time-headway in car-following, they found evidence for a relationship between braking skill and choice of time-headway. Since time-headway (THW), defined as the time interval between two vehicles in car-following, represents the time availa­ble to reach the same level of deceleration as the lead vehi­cle in case it brakes, they studied whether choice of THW is related to time-critical skills underlying braking performance. THW was constant over the range of speeds studied. Drivers were con­sis­tent in their choice of THW, evidencing systematic individual diffe­rences in choice of THW during car-following. The results sugge­sted differen­ces in skills related to the motor control of braking as a function of preferred time-headway. What was lacking in this study was a specific model of the braking skills so that it was difficult to pinpoint in which respect drivers with short THW differed from those with long THW. In the present study a model for the decomposition of perceptual-motor processes in braking is proposed and individual differen­ces in the effi­ciency of such processes are related to the choice of time-headway in a free field situation.

Braking for a decelerating lead vehicle requires substan­tial perceptu­al-motor skills because of the dynamic task environ­ment. Lee and his co-workers have shown in a number of publications that a perceptu­al varia­ble, named tau, which is the inver­se of the expansion of the retinal image, is used in action. This variable directly specifies time-to-contact in dynamic situations. Thus, perception is assumed to guide action and this relation between tau and action has been established in a number of different tasks such as long jum­pers running up to a take-off board (Lee et al., 1982) and jum­ping up to hit a falling ball (Lee et al., 1983). Also, Bootsma and Van Wieringen (1988) found that time-to-contact plays an important role in the guidance of actions of an experienced table tennis player. In car driving and braking the equivalent of time-to-contact is time-to-collision (TTC). Lee (1976) suggested that TTC information is used by the driver in the initia­tion and control of braking. Van Winsum and Heino found that the initiation and control of braking for a decele­rating lead vehicle was very sensitive to TTC information. The timing of the initiation of the braking response was equally sensitive to TTC for short follo­wers and long follo­wers. However, short followers were more efficient in the control of braking, braked harder and adju­sted the inten­sity of braking better to the critica­li­ty (as measured by TTC) at the moment the lead vehicle started to decelerate, compared to long followers. So, it appeared that the difference between the short and long followers was in the execution of the motor response.

­   Substantial individu­al differen­ces in the ability to accu­rate­ly estimate TTC have been reported in the literature (for instance see Schiff and Detwiler, 1979). A general finding is that TTC is unde­res­timated with a constant proporti­on. TTC esti­mation is more accurate for smaller TTC’s, see for in­stance, McLeod and Ross (1983), Cavallo et. al (1986), Hoffm­ann and Mortimer (1994). Given this evidence, the re­sults of Van Winsum and Heino could have been af­fected by the fact that the TTC at the moment the lead vehicle started to brake (t0) was smaller for short follo­wers alt­hough the distan­ce at t0 was the same for all subjects. This was caused by a higher speed on average for short follo­wers on t0. Theoreti­cally, short followers were thus in a positi­on to estima­te TTC more accurately. In additi­on, since TTC was smaller for short follo­wers they may have been forced to brake harder and more accu­ra­tely. In order to control for effects of diffe­rential criti­cality and accu­racy of TTC estimation, all dri­vers will be subjec­ted to the same high level of criticali­ty in the present study.

In dynamic situations such as braking for a decele­rating lead vehicle, following the initial reaction of releasing the accelerator, the motor response is assumed to consist of two phases, i.e. an open-loop and a closed-loop phase. We attempt to separately assess these three processes by analyzing the braking response in terms of Reaction time (RT), the Brake Initiation Movement Time (BIMT) and the Brake Control Movement Time (BCMT) (see Figure 1). Starting from the adaptation theory we expect that the quality of these processes is related to preferred time-headway: specifically we hypothesize that preferred time-headway (behaviour on the tactical level) can be predicted from the BIMT and the BCMT (performance on the operational level). To be able to assess the reliability of preferred THW as an indicator of a stable individual characteristic, it is measured at four different speeds. It is expected that the results of Van Winsum and Heino concer­ning the con­sistency of THW and the constancy of THW over speed will be replica­ted.

Reaction time (RT) represents the interval between stimulus presentation and movement initiation. Several information processing stages including response selection and response preparation, together called motor program­ming, occur within this interval. Motor programming time, as a part of RT, is assumed to be related to temporal complexity and organization of the movement to be executed, but not with physical task dimensions such as distance (Kerr, 1978). This suggests that the time associated with parameter setting for a generalized motor program does not vary for different parameter values. Thus, TTC is not expected to affect RT because TTC is assumed to determine the speed parameter value for the generalized motor program.

The Brake Initiation Movement Time (BIMT) is used to operationally define the open-loop phase under the control of the genera­lized motor program for braking of which the speed parameter is set by TTC informa­tion. During this phase the influence of feedback is absent. Because of the time charac­teristics of the braking response the open-loop phase is defined here as the interval between the moment the driver withdraws the foot from the accele­rator pedal and the moment the brake pedal is touched. The duration of this phase is then assumed to be depen­dent on TTC at the moment the driver de­tects the deceleration of the lead vehicle or at the moment the driver decides to brake.

Error detection and error cor­rection are assumed to take place during the closed-loop phase, operationally defined here as the Brake Control Movement Time (BCMT). This is the inter­val between the moment the brake pedal is touched and the moment the brake maximum is reached. Since the environmen­tal goal of the movement is to avoid a collision and to keep sufficient dis­tance to the lead vehicle, TTC infor­mation is possibly used during this feedback process. Accor­ding to Hayes and Marteni­uk (1976) movement control complexity can be viewed as the informa­tion load imposed on the performer by the necessity to detect and correct movement errors. For more skilled operators movement time decreases because of a decrease in the number of movement corrections (Keele, 1968). During the closed-loop phase of the braking response, movement time is then expected to be related to the number of movement correcti­ons.

 

7.2 Method

 

Subjects. Sixteen (8 male, 8 female) subjects participated in the experi­ment. The average age was 33.6 years (sd. 6.1, range 22-47). They had held a driving license on average for 11.6 years (range 2-27). The average annual kilometrage was ap­proximately 10083 kilometers (range 1500-30000).

Apparatus. The experiment was performed in the Traffic Research Centre (TRC) fixed-based driving simulator. It consists of a car (BMW 518) with a steering wheel, clutch, gear, accele­ra­tor, brake and indica­tors connected to a Silicon Graphics Skywriter 340VGXT compu­ter. A car model converts driver control actions into a displacement in space. On a 2 x 2.5 meter projection s­creen, placed in front of the car mockup, an image of the outside world with a horizon­tal angle of 50 degrees is projec­ted by a graphical videopro­jector, controlled by the 3D-grap­hics software. Images are presented with a rate of 15 to 20 frames per second, resulting in a sug­gestion of smooth move­ment. The visual objects are buil­dings, roads, traffic signs, traffic lights and artificially intelligent traffic. The sound of the engine, wind and tires is presented by means of a digital soundsampler recei­ving input from the simulator computer. The simula­tor is described in more detail elsewhere (Van Wolffe­laar & Van Winsum, 1992 and Van Winsum & Van Wolf­felaar, 1993).

Procedure. A circuit of two-lane roads (lane-width 3 meters) with broken center lines and continuous edge lines was used. Since the subjects had participated in another simulator experiment not involving car-follo­wing prior to the present experiment, they were already sufficiently practi­ced. First, preferred time-headway was measured as a function of different speeds. Subjects were instructed not to overtake other vehicles, to respect the speed limit of 80 km/h and to follow other vehicles at a safe distance. While driving, the sub­jects approached vehicles that were parked on the right shoulder. At a distance of 200 meter these vehicles accelerated to a fixed cruising speed and merged in front of the simulator car. There were four of these trials that differed in the cruising speed of the lead vehi­cles. The order of speeds was 60, 40, 70 and 50 km/h for all subjects (speed condition). In every trial, time-headway was measured during 5 minutes.

After this, a vehicle, driving with a speed of 60 km/h, was approached. Prompted by the experi­menter the subjects were asked to rate the danger of the present headway on a scale from 1 to 5. Then they were requested to drive a bit closer and again asked to give a rating. This continued until a time-headway of 0.6 seconds was reached. At that moment the lead vehi­cle suddenly decelerated unexpectedly from 60 km/h to 30 km/h with a decelera­tion of 6 m/s². This constitutes the braking condition. The aim of this procedure was to ascertain a fixed time headway at the moment the lead vehicle started to brake for all subjects.

Data registration and analysis. Speed, brake and accele­rator pedal signal (percenta­ge pres­sed), distance-headway, time-headway and time-to-collision were sampled with a fre­quency of 10 Hz. Average THW was computed, for the four trials in the speed condition, from the moment the simula­tor car reached the same velocity as the lead vehicle until the lead vehicle left the road. THW’s were averaged over the four speed trials to compute preferred time-headway (THWpref). The effect of speed on THW was tested with multiva­riate analysis of variance with repeated measurements.

Figure 1 shows the time-history of braking together with a number of dependent variables. In the braking condition, t0 represents the moment a THW of 0.6 was reached. On t0 the lead vehicle started to brake. The moment the acce­lerator pedal was 5% less than the position on t0 represents tacc. Reaction time (RT) was calculated as the interval between t0 and tacc. The moment the brake pedal was pressed more than 5% is indicated as tbr. The interval between tacc and tbr represents the open-loop phase of the movement and is referred to as brake initiation movement time (BIMT). The moment the maximum brake position was reached is indicated by tmaxbr. The duration of the closed-loop phase, brake con­trol movement time (BCMT), was computed as the inter­val between tbr and tmaxbr. Move­ment time (MT) was compu­ted as the sum of BIMT and BCMT. TTC on tacc is referred to as TTCacc.

During the closed-loop phase the brake pedal is pressed. A typical time-history of this is presented in figure 2. It shows the percentage at which the brake pedal is pressed toge­ther with the velocity of pressing the brake pedal and accelera­tion of brake pedal signal as a function of time. The number of decelerations in this signal reflect the number of movement (speed) corrections. The effect of driver reactions to braking by the lead vehicle on THWpref were tested with multiple regression analysis.

 

Figure 1. Time-history of the braking maneuver. Vlead represents speed of lead vehicle in m/s, accel represents accelerator pedal position.

 

 

Figure 2. Brake pedal signal, velocity and acceleration of bra­king as a function of time during the closed-loop phase.

 

7.3 Results

Before testing the main hypothesis a preliminary assessment is made of the reliability of THWpref.

Reliability of preferred THW. THW was not significantly affected by the speed of the lead vehicle (F(3,15)=1.20, p>=0.352), where­as distance headway significantly increased with speed (F(3,15)=­20.20, p<0.00­01). This means that THW was constant over speed. The test’s reliability index (Cron­bach’s alpha) for the four measurements of THW over different speeds was 0.89, while the standardized alpha was 0.90. This was taken as eviden­ce that all THW’s are an expression of a subjects’ general preferred THW. THWpref was computed as the average THW over the four speed trials.

The correlations of THWpref with annual kilometrage and number of years licensed were not statistically significant (R=0.13 resp. -0.02). This means that preferred time-headway was not related to driving experience. Also, none of the braking-related variables correlated significantly with driving experience.

 

The relationship between preferred THW and braking skill. It is first tested whether the duration of the open-loop phase is determined by TTC after the RT interval and whether the duration of the closed-loop phase is affected by the number of movement corrections as predicted by the braking model presented in the introduction. The regression coefficients (Beta weights) are presented in table 1.

 

Table 1. Effects of regression analyses of TTCacc and movement corrections on

the duration of the open-loop (BIMT) and closed-loop (BCMT) phases.

 

Dependent   Independent       R(=Beta)  F      

 

BIMT           TTCacc                0.81          27.11 **

BCMT          nr corr               0.83          30.09 **

 

nr corr = number of movement corrections

** = p < 0.01;

 

From inspection of table 1 it appears that the duration of the open-loop phase was strongly determined by TTC at the moment the accelerator was released. The duration of the closed-loop phase was strongly determined by the number of movement corrections. This confirms the model of braking discussed in the introduction.

It was then tested whether operational braking performance affected choice of time-headway. For this the regressions of RT, BIMT and BCMT on THWpref were analyzed, controlling for the effects of TTCacc and number of movement corrections. In this way the direct effects the independent variables on THWpref could be established. Table 2 lists the effects of RT, BIMT and BCMT on THWpref. This table should be read as follows. The first column lists the dependent variable. The second column lists the indepen­dent variables in order of inclusion in the regressi­on equati­on. R represents multiple correlation after addition of the independent variable, and F represents the accompany­ing F statistic for the whole regression equation. Beta and T represent the Beta weight and t value when all dependent variables are included in the equation.

 

Table 2. Effects of multiple regression analyses of brake-related times on THWpref.

 

Dep              Indep         R                F               Beta         T           

 

THWpref       RT             0.31         1.50          -0.31                  -1.23N.S.

 

THWpref       BIMT        0.53         5.37          1.12                  3.25**

TTCacc       0.68         5.64          -0.74                  -2.13*

 

THWpref       BCMT       0.49         4.39          0.49                  2.09*

nr corr      additional contribution too small

for inclusion

THWpref       MT            0.67          11.24         0.67                  3.35**

 

* = p < 0.05; ** = p < 0.01; N.S. = not significant

 

Figure 3. Path diagram with partial regression coefficients.

*=p< 0.05, **=p< 0.01, ns=not significant.

 

There was no significant effect of RT on THWpref. The effect of BIMT on THWpref, with TTCacc controlled for, was statistically significant. This means that subjects with a faster open-loop motor reaction preferred a smaller THW. This was not simply caused by a smaller TTCacc for subjects with a smaller THWpref, since there also was a significant direct effect of TTCacc on THWpref indicating that drivers with a larger TTCacc tended to have a smaller preferred time-headway. The effect of BCMT on THWpref also was statistically significant. There was no significant direct effect of number of movement corrections on preferred time-headway. This means that the faster closed-loop motor response of drivers with a smaller preferred time-headway was caused by a smaller number of movement corrections. The effects of total movement time (MT) on THWpref also was statistically significant. The path diagram of dependent variables on preferred time-headway is presented in figure 3.

 

7.4 Discussion and conclusions

The hypothesis that preferred time-headway is consistent within the driver and constant over different speeds during steady-state car-following was confirmed for the range of speeds (40, 50, 60, 70 km/h) examined in the present experi­ment. This replicates the results of Van Winsum and Heino (1996).

Reaction time, i.e. the difference between the moment the lead vehicle started braking and the moment the accelerator was released, was not related to preferred time-headway. This confirms the results of Van Winsum and Heino that short follo­wers do not differ from long followers in perceptual mechanisms related to time-to-collision (TTC) detecti­on. The open-loop phase of the motor response was very sensitive to TTC, and especially to TTC at the moment the foot was released from the accelerator pedal. This suggests that as soon as the driver detects the deceleration of the lead vehicle, the speed parameter of the generalized motor program is set as a function of TTC. Drivers who moved their foot faster to the brake pedal had a smaller preferred time-headway. The direct effect of TTC at the moment the accelerator was released on preferred time-headway indicates that the effect of the duration of the open-loop phase on preferred time-headway was not caused by a smaller TTC for short followers. This suggests that drivers with a smaller preferred time-headway program the movement speed of the foot to a higher level than drivers with a longer preferred time-headway. This suggests differences in the transfor­mation of perceptual information into the adjustment of the speed parame­ter.

The duration of the closed-loop phase of the motor response was strongly related to the number of movement corrections. This confirms the expectations, discussed in the introduction, of separate influences on the duration of the open-loop and the closed-loop phases. Subjects who moved their foot faster to the maximum during the closed-loop phase and who exhibited fewer movement corrections had a smaller preferred time-headway. This suggests a higher skill level in these subjects. Subjects with a larger preferred time-headway appear to be more uncertain about the requi­red braking response.

An important result in the present experi­ment was the strong effect of total movement time on preferred time-headway. This strengthens the conclusion that short and long followers differ in both the open- and closed-loop phases of movement. Short follo­wers may be more sensitive to the task requirements in emergency braking situations. Short and long followers differ in the efficiency of the control of braking. This was also found in Van Winsum and Heino, but there results could have been affected by differences in absolute levels of TTC between drivers with different preferred time-headway.

Together, the results suggest that individual differences in choice of time-headway are related to individual difference in braking performance. This supports the hypothesis that drivers adapt their tactical level behaviour to their opera­tional skill level. However, the mechanism could also be the other way around: Short followers may have had more opportunities to acquire emergency braking skills, simply because they had to brake hard more often. It can be argued that this would be particularly the case in relatively inexperienced drivers. A very experien­ced driver, even when being a long follower, will probably have experienced a substantial number of emergency stops anyway. Also experience in other situations may count, e.g. for stationary objects such as traffic lights. Braking for stationary objects also requires a tuning of the braking response to perceptual information. In the present study, none of the braking related variables were affected by driving experience. This does not support the idea that short followers have learned to brake more efficient­ly because they have been exposed to critical encoun­ters more often. However, this is not enough evidence to rule out the alternative hypothesis and this issue will have to return in future research.


EXPERIMENT 3: Choice of time-headway in car-following and the role of time-to-collision information in braking

0 comments
EXPERIMENT 3: Choice of time-headway in car-following and the role of time-to-collision information in braking

This is chapter 6 from the thesis “From adaptive control to adaptive traffic behaviour” about traffic psychology and behavioural adaptation of drivers, by Wim van Winsum. The thesis is from 1996. It describes a number of behavioural experiments into car driving that were performed in a research driving simulator.

Other chapters of this thesis can be found here:

 

Time-headway (THW) during car-follo­wing and braking response were stu­died in a driving simulator from the per­specti­ve that behaviour on the tactical level (e.g. choice of THW) may be linked to ope­rational compe­tence of vehicle con­trol (e.g. bra­king) via a process of adapta­tion. Time-headway was con­sis­tent within dri­vers and con­stant over a range of speeds. Since time-headway repre­sents the time avai­lable to the driver to reach the same level of dece­lerati­on as the lead vehi­cle in case it brakes, it was studied whether choice of time-headway was related to skills underly­ing bra­king perfor­mance. The initiation and control of bra­king were both affec­ted by time-to-collision (TTC) at the moment the lead vehi­cle started to brake. This stro­ngly suppor­ted the idea that time-to-collisi­on informa­tion is used for jud­ging the moment to start bra­king and in the control of bra­king. No evidence was found that short followers differ from long follo­wers in the abili­ty to accura­tely per­ceive TTC. There was howe­ver eviden­ce that short follo­wers are better able to program the intensity of braking to required levels. Also, short followers tuned the control of braking better to the development of critica­lity in time during the braking process. It was conclu­ded that short follo­wers may dif­fer from long followers in programming and execu­tion of the braking respon­se.

 

6.1 Introduction

Close car-following has been associated with traf­fic accident invol­vement. Rear-end collisions accounted for about 24% of all acci­dents involving two or more vehi­cles in the U.S.A in 1990 (McGehee et al., 1992). These accidents are usually attribu­ted to maintaining insuffi­ciently long head­ways and/or to inatten­tive driving resulting in respon­ding too late to a decelerati­on of a vehicle in front. In the literature, headway is expres­sed either as distance headway (DHW) or as time head­way (THW) (Fuller, 1981). DHW is the bumper to bumper dis­tan­ce between the lead vehicle and the following vehi­cle­. THW is the time interval between two vehicles in car-follo­wing, calcula­ted as DHW divided by the speed (in m/s) of the follo­wing vehicle. When the follo­wing and the lead vehi­cle drive at the same speed (steady-state following), THW repre­sents the time available to the driver of the follo­wing vehi­cle to reach the same level of deceleration as the lead vehi­cle in case it brakes. This avai­la­ble time is independent of speed. A faster braking response is then requi­red with a smaller THW. Also, the con­trol of braking is more critical in that case. In this article, the THW during steady-state car-follo­wing will be referred to as THWpref (preferred time head­way).

Evans and Wasie­lewski (1982) found that drivers with a larger THWpref had a history of fewer traffic violati­ons and traffic acci­dents. However, the same authors also argued that acci­dent invol­vement did not have a reliable relation with THWpref by itself (Evans and Wasielew­ski, 1983). Especially younger drivers employed smaller THW­’s, as did drivers of newer cars and of vehicles with medium mass.

Several factors have been identified that influence choice of THW. Choice of THW has been associated with personality factors by some authors. Sensation seeking as a personality trait is assumed to be rela­ted to risky behaviour (Zuc­ker­man, 1979). For example, Zuckerman and Neeb (1980) found a positive corre­lation between the sensation seeking score and reported dri­ving speed, whereas Heino et al. (1992), using a realistic car-fol­lo­wing task, reported a smal­ler THWpref for sensation see­kers than for sensation avoi­ders. Ota (1994) studied car-following behaviour in relation to personality trai­ts. He suggested social maladjustment as an important factor in choice of THW, alt­hough correlations between THW and persona­lity test scores were not significant.

O­ther authors have stressed the importance of task-related factors with regard to THWpref. Fuller (1981) stu­died THW of truck drivers in convoy situations. During the late shift, cove­ring a large period of driving in the dark, THWpref was signi­ficantly larger than during daytime dri­ving. This was explai­ned as an effect of visual conditions. Brookhuis et al. (1991) reported an increase in THW when using a car telep­hone while dri­ving, which can be regarded as an additi­onal task compe­ting for attention. This sug­gests the driver is aware of effects of task demands on the ability to detect a dece­lera­tion of a lead vehicle and adapts THW according­ly.

Choice of THW also has been associated with temporary state-related factors. Fuller (1984) reported a time-on-task effect on THW for older truck drivers in the late shift. After seven hours of driving, THW increased quite strongly, accompa­nied by verbal reports of performance decre­ments, drowsiness and exhaustion. In an expe­riment reported by Smiley et al. (1981) in an inter­active dri­ving simulator, mariju­ana resulted in increased headways during car-follo­wing. Smiley et al. (1986) studied the effect of marijua­na on sever­al car-driving tasks on the road. Again marijuana significant­ly increased headway in a car-following task. In another simula­tor study, Smiley et al. (1985) found that marijuana increased headway while alcohol decrea­sed headway. These results strongly sug­gest effects of temporary states such as fatigue or states induced by marijua­na and alcohol on THWpref; fatig­ue and marijuana increase THWpref which may be a reflecti­on of an adaptation of THW to adverse effects on the brake reaction, whereas alcohol decreases THWpref, possi­bly because drivers overestimate their braking competen­ce under alcohol.

The effects of task-related factors and trans­ient states refer to intra-individual diffe­rences. The re­sults stron­gly suggest a process of adaptation of THW to changes in operati­onal level competen­ce which is influ­enced by task-related and state-related fac­tors. From the same perspec­tive, inter-indi­vi­dual differences in following behaviour, may be related to inter-individual diffe­rences in opera­tional level competence, such that THWpref is adapted to limitati­ons in braking-related competence. These limitations in braking compe­tence may then be determined by specific skills required for opti­mal braking performance. For this to be the case, THWpref must be consis­tent within the individual driver, while it differs between drivers as a function of operational skill. Since THWpref represents the ultimate reaction time in case of a deceleration by the lead ­ve­hicle, THWpref must be invariant over spe­ed. However, in spite of years of research into car-following it is still not clear whe­ther this time headway constancy holds over speed and whether it is consis­tent within drivers.

Fuller (1986) reana­lyzed the results of previous car-follo­wing experiments and found nega­tive correla­tions between speed and THW. Following distan­ce increased with speed but not enough to maintain THW at a constant le­vel. However, the conditions resulting in different speeds varied widely. High speeds were associated with rural open-road conditions with low traffic density and the absence of juncti­ons, pedes­trians and other hazards. Low speeds, on the other hand were associa­ted with oppo­site condi­tions. Conditions that resulted in lower spee­ds, and an accompanying larger THWpref, were characte­rized by multiple tasks compe­ting for atten­tion, possibly resulting in performance decrements in braking. Ota (1994) studied THW while drivers were required to drive with a speed of 50, 60 or 80 km/h and follow under dif­ferent instructions such as ‘follow at a comfortable dis­tance’ and ‘follow at a minimum safe distance’. No effects of speed on THW were found while instruction signi­ficantly affec­ted choice of THW. This sug­gests that THWpref is constant over different speeds.

In the present study, an important hypothesis is that THWpref is constant over speed and consistent within the driver. In order to test consistency of THWpref and con­stancy over speed, it is required that, besides speed, all other factors that might affect braking performance are con­stant.

Accor­ding to Lee (1976) dri­vers are able to control braking based on time-to-collision (TTC) information from the optic flow field (visual angle divided by the angular veloci­ty). This would enable the driver to judge the mo­ment to start braking and to control the braking pro­cess. The initia­tion of braking includes the timing of releasing the accelera­tor pedal after a deceleration of the lead vehicle has been detected as well as the interval between release of the acce­lerator pedal and the moment the foot touches the brake pedal. The control of braking includes braking intensity and the interval between the moment the brake is touched and the moment the maximum brake pressure is reached. Brake reaction time (BRT) is usual­ly measured as the interval between the onset of the stimulus, such as the brake ­lights of the lead ve­hi­cle, and the moment the brake is tou­ched. Therefore, BRT is an important measure for the initiati­on of braking. BRT to anticipated events is faster than to unexpec­ted events (Johan­sson and Rumar, 1971) and the DHW at the moment the lead vehicle brakes has a strong effect on BRT (Brook­huis and De Waard, 1994; McKnight and Shi­nar, 1992;, Sivak et al., 1981). An important skill that has been associa­ted with the initi­ation of bra­king rela­tes to the perception of time-to-collisi­on (TTC). TTC is defined as the time requi­red for two vehicles to colli­de if they continue at their present speed and on the same path (see Van der Horst, 1990). TTC is computed as DHW/Vr, where Vr is the relative velocity or speed diffe­rence which must be larger than zero. While the ability to accurately perceive TTC is often mentio­ned as an important factor for judging the moment to start bra­king, studies that related TTC to actual braking are scar­ce. However, Van der Horst (1990) reported evidence that both the decision to start bra­king and the control of braking are based on TTC information available from the optic flow field. If TTC is an important factor in the initiation of braking, a relati­on is expected be­tween the TTC at the moment the lead vehicles starts to brake (TTCt0) and BRT. Since TTCt0 is an index for criti­cality, it is expected that BRT is faster if criticality is higher, i.e. when TTCt0 is smaller. A con­sistent finding in the literature is an underes­timation of TTC, espe­cially at higher TTC’s. Schiff and Detwi­ler (1979) found substantial individual differences in the ability to give accurate judgme­nts of TTC and an avera­ge unde­restima­tion of TTC of 39%. McLeod and Ross (1983) found that men gave higher and more accurate judgments than women. They reported an unde­restimation of TTC of 42%. Cavallo et al. (1986) found that experienced drivers produced better estima­tes of TTC than inexperienced drivers. They reported a general unde­restimation of 35%. Hoffmann and Morti­mer (1994) found that both estima­ted TTC and standard deviati­on of esti­mated TTC were linear­ly related to actual TTC. They reported an unde­restima­tion of TTC of 20% on average, while other studies typi­cal­ly report an unde­restima­tion of around 40%. This better perfor­mance in TTC estimati­on was attributed by Hoffmann and Mortimer to the fact that in their expe­ri­ment both vehicles were in moti­on, while other experi­ments typically measured estimated TTC to a static ob­ject. The studies on TTC estimati­on give substantive eviden­ce for underesti­mation of TTC and for individual differences in the ability to accurately esti­mate TTC. Differences in ability to accurately estima­te TTC are assumed to be expressed in the initiation of braking. BRT of drivers with better TTC estimation skills is assumed to covary more with TTCt0 than BRT of less skil­led drivers. This is because better skilled dri­vers are more sensitive to varia­tions in TTCt0. A hypo­thesis in the present study is that THWpref is related to sensi­tivi­ty of the initiation of braking to TTC informa­tion. Drivers who are more sensi­tive to TTC are then better able to judge the moment to start bra­king, while dri­vers who are less sensitive to TTC informa­tion run a higher risk of starting to brake too late. This might result in a larger safety margin and thus a higher THWpref for these dri­vers.

Drivers may not only differ in the initiation of the bra­king response but also in the control of braking. Van der Horst (1990) studied the control of bra­king by the maximum deceleration reached by the driver (DEC­max), the mini­mum TTC reached during braking (TTCmi­n), and the time differen­ce be­tween the moment of TTCmin (tTTCmin) and the moment of DECmax (tDECmax). TTCmin des­cribes how immi­nent a colli­sion has been during the braking process. According to Van der Horst, tDEC­max gives an indi­cation of the moment the driver knows a collision will be avoided. During the time before tTTCmin is reached, TTC is still decreasing resulting in increasing criticality. If tDECmax occurs some time before tTTCmin, critica­lity is still incre­asing at the moment the driver already relaxes the dece­lerati­on. If tDECmax occurs some time after tTTCmin is reached the driver keeps increasing the deceleration when it is no longer necessary. A close relation in time between tDECmax and tTTCmin then suggests a more efficient control of braking, where the control of braking is better tuned to the deve­lopment of criticality in time. In the present expe­riment it will be examined whe­ther THWpref is related to braking control as indicated by these measures. In addi­tion to this, the maximum percentage brake pressed (M­AXB­R), and the interval between touching the brake­pedal and the moment the brake pedal is pressed to the maximum value are measu­red. Fur­thermo­re, it will be exami­ned whether the intensi­ty of the braking reacti­on, measu­red by MAXBR, is more sensi­tive to TTC at the moment the lead vehicle starts to brake for short follo­wers compa­red to long follo­wers. A higher sensiti­vity of the in­ten­sity of braking to TTCt0 suggests that the braking response is more adapted to critica­lity at the moment the driver de­tects the braking of the lead vehicle.

In summary, the following hypotheses will be tested in the present experiment.

1) Preferred time-headway is constant over different speeds.

2) Preferred time-headway is consistent within indivi­dual drivers, but differs between drivers.

3) The initiation of braking, measured by BRT, is more stron­gly related to TTC at the moment the lead vehicle starts to brake for short followers compared to long followers. This is assumed to be related to diffe­rences in the ability to percei­ve TTC information.

4) Preferred time-headway is related to the intensity of braking and quality of braking control. The intensity of braking is measured by MAXBR while the quality of bra­king control is measured by the sensitivity of the bra­king intensi­ty to criticality (as measured by TTC) and by the time diffe­rence between tTTCmin and tDECmax.

 

6.2 Method

 

Apparatus. The driving simulator of the Traffic Research Centre (TRC) was used for the present experi­ment. This fixed-based simu­lator consists of two inte­grated subsys­tems. The first subsystem is a conventional simulator composed of a car (a BMW 518) with a steering wheel, clutch, gear, accelera­tor, brake and indica­tors connected to a Silicon Graphics Skywriter 340VGXT compu­ter. A car model converts driver control actions into a displacement is space. On a 2 x 2.5 meter projections­creen, placed in front of the car mockup, an image of the outside world with a horizon­tal angle of 50 degrees is projec­ted by a graphical videopro­jector, controlled by the graphics software. Images are presented with a rate of 15 to 20 frames per second, resulting in a sug­gestion of smooth move­ment. The visual objects are buil­dings, roads, traffic signs, traffic lights and other vehicles. The sound of the engine, wind and tires is presented by means of a digital soundsampler recei­ving input from the simulator computer.

The second subsystem consists of a dynamic traffic simula­tion with interacting artificially intelligent cars. For experimental purpo­ses different traffic situa­tions can be simulated. The simula­tor is described in more detail elsewhere (Van Wolffe­laar & Van Winsum, 1992 and Van Winsum & Van Wolf­felaar, 1993). De Waard et al. (1994) reported a significant correlation (r=0.67) be­tween THW measured in this simulator and ratings of preferred headway on a photo-preference test. In this test subjects rated preferred headway from a series of photographs with a view of a lead vehicle through the wind­screen on a motorway. This supports the validity of this simulator for measuring car-following behaviour. Also, TTC has been reported to be directly available from the optic flow field without requiring speed and distance estimation. Since visual angle and angular velocity are identi­cal in the simula­tor and in real world driving, this simulator was assumed to be a valid instrument for estimation of TTC.

 

 

Procedure. The circuit was made of two-lane roads with a lane-width of 3 meters. All roads had delineation with broken center lines and closed edge lines. Side­roads connected with an angle of 45 degrees to the main road, allowing other vehi­cle to merge in front of the simulator car and leave the main road. The length of the circuit was 7600 meter.

Before the experiment started, subjects completed a ques­tionnaire on driving experience and age. After this, subjects were instructed to drive as if they had to reach their desti­nation as soon as possible, without overta­king other vehicles, to drive safely and to res­pect the speed limit of 80 km/h. The experimen­t started after a ten minutes practice drive.

The experiment consisted of two parts, separated by a 15 minutes break. During the first part choice of head­way was measu­red as a func­tion of speed. Lead vehicles drove with a con­stant veloci­ty of either 40, 50, 60 or 70 km/h. These diffe­rent speeds are referred to as ‘sp­eed conditions’. Sub­jects were required to drive around the cir­cuit twice. The first drive around the circuit was used to familiarize sub­jects with other traf­fic. Vehi­cles merged in front of the simulator car, control­ling their speed such that when the simula­tor car was 50 meter from the inter­secti­on, the lead vehicle was 100 meters in front of the simulator car.

During the second part of the experiment braking behaviour was measured. Vehicles merged in front of the simulator car in the same way as des­cri­bed above. Lead vehicles drove with a constant speed of either 60 or 50 km/h, resulting in two ‘braking conditions’. As soon as the lead vehicle was 50 meter in front of the simulator car (t0), it dece­lera­ted with -2 m/s², with its brake­lights on, to a speed 20 km/h below the initial cruise speed.  As soon as the simulator car reached this speed (40 of 30 km/h) the lead vehicle pulled up again. The two braking conditions (50 vs 60 km/h) were used to study wit­hin-subjects differences in braking as a func­tion of TTCt0.

 

Data registration and analysis. Speed of the simu­lator car (V) and lead vehicle (Vlead) in m/s, distance head­way (DHW) in meters, acceleration in m/s² and brake pedal signal expres­sed as per­cen­tage pressed were sam­pled with a frequency of 10 Hz. THW was calcula­ted as DHW/V. TTC was calcu­lated as DHW/Vr, with Vr being the relative speed (V-Vlead). Average THW was computed from the moment the simu­lator car and the lead vehi­cle drove with the same speed until the lead ve­hicle left the main road. THWpref was computed as the average THW over the four speed conditi­ons.

In the second part of the experiment t0 represents the moment a DHW of 50 me­ters was reached. On t0 the lead vehicle started to brake. TTCt0 repre­sents the TTC on t0. BRT was compu­ted as tbr – t0, where tbr refers to the moment the brake ­pe­dal was pres­sed more than 5%. TTCbr repre­sents TTC on tbr. On tmaxbr the maximum brake pressu­re, MAXBR, was reached. TTCmaxbr represents TTC on tmaxbr. Brake control move­ment time, (BCMT) was calcu­la­ted as tmaxbr-tbr. The moment the maxi­mum dece­lerati­on, DECmax, was reached is indica­ted as tDECmax. The moment the minimum TTC, TTCmin, was reached is indicated as tTTCmin. The abso­lute time difference between the moment of maximum dece­le­ration and the moment of minimum TTC was computed as ABS(tDEC­max-tTTCmin) and is referred to as tdif. Figure 1 shows a time history of bra­king, together with a number of depen­dent variables.

Analy­sis of covariance was applied to test differen­ces in sensiti­vity to TTC as a function of THWpref. For this, diffe­rences between the two braking conditions were studied to test whe­ther braking-related variables covaried with TTC. The diffe­ren­ce in TTCt0 between bra­king condition 60 (lead vehicle braked from 60 to 40 km/h) and braking condition 50 (lead vehicle braked from 50 to 30 km/h) is expressed as dTTCt0. The diffe­rences in MAXBR and BRT between these two conditions are ex­pressed as dMAXBR and dBRT. The regression coefficient of dBRT and dMAXBR on dTTCt0 is an indicator for the sensitivity of BRT and MAXBR to TTCt0. Higher sensitivity is expres­sed as a steeper slope (larger coefficient of regressi­on). Analysis of covariance was used to test differences in slope as a function of THWpref.

 

 

Figure 1. Time-history of braking and dependent varia­bles.

 

Effects of THWpref and braking conditions on dependent variables were tested with repeated measu­rements multi­variate analysis of variance (MANO­VA) with braking con­dition as a within-subjects factor.

Subjects. Fifty-four male subjects participated in the experi­ment. The average age was 29 years (sd. 8.12, range 19-48) with 65% of the subjects being younger than 30 years of ages. They had held a driving license for 9 years on average (range 1-29).

 

6.3 Results

 

Stability of THWpref. THW was not significantly affected by speed of the lead vehi­cle (F(135,3)= 1.27, p>=0.2­5), see figure 2. This sup­ported the hypothesis that THW is constant over speed.

 

Figure 2. THW as a function of speed.

A high correlation between THW’s in the four speed con­di­tions suggests consistent following behaviour. THW’s in all speed conditi­ons were signifi­cantly correlated (p < 0.001), as shown in table 1. Additio­nal evidence for consis­tency in following behaviour was obtained by con­sidering each THW as an “item” in a (4-item) “following behaviour” test (Hendrickx, 1991). The test’s reliability index (Cron­bach’s alpha) was found to be as high as 0.91. This was taken as evidence that all THW’s were an expression of a subjects’ general THWpref.

 

 

Table 1. Correlation matrix for THW’s in the four speed condi­tions

 

THW50       THW60    THW70

 

THW50     0.69**

THW60     0.76**        0.63**

THW70     0.67**        0.69**       0.60**

 

(** indicates p < 0.001).

THWxx : THW = time headway, xx = speed (km/h) of lead vehicle

 

These results supported the hypothesis that THW is con­sistent within drivers, but differs between drivers. For further analysis, the average THW over the four speed conditions was computed as THWpref. Based on the frequency distribution of THWpref, three groups of equal size were created. These groups are referred to as ‘THWpref groups’. These groups served as a between-subjects factor in subsequent analyses. Four subjects were not included because they failed to reach a stable THW in the 70 km/h condition. Table 2 shows number of subjects, average THW and stand­ard deviation of THW for the THWpref groups.

 

Table 2. Size, mean THW and sd of THW for THWpref groups

 

THWpref group      N               mean THW(s)   sd of THW

 

short                      17              0.67                   0.19

medium                 16              1.08                   0.09

long                       17              1.52                   0.27

 

Braking responses. Two additional sub­jects failed to display a clear brake res­pon­se in one of the two braking conditions. Therefo­re, the total number of sub­jects in the analyses was 48.

Figure 3 shows the time history of TTC for the three THWpref groups in both braking conditions. Four data­points are dis­played. The first point represents TTCt0, the second TTCbr, the third TTCmin and the fourth TTCmaxbr. The time interval between TTCt0 and TTCbr repre­sents BRT, while the time inter­val between TTCbr and TTCmaxbr repre­sents brake control movement time (BCMT).

 

The initiation of braking. Table 3 gives the MANOVA effects of THWpref group and braking condition on variables related to the initiation of braking.

TTCt0 and TTCbr were significantly smaller, while the relative speed (Vr) at t0 and tbr was signi­ficantly larger for subjects with a smaller THWpref. At t0 long followers already had lowered their speed to a greater extent than short followers. BRT was not significantly diffe­rent for short followers compared to long followers.

Table 3. Effects of THWpref group and braking condition on variables related to the

initiation of braking (F-sta­tis­tics)

 

Effect

Dependent               THWpref group           Braking con.               interac­tion

 

TTCt0                          8.57**                      0.16                           0.52

TTCbr                        18.05**                      0.59                           1.14

Vrt0                           15.83**                      6.79**                       1.90

Vrbr                           24.72**                      8.07**                       1.26

BRT                            0.62                          20.57**                      1.01

 

THWpref group effect : df = 45,2;

Braking condition  effect : df = 45,1;

Interaction effect: df= 45,2

** = p < 0.01

 

 

Figure 3. Time history of TTC as a function of THWpref group­s for braking

condition 50 (left) and braking con­dition 60 (right).

 

Braking condition had a significant effect on BRT. BRT was faster in the conditi­on where the lead vehicle decele­rated from 50 to 30 km/h. This was accompa­nied by a larger relative velocity on t0 and tbr in this condi­ti­on. None of the interac­tions were signi­ficant.

Table 4 presents the correlations of BRT with TTCt0 and TTCbr.

 

Table 4. Correlation of BRT with TTC in braking conditi­on 50 and 60

 

Condition 50     Condition 60

 

TTCt0            0.66**               0.61**

TTCbr            0.01                   -0.21

 

** = p < 0.01

 

The correlations of BRT with TTCt0 were highly signifi­cant. The correlations of BRT with TTCbr were not signi­ficant. Thus, BRT decreased as TTCt0 decreased for both braking condi­tions. This was taken as evidence that the initiation of braking, indica­ted by BRT, was sensitive to TTC information as an index for criticality. The significant effect of THWpref group on TTCt0 and the absen­ce of a significant effect of THWpref on BRT suggests the TTC criterion for initiating the braking response is lower for short followers.

One of the hypotheses was that the initiation of braking was more sensitive to TTC for short followers compared to long follo­wers. Sensiti­vity was expressed as the extent to which BRT covaries with TTCt0. Analysis of covariance revealed that dBRT was depen­dent on dTTCt0 (F(42,1) = 14.77, p<0.001). This means that, within Ss, a smaller TTCt0 resulted in a faster BRT. Since dBRT was compu­ted as the difference between BRT’s in the two braking conditions, the effect of braking condition on BRT is partly explained by within-subjects differences in TTCt0. Thus, the initiation of the braking response was very sensi­ti­ve to between-subjects as well as wit­hin-subjects variations of TTC at t0. The slope of the regres­sion of dBRT on dTTCt0 repre­sents the sensi­tivity of BRT for TTCt0. The magnitu­de of the slope as well as the correlation coeffi­cients are shown in table 5 for the three THWpref groups. Although the correla­tion and regres­sion coeffi­cients suggest a stronger relation be­tween dBRT and dTTCt0 for short followers, this was not con­firmed by analysis of covariance because the inter­action with THWpref groups was not signifi­cant (F(42,2­)=1.­62, p=0.210). Thus, the hypothesis that short follo­wers are more sensitive to TTC information in the initi­ation of the braking response was not confirmed.

 

 

Table 5. Correlation and sensitivity of BRT to TTCt0

 

THWpref group               R               coefficient of regression

 

short                               0.72**      0.19

medium                          0.63**      0.12

long                                0.51*        0.06

 

** = p < 0.01; * = p < 0.05

 

The control of braking. Table 6 shows the effects of THWpref group and braking condi­tion on variables rela­ted to the control of braking.

 

Table 6. Effects of THWpref group and braking condition on variables related to

the control of braking (F-statis­tics)

 

Effect

Dependent               THWpref group           Braking con.               interac­tion

 

TTCmin                      18.78**                      0.30                            1.23

TTCmaxbr                   16.13**                      0.01                            0.51

BCMT                        0.86                          2.01                            2.19

MAXBR                    6.24**                      7.12**                        0.33

DECmax                     4.54*                         2.49                            0.02

tdif                              3.88*                         0.75                            0.09

 

THWpref group effect      : df = 45,2

Braking con. effect          : df = 45,1

interaction effect             : df = 45,2

** = p < 0.01; * = p < 0.05

 

The minimum TTC during braking (TTCmin) was signi­ficantly smaller for short followers, as was the TTC at the mo­ment the brake was pressed to the maximum (TTCmaxbr). Short followers generated a more intense brake reaction than long followers : MAXBR was signifi­cantly larger for short followers. Also DECmax was larger for short follo­wers. This sup­ported the hypothesis that short followers differ from long followers in the inten­sity of the bra­king response. BCMT, the time wit­hin which the brake maximum was reached, was not affec­ted by THWpref groups.

The absolute time difference between tDECmax and tTTCmin, tdif, was seen as an indicator for the effi­ciency of bra­king con­trol. There was a significant effect of THWpref group on this measure. Tdif was smaller for short follo­wers compared to long follo­wers, see figure 4. This supported the hypothesis that short follo­wers differ from long followers in the quality of braking control.

 

In order to test the sensitivity of the intensity of braking to criticality, an analysis of covariance was performed on dMAXBR (diffe­rences in MAXBR between the two braking conditi­ons) as a function of dTTCt0 (diffe­rences in TTCt0 between the two braking conditi­ons), with THWpref group as a between-sub­jects factor. A smaller TTCt0 generally resulted in a larger MAXBR (F(42,1)=22.37, p=0.000). This means that the intensity of the braking reacti­on strongly depended on TTCt0. The inter­acti­on with THWpref group was significant as well (F(42,2) = 4.63, p=0.015). In table 7 it can be seen that MAXBR decreases more as a functi­on of TTCt0 for short followers compared to long followers. The differen­ces in slope indicate that the intensity of the bra­king res­ponse is more sen­si­tive to TTCt0 for drivers with a smaller THWpref, alt­hough the correlations between dMAXBR and dTTCt0 are comparable for the three groups.

This again supported the hypothesis that short follo­wers differ from long followers in the quality of braking control.

 

Figure 4. Difference between tTTCmin and tDECmax as a functi­on of THWpref group­s

and braking condi­tion.

 

 

Table 7. Correlation and sensitivity of MAXBR to TTCt0

 

THWpref group      R                       coefficient of regression

 

short                      -0.69**             -6.52

medium                 -0.57*                -3.13

long                       -0.58*                -1.13

 

** = p < 0.01; * = p < 0.05

 

 

6.4 Discussion and conclusions

 

The hypothesis that THWpref is consistent within the driver and the hypothesis of constancy of THWpref over speed during steady-state car-following were confir­med for the range of speeds examined in the present expe­ri­ment. The brake reaction of drivers was analyzed in order to investigate whe­ther diffe­ren­ces in THWpref during steady-state car-following are related to differences in braking performance and under­lying skills. Since THW during steady-state following represents the time avai­la­ble to the driver to give an appropriate braking res­pon­se in case the lead vehicle decelerates, THW may be the result of an adaptation of the driver to indi­vidual differences in braking competence. Braking perfor­mance was assumed to be related to the ability to percei­ve time-to-collision (TTC) and the ability to generate an efficient braking response, depen­ding on the criticality of the situation. The initiation of braking, as measured by brake-reaction time (BRT) was stron­gly related to TTC at the moment the lead vehicle started to brake (TTCt­0 ) and thus to criticality. This strong relation was apparent between subjects as well as within subjects. This conforms with the sugge­stion in the literature that TTC infor­mation is used by the driver to judge the moment to start braking. Howe­ver, dri­vers with a smaller THWpref during steady-state follo­wing start to brake at a lower TTC, i.e. when the criticality is higher. This suggests a different TTC criteri­on for the initiation of braking, depending on preferred time-head­way. Although the initi­a­tion of bra­king was very sensi­tive to TTC information, there were no diffe­rences between short followers and long followers in sensitivi­ty of BRT to TTCt0. Thus, the hypothesis that diffe­rences in THWpref during steady-state following are related to the ability to accurate­ly per­ceive TTC was not confirmed since a differential ability related to TTC perception was assumed to be expressed in BRT.

The minimum TTC during braking was smaller for short follo­wers. This indicates that a collision was more imminent for short follo­wers than for long followers. There were howe­ver differences in the control of bra­king. Firstly, short follo­wers pressed the brake pedal to a higher maximum, resul­ting in a larger deceleration. Secondly, for short followers the inten­sity of the bra­king response was more strongly depen­dent on the criti­cality at the moment the lead vehi­cle started to brake. This suggests that the intensity of braking is at least partly programmed before response execution and confirms the suggestion in the literature that TTC information is used in the control of braking. Short followers are then better able to ­program this response to the appropriate level, depending on criticality. However, at the moment the lead vehi­cle starts to decele­rate, the driver does not know how strong it will decelera­te and for how long. Therefore, visual feedback during the braking maneuver is important for conti­nuously adapting the braking res­ponse to the required level. The programmed braking intensity may then have to be adjusted to another level depending on the development of criticality in time. The moment of maximum deceleration (tDECmax) was assumed to indicate when the driver knows a collisi­on will be avoi­ded. A closer corres­pondence in time with the moment of minimal TTC (tTTCmin) sug­gests a better ability to adjust the control of braking to requirements. In this res­pect, the third difference was found between short en long follo­wers. For short followers the absolute difference between tDECmax and tTTCmin was smaller, indi­cating a more efficient bra­king control where the timing and intensity of bra­king is better tuned to the development of criti­cality in time during the braking process.

These results suggest differences in skills related to the res­ponse programming and response execution of braking between short and long followers. On the other hand, the absolute levels of TTCt0 were different between THWpref groups. An alter­native explanation may then be that short followers had to generate more efficient braking responses that were better tuned to criticality because criticality was higher for them to begin with. In other words, they may have been forced to perform better. Also, since TTC during the braking process was lower for short followers, and, as discussed in the introduc­tion, the estimation of TTC is more accurate for smaller TTC’s, the differences between short and long followers in braking con­trol may have been caused by a more accurate esti­ma­tion of TTC by short followers. In both cases, however, the sensi­tivity of BRT to TTCt0 is expected to be higher too for short follo­wers. Since this was not the case, the evidence presented suggests diffe­rences in skills related to the pro­gramming of the intensity of braking and the con­trol of bra­king between short and long followers.


EXPERIMENT 1: Speed Choice and Steering Behaviour in Curve Driving

0 comments
EXPERIMENT 1: Speed Choice and Steering Behaviour in Curve  Driving

This is chapter 4 of the thesis “From adaptive control to adaptive traffic behaviour” about traffic psychology and behavioural adaptation of drivers, by Wim van Winsum. The thesis is from 1996. It describes a number of behavioural experiments into car driving that were performed in a research driving simulator.

Other chapters of this thesis can be found here:

 

The relation between speed choice and steering performance during curve negotiation was studied in a driving simulator. The hypothesis was that curve radius and steering competence both affect steering error during curve driving resulting in compensatory speed choice. In this, the control of safety margins was assumed to operate as a regulatory mechanism. Smaller curve radii resulted in a larger required steering wheel angle while steering error increased linearly with required steering wheel angle. This was compensated for by choosing a lower speed, such that the time-to-line crossing to the inner-lane boundary was constant over all curve radii examined. Steering competence was measured during straight road driving. Poorer steering competence also resulted in larger steering errors that were compensated for by choosing a lower speed such that the safety margin to the inner-lane boundary was unaffected by steering competence.

4.1 Introduction

 

Car driving behaviour in curves may be regarded as an interesting case where steering, as an example of operational performance, is intimately related to behaviour on the tactical level, in this case the choice of speed as a function of curve radius. The distinction between the operational and the tactical level of car driving behaviour has been made by several authors (c.f. Michon, 1985) and might form a fruitful basis for the development of modern driver behaviour theories (see c.f. Ranney, 1994). Until now, studies of car driving behaviour in curves have focused either exclusively on speed choice or on steering behaviour while no attempts have been made to integrate these two lines of research.

A consistent finding in studies on speed choice in curves is that speed has a curvilinear relation with curve radius (see c.f. Kanellaidis et al., 1990) and an inverse relation with lateral acceleration. This means that with smaller radii speed is lower but lateral acceleration is higher compared to larger radii (c.f. McLean, 1981). Sometimes an inverse linear relation is reported (Ritchie et al., 1968) while other studies have found an inverse non-linear relation between speed and lateral acceleration (Herrin and Neuhardt,1974; Macura, 1984). These results have encouraged the idea that lateral acceleration is used by drivers as a cue in speed choice in which a smaller lateral acceleration is accepted as a safety margin at higher speeds (and thus larger radii).

In studies of steering behaviour during curve negotiation, speed is usually held constant. Donges (1978) presented a two-level steering control model that incorporated negotiating curves. Anticipatory open-loop control starts with a steering action some time before the curve is entered followed by a steering-wheel angle maximum, dsa, in the curve. Then a period of stationary curve driving begins during which the driver generates correcting steering actions in a compensatory closed-loop mode. In a survey of models of steering behaviour Reid (1983) argued that driver models should incorporate both lane tracking and speed control. In Donges’ model the parameters estimated to fit the model on experimental data were influenced by vehicle speed and confounded with road curvature. Curve radius and speed during curve negotiation affect required operational performance because both factors affect the required steering-wheel angle. Godthelp (1986) described this phenomenon as follows: the required steering-wheel angle for a particular curve can roughly be characterized as dsr = GL(1+Ku²)/Rr. In this, dsr represents required steering-wheel angle, Rr the road radius in meters, G the steer-to-wheel ratio, L the wheel base, K a vehicle related stability factor and u represents longitudinal speed in m/s. For any given speed, required steering-wheel angle then increases with smaller radii, but for a given radius it increases with higher speed, if K is larger than zero, which is the case for a normal understeered car.

If the steering-wheel angle during curve negotiation matches the required steering-wheel angle perfectly, speed is only restricted by an upper limit where the vehicle starts skidding. The speed at which this occurs is generally much higher than actual speed in curves. The hypothesis of the present study is that steering errors play an important role in speed choice, such that speed is adapted to operational performance. There is some evidence that steering errors increase linearly with required steering-wheel angle, see c.f. Godthelp (1985, 1986). Since negotiating curves with a smaller radius requires a larger steering-wheel angle, the implication is that steering error is larger in curves with smaller radii compared to wider curves. If steering error is a linear function of required steering-wheel angle, the fraction defined as steering error divided by required steering-wheel angle should be constant over radii.

There is also evidence that steering error is affected by steering competence. Cavallo et al. (1988) found that, under visual occlusion, experienced drivers estimated the correct required steering-wheel angle better than inexperienced drivers. Also, experienced drivers exhibited less variation in steering-wheel amplitude during closed-loop control compared to inexperienced drivers. These results suggest that experienced drivers generate smaller steering errors.

If the driver compensates for larger steering errors induced by smaller radii or poorer steering competence by choosing a lower speed, some regulating mechanism or safety margin is required that determines how speed is adapted. It is suggested here that the time-to-line crossing (TLC), developed by Godthelp et al. (1984), is such a safety margin. TLC represents the time available for a driver until the moment at which any part of the vehicle reaches one of the lane boundaries. In a study of Godthelp (1988) drivers were instructed to generate correcting steering actions when vehicle heading could still comfortably be corrected to prevent a crossing of the lane boundary. Drivers made a corrective steering action at a constant TLC irrespective of vehicle speed.

The model on the relation between speed choice and steering performance may then be summarized as follows. Required steering-wheel angle is determined by curve radius and speed, while steering error is determined by required steering-wheel angle and steering competence. It is assumed that the driver has learned the effect of curve radius and speed on required steering-wheel angle and on steering error from previous experiences. Also, it is assumed that steering error is consistent and the driver is aware of his or her steering competence. When the driver approaches a curve, both radius and steering competence cause an anticipatory adjustment of speed, much like the anticipatory avoidance response in the threat avoidance model of Fuller (1984), in which the effects of radius and steering competence on steering error are traded off with speed such that the safety margin TLC remains constant and independent of radius and steering competence. Although mathematically TLC is determined by steering error as well as speed, the higher steering errors associated with smaller radii and poorer steering competence are assumed to result in lower speeds because of the constancy of preferred TLC as a guiding principle. This principle will then result in low or non-significant correlations of speed and steering error with TLC. The relation between lateral acceleration and speed is then assumed to be a by-product of this mechanism.

In the experiment steering competence was measured separately during straight road driving. Road radius was manipulated within-subjects with radii of 40, 80, 120 and 160 meters. Originally, lane-width was manipulated within-subjects as well, since lane-width was expected to affect TLC. However, the effects of lane-width are not reported since these are not of crucial importance to the issue addressed here. Also, subjects used only a part of the lane-width because they drove relatively close to the inner lane boundary. This counteracted possible effects of lane-width on TLC and speed choice. There is also evidence in the literature that drivers use the inner lane boundary as a reference for vehicle guidance, see c.f. Shinar et al. (1980), McDonald and Ellis (1975), Cohen and Studach (1977). Therefore, only TLC and steering behaviour data towards the inner lane boundary are reported in the present article.

 

4.2 Method

 

Apparatus. The experiment was performed in the Traffic Research Centre (TRC) fixed-based driving simulator. It consists of a car (BMW 518) with a steering wheel, clutch, gear, accele­ra­tor, brake and indica­tors connected to a Silicon Graphics Skywriter 340VGXT compu­ter. A car model converts driver control actions into a displacement in space. On a 2 x 2.5 meter projection s­creen, placed in front of the car mockup, an image of the outside world with a horizon­tal angle of 50 degrees is projec­ted by a graphical videopro­jector, controlled by the 3D-grap­hics software. Images are presented with a rate of 15 to 20 frames per second, resulting in a sug­gestion of smooth move­ment. The visual objects are buil­dings, roads, traffic signs, traffic lights and artificially intelligent traffic. The sound of the engine, wind and tires is presented by means of a digital soundsampler recei­ving input from the simulator computer. The simula­tor is described in more detail elsewhere (Van Wolffe­laar & Van Winsum, 1992 and Van Winsum & Van Wolf­felaar, 1993).

 

Procedure. A circuit of two-lane roads with a lane-width of either 3.0, 3.5 or 4.0 meters was used. Roads had deline­ation with broken center lines and continuous edge lines. Four left-turning curves with 90 degrees angle and radii of 40, 80, 120 and 160 meters were separated by straight road segments. After completing a questionnaire on driving experience and age, subjects practiced driving in the simulator for ten minutes. They were instructed to choose their own preferred speed but to adapt the speed for curves as they normally would and to stay in the right lane. There were three trials, one for every lane-width. Each trial consisted of five roundtrips. This means that in every trial all four curves were negotiated five times. The three trials are treated as multiple measurements here.

 

Data registration and analysis. Sample measurements (10 Hz) were taken on speed (m/s), lateral position, steering-wheel angle (degrees), TLC (seconds), and steering error (degrees).

The steering integral (Ids) during straight road driving was used as a measure for steering competence. This was computed as follows. The steering-wheel signal was divided into periods where the steering wheel was turned to left and periods where it was turned to right (relative to the zero angle). For every period the amplitude was integrated over time and these integrals were averaged resulting in Ids. Thus, this measure is affected by both steering-wheel amplitude and frequency. A smaller steering integral represents better steering performance. Steering error in curves, dse, was defined as the difference between the actual steering-wheel angle and required steering-wheel angle (ds – dsr).

Figure 1 presents a time-history of steering error and TLC during curve negotiation. The curve is entered at time 0. Positive values of steering error and TLC represent steering to the inner lane boundary (left) while negative values represent steering to the outer lane boundary (right). The steering error fluctuates around zero. If steering error is zero then the steering-wheel angle equals the required steering-wheel angle. The open-loop phase ends when the maximum steering-wheel angle, dsa is reached. In Figure 1 this is indicated by the first maximum for dse. This is followed by closed-loop steering control during which deviations from the required steering error are minimized by the driver.

The following variables were analyzed:

– The steering error dse on the moment dsa is reached. This represents the steering error during the open-loop phase.

– The required steering-wheel angle dsr. This was measured as the steering-wheel angle on the moment that steering error was zero just before dsa was reached.

– The steering error ratio, computed as dse/dsr. This ratio is a measure for the relative steering error.

– The steering error integral, Idse, during the closed-loop phase. This was computed as the average integral of all periods where the steering error was directed toward the inner lane boundary.

– The minimum TLC’s to the inner lane boundary, TLCmin during the closed-loop phase. These were determined and averaged for every radius/trial combination.

– The minimum speed during curve negotiation. This was determined and averaged for every radius/trial combination.

The effects of radius were analyzed with repeated measurements analysis of variance. The effects of steering competence were analyzed with correlation and regression analyses. The confidence level for significance was set at p£0.05.

Figure 1. Steering error and TLC time-history during curve negotiation.

Subjects. 16 subjects, 8 male and 8 female, participated in the experiment. The average age was 34 years (sd. 6.3, range 22-47). They were licensed drivers for 12 years on average (sd. 6.3, range 2-27). The average annual kilometrage was 10594 (sd. 8267, range 1500-30000).

 

4.3 Results

 

The correlation between steering integral Ids and drivers’ total kilometrage was -0.62 (p<0.01). This means that more experienced drivers steered more accurately on straight road segments.

The minimum speed during curve negotiation was significantly affected by radius (F(3,15)= 58.17, p<0.01). Required steering-wheel angle (dsr) was significantly affected by radius (F(3,15)=188.24, p<0.01) as was the steering error (dse) during the open-loop phase (F(3,15)=28.28, p<0.01) and the steering error integral (Idse) during the closed-loop phase (F(3.15)=14.29, p<0.01). The effect of radius on steering error ratio was not statistically significant. Also, the effect of radius on the minimum TLC (TLCmin) during the closed-loop phase was not significant. The averages of these dependent variables as a function of radius are presented in Table 1.

Table 1. Averages of dependent variables as a function of radius

 

Radius (m)

Dependent variable                40              80                120             160

 

speed (m/s)                            11.23        14.92        17.58              17.99

required angle (degrees)     121.44        74.64       56.56             43.47

steering error:

-open loop (degrees)           14.20         7.47           5.54              4.75

-closed loop (integral)        14.02         6.55           5.26              4.67

steering error ratio                 0.12         0.10           0.10           0.11

minimum TLC (s)                   2.52         2.70           2.89           2.79

 

 

It can be seen that a smaller radius resulted in a larger required steering-wheel angle, larger steering errors and a lower speed. However, TLC and the steering error ratio were constant over all radii. Both steering errors during the open and closed-loop phases were affected by radius in the same manner.

 

Table 2. Standardized alpha coefficients of dependent variables

 

Dependent variable              standardized alpha

 

Speed                                    0.93

required angle                      0.91

steering error:

-open loop                            0.88

-closed loop                         0.86

steering error ratio              0.91

minimum TLC                      0.90

 

In order to test effects of individual differences in steering competence on dependent variables it is required that these variables are consistent within the driver. In that case, it is justified to average over all measurements (4 radii x 3 repetitions). In that way, the effect of radius is canceled while the effect of individual differences is preserved. The reliability, or consistency, of the dependent variables was tested with the standardized alpha coefficient. This represents the estimated square of the correlation of scores on a collection of items, in this case the 12 measurements, with true scores (Nunnally, 1978). For basic research a reliability of 0.80 is generally regarded as a satisfactory level.

Table 2 presents the standardized alpha coefficients for all dependent variables. It can be seen that all variables are reliable and most alpha’s are higher than 0.90. The minimum speed, TLC, steering errors, required steering-wheel angle and steering error ratio were averaged over radii and repetitions. Figure 2 presents the results of multiple regression analyses. Only significant partial regression coefficients are displayed.

 

Figure 2. Path diagram with partial regression coefficients.

*=p<0.05, **=p<0.01, ns=not significant.

 

It can be seen that the measures for steering errors in the open-loop and the closed-loop phase are strongly intercorrelated, indicating that they measure the same phenomenon. Steering error is determined by required steering-wheel angle, while there is no direct path from speed to steering error. Required steering-wheel angle is strongly determined by speed. In addition to this, steering error is strongly determined by steering competence (Ids). But while a higher steering competence results in lower steering error it also results in higher speed. Because steering competence is an intermediary factor, there is no effect of speed or steering error on TLC. Also, there is no path from steering competence to TLC. This suggests that subjects with poorer steering performance maintain the same safety margin as subjects with better steering performance, and that they choose a lower speed in order to maintain that safety margin. The correlation between Ids and the steering error ratio was 0.74 (p<0.01).

 

4.4 Discussion and conclusions

The effects of curve radius as a road design factor and steering competence as an individual driver characteristic on speed choice in curves were studied from the perspective that effects on operational performance are compensated for on the tactical level. The implied mechanism in the case of curve negotiation is that both curve radius and steering competence affect steering errors on the operational level. In this, the preferred TLC was assumed to be a regulating mechanism that determines how speed is controlled in order to compensate for larger steering errors. Since TLC is mathematically determined by speed and steering error, higher steering errors can be compensated for by choosing a lower speed such that TLC is unaffected by radius or steering competence. The results supported this model. It was found that both required steering-wheel angle and steering error during the open and closed-loop phases increase with smaller radii, but that the relative steering error, defined as steering error divided by required steering-wheel angle, is constant over radii. This strongly suggests that steering error is linearly related to required steering-wheel angle and is consistent with the results of Godthelp (1985, 1986). Smaller radii resulted in the choice of a lower speed, but the minimum TLC’s during curve negotiation were not affected by radius. This suggests that larger steering errors are compensated for by choosing a lower speed such that a constant minimum TLC is maintained. This finding confirms the ideas of Summala (1988) and Rumar (1988) that drivers control safety margins that can be operationalized as distance or time related measures. The TLC as a safety margin then is controlled by the drivers’ speed choice. The results suggest that speed choice and steering performance are both intimately related in negotiating curves.

In this study, individual differences in steering competence strongly determined speed choice and steering performance in curves. Steering competence was measured with the steering integral during straight road driving. A larger steering integral is an indication of poorer steering performance. The quality of steering performance was related to driving experience. Steering performance, speed choice and minimum TLC were consistent within drivers during curve negotiation. Steering error was determined both by steering competence and by required steering-wheel angle while required steering-wheel angle was determined by speed. This confirms the model discussed in the introduction. Because drivers with poorer steering performance drove slower, while their steering errors were larger, no significant relations of speed and steering errors with TLC were found. This suggests that drivers with poorer steering competence compensated their larger steering errors, which decreased TLC, by choosing a lower speed, which increased TLC. Since steering competence did not affect TLC, it can not be concluded that drivers with poorer steering competence were less safe drivers. Steering error ratio correlated significantly with steering-competence as measured by the steering integral. The strong effect of steering competence on the steering errors during curve negotiation suggests that the steering integral is a good indicator for the quality of steering performance and that steering performance is consistent within the driver.

Based on the finding that steering error is a linear function of required steering-wheel angle and on the constancy of the minimum TLC to the inner lane boundary, the speed in curves as a function of radius was calculated using a mathematical model. From this, lateral acceleration was computed. Lateral acceleration proved to be an inverse function of speed as a by-product of the presented driver strategy.

Thus it appears that both radius as a road design element and steering competence as a driver characteristic exercise their influence on driving behaviour in the same manner. Both affect operational performance resulting in an adaptation of behaviour on the tactical level in an attempt to control safety margins. This is of theoretical significance for driving modeling in general since it suggests that effects of various factors related to the vehicle, weather, road, traffic, temporary states and the driver on behaviour on the tactical level (c.f. speed choice) may exercise their influence via an effect on operational performance. Most driver models are exclusively directed at either the operational or the tactical level. However, it is suggested that the relation between operational performance and behaviour on the tactical level should be a fundamental element in driver modeling.

 


Instrumentation: the driving simulator

0 comments
Instrumentation: the driving simulator

This is Chapter 3 of the Thesis from Adaptive control to adaptive driver behaviour, by van Winsum, 1996.

Other chapters of this thesis can be found here:

 

3.1. Background

The preparation of the experiments discussed in this thesis required a substantial amount of software design and implementation for the TRC driving simulator. A full description of the functionality and implementation of the simulator is beyond the scope of this chapter. The reader is referred to other documents for more detail and background (for example Van Wolffelaar & Van Winsum, 1992; Van Wolffelaar & Van Winsum, 1994).

The driving simulator of the Traffic Research Center (TRC) was developed as an instrument for behavioural research of driving. The GIDS project in which the TRC was involved at that time required a simulation testbed for mathematical driving modeling. Because of the dynamic complexities of driver tasks in multi-vehicle traffic, a dynamic traffic simulation was required as a test tool (Van Winsum and Van Wolffelaar, 1993). The objective of GIDS, an acronym for Generic Intelligent Driver Support, was ‘to determine the requirements and design standards for a class of intelligent co-driver (GIDS) systems that are maximally consistent with the performance requirements and performance capabilities of the human driver.’ (Michon and Smiley (1993). A prototype system was developed as a demonstrator for the essential features of the GIDS concept. One of the functions of the GIDS system was to compare required driving behaviour with actual behaviour. Required behaviour was modeled for a number of driving tasks and implemented in a computer system (Van Winsum, 1991; McLoughlin et al., 1993). It was decided at that time to design and implement a dynamic traffic simulation model and connect this with the driving simulator under development. From that moment the driving simulator evolved as a dynamic driving simulator with an interacting traffic world that could be connected to the GIDS system to serve as a test bed. After this, the simulator was also used as a testbed for other external driver support systems.

However, most importantly, the simulator is an instrument for the study of driver behaviour. Because it is used by researchers with very different questions and requirements, flexibility in software design has been considered to be important. This was accomplished by using the object-oriented computer language C++, and a multi-purpose UNIX machine instead of a single purpose dedicated simulation machine. To further increase flexibility for the researchers and to facilitate the design and testing of the experiments reported in this thesis, a fourth generation simulation language, SSL (Scenario Specification Language), was developed for the specification of experiments, together with a specification language (NSL, Network Specification Language) and software tools for roadnetwork creation. Data-sampling and data-processing facilities were added to facilitate experimentation.

 

3.2 The structure of the simulator

The simulator is composed of a number of software and hardware components that are connected via interfaces. The ‘conventional’ driving simulator consists of a physical car mockup, a car model implemented in software and a graphics system, together with a static road network environment. A dynamic traffic environment is added to this. The structure of these components as well as the object relations are shown in figure 1. In this figure several types of relations can be seen. An “is-a” relation specifies that a certain object type is a subtype of an other more abstract object type. For example, a BMW-car is some kind of car. This means that it inherits the functionality of the more abstract object type car. A “has-a” relation specifies that a certain objects has another object as a member. For example, a car has a traverser. The heavy printed arrows specify the direction of the flow of information. For example, there is an information flow from the object roadnet to the object sensor. This means that a sensor requests information from a certain instantiation of the object roadnet.

 

 

Figure 1. Logical structure of components of the driving simulator and relations between objects.

 

In addition to this, a number of facilities related to data-sampling and processing and experimental control are added to the simulator.

 

Car cabin. The steering wheel, clutch, gear, accelera­tor, brake and indica­tors of the car (a BMW 518) are connected to a Silicon Graphics Skywriter 340VGXT compu­ter (IRIS). Electromotors and other electronic appliances are built in the car to excert forces on the pedals and steering wheel and to send data on the steering wheel, accelerator pedal, brake pedal and indicators to the IRIS computer for further processing by the car model.

 

Car model. The IRIS computer processes these signals in a separate process referred to as the car model. The car model is described in more detail in Spaargaren (1994). It computes the longitudinal and lateral speed and acceleration that are the result of physical characteristics of the car and the input from the car cabin. From this the new coordinate position in the artificial world the car is driving in is computed. The output of the car model is used by the car traverser and by the graphics system. The traverser constitutes the link with the dynamic traffic process while the graphics system presents the output of the full system in a real-time visual format to the driver.

 

Graphics system. On a projection screen, placed in front, to the left and to the right of the driver, an image of the outside world from the perspective of the driver with a horizontal angle of 150 degrees is projected by three graphi­cal videopro­jectors that are controlled by the graphics software. Images are presented with a rate of 15 to 20 frames per second, resulting in a sug­gestion of smooth move­ment. The visual objects are buil­dings, roads, traffic signs, traffic lights and other vehicles.

In addition to this, the sound of the engine, wind and tires is presented by means of a digital soundsampler recei­ving input from the simulator computer.

 

Logical network (Roadnet). The logical network is the static environment in which the simulator car and traffic operate. The static environment consists of a database with a network of roads, traffic signs, traffic lights and buildings. This database is used for the visualization of the environment by the graphics system and by the artificially intelligent traffic to evaluate the present situation. The database can be generated in two ways:

– by NSL (Network Specification Language). This is a user specification language, created for the TRC simulator (Van Winsum, SSL/NSL specification release 1.2, 1994), by which a network of roads can be specified as an ASCII text. This text is processed by an NSL interpreter program that generates a road network database that is used by the simulator (Van Winsum, 1994, NSL scanner/ parser/interpreter computer program).

– by means of an interactive graphical program written in C++/OSF Motif (Van Winsum, 1993, program WORLDED). The user can specify a network of roads by means of click and point operations. The output of the NSL interpreter can also be used as input for this program to visualize and change the network.

 

The network consists of a structure of three base tables: a table with intersections, a table with paths and a table with segments. An intersection is a point in the network coordinate system with 1..n, {n >= 1}, outgoing paths. Coordinates are in meters. The following relations hold:

 

– n = 1: the intersection forms a terminal point in the network. If cars approach this intersection they cannot proceed beyond the intersection and provisions are made to ensure that the car turns around in the opposite direction as soon as the intersection is reached. The intersection has no physical layout and has the appearance of an ending road. The implication is that it is not possible for cars to move off the logical world.

– n = 2: the intersection is a virtual intersection in the sense that it has no specific layout and is not treated as an intersection by the traffic. The only purpose of creating such an intersection is for the convenience of the network constructor.

– n > 2: the intersection has more than two branches.

An intersection is of a certain type (f.i. a roundabout), it can be controlled by traffic lights with a certain control strategy, and it contains a list of references to outgoing paths. This list is ordered such that the path connections to the intersections are counterclockwise. In addition to this the intersection contains information about the layout, which is used by the graphics system and by the traffic.

 

A path is a logical connection between two intersections and always has one direction. It must start at one intersection A and end at one intersection B, where A may be equal to B. If A=B then the path is logically a circular path. All paths  have precisely one path in the opposite direction, referred to as a counterpath. It has a list of references to segments with 1..n elements, {n>=1} . This list is ordered such that the segments are in successive order. A path also contains information on right-of-way at the intersection at the end of the path, whether entry into this path is allowed, a reference to a traffic light at the end of the path if there is one, and information on buildings on the right side of the segments on the path.

To every path an ordered list with references to cars is attached. This list is ordered such that it reflects the order of the cars on the path and it may be empty. Cars can be added or removed at any time during the simulation process. In this way the simulator car and the computer controlled cars are connected to the static environment. Because every car is an object in the software-engineering conception that it has its own functions and data-structures, every car performs its own administration of detailed position (coordinates, distances from the last intersection and from the edge of the road etc.) in relation to the logical network.

The concept of path corresponds to the terminology of graph theory. Using that terminology, intersections are nodes.

 

The combination of nodes and paths may be described as a directed graph with the following properties:

– Suppose the network is represented as the graph G=(P,Z), with P being the set of intersections or nodes and Z being the set of ordered relations between the intersections, then P = {0..n} with n > 0. The fact that all intersections are member of a set ensures that all members occur once. The number of the intersections are in successive order. A set of intersections is, for example, {0,1,2,3}, meaning that there are 4 intersections. The set {0,1,3,4} is incorrect because the number 2 is missing.

– Z contains the relations between two nodes A and B, for example {{1,2}, {1,3}, {1,1}}. If {A, B} is a member of Z then {B,A} is also a member. This shows that all paths have a counterpath. A road can be traveled in two directions and this is the reason that every path has a counterpath. If only one-way traffic is allowed there are still two paths because physically it is possible to enter a one-way street into the wrong direction although legally it is not allowed.

– The fact that Z is described as a set suggests that the member {A,B} may occur only once. This restriction has been abandoned for practical purposes. There may be more than one instantiation of the relation {A,B}. In that sense Z is not a set but a collection. This restriction was loosened because sometimes there is more than one road between two intersections.

– A further restriction to the graph specification is that all nodes must occur in at least one relation, that is, a node that is fully unconnected is not allowed.

 

A segment is represented as a line through the middle of a roadpiece. It can be either straight or curved and is undirected. Segments are members of ordered lists connected to a path and the ordered list must contain al least one segment. A segment must be a member of one and only one ordered list. Segments represent the physical layout of the road, while a path represents the logical presence of a road. The direction depends on the path the segment is in. If the segment is straight the two end points are given in coordinates. If it is curved the segment contains the necessary information on the curvature, such as the radius, the centerpoint of the arc etc. A segment has a certain lane-width. At present only two-lane segments are allowed.

Traffic signs, buildings and traffic lights are connected to the network and have a certain position, angle, and type. Within the simulator program this roadnet representation is implemented as the separate object class in the roadnet module (Van Winsum, 1992, computer program class c_roadnet, roadnet.c). This object performs its own administration and can be queried from outside via an interface.

The following is an example of a definition of a simple network with NSL.

Define Inter[0] {

X := 100; Y:= 100;

}

Define Segment[0] {

Type   := Straight;

StartX := Inter[0].X;

StartY := Inter[0].Y

Length := 100;

Angle  := 90;

}

 

In this definition a straight road of 100 meters with an absolute angle of 90 degrees is created, starting at coordinate position [100, 100]. Paths are added automatically by the system. Since this definition of a network would result in a path without an end node, the system creates an end node (intersection number 1). Since the lane-width is not specified, the segment is assigned the default lane-width of 3 meters by the NSL system. In this case the NSL interpreter creates 2 intersections, 2 paths and 1 segment, no traffic signs, traffic lights or buildings. NSL contains a number of geometric transformation methods and rules to assists the user and to make it easier to build the network.

 

Traffic. Traffic consists of a list of cars that may be controlled by a human driver (the simulator car) or by an artificially intelligent ‘driver’. Every car has a number of properties, such as a length, a width, a wheel-base and so on and a number of objects that are needed for driving in the logical world. These objects are a traverser, a sensor and a decision (control) mechanism. In the case of a human-controlled car the decision mechanisms is of course the human driver who, together with the car model, determines the movement of the car. In the case of a computer controlled car the decision mechanism is composed of a set of decision rules. Traffic is implemented in the simulator program as a separate object container class (Van Winsum, 1992, computer program class c_traffic, traffic.c). It contains all kinds of methods for adding or removing cars from a traffic list. The class traffic contains references to cars which may be very different in type. The mechanisms of late binding and virtual classes and inheritance, which are properties of the object-oriented methodology used, ensure that in the future all kinds of other moving objects such as pedestrians and bicyclists may be added to traffic. Every car has its own instantiation of a traverser, sensor and control object. These objects also may be of different types. For example, a human controlled car (the simulator car) would need a somewhat different traverser than a computer controlled car or maybe a pedestrian.

In the case of a human driver, the output of the car model, i.e. the speed and the angle of lateral displacement, are input for the traverser. For computer-controlled cars, the output of the artificially intelligent decision mechanism is the input for the traverser. The traverser calculates the lateral position (with respect to the right side of the road), the longitudinal displacement with respect to the road, it connects the car to the network of roads, checks which path is selected if the car is on an intersection and performs a number of other checks to maintain the position of the car accurate with respect to other traffic. It uses deadreckoning techniques in this process. The traverser is the interface between traffic and the road network and it also connects the simulator car with the interactive traffic world. The traverser is implemented as a separate object class in the simulator program, such that every car has a reference to its own instantiation of a traverser object (Van Winsum, 1992, computer program class, c_traverser, travers.c)

The sensor is an object that really consists of a set of sensors. Both the human controlled car and the computer controlled cars have a sensor object but they use it differently. In general, the sensor is used to ‘look’ into the network. In this way every car, including the simulator car, can evaluate the present type of road and curvature, evaluate the distance and speed of traffic in front etc. This means that the sensor is an interface between the network and the car in that it requests information from the network. The human controlled car uses this information for data storage purposes and to give input to driver support systems. The computer controlled cars use this information for the decisions they are required to make concerning their speed and course. Sensor is implemented as a separate object class in the simulator program (Van Winsum, 1992, computer program class c_sensor, sensor.c). Every car has a reference to its own instantiation of a sensor object.

The control mechanism for the human driver is the human information processing system that uses visual information received via the graphics system, to exert the controls in the car cabin. These car control signals are processed by the car model. The output of the car model is used to update the graphics and as input for the traverser that connects the simulator car to the network. The control mechanism of the computer controlled cars consists of a set of decision rules. Every computer controlled car has rules for different driver tasks on the tactical level. These tasks are related to curve negotiation, car-following, overtaking, negotiating intersections, speed choice on straight roads and processing road sign information. The car evaluates which tasks are presently performed and computes a required speed and lateral position. Since multiple tasks can be performed at the same time, a decision mechanism selects the appropriate speed and lateral position together with the required acceleration and wheel-angle to reach this state, after all tasks have been evaluated for the present car . This results in a natural and human-like behaviour that contributes in an important way to the fidelity of the simulator. For computer controlled robot cars the artificial intelligence is implemented in a separate object class in the simulator program (Van Winsum, 1992, computer program class c_control, control.c).

3.3 Data collection and processing

A large quantity of performance data can be collected with any sampling frequency. Examples are time-to-collision, time-to-intersection, time-to-line crossing, lateral position, speed, acceleration, brake force and so on. The user creates an ASCII text with keywords that specify the sample frequency and the data to sample with that frequency. The data are then sampled during a simulation run and stored into a binary file. The real-time handling of data-storage during a simulator run is controlled by a separate object class c_data that is implemented in the simulator program (Van Winsum, 1992, computer program class c_data, newdata.c).

After a simulator run the data can be visualized and preprocessed with a graphical program written in C++ and X-windows/OSF motif (Van Winsum, 1994, program DATAPROC).

 

For the experiments described in this thesis the real-time sampling of time-based information was required. The variables used are TTC (time-to-collision), TLC (time-to-line crossing) and THW (time-headway during car-following). These measures are defined and implemented as follows:

 

– THW is defined as D/u

for u > 0, else THW = infinite (undefined)

with D = bumper to bumper distance in meters along the path between

the simulator car and the lead vehicle, and

u = speed of simulator car in m/s

 

– TTC is defined as D/(u – ulead)

for (u – ulead) > 0, else TTC = infinite (undefined)

with ulead = speed of lead vehicle in m/s

 

– TLC is calculated differently depending on whether the car is on a straight road or in a curve.

In general, TLC = DLC/u,

for u > 0, else TLC = infinite (undefined)

with DLC = distance to line crossing along the vehicle path and

u = speed of simulator car in m/s.

 

DLC is solved goniometrically using the cosine rule. Normally, the car is not driving in a straight line but it alternates between curves to left and to right. The radius of the vehicle path is calculated using the coordinates of the centerpoint of the curve the car is driving. This centerpoint is calculated as the point where the perpendicular lines through the frontwheel and the rearwheel, with slipangles added to the wheelangles, intersect. Rv, the vehicle radius, is then computed as the distance between the center of gravity of the car and the centerpoint of the vehicle curve. Rv1 then is the distance between the front (left or right) wheel and the centerpoint of the vehicle curve. TLC then measures the time until either the left or right front wheel crosses the left or right lane boundary, given the current vehicle path and speed.

First the case for straight roads is described if the vehicle makes a left turning curve, see figure 2. DLC is computed as a*Rv1. Since Rv1 is known, only a has to be computed, using the cosine rule.

 

 

Figure 2. Determination of the length of the arc DLC for driving on straight roadsections.

– The length of the linepiece A is computed as Rv1-(dleft/cos(ha)), with dleft being the distance between the left frontwheel and the lane boundary (in a line perpendicular on the road) and ha the angle between the line perpendicular on the road and the line from the front wheel to the centerpoint of the vehicle curve.

– The length of the linepiece C is computed as  (2*A*cos(ß)+Ö((2*A*cos(ß))2-4*(A2-Rv12)))/2

Then a = arccos((A2 + Rv12 – C2)/(2*A*C))

and DLC = a*Rv1

 

Figure 3. Determination of the length of the arc DLC for driving on curved roadsections.

 

Figure 3 shows the situation for determining the TLC while the car is negotiating a road curve. Again, DLC is determined as a*Rv1. In this case a is computed differently.

– The length of linepiece A represents the distance between the centerpoint of the roadcurve and the centerpoint of the vehicle curve.

– Angle ß is computed as the angle difference between the line from the centerpoint of the vehicle curve to the centerpoint of the roadcurve and the line from the centerpoint of the vehicle curve to the left front wheel (if the vehicle turns towards the inner lane boundary).

– Angle a1 is computed as arccos((A2 + Rv12 – Rr2)/(2*A*Rv1))

– a= ß – a1 and DLC = a*Rv1

 

In addition to this, vehicle control information was required for the experiments. The position of the accelerator pedal, expressed as a percentage pressed, the position of the brake pedal and the force excerted by the foot on the braking pedal were used in the studies on car-following, while steering wheel angle was used in the study on steering performance and curve negotiation. The results of these control actions, such as speed, acceleration, heading angle and lateral position, were sampled and processed as well.

For every experiment automatic data processing programs were written to extract and process the required data. These data were then transformed into a format suitable for processing by SPSS.

 

3.4 Scenario Specification Language (SSL)

SSL is a user specification language that was defined and implemented as a tool for specification and design of experiments. It contains most of the functionality of the simulator. A description of SSL then essentially gives a description of the functionality of the TRC simulator. For a full specification of the language the reader is referred to the SSL/NSL specification document (Van Winsum, 1994).

An ASCII file with SSL commands is analyzed by a scanner and parser module during initialization of the simulator program and syntactical errors are reported to the user. (Van Winsum, 1994, SSL scanner/parser/interpreter modules). If no syntactical errors are found, the SSL commands are converted to an internal data-structure that is interpreted in real-time by the SSL-interpreter during execution of the simulation process. Since the simulation process is a dynamic process in which the state is determined by SSL specifications, the human driver, the behaviour of traffic and by the process operator who interacts with the computer via the user interface, the course of events is not deterministic. However, SSL commands can be used to force identical situations for all subjects in an experiment. Since SSL commands are often conditional, the state of the traffic world can be queried and events can be triggered if some condition is true or false.

Scenarios are defined in a SSL text file. A scenario is a predefined list of situations with a start and an end condition: the scenario starts when the start condition is fulfilled and terminates when the end condition is fulfilled. A scenario may involve 0..n cars, referred to as participants, in addition to the simulator car. A participant is a car that performs conditional actions. A scenario may be used for controlling traffic and traffic lights, for indicating when data must be stored, for communication with the driver with spoken or written messages, for sending messages to other devices, and so on. SSL is not exclusively a language for specification of traffic situations during an experiment. It also may be used for rapid prototyping of driver support systems, for creating test situations and for debugging. It is important to note that SSL is often used to override default settings and default behaviour. For example, if a computer-controlled car is created with SSL, the car follows its own rules unless specified differently with SSL.

The following is an example of an SSL description.

 

Define Scen[1] {

Var { time; }

Start {

When ( Part[MainTarget].LeadCar != Absent and

Part[MainTarget].DisToLeadCar < 50 );

Scen[].NrTimes := 1;   time := runtime();

}

End {

When ( runtime() – time > 20 );

}

Define Part[1] {

Start {

Part[].CarNr  := Part[MainTarget].LeadCar;

Part[].MaxVelocity := 50/3.6;

}

End {

Part[].MaxVelocity := 100/3.6;

}

}

}

 

This scenario specifies that if there is a lead vehicle and the distance to it is less than 50 meters then this lead vehicle starts to drive with a maximum speed of 50 km/h during 20 seconds. After 20 seconds (at the end of the scenario) this vehicle pulls up to a speed of 100 km/h.

SSL files contain the full script for an experiment and are thus complete specifications of an experiment. This ensures repeatability and detailed documentation of experiments. Since researchers are able to make their own SSL script files they can design and test experiments with a minimal dependency on technical staff and computer programmers.

 

3.5 The use of the simulator in the experiments

The driving simulator offers a number of advantages compared to studying driver behaviour on the road.

1) The sensors of the simulator car and of other cars used in the car-following experiments contain important information that is much harder to obtain with current technology in a real world test situation. This information is vital as input for the control of experiments and data-sampling. For the experiment on curve driving the measurement of TLC information during curve negotiation would be very hard to obtain in real world driving. A simulator is the only practical way to obtain complex measures such as the TLC in curves. Although time-to-collision information may be obtained during on-road experiments it is measured more practically and efficiently in the simulator.

2) All kinds of situations can be generated and tested that would be very hard or impossible to generate in the real world. In the curve negotiation experiment the drivers are required to negotiate a number of different road curves with a specific lane width and radii. Roads with the precise characteristics required by this experiment are very hard to find in the real world. During the car-following experiments the lead vehicle was sometimes required to drive with a certain fixed time-headway in front of the simulator car. This would be difficult to establish in on-road experiments.

3) The responses of drivers to maneuvers too dangerous to be tested in real world driving can easily be tested in the simulator. This is especially important in the car-following and braking experiments discussed in the chapters 5 to 9.

4) Situations can be brought under experimental control. This is important for the comparability of the results since all subjects have encountered precisely the same situations. In on-road experiments traffic density and weather conditions are hard to control. In this respect a simulator has important advantages compared to real world experiments.

 

In the experiments performed in the context of this thesis, the time-based safety margins TLC and TTC play an important role. The perception of TTC has been studied in a large number of experiments (see chapter 6). These studies have given strong support for the idea that TTC information is extracted from the optic flow field. The expansion of the image on the retina gives sufficient information for the extraction of TTC information without requiring the driver to assess speed or distance information. Since the graphical properties of optical perspective, visual angle and optical expansion rate are the same in the TRC simulator as in real world driving, there is reason to assume that the driving simulator is suitable for the type of research discussed in the chapters 5 to 9. An important prerequisite for a smooth optic expansion is a high graphical frame rate. In order to obtain a high frame rate, the visual scenes in all experiments are limited to the essential components while substantial effort has been invested in the design of fast algorithms for traffic handling and experimental control.

 


Chapter 2: Thesis traffic psychology

0 comments
Chapter 2: Thesis traffic psychology

This is chapter 2 of the thesis “From adaptive control to adaptive traffic behaviour” about traffic psychology and behavioural adaptation of drivers, by Wim van Winsum. The thesis is from 1996. It describes a number of behavioural experiments into car driving that were performed in a research driving simulator.

Other chapters of this thesis can be found here:

 

                                                Models of driving behaviour

 

2.1 Introduction

 

A wide range of models of driving behaviour has been described in the literature that typically emphasize a specific aspect of car driving. Some models emphasize operational performance while others stress the importance of behaviour on the tactical level. Also, some models focus on individual differences while others emphasize situational factors. The number of serious attempts at categorization of driver behaviour models is limited. The problem is that the categorizations are almost always too limited, exclude important models or are wrong according to the advocates of some models. However, an important attempt is the one proposed by Michon (1985). Michon has made a distinction between taxonomic and functional driver behaviour models. Taxonomic models are inventories of facts while functional models specify relations between components. The best known example of a taxonomic model is the task analysis of the driving task developed by McKnight and Adams (1970). This task analysis specifies the driving task in terms of behaviour requirements (for a distinction between several types of task analyses, see Hackman, 1969). This means that it describes what the driver should do. The task analysis of McKnight and Adams is not aimed at understan­ding driver behaviour or at describing how the driver actually drives. Given the purpose of driver education, this is not surprising. However, since it is not aimed at under­standing driver behaviour, it is not discussed here.

In the sixties, the early concept of accident proneness was replaced by studies of indivi­dual differences in accident involvement that focused on psychologi­cal abilities. Together, these studies have become known as the skill model. The term ‘skill model’ is actually a misnomer, because the model centers around psychological abilities instead of skills. The use of the word ‘skill’ suggests that the model focuses on car driving performance while in fact psychological abilities are tested and correlated with accident involvement. Proponents of this model assume a relation between psychological abilities and car driving skills but generally fail to test this relation explicitly. Still, because of the general use of the term in traffic psychology, skill model is used in this thesis. Situational factors were hardly given any consideration in this approach. The dependent variable has been accident involvement instead of actual car driving behaviour. The studies in the tradition of accident proneness and differential accident involvement belong to the taxonomic models and focus on indivi­dual differen­ces measured by psychological ability tests.

Individual differences are also stressed in studies on the effects of aging. Since elderly road users differ in car driving performance from younger drivers and are assumed to suffer from skill degradation having an effect on accident involvement, the studies on this issue are grouped under the heading of the skill model in this thesis. However, it must be stressed that these studies differ strongly from the correlation approach of the differential accident involvement studies. Also, behaviour on the strategic and the tactical level of car driving is expli­citly incorporated.

The effects of temporary states induced by alcohol and drugs are assumed to affect car driving skills. Traditionally, studies on the effects of temporary states are not part of the skill model if this  is envisaged as being equiva­lent to the correlation approach. Also, the focus is neither on individual diffe­rences nor on situational factors. Since temporary states may be seen as within-subjects manipulations of skill level they may give some insights in the workings of skill level on driver behaviour. This is why studies on the effects of temporary states are discussed under the heading of skill models in the present thesis.

Michon (1985) has categorized the motivational models as functional models. In most motivational models the emphasis is on risk. Since they have been introduced to a certain extent as a reaction to the skill model, the emphasis is on situational factors instead of individual differences. Skills and abilities are not regarded as important in motivational models and behaviour on the operational level has a low priority in this approach, or as Näätanen and Summala (1976) put it: “crucial to traffic safety is what the driver actually will do in any given situation, rather than his maximum level of performance and the environmental demands”. Motivational models mainly study behaviour on the tactical level, especially speed choice. Although individual differences have a low priority in motivational models, young (male) drivers are regarded as a subgroup of the driving population that deserves special attention because of their high accident involvement rate. Studies of the young male driver are grouped under the heading of motivational models, since in the literature motivati­o­nal factors are emphasized in the behaviour of the young driver.

The adaptive control models, also categorized as functional models by Michon (1985), focus on behaviour on the operatio­nal level and especially on steering behaviour. Individual differences are ignored while effects of situational and vehicle-related factors on operational behaviour are emphasized.

In the next paragraphs, and especially in the paragraphs 2.2.2 and 2.2.3, a number of experimental studies on driver behaviour are discussed.  However, it is not intended to give a comprehensive review. The results of studies on driver behaviour are merely referred to as illustrative examples for the model of driver behavior that is developed during the course of the next paragraphs.

 

2.2 Skill models

 

  • From accident proneness to differential accident involvement.

 

 The concept of accident proneness has been in vogue from the 1920s up until the 1960s, and played an important role in theories of driver behaviour. McKenna (1983) presented a conceptual analysis of accident proneness. The idea was that some individuals are more liable to be involved in accidents than others. The statistical techniques that have been applied to resolve this issue have given rise to substantial contro­versy. One of the problems mentioned by McKenna is that differential accident liability can always be attributed to differences in exposure to risk. Moreover, the lack of a clear definition of accident proneness has resulted in confusion. Several meanings have been assigned to the concept of accident proneness. Some have understood it as that most accidents are caused by a few people. This is associated with the definition of accident proneness as a disproportionate involvement in accidents in a statistical sense. However, the mere randomness of accidents suggests that some people have been involved in more accidents than others because of ‘bad luck’. Others have regarded it as an individual property, or as a persona­lity characteristic or disposition leading to a disproportional accident involvement. In that case accident proneness is a trait. However, the connection between these (personality) characteristics and actual car driving behaviour resulting in a higher accident involvement is unclear.

McKenna (1982) proposed the differential accident involvement approach as an alternative to the concept of accident proneness because this would offer a better theoretical understanding of the psychological abilities and characteristics associated with human error. Further advantages of this approach are that it does not suffer from the moral and emotional connota­tions associated with accident proneness, and that it is based on psycholo­gical testing instead of statistical modeling. The differential accident involvement approach evaluates the contribution of psychological abilities instead of personality factors to accident involvement. Although this approach has become known as an important representative of the so-called skill model, it is important to note that it is not driving skill as such that is being evaluated but psychological abilities that are assumed to be related to driving skills. Efforts were made to identify the psychological abilities critical to safe car driving. A substanti­al amount of research was devoted to the study of corre­lati­ons between performance on perceptual-motor tasks that were assumed to measure abilities required for safe driving on the one hand and acci­dents on the other hand.

Unfortunately, because this approach is purely correlational, the nature of the relation between psychological abilities and accident involvement is not made explicit at all. The existence of such a relation is assumed on intuitive grounds and based on face validity. Because the process controlling this assumed relation was not investigated, the effects of psychological abilities on operational driving performance and on behaviour on the tactical level have not been examined. Therefore, accident involvement has been the only dependent variable in this line of research. The results were generally disappointing. A small overview of some of the extensive relevant literature gives the following results:

Vision is generally accepted as being of central impor­tance in driving. Yet correlations between several visual performan­ce tests such as static acuity, dynamic acuity, visual field, glare recovery and recognition on the one hand and accident rate on the other, are typically lower than 0.05 (Rumar, 1988).

The psychological test that has probably been studied most often in relation to accident involvement is the embedded figures test (EFT) of Witkin. This test measures the cognitive style of ‘field independence’ and it requires that a simple form is found within a background. The EFT has been presented as predicting accident rate. Mihal and Barrett (1976) reported a correlation of 0.24 between EFT performance and accident involvement. Loo (1978) obtained a correlation of 0.42 with self-reported accident rate. However, Harano (1970) found a correlation of only 0.001 and McKenna et al. (1986) found a non-significant correlation of 0.19 between EFT per­formance and accident rate. Also, Quimby and Watts (1981) failed to obtain a significant correlation with accident involvement.

Other psychological tests, such as the dichotic listening test, Stroop test and reaction time tests also have been reported to be poorly related to accident involvement (McKenna et al., 1986; Quimby and Watts, 1981).

Noordzij (1990) reviewed the German literature on individual differences and accident liability. Perfor­mance measures on a wide range of tests failed to predict safe driving in any of the reviewed studies. Some studies even reported relations contrary to the expected direction, such that better performance in the laboratory and on the road was associated with poorer accident histories.

 

McKenna et al. (1986) gave two explanations for the low correlations. The reliability of accident scores is low when these are obtained over only a few years. This makes it impossible to obtain high correlati­ons between accident rate and test per­formance. Furthermore, accident rate probably reflects diffe­rent psychological abili­ties that cannot be captured in a limited number of tests. Häkkinen (1979) demonstrated that the reliability of acci­dent scores increases by lengthening the time over which accidents are measured. He argued that the lack of significant relations between test scores and accident involvement in so many studies was caused by short exposure periods and poor control of environmental risk. Häkkinen studied accident involvement of professional bus and streetcar drivers and found significant differences between safe drivers and acci­dent involved drivers on a number of psychological tests measuring, for example, eye-hand coordination, choice reaction time and psychomotor personality factors. The correlations were over 0.40. The study of Häkkinen has often been referred to a evidence for the skill model, and it is one of the few studies that supports the model.

In summary, psychological abilities assumed to be related to driving skills have proven to be unrelated to accident involvement, except perhaps for professional drivers. Summala (1985) explained the results of Häkkinens’ study by the forced-paced nature of the driving task for this group of drivers. The task of pro­fessional bus drivers is paced by time-schedules and differs from the task of private drivers who are able to decrease the speed, overtake less often or avoid bad conditi­ons. The explanation suggested in this thesis is that the driver adapts behaviour on the tactical level to the level of operational performance if the driving task is self-paced. This prevents a higher accident involvement for drivers with poorer psycho-motor abilities. This ofcourse assumes that drivers with poorer psycho-motor abilities are characterized by poorer operational performance. However, when the task is forced-paced adaptation is not possi­ble. Unfortunately, the effects of psycho-motor abilities on operatio­nal performan­ce and on tactical behaviour, such as speed choice, have not been studied, thus making it impossible to prove the existence of such an adaptive mechanism from the data presented so far. There is however evidence that adaptive processes play an important role in accident causation of elderly drivers, who suffer from age-related performance decrements and in some transient state-related performance decrements. In that case the term compensati­on is applied instead of adaptation.

 

  • Individual differences in skill: the elderly driver.

 

It is well documented that older drivers have to cope with declining vision and exhibit poorer performance on a wide range of tests of perceptual and motor ability and response speed (see for example Ysander and Herner, 1976). Ranney and Pulling (1990) found that older drivers (74-83 years of age) score lower on laboratory tasks requiring rapid switching of attention. Rackoff and Mourant (1979) reported poorer performance of older drivers on motor tests and especially on the embedded figures test.

Yet the accident rate of elderly drivers is lower than expected on the basis of the skill model, although the fatality amongst elderly drivers is quite high due to their physical frailness (Evans, 1988; Brouwer, 1989). Hakamies-Blomqvist (1994) found that older drivers had fewer accidents at nighttime and under bad weather and road-surface conditions compared to younger drivers. Older dri­vers were also less often in a hurry, alcohol intoxicated or distracted by non-driving activities compared to younger drivers. These results were interpreted as evidence that older drivers avoid more diffi­cult conditions. Ranney and Pulling (1990) reported that complex traffic situations pose problems for elderly drivers. They are more often involved in multiple vehicle intersection accidents, while they are less involved in single-vehicle accidents. They questioned the idea that older drivers have higher accident rates than middle-aged drivers. Although drivers over 65 make up 11.2% of the driving population in the United States, they are involved in only 7% of all accidents. A study of Cerelli (1989) was cited reporting that drivers over 75 have a crash involvement rate that is 2.5 times lower than that of drivers aged 40, and 5 times lower than that of 20 year old drivers. According to Brouwer and Ponds (1994) the fatality risk for drivers of age 70 is about three times as high compared to drivers at age 20, due to physical changes such as osteoporosis and decreased cardio­vascular efficiency resulting in an increased physical vulnerability. Correction for this increased vulnerability gives a better impression of actual accident involvement of older drivers compared to younger drivers. Application of this correction factor resulted in almost equal casualty risks for 35 and 70 year old drivers in the Netherlands in the eighties. Evans (1988) also found that when correcting for increased vulnerability, fatalities for older drivers are less than for male drivers under 20.

The results suggest that, although older drivers suffer from decreased performance on most tests of psycho-motor and attentional abilities, their accident risk is not dramatically different from drivers of other age groups. In situations with high time-pressure and situations beyond the control of the driver accident risk appears to increase for older drivers. A possible cause for this phenomenon may be found in the distinction between self-paced and forced-paced driving situations. When the driving task is self-paced, the situation allows the driver to compen­sate for performance deficits. However, compensation is impossible in forced-paced situations. In that case the driver is subjected to higher levels of time-pressure. The results may then be explained in terms of a process of adaptation: older drivers may compensate for their degradations of psycho-motor abilities by changing their behaviour both at the strategic level and the tactical level. There are a number of research findings in support of adaptive mechanisms.

The ultimate decision at the strategic level is to give up driving. Kosnik et al. (1990) found that older drivers who had recently given up driving reported more visual problems compared to older drivers who had not given up driving. The results suggested that older drivers are aware of their visual deficits and that this awareness influenced decisions about driving. At the strategic level decisions are also made regarding the time of driving. Planek and Fowler (1971) and Ysander and Herner (1976) found that older drivers avoided driving in the dark, on icy roads and in unknown cities more than younger drivers. According to these authors, self-selecti­on seems to be a factor of great importance when judging the traffic safety risks of elderly drivers. Older drivers also may compensate for their age-related impair­ments by limiting their driving and avoiding risky situations and rush hours (Ranney and Pulling, 1990). In addition to this, there is some evidence in support of compensation at the tactical level. Ranney and Pulling found that older drivers drive slower compared to younger drivers. This was also reported by Rackoff and Mourant (1979). They cited the studies of Case et al. (1970) and Rackoff (1974) in which it was found that the vehicle speed of older drivers, in an instrumented vehicle, was about ten percent less than the speed of younger drivers. The tendency of older drivers to drive at lower speeds was also referred to by Rumar (1987). The proportion of accidents where speed is below average increases as a function of age.

 

  • Variations in skill as a function of temporary states.

 

Both the consumption of marijuana and alcohol result in temporary state changes. This is generally assumed to temporarily affect perceptual-motor abilities. Because of this, the literature on temporary states is discussed under the heading of the skill model, although it must be stressed that in practice this field of research is treated as a separate problem domain, while the results of these studies are normally not related to a specific driving model. The line of reasoning in this thesis is that marijuana and alcohol may effect psycho-motor abilities which may affect operational driving performance. The factors marijuana and alcohol may be considered as a natural experiment in which perceptual-motor abilities are manipulated within the driver. This then offers interesting opportunities to study the effects on tactical behaviour and the relation with accident involvement. The effects of this within-subjects manipulation on driving skills and driving behaviour may then give important information about the workings of the process of adaptation.

 

 

Marijuana. Moskowitz (1985) reviewed a large number of studies on the effects of marijuana on psychological abilities. In reaction time experiments neither the speed of initial detection nor the speed of responding appears to be affected by marijuana, although the frequently reported increase of RT variability suggests that attentional mechanisms are impaired by marijuana. Tracking is significantly affected by marijua­na. Also, perceptual functi­ons and vigilance are negatively affected by this drug.

However, based on a review of a number of epidemiological studies, Moskowitz (1985) concluded that there is little evidence for an increased risk of accident involvement under marijuana. Robbe (1994) reviewed the epidemiological literature as well and concluded that some people do drive after cannabis use and that drivers involved in accidents often show the drug’s presence. However, because alcohol has been a severe confounding factor in all surveys of accident-involved drivers, the independent contribu­tion of marijuana to accidents remains unclear.

The effects of marijuana on driving behaviour has been examined in many experiments. According to Robbe (1994), the foremost impression one gains from reviewing the literature is that no clear relationship has been demonstrated between marijuana and either seriously impaired driving performance or the risk of accident involvement. Smiley (1986) compared simulator and on-road studies of marijuana effects on car driving perfor­mance. In simulator studies with realistic car dynamics and in interactive simulators strong effects of marijuana on operational performance were found. In a study of Smiley et al. (1981) in an inter­active driving simulator variability of veloci­ty and lateral position increased during curve negotiation and while following cars and in windgusts. Variability of headway and lateral position while following cars also increased under marijuana. However, a larger headway was chosen during car-following under marijuana. In a study by Stein et al. (1983) with an interactive simulator, performance effects of marijuana were examined in a number of driving tasks such as car control during windgusts, curve following and lane changes. Although there were effects on steering performance, mean driving speed was lower under marijuana.

Several other studies have presented behavioural evidence suggesting that drivers may adapt their tactical behaviour to deteriorated operational performance by choosing a lower speed or by increasing headway in car-following. In an on-road study by Caswell (1977) drivers under marijuana drove more slowly. In an on-road study by Smiley et al. (1986) the effects of marijuana on several tasks such as car-following, curve following, open road driving, emergency decision making and obstacle avoidance were measured. Marijuana only had a few effects, but it signifi­cantly increa­sed headway in the car-follo­wing task. Smiley (1986) concluded that all studies indicate that when the driver under marijuana has the possibi­lity to choose a lower speed, there are no effects on lane position control while speed is reduced. Stein (1986) studied the effects of marijuana on driving behaviour in a number of driving tasks in a simulator. A dose dependent effect of marijuana on speed was found; drivers decreased speed more with higher doses. In a task requiring the driver to compensate for random wind gusts, a strong effect of marijuana was found on mean speed and speed variability. Drivers were also required to control speed and steering during the negotiation of curves. Again, marijuana decreased speed. The speed reduction was also found in an obstacle avoidance task. No effects of marijuana on steering behaviour were found.

Robbe (1994) performed three on-road experiments in which the effect of marijuana on car driving was examined. In a study with driving on a restricted highway it was found that marijuana affected steering performan­ce as indicated by an increased standard deviation of lateral position (SDLP). Subjects were instructed to maintain a constant speed of 90 km/h, or less if they felt incapable of driving safely at that speed. The greater the dose, the harder the subjects attempted to compensate as indicated by perceived effort and increased heart rate. Despite the instruction, there was a small reduction in mean speed under marijuana. Drivers rated the quality of their own driving performance lower with higher doses, suggesting that they were aware of the effects of marijuana.

In another experiment, Robbe (1994) had subjects drive on a highway with other traffic under the instruction to maintain a speed of 95 km/h. This also involved a car-following test in which subjects were instructed to maintain a 50 meter headway. A marijuana dose-dependent increase in SDLP was found and a decrease in speed under marijuana. Also, under marijuana headway increased although the increase was highest with the smallest dose. Reaction time to speed changes in the preceding vehicle increased under marijuana. However, reaction time was confounded with headway, such that RT increased with increased headway.

In a third experiment, Robbe (1994) examined the effects of marijuana in a city driving task. Driving performance was evaluated by trained observers (driving instructor). No effects of marijuana were found on driving performance. Under marijuana it took more time to complete the circuit, suggesting a lower speed, although this was not significant. Drivers under marijuana perceived their driving quality as poorer compared to placebo and perceived their effort as higher.

In conclusion, the studies of the effects of marijuana suggest that, firstly, it affects perceptual and psycho-motor skills, secondly, it affects performan­ce on the operational level, and thirdly, it affects behaviour on the tactical level, especially when the task is self-paced. Evidence was presented that the drivers are aware of performance decrements under marijuana. It may be hypothesized that the perception of feedback of these performance decrements is a necessary prerequisite for such a compensation strategy. However, the nature of the perception of feedback, whether it is conscious or unconscious, is at present unclear. When the task is self-paced instead of prescribed by the experimenter (by instructing the subject to maintain a fixed speed), effects of marijuana on operational performance may be limited due to compen­sation for decreased skills: when drivers are allowed to choose their speed, effects of marijuana on steering behaviour are generally absent, while effects on steering behaviour are found when speed is prescribed by the experimenter. This compensation mecha­nism may explain why epidemiolo­gical studies have been unable to find a relation between marijuana and acci­dent involvement.

 

 

Alcohol. A substantial part of the literature on accidents and driver behaviour concerns the effects of alcohol. The effects of alcohol on performance are well documented for a large number of tests. Only a few examples are given here. Moskowitz and Robinson (1986) reviewed the literature on the effects of alcohol on task performance. They analyzed the results of 178 studies that fulfilled regular methodological criteria. Forty-five percent of the studies indica­ted impairment at 0.04% BAC (blood alcohol concentration) or less. The majority of studies reported impair­ment at below 0.07% BAC. Impairments were found in tracking, divided atten­tion, information processing, eye movements and psycho-motor skills, especial­ly in tasks requiring skilled motor performan­ce and coordination. Divided attention deterio­ra­ted already at very low BAC levels. Signal detection, visual search and recognition tasks also showed impairments at low BAC levels. Kennedy et al. (1989) measured the effect of BAC level on performance in a battery of nine tests measuring motor speed, symbol manipulation/reasoning, cognitive proces­sing speed and speed of response selection. Performance on eight out of nine tests was strongly and monotonously affected by BAC.

Evans (1991) estimated that 47% of fatal accidents, 20% of injuries and 10% of property damage are attributa­ble to alcohol. This means that alcohol contributes importantly to traffic accidents with the contri­bution increasing as crash severity increases. Evans (1989) concluded that eliminating alcohol would reduce traffic fatalities in the United States by 47±4 percent. Guthrie and Linnoila (1986), suggested that epidemiological studies indicate a disproportionate number of alcohol related fatal crashes involving young male drivers below 24 years of age. The majority of alcohol related accidents occur during the weekend, especially at evening hours, and in summer. According to Smiley (1989), alcohol is involved in 62 percent of all fatal single vehicle accidents.

There is also overwhelming evidence that alcohol affects operational driving performance. Louwerens et al. (1986) studied the effects of four doses of alcohol in a task where subjects were required to drive with a constant speed of 90 km/h with a constant lateral position between the right lane bounda­ries. Standard deviation of lateral position (SDLP) increased in a dose dependent manner as a function of alcohol. The subjective assessment of driving performance by the driver correlated poorly with SDLP and BAC level. This suggests that drivers were unaware of performance decrements under alcohol. In a simulator study with several driving tasks, Stein (1986) found that alcohol increased the number of accidents. Also, in a task requiring the driver to compensate for windgusts while following a winding road, steering behaviour was significantly affected by alcohol, and lane position variability was increased under alcohol. No effects of alcohol on mean speed were found, although speed variability increased under alcohol. Stein and Allen (1986) reported the results of an experiment that aimed to unravel the effects of alcohol on performance and risk taking. This is important because the effect of alcohol on accident involvement has often been attributed to an increase in deliberate risk taking. The effects of alcohol on driver behaviour was studied in a driving simulator and on a closed course. Both methods gave essentially the same results. Alcohol increased speed variability and the number of times the speed limit was exceeded. As drivers were well aware of the speed limit and the probability of detecti­on, and since speed feedback was available both visually and aurally, the increased variability suggested decrements in the driver’s perception and/or speedometer monito­ring. Also the frequency of running red lights was increased by alcohol. The subjective probability of running a red traffic light was affected by alcohol while risk acceptance was not affected by alcohol. Stein and Allen saw these results as evidence that the driver’s perception of speed and distance was impaired by alcohol, and that the drivers were unaware of this impairment. They concluded that the locus of effect of alcohol on risk taking is on the perceptual level instead of the risk acceptance level. Wilde et al. (1989) investigated the effect of BAC on performance on a response timing task and a general knowledge quiz. The findings did not support the hypothesis that alcohol increases deliberate risk taking. A significant increase in overconfidence in the cognitive task was observed under alcohol, but overconfidence and risk taking were not correlated.

In an on-road study by Caswell (1977) drivers performed several tasks such as overtaking, driving on straight road sections and curves and through narrow gaps while responding to road signals, traffic signals and auditory signals in a subsidiary task. Alcohol resulted in increased speeds and poorer tracking performance. In an on-road study of Smiley et al. (1986), alcohol at 0.05% BAC was asso­ciated with significantly higher speed on straight roads and in curves. Also, alcohol decrea­sed the number of peripheral stimuli detected. According to Smiley (1986), in three of the four studies reviewed, where effects of alcohol on speed were recorded, alcohol was associated with an increase in speed while it significantly affected steering performance in a number of studies (Smiley, 1989). In a study of Hansteen et al. (1976), alcohol increased the number of cones hit and the amount of ‘rough vehicle handling’ while it increased speed. Robbe (1994) tested the effect of alcohol on driving performance during city driving. Alcohol decreased performance in ‘vehicle handling’ and ‘action in traffic’, while speed was increased. Subjects thought, however, that they had driven as well as following placebo and there was no effect of alcohol on effort invested in the driving task.

In summary, alcohol strongly affects perceptu­al and psycho-motor skills as well as performance on the operational level of car driving. At the same time, alcohol increases speed. In this the effect of alcohol is opposite to the effect of marijuana. It may then be hypothesized that a lack of compensation for impair­ments in performance is the cause for the very strong role of alcohol in accident involvement. Evidence was presented that suggests that drivers are unaware of performance decrements under alcohol. This may be somehow related to the absence of compensatory speed changes and effort.

 

  • Conclusions and consequences for the present model.

 

Correlation studies in the tradition of the differential accident involvement approach have been unable to demonstrate a relation between psycho-motor abilities and accident involvement, with the possible exception of professional drivers. This may be explained by the self-paced nature of the driving task for private drivers which allows compensation for poorer skills, such that effects of psycho-motor abilities on accident involvement are decreased. Because the driving task for professional drivers is often controlled by time schedules and fixed driving times and routes, their task is more of a forced-paced nature. In the differential accident involvement approach a relation between psycho-motor abilities and driving performance was assumed instead of tested and driving behaviour on both the operational and the tactical level typically was not examined by this approach.

The effects of individual differences in psycho-motor abilities on behaviour on the strategic and the tactical level was illustrated by the case of the elderly driver. Accident involvement of the elderly driver is much lower than expected from the skill model. A possible cause for this may be that behaviour on the strategic level and the tactical level is adapted to poorer psycho-motor abilies of elderly drivers, in the sense that these drivers often refrain from driving in the dark, under bad weather conditions and so on, and that they drive with lower speeds.

Variati­ons of skill within the driver as a function of temporary states were illustrated by examining the effects of marijuana and alcohol. These studies have supplied strong evidence for effects on psycho-motor abilities and operational performance, together with effects on tactical behaviour. Evidence was presented that supports the hypothesis that effects of marijuana on accident involvement are tempered because of compensation of behaviour on the tactical level for degradations of perceptual-motor abilities and operational performance under marijuana, and that the driver perceives the feedback of poorer operational performance. Alcohol also strongly affects perceptual-motor abilities but the driver appears to be unable to perceive this. This may explain why the driver does not compen­sate behaviour on the tactical level which may be the cause for a strongly increased accident risk. The results are consistent with the idea that behaviour on the tactical level is adapted by decreasing speed or increasing headway when feedback of operational performance decrements is perceived by the driver.

The results presented so far, are summarized in figure 1 as a first attempt to describe a model of driver adaptation. Individual differences in psycho-motor abilities, for example as a function of age, and effects of temporary states induced by marijuana or alcohol on these abilities affect operational car driving performance.

 

 

Figure 1. Model of driver adaptation, derived from the discussion of skill models

 

Normally, these effects are monitored by the driver and the driver perceives the feedback of these effects, although this may be inhibited by alcohol. Two kinds of adaptation may occur. Either behaviour on the tactical level is adapted to compensate for decreased operational performance, if driving is self-paced. If driving is not self-paced the driver may also increase effort to improve operational performance or, alternatively, the driver may adapt behaviour on the strategic level by deciding to give up driving for a while or altogether, or by avoiding driving in bad conditions.

Thus far, only the effects of poorer operational performance have been examined, resulting mainly in decreased speeds. The reverse, better performance resulting in increased speed has not been examined. However, in the motivational models, discussed in the next paragraph, it is argued that behaviour on the tactical level, such as higher speed or smaller headway during car-following, are the result of motiva­tional factors instead of improved performance.

 

2.3 Motivational models of car driving

 

During the seventies driver behaviour modeling shifted to motivational approaches as alternatives for the skill model. The main reason for the rise of motivational models was the rejection of accident proneness as an explanatory concept and the disappointing results of the differential accident involvement approach (Summala, 1985; Näätänen and Summala, 1974). The fact that increasing driver skills and decreasing environmental demands did not result in increased traffic safety in a straightforward manner was attributed to the self-paced nature of the driving task in which the driver is able to control task difficulty. The emphasis of the motivational models on transient or situational factors came as a response to the individual difference approach of the skill model (Ranney, 1994). The actual behaviour of the driver in any given traffic situation was given more importance than the maximum level of performance. The motivational models that emerged in the seventies were based to a large extent on a few articles, written in the sixties, that stressed the self-paced nature of the driving task. Taylor (1964) claimed that galvanic skin response (GSR) per unit of time was constant. He regarded GSR as a measure for subjective risk and hypothesized that the driver adjusts the level of risk taking to keep emotional responses on a constant level. In this view, speed was adjusted to keep subjective risk on a constant level. This notion of compensation by adjusting speed differs from compensation as discussed in previous paragraphs. Taylor (and the motivational models in general) conjectured that speed was adjusted to compensate for subjective risk, while the viewpoint expressed in previous paragraphs suggests that speed is adjusted to compensate for degradations or improvements in operational performance. Both viewpoints stress the importance of the self-paced nature of the driving task. The Risk Homeostasis Theory of Wilde can be seen as a descendant of the principle formulated by Taylor. Cownie and Calderwood (1966) argued that accidents are the product of a simple closed-loop process in which feedback from the consequences of driver actions and decisions play an important role. They emphasized the importance of finding a good balance between motivating and inhibitory forces of positive and negative motivating events. This viewpoint has played an important role in the model of Näätänen and Summala.

Motivational models of driver behaviour have become synonymous with models of risk taking. The most important variants are Wilde’s Risk Homeostasis Theory, Näätänen and Summala’s Zero Risk Model and Fuller’s Threat Avoidance Model. The relation between motivations other than risk taking and driving behaviour has been examined in a limited number of studies, for example French et al. (1993), Rothengatter and de Bruin (1986) and Rothengatter (1988). French et al. (1993) investigated the relation between decision-making style, driving style and accident rates. The results of this study do not give an indication how behaviour on the tactical level is affected by motivational factors. Speed is described as an aspect of driving style together with more motivational concepts such as social resistance and deviance. This indicates that the concept of “driving style” is not clearly defined since it mixes overt behavioural manifestations with covert motivational constructs. Because of this it is difficult to integrate with other approaches discussed in this thesis. Rothengatter and de Bruin (1986) and Rothengatter (1988) examined the relation between speed choice and motivational factors within the framework of Fishbein and Ajzen’s model of reasoned action. It was found that speed choice is determined by four motivational factors: pleasure in driving, risk, travel time and costs. Pleasure in driving proved to be the strongest determining factor of speed choice, such that the subjects with the highest speed scored highest on pleasure in driving. However, pleasure in driving was also related to the top speed of the vehicle and thus to vehicle characteristics: drivers of high performance cars scored higher on pleasure in driving compared to drivers of low performance cars. This was explained by suggesting that drivers with more pleasure in driving, as a characteristic of the person, are more inclined to buy high performance cars. However, the reverse could also be true: drivers of high performance cars may enjoy driving fast more than drivers of low performance cars because the car allows better control at higher speeds. This issue should be considered in further studies.

 

  • Risk Homeostasis Theory.

 

Risk compensation models propose a general compensatory mechanism whereby drivers adjust their driving (e.g. speed) to establish a balance between what happens on the road and their level of accepted subjective risk. Wilde’s Risk Homeostasis Theory (RHT) is based on the assumption that the level of accepted subjective risk is a relative­ly stable personal parameter. An important implication is that drivers will nullify the effects of safety improvements by driving faster or behaving less cautious in general. This has resulted in considerable controversy. RHT (see for example Wilde, 1982), previously known as risk compensation theory (for example Wilde, 1976), consists in fact of two models; an individual model of driver behaviour and an aggregate model that relates driver behaviour to accident rate. In the individual model the driver is assumed to have a target level of risk that represents the amount of accident risk the driver accepts. This is continuously compared to the perceived level of risk which is an estimate of the accident risk in the immediate future. The perceived level of risk is determined by the vehicle path, the road environment and paths of other road users (the stimulus situation) and anticipations regarding the development of the stimulus situation in time. When there is a discrepancy between perceived risk and target level of risk the driver makes a behavioural change, either in the direction of reducing the level of risk if perceived risk is larger than target level of risk, or in the direction of increasing the level of risk if the reverse is true. This results in a homeostatic process in which the driver aims to match the perceived level of risk with the constant target level of risk.

The name ‘risk homeostasis’ theory is, however, essentially derived from the aggregate model, see figure 2. Again, drivers compare the perceived level of risk with the target level, resulting in adjustment actions. Aggrega­ted over all road users over a given time span, these adjustment actions will produce the rate of accident frequency and severi­ty. This has a lagged feedback on perceived risk. Decreased accident rates then decrease percei­ved risk after some time, resulting in adjustment actions that increase accident risk in a homeostatic way. The only factor that affects accident risk then is the target level of risk. In the aggregate model, skills play some role, although improvements in skills are unlikely to have a lasting effect on accident frequency and severity (Wilde, 1981). Perceptual skills determine the extent to which subjective risk corresponds to objective risk. It is important to note here that Wilde does not think that improving risk perception skills will improve traffic safety. Decision skills refer to the drivers’ ability to bring about the desired adjustment, while vehicle handling skills determine the extent to which the planned actions are executed properly. In the individu­al model of risk compensation skills only have a modest influence on perceived level of risk and on the transformation of sensory input. Thus, the concept of skill, as discussed previously, plays no role in Wilde’s theory. Individual differences are restricted to individual differences in motivational states that may affect the target level of risk.

 

 

Figure 2. Aggregate model of Risk Homeostasis Theory (from Wilde, 1982).

 

The effects of motivations on choice of time-headway (THW) during car-following has been studied by Heino et al. (1992). Drivers were classified as sensation seekers or sensation avoiders depending on their scores on personality inventories. It was expected that sensation seekers seek more risk and are thus characterized by a higher level of target risk. It was found that sensation seekers followed at a smaller THW which is associa­ted with higher objective risk by several authors. However, Heino found that this smaller THW was not associated with a higher subjective risk. It appeared then that both sensation groups accepted the same level of risk, and thus did not differ in target level of risk, but for sensation seekers this perceived level of risk was achieved at a smaller THW compared to sensation avoiders.

Meanwhile the discussions concerning the validity of RHT have gone on for years. Evidence provided by Wilde supporting RHT has been refuted by various counterexamples (for example, McKenna, 1982; Evans, 1985). Even more controversy has arisen over Wilde’s hypothesis that safety improve­ments will not work unless it affects the target level of risk (McKenna, 1988). The ability of drivers to monitor accident risk has been questioned and the assertion that drivers experience or accept risk has been challen­ged (for example, Evans, 1991). The plausibility of seeking some level of risk has been seriously doubted and according to several authors drivers seek the lowest possible, or zero, level of risk.

 

2.3.2 Zero Risk Theory.

 

Näätänen and Summala (1974, 1976) have presented a model of the driver’s decision making that has become known as the zero-risk theory. In this model motivational factors such as subjective risk, other motivations and vigilance determine driver decision-making and behaviour, see figure 3. The subjective risk monitor is a crucial element in this model. It was conceptualized as a monitor that generates subjective risk or fear depen­ding on the experienced risk in the present or expected traffic situation. Activation of subjective risk inhibits ongoing behaviour in the sense that it results in behaviour such as slowing down. It also has an inhibitory effect on subsequent behaviour in the sense that drivers learn to behave more cautiously in similar situations. However, most of the time subjective risk equals zero. Other motivations provide an excitatory component resulting in increased speed. These other motives are affected by persona­lity factors such as aggressiveness and the state of mind of the driver. Several changes in the traffic environment may affect subjective risk. Drivers may drive faster or overtake other cars more frequently before subjective risk is experienced. The best traffic safety measures are those that decrease objective risk but increase subjective risk.

 

 

Figure 3. The zero-risk model of Näätänen and Summala (from Näätänen and Summala, 1976).

 

An important difference with the risk homeostasis theory is that in the zero-risk theory the driver is assumed to accept no risk at all, that is, the target level of risk is zero. Näätänen and Summala state on the one hand that subjective risk is an important determinant of driver behaviour (as an inhibitory factor) while on the other hand most of the time subjective risk equals zero. Fuller formulated his Threat Avoidance Model, to be discussed later, partly in response to this inconsistency. In later publications by Summala (1985, 1988) the concept of subjective risk was more or less abandoned as an explanatory factor in driver behaviour: it is not risk that the driver attempts to control, but instead, drivers control and maintain safety margins, since normally the driver gives no consideration to risk (Summala, 1988). The concept of subjective risk should be reserved for ‘arm-chair estimates’ of the risks of, for example, traffic scenes shown to subjects for research purposes (Summala, 1988). He stated that the output of the subjective risk monitor was meant to represent a fear response resulting from the perception or expectation of loss of control over the car or of being on a collision course (Summala, 1986). The concept of ‘perceived loss of control’ thus replaced the continuous control of subjective risk. Taylor (1976) postulated that subjective risk is equivalent to the percep­tion of loss of control. The relative unimportance of subjective risk in the zero-risk theory was further exemplified by Summala by his observation that as drivers become more experienced driving becomes more automated and feelings of uncert­ainty or fear related to perceived loss of control decrease because confidence in control skills increases. He regarded the fact that drivers are not very well able to take account of the objective variance in the traffic system as the most important point of the zero-risk theory. The second main point is that different kinds of motives push drivers towards higher speeds and if the traffic system provides (environ­mental) opportunities to satisfy these motives, drivers are inclined to use them. This is not risk compensation but merely the result of a tendency for satisfaction of motives. Examples of such motives are that high speed as such is motivating and higher speeds mean shorter travel times. Also the conserva­tion of effort is seen as an important motive resulting in a reluctance to slow down. Speed also provides outlets for many other ‘extra motives’ such as the motive to demonstrate driver skills to peers. In order to improve traffic safety, drivers should be prevented to satisfy their motives by introducing speed limits.

Thus, the zero-risk theory has undergone significant changes in time. The role of the inhibitory forces associated with subjective risk has diminished while the excitatory motivational components are emphasized more strongly. Also, the concept of safety margins has replaced the concept of subjective risk and ‘perceived loss of control’ has become an important factor in the control of safety margins. Although this is not stated explicitly by Summala, loss of control is strongly related to performance on the operati­onal level. Because safety margins appear to be underlying controlling variables of driver behaviour, different subtasks such as lane keeping, car-following, curve negotiation, gap acceptance and overtaking should be analyzed in more detail (Summala, 1988). The concept of safety margin will be discussed later.

Yet, a large number of studies have focused on one aspect of the early version of the zero-risk model: the assumption of a discrepancy between subjective and objective risk. A threshold for risk perception is assumed, and risk compensation occurs only if this threshold is exceeded. Below this threshold subjective risk is experienced as zero. The idea then became in vogue that risk perception may be distorted as a function of several factors such as age or driving experience. Most studies on risk perception have focused on the young driver and these will be reviewed later.

 

  • Threat Avoidance Model.

 

Although the Threat Avoidance Model of Fuller is typically classified as a motivational theory it is actually more a theory of learning applied to car driving. Fuller (1984) has formulated his threat avoidance model of driving behaviour partly in an attempt to solve two problems associated with the zero-risk theory. The first problem was the dissociation between subjective and objective risk in the zero-risk model. The second problem was that subjective risk reactions constitute an important determinant of decision making while at the same time the driver feels no subjective risk most of the time. Since the experience of subjec­tive risk is aversive drivers are motivated to escape from situations that elicit subjective risk or to avoid those situations. Thus, subjective risk reactions are important determinants of behaviour assuming that drivers are able to anticipate and make appropriate adjustments to upcoming hazards. If driving consists to a large extent of learned avoidance reactions drivers will rarely experience any subjective risk at all.

Figure 4 gives a representation of the model. Because, in general, the driver’s own actions determine whether or not interactions with the road environment will be punishing, stimuli in the road environment have an aversive potential. A discriminative stimulus is some precursor of a potential aversive stimulus which has been learned by association. Several consequences are experienced as aversive. These may be very common consequences such as loss of self-pacing in the driving task and a state of high arousal, or less common consequences such as loss of vehicle control, physical injury, material damage, loss of self esteem and so on. The driver then is motivated to prevent these negative consequences and not just to avoid the experience of subjective risk. The discriminative stimulus is a function of the drivers’ perception of speed, the road environment and skill. It is an integration of these features projected into the future. For example, the combination of a particu­lar speed, a curve that is approached and an estimate of present vehicle handling skills determine together whether this constitutes a discriminati­ve stimulus. When perceived capability is a primary factor underlying the discriminative stimulus some compensatory response is generated that may consist of raising the performance level or some behavioural adjustment such as lowering the speed or increasing headway during car-following. It is then important to note that the issue of feedback and compensation is solved in Fuller’s model by assuming behavioural anticipatory avoidance responses to a discriminative stimulus. A delayed avoidance response may occur if the anticipatory avoidance response is inadequate. If there is a discriminative stimulus, the probabi­lity of an anticipatory avoidance response or a non-avoidance response is both determined by the subjective probability of expected threat and by the rewards and punishments of the response alternatives. If no discriminative stimulus is detected then either no threat is realized or a threat occurs demanding a delayed avoidance response of the driver.

 

 

 

Figure 4. Fuller’s threat avoidance model (from Fuller, 1984).

 

 

This may occur because of perceptual errors, inadequate learning to recognize a discrimi­native stimulus, unpredictable behaviour of other road users or sudden mechanical failure. If the driver makes no anticipatory avoidance response a delayed avoidance response represents an escape from an aversive stimu­lus. Learner drivers are more likely to make delayed avoidance rather than anticipatory avoidance responses than experienced drivers because they have not yet developed associations between discriminative and potential aversive stimuli. The higher accident involvement of young drivers was explained by this lack of associations, while Näätänen and Summala (1976) attributed the high accident involvement of young drivers to the relative strength of ‘extra motives’ in this group, such as the desire to drive at high speeds.

Based on this framework, Fuller suggested that there may be drivers who are predominantly anticipatory avoidance responders while others are predo­minantly delayed avoidance responders. This may then be related to indivi­dual differences in the ability to detect hazards. This ability has been referred to as ‘hazard cognition’ by several authors and it is believed to be related to accident involvement of young and inexperienced drivers.

 

 

  • Individual differences in motivations: the young driver.

 

Young drivers, especially males, from 18 to 24 are dramatically more often involved in accidents compared to drivers of other age groups (Evans, 1991). This overinvolvement of young male drivers in the accident statistics is one of the most consistently observed phenomena in traffic throughout the world. A confounding factor is that young drivers usually are the least experienced. Simpson (1986) stated that the reason for the high involvement of young drivers in vehicle accidents, even when exposure to risk is controlled for, is not clear. While young people from 16 to 24 years of age represent 17% of the Canadian population, they account for 31% of all traffic fatalities, 33% of all traffic injuries and 58% of all driver fatalities in Canada. Because risk is usually applied as an explanatory concept for the high accident involvement of young drivers, studies on this issue are discussed here.

The meanings of the risk-related concepts will be discussed first as they are applied in the case of the young driver. Risk-taking is something which is usually inferred from observation of behaviour (Saad, 1989). Traffic researchers often assume that high speed and close following carry a higher objective risk. Drivers who display such behaviours are then assumed to take more risks. Jonah (1986) has given several examples of higher risk-taking in young drivers. Young drivers have been reported to drive at higher speeds (for example Wasielewski, 1984; Soliday, 1974), although the correlation between speed and age is generally very low. Also, younger drivers have been reported to follow at smaller headways (Evans and Wasielewski, 1983). This behaviour associated by a number of researchers with higher risk taking in young drivers, is often seen as evidence that young drivers either deliberately seek more risk or accept a higher target level of risk, and thus have a higher risk acceptance or risk utility, or have a deficient risk perception, i.e. they fail to see the risk involved with such behaviours. The former concept is associated with Wilde’s model while the latter is more closely associa­ted with the models of Näätänen and Summala and Fuller. Both concepts have been used as expressi­ons of subjective risk. One of the problems with risk research centers around the conceptual vagueness of the term ‘subjective risk’. It is not always clear whether it refers to a failure to perceive the potential danger (hazard perception), to an underestimation of the probability of a certain event (subjective estimation of objective risk), to the driver’s poor appreciation of his or her ability to cope with the situation, or to attitudes and motives regarding safety (risk acceptance) (Saad, 1989). Haight (1986) argued that the only valid meaning of the term ‘risk’ refers to empirical probability or expected cost. In that case risk is a statisti­cal concept referring to the outcome of behaviour on a highly aggregated level. In such a view there is little room for terms such as subjective risk, risk perception or risk acceptance. Another problem associated with some risk research is the circularity in reasoning. The explanati­on for behaviour associated with a higher objective risk, resulting in more accidents, is that drivers deliberately want a higher objective risk or fail to see the objective risk involved. So the behaviour to be explained is explained in terms of the outcomes of precise­ly the same behaviour.

The high accident involvement of young drivers has often been attributed to poorer risk perception, resulting in a larger discrepancy between subjective risk and objective risk for young male drivers. Jonah (1986) stated that, even though young drivers may perceive as much risk while driving as older drivers and thus do not deliberately seek more risk, they may be more confident in their ability to avoid an accident. In Jonah’s review, risk perception was meant to reflect the subjective estimation of objective risk. He presented some evidence that younger drivers had poorer risk perceptions in the sense that they estimated objective risk lower compared to other age groups. However, it is not clear what this means. Basically, the subjects were asked about their knowledge of statistical facts over which even traffic researchers are still debating. Wilde’s model is the only risk model that assumes that knowledge of drivers concerning statistical accident risk affects behaviour. It has been objected by many authors that it is highly unlikely that drivers are aware of accident statistics or that these play any role in driving behaviour. Finn and Bragg (1986) also measured subjective risk or risk perception as the estimation of objective risk as a statistical phenomenon by asking questions such as ‘how many people were killed in traffic accidents in Massachusetts last year’. Although it was found that young drivers see driving as more dangerous when general questions about accident risk were asked, and they recognize that their age group is at greater risk of accident involvement compared to other age groups, they see their own chances to be involved in an accident as lower compared to their own age group and older drivers when specific questions about their own risk are asked. Finn and Bragg saw this as evidence that young drivers differ from older drivers in lower risk perception and not in risk acceptance and that risk perception, or at least seeing less risk in driving situations compared to older male drivers, may account for the high accident involvement of young male drivers. Bragg and Finn (1982) found that specific behaviours such as speeding and tailgating were perceived as less risky by young drivers. They hypothe­sized that the lower perception of risk in young drivers may be attributa­ble to the greater confidence in their skill or belief in their ability to handle a particular hazardous situati­on. Risk perception was thus connected with confidence in driver skills.

Matthews and Moran (1986) assessed the relationship between perceived skill and perceived risk. In their study young (18-25) and middle-aged (35-50) male drivers completed a question­naire on accident risk and driving ability and gave subjective ratings of risk to videotaped traffic situations. Young drivers gave lower ratings of accident risk for driving situations which demanded fast reflexes or substantial vehicle handling skills. They rated their own risk of an accident and driving abilities as being the same as for older drivers. However, they saw their peers as being significantly more at risk and as having poorer abilities than themselves. The data suggested that risk perception is strongly related to perceived ability. Spolander (1982) found that drivers with three years of experience judged themselves to have better driving skills compared to other drivers. The drivers who gave the highest ratings on skill also reported  faster driving.

Brown and Groeger (1988) distinguished two inputs to the process of risk perception: information on potential hazards in the traffic environ­ment and information on the joint abilities of driver and vehicle to prevent that hazard potential being transformed into actual accident outcomes. Risk perception is the detection of any shortfall in the ability to avoid realizing the potential of immedia­te task and environmen­tal hazards.

This short review makes clear that the concept of risk perception has more than one meaning which makes the interpretation of results from these studies difficult. On the other hand, subjective risk has been linked more and more with (perceived) driving skills. This suggests that, at least in the mind of the driver, subjective risk really means fear of loss of control, as was suggested during the discussion of the zero-risk theory.

 

In another line or research, the high accident involvement rate of especially young male drivers has been associated with the use of alcohol and drugs as a lifestyle-related phenomenon. Although as many as 50% of fatally injured young drivers have been found to be positive for alcohol, this is slightly lower than the frequency for older drivers. Also, it has become clear from surveys that drinking and driving is widespread among younger drivers although they had typically consumed less alcohol than older drivers. In alcohol related crashes younger drivers tend to have lower BACs than older drivers (Simpson, 1986). Yet, the high accident involvement among young drivers has been attributed to risky driving behaviour as an aspect of adolescent lifestyle that is embedded in the same set of personality and behaviour aspects as other kinds of adolescent problem behaviour such as delinquency, problem drinking and illegal drug use and smoking (Jessor, 1986). Also, Beirness and Simpson (1986) found that accident involved young drivers score higher on thrill and sensation seeking, alcohol consumption and frequency of drinking while they score lower on traditional values and usage of seat belts. In short then, some authors believe that the high accident involvement of young, and especially male, drivers is a lifestyle related phenomenon resulting in a higher deliberate risk acceptance or higher target level of risk, using the terminology of Wilde. But in that case it would be expected that a higher percentage of accident involved young drivers are positive on alcohol and have higher BAC levels compared to older drivers. This obviously is not the case.

It has frequently been reported that the relative risk of becoming involved in a fatal accident rises faster as a function of BAC level for younger drivers compared to older drivers (Simpson, 1986; Kretschmer-Bäumel and Kroj, 1986). In other words, with increases in the amount of alcohol consumed, the accident risk increases for all age groups, but much more rapidly for the young. Although the typical explanation for this has been the relative inexperience of young drivers with alcohol, driving and the combination of these, there is no scientific evidence that inexperience with drinking and/or driving is the cause for the stronger impact of alcohol on accident rate for the young (Simpson, 1986; Mayhew et al. 1986). Although the reason for the interaction between age and BAC level on accident involvement is not clear, it suggests that both factors share a common locus of effect, in the sense that the factor that causes the higher accident rate of young drivers is aggravated by alcohol. In the discussion of the effects of alcohol it was suggested that the lack of compensation for impaired performance may be the cause for the large role of alcohol in accident causation. Evidence was presented that drivers are unaware of performance decrements under alcohol which is possibly the cause for the absence of compensatory speed changes and effort. From the same perspective it may be suggested that young and inexperienced drivers have not yet learned to recognize the effects of situational factors on their perfor­mance and thus fail to compensate for these effects resulting in speeds that are too high for the circumstances.

 

2.3.5 Conclusions and consequences for the present model.

 

In the literature on risk perception it is often suggested that young drivers are more involved in accidents because the discrepancy between subjective risk and objective risk is higher in this group. The reason proposed in a number of studies is that young drivers tend to overestimate their own abilities, although the overestimation of one’s abilities appears to be a general phenomenon for all age groups. A drawback of studies on young drivers is that the problem is often examined without simultaneously measuring operational and tactical behaviour. Following the line of reasoning of this thesis, it may be hypothesized that the high accident involvement of young drivers is caused by a failure of young drivers to adapt their speed, or tactical behaviour in general, to the traffic situation, because they overestimate their ability to cope with hazardous situations and fail in the perception of feedback of operational performance decrements induced by traffic situations, vehicle characte­ristics or environmental variations in general. The interaction between BAC level and age on accident involvement may then be suggestive of a common locus of effect for both the factors alcohol and young drivers. However, it must be stressed that the lack of studies that have examined explicitly the operational and tactical behaviour of young drivers prevents firm conclusions.

From the discussion of the motivational models, a second version of the model of driver adaptation is presented in figure 5.

 

 

 

Figure 5. Model of driver adaptation, derived from the discussion of motivational models.

 

Various situational factors may affect operational performance resulting in a sense of ‘loss of control’. In general, this effect is monitored, although for young drivers monitoring or recognition of these effects on operational performance may be hampered in some way. Associations between situational factors and the monitored effects on operational performance are formed that result in adaptations of behaviour on the tactical level (anticipatory avoidance responses, according to Fuller). There also is a direct effect of performance monitoring on tactical behaviour. The effects of age on accident involvement may also be explained in terms of fewer associations between situational effects and operational performance because of limited experience, resulting in fewer anticipatory avoidance respon­ses. However, the comments made on the young driver are highly hypothetical and need to be verified by more rigorous experimentation.

 

2.4 Adaptive control models

 

The adaptive control models, referred to by Michon (1985), deal primarily with the operati­onal level of car driving behaviour. These models have been inspired by the principle of adaptive control in which the human operator adapts his control behaviour to the characteristics of the system to be controlled. This concept resembles the use of the term adaptation in this thesis. An important difference lies in the behavioural level at which this process of adaptation occurs: in adaptive control models adaptation occurs on the operational level while in the model of adaptation discussed in this thesis adaptation occurs primarily on the tactical level.

Michon (1985) distinguished between two different classes of adaptive control models; the servo-control models and the information flow control models. The first class is primarily concerned with manual control in the context of signals that are continuous in time, while the second involves discrete decisions. In practice, the distinction has somewhat vanished, resulting in hybrid models. Servo-control models consider driving as a continuous tracking task. These models have been applied to operational performance of steering on straight roads and curves and to obstacle avoidance maneuvers. Input signals are transformed by transfer functions into a vehicle output. Transfer functions represent both driver and vehicle dynamics and contain lead components to account for preview or anticipation of the driver and lag components representing driver and vehicle inertia.

Young (1969) discussed a number of different types of adaptation to the system to be controlled. Input adaptation refers to the ability of the operator to detect familiar or repeated patterns in the input and track these in a predictive or open loop fashion.  The adaptive control models applied to the driver task mainly refer to input adaptation. The best known is the STI model described by McRuer et al. (1977). One variant of this model, see figure 6, refers to compensatory steering control on straight roads. In this model the driver is assumed to act as a regulator against external disturbances that arise from wind and road surface effects. Thus, operational performance is continuously adapted to system disturbances and vehicle characteristics. The steering wheel output is determined by transfer functions while the visual inputs to the model are lateral position and vehicle heading errors. In the ‘input adaptation’ models the predictable aspects of  the steering task, such as the required steering angle as determined by the road curvature, are described as precognitive tracking while the random components in the input signal are handled by compensatory tracking. Another important type of adaptation is referred to as controlled element adaptation.  This occurs when the operator changes his control strategy as an adaptation to changes in the dynamics of the system. If a driver normally drives a sedan but changes to a sports car he has to adapt his steering behaviour to the different steering ratio. In general, any change in vehicle characteristics or vehicle dynamics requires some form of controlled element adaptation.  In chapter 4 an experiment on steering during curve negotiation is discussed. In the introduction of chapter 4 it is stated that required steering wheel angle during curve negotiation is determined by curve radius and by speed.  Speed then changes the dynamics of the system to be controlled and requires an adaptation of steering wheel angle as a form of controlled element adaptation. In that sense speed  is considered as a ‘property’ of the system to be controlled  instead of  a form of  operator adaptation on its own.  In the adaptive control models, the operator is described as someone who responds to the task-characteristics  instead of someone who actively creates the task. However, because the driving task is self-paced most of the time, the behaviour of the driver affects the dynamics of the task.

 

 

 

Figure 6. STI compensatory steering model (from Reid, 1983).

 

  • Some properties of the adaptive control models.

 

A consistent feature in attempts to validate these models with human drivers is that subjects are instructed to drive with a fixed speed, thereby excluding possible effects of tactical behaviour on operational performance. Also, the parameters that are found using human drivers often apply to only one situation. Variation in speeds and curve radii will affect the parameters of the models (see for example Donges, 1978). It is argued here that operational performance and behaviour on the tactical level are interdependent and should both be incorporated into a single model.

There are a large number of examples that suggest that speed is used to compensate for detrimental effects of various task-related and situational factors on operational steering performance. For example, Good and Baxter (1986) used the STI model to study steering performance as a function of roadway delineation. The quality of steering was expressed, among other things, by the remnant that accounts for that part of the manual control output that is uncorrela­ted with the input. A smaller remnant then indicates better steering performance. Wider edge lines resulted in a smaller remnant because of improved vehicle guidance. However, wider edge lines also resulted in higher speed. Also, day time driving resulted in better steering performance and higher speed compared to night time driving. Thus, it appears that factors that improve steering performance result in higher speeds. However, the effects on speed are not accounted for by the model and are considered undesirable artifacts.

Tenkink (1988) studied the effects of sight distances of 27, 37 and 183 meters with fixed speeds. Standarddeviation of lateral position (SDLP) increased with higher fixed speeds over all sight distances with steeper increases for smaller sight distances. A smaller sight distance resulted in a larger SDLP at a given speed. Lowering sight distance thus deteriorated steering performance and this was aggravated with higher speeds. However, if drivers were allowed to choose their own speed, reductions in sight distance resulted in the choice of lower speeds while SDLP was maintained on a relatively constant level, except for very short sight distances of 27 meters where speed was not reduced enough to prevent an increase in SDLP. According to Tenkink, a safety margin based on time may have caused the speed reduction under reduced sight distance, because the speed-distance curve appeared to approach a line through the origin, with a slope corresponding to a minimum time of 1.2 seconds for driving on straight roads. Harms (1993) also studied the effect of reduced sight distance on speed choice and lane keeping. She found that reduced sight distance resulted in the choice of a lower speed, while SDLP was unaffected, even with the shortest sight distance of 30 meters. She suggested that the speed reducti­on had prevented a deterioration of lateral control performance as a function of sight distance.

These studies suggest that situational factors that affect operational steering performance are compensated for by speed choice if task conditions are self-paced. If drivers are not given the opportunity to adapt behaviour on the tactical level they are forced to improve behaviour on the operational level, and it is under these conditions that the adaptive control models are normally tested.

 

In most adaptive control models lateral position deviations, heading angle and anticipated curvature are treated as the input variables that are continuously transformed into a steering wheel angle. The validity of the input variables and the assumption of continuous minimization of errors has been challenged by a number of authors. Riemersma (1987) performed a number of experiments to find the visual cues that are used by the driver in steering control. He found that control of lateral position alone is not sufficient for lane keeping in straight road driving and that heading angle is not directly used as an input variable in steering control, in contrast to the assumption of adaptive control models.

Blaauw (1984) studied the multitasking aspects of car driving. A monitoring function was assumed to supervise manual control associated with steering and speed control on the operational level. Because of a supervisory function, perceptual and control actions are not executed continuously, in contrast to the assumption of the adaptive control models, thus allowing free time in-between control actions. Experienced drivers adjusted their steering control better to increased task demands invoked by driving with a constant speed or night time driving compared to inexperienced drivers. Also, in self-paced conditions where drivers were free to choose their own speed, increasing task demands by occlusion or night time driving resulted in the choice of lower speeds.

Godthelp (1984) questioned the assumption of the adaptive control models that the driver behaves in a closed-loop error-correction mode in which continuous attention is allocated to the steering task. He applied the Time-to-Line-Crossing (TLC) as a measure that reflects the time available for the driver before a correcting steering action is needed to prevent a lane boundary exceedence. The amount of time the driver voluntarily refrains from using visual feedback (occlusion time) correspon­ded closely with TLC values. This means that when the driver has less time available to postpone correcting steering actions, a request for visual feedback is made sooner. This implies that the driver is aware of the time available and that correcting steering actions are generated when some TLC criterion has been reached. Drivers chose occlusion times of about 40% of the available time, irrespective of speed. Also, if steering corrections during the occlusion interval were larger, the driver requested visual feedback sooner, suggesting awareness of the driver’s own steering behaviour and a compensatory effect on visual sampling. When, in Godthelp (1984), drivers were asked to switch to error-correction when vehicle motion could still comfortably be corrected to prevent a crossing of the lane boundary, it appeared that drivers chose a strategy where TLC on the moment of steering correction was about constant over different (fixed) speeds. This constancy of TLC over speed was obtained without occlusion, while the strategy of requesting visual feedback when 40% of available time was reached occurred under occlusion. This difference was explained as a result of the degree of uncertainty regarding the vehicle trajectory. Thus, Godthelp found strong evidence that steering control is not continuous, that drivers are sensitive to TLC and that TLC information is used in steering control.

 

The relation between vehicle dynamics and operational behaviour constitutes an important aspect of adaptive control models. Godthelp and Käppler (1988) found that changing the vehicle characteristics to heavy understeering resulted in increased steering control effort but similar lateral control performance, as evidenced from TLC control performance, compared to a normally understeered car, because drivers were able to develop an accurate internal representation of the vehicle dynamics. In both normal and heavy understeered cars the accepted occlusion times were about 40% of available time, independent of (fixed) speed. This suggests that drivers adapt their visual information intake and steering behaviour to the dynamic characteristics of the vehicle such that the same strategy is maintained. From the results of Godthelp and Käppler it may be inferred that drivers are sensitive to vehicle handling properties and change their operational behaviour as a function of this if the driver is required to drive with a fixed speed. This may be considered as an example of controlled element adaptation and thus as an example of adaptation of operational behaviour. A number of other studies have revealed effects of vehicle characteristics on tactical driver behaviour. Rumar et al. (1976) studied the effects of studded tires on speed choice in curves. Drivers with studded tires drove faster compared to drivers with unstudded tires in icy road conditi­ons. This did not result in lower safety, since the ‘safety margin’, defined as the difference between real and critical lateral acceleration, was larger with studded tires. Summala and Merisalo (1980) also found that drivers with studded tires chose higher speeds in curves in low-friction conditions and that the safety margin was greater for drivers with studded tires in slippery conditions. The higher speeds with studded tires in low friction conditions may be regarded as an adaptation of tactical behaviour to the increased friction coefficient induced by studded tires. Also, the acceleration capability of cars has been shown to affect behaviour. Evans and Herman (1976) found that drivers accepted smaller gaps with oncoming cars while negotiating intersec­tions if the acceleration capability of the car was higher. However, the physical safety margin was not negatively affected by acceleration capability. Also, newer cars used higher levels of deceleration compared to older cars when they stopped at signalized intersections (Evans and Rothery, 1976). This was explained as a possible adaptation of behaviour (on the tactical level) to compensate for reduced mechanical conditions in older vehicles. Evans and Wasielewski (1983) found that drivers of newer cars and cars with intermediate mass followed with a smaller time-headway. This may also be the result of better deceleration capabilities of newer cars. Evans (1991) postulated that improved braking and vehicle handling characteristics result in increased speeds, closer following and higher speeds in curves. When safety changes are invisible to the user as may be the case with seat belts and increased crashwor­thiness, there is no evidence of any measurable human behaviour feedback. A similar point was made by Lund and O’Neill (1986). Design changes that reduce the likelihood of a crash do have an effect on behaviour. They stated that how a car is driven depends on feedback to the driver about the car’s handling characteristics. Vehicle-related factors may then affect both operational and tactical driver behaviour depending on the visibility of the feedback.

 

  • Conclusions and consequences for the present model.

 

Adaptive control models study the effects of system characteristics on operational behaviour without establis­hing a link with behaviour on the tactical level. Also, adaptive control models assume that continuous attention is being allocated to the steering task resulting in continuous error correction.

 

 

 

Figure 7. Model of driver adaptation, derived from the discussion of adaptive control models and related research.

 

Under forced paced conditions effects of vehicle characteristics and situational factors generally affect operational behaviour as is predicted by the adaptive control models. However, car driving is a self-paced task most of the time and it is under these conditions that speed reducti­ons generally occur, possibly as an attempt to compensate for effects on operational performance. Evidence was presented that a time-based variable, TLC, is used by the driver as a criterion for generating corrective steering actions. TLC is determined by operational steering perfor­mance, vehicle characteristics, speed and lane width. Effects of situational and vehicle-related factors on steering performance and vehicle dynamics may then be compensated for by a speed reduction, such that a constant safety margin is maintained.

From the discussion of the adaptive control models, a third version of the model of driver adaptation is presented in figure 7. Various situational factors, driving experience and vehicle characteris­tics affect operational performance. This effect is monitored and adapted for either via allocation of effort in order to improve operational performance, or via an effect on behaviour on the tactical level.

 

 

2.5 Connecting operational and tactical behaviour: a driving model based on safety margins

 

The adaptation model as it emerges from the discussion of the literature on driver models and driving behaviour is presen­ted in figure 8. This model states that several factors affect operational performance. For example, temporary states, induced by alcohol or marijuana, affect psycho-motor abilities while psycho-motor abilities affect operational performance. Also, vehicle related factors situational factors and driving experience may affect operational performance in accordance with the adaptive control models. The effects on operational performance are perceived via a feedback loop by the driver, although alcohol and young age may inhibit this. If driving is self-paced, the driver adjusts behaviour on the tactical level by either increasing speed or decreasing headway during car-following if operational performance is improved, or by decreasing speed or increasing headway if operational performance deteriorates. If there are no opportunities to adapt behaviour on the tactical level, i.e. when the driving task is forced-paced, the driver may elect in allocate more effort to increase operational performance. Adaptation of tactical behaviour or effort allocation does not only occur as a response to momentary changes, but also in the form of an anticipatory response. This response is the result of learned associations between various factors and effects on operational performance allowing an adaptation of tactical behaviour in the absence of an effect on operational performance. For example, if the driver has learned the effects of rain on road friction and on operational steering performance, he may already choose a lower speed before these effects are actually experienced during a particular period of rain

 

 

 

Figure 8. Adaptation model of car driving.

 

However, the mechanisms by which this process works are still unclear. The extent to which speed is adapted cannot be predicted because of the lack of a unitary measure that incorporates both behaviour on the operational level and the tactical level. An organizing principle may be found in the operation of safety margins. Earlier it was mentioned that, according to Summala (1985), drivers maintain safety margins and that this process should be analyzed in more detail in subtasks such as lane keeping, car-following, curve negotiation, gap acceptance and overtaking. Rumar (1988) shares the point of view that drivers control safety margins instead of risk. He proposed that a safety margin may be operationally defined as an area of safe driving around the car, equivalent to the old idea of the subjective dynamic field that expands in front of the car if speed is increased (Gibson and Crook, referred to by Rumar, 1988). Safety margins can be operationally defined as distance or time related measures (Summala, 1988), although they have also been described in other terms such as a diffe­rence between actual and critical lateral acceleration. Summala has mentioned the time-to-line-crossing (TLC) and time-to-collision (TTC) as examples of safety margins.

Operational control in car driving is usually separated into lateral control and longitudinal control. Lateral control refers to keeping the car within the lane boundaries or to steering away from objects that block the path of the vehicle. Longitu­dinal control refers to activities related to the control of speed, such as braking and use of accelerator and clutch. It is proposed here that the driver uses TLC as a safety margin during lateral control, while TTC, or more general­ly time-to-object (TTO), is used as a safety margin during longitudinal control. Ofcourse, TLC is the same as TTO to either of the lane boundaries. Thus, safety margins are proposed to represent time-related measures. This has a number of advantages. Because driving is a dynamic task in which the driver and other traffic participants move with varying speeds, time may be used as a relatively constant parameter that can be controlled by means of tactical adaptations of speed or headway. In addition to this, there is abundant evidence that humans are very well equipped to perceive time to static and dynamic objects in dynamic situations.

Lee (1976) argued that dri­vers are able to control braking based on time-to-collision (TTC) information from the optic flow field (visual angle divided by the angular veloci­ty). This would enable the driver to judge the mo­ment to start braking and to control the braking pro­cess. The ability to use TTC informa­tion and the actual use of this information has been established in a number of studies, referred to in subsequent chapters on car-following and braking. Van der Horst (1990) showed that time-to-intersecti­on (TTI) is used by the driver in the decision when to start braking as well as in the control of braking. The TTI at which the driver starts braking appeared to be rather constant over speed. In stopping for a stationary object the minimum TTC during the approach was also about constant over different approach speeds. This suggests that time-to-object may be used as a safety margin the driver is not willing to exceed in longitudinal control tasks. Behavioural manife­stations of adaptation on the tactical level in longitudi­nal control tasks are adaptation of speed and of time-headway during car-following. It may be argued that poorer perfor­mance in operatio­nal control increases the chance that a TTO safety margin is exceeded. In approaching a stationary object such as a traffic light, for which the driver has to stop, the driver may decrease his speed earlier in order to compensate for this. During car-following, the driver may choose a larger time-headway. This allows more time to react if the lead vehicle decelerates and thus minimizes the chance that a critical TTC is exceeded.

As was already mentioned in the previous paragraph, drivers appear to be able to estimate the TLC in lateral control tasks, and there is evidence that TLC plays an important role in steering control. If TLC is too small, it can be increased by choosing a lower speed. Thus speed adaptations allow control of a TLC-based safety margin.

Several factors related to operational performance, vehicle characteris­tics, environment and behaviour on the tactical level affect these time-based safety margins. TLC is affected by vehicle dynamics, steering performance, speed, road width and curve radius. TTC is affected by braking characteristics of the vehicle, braking performance of the driver, initial headway, and behaviour of the lead vehicle. Thus, these measures of safety margin integrate many different aspects of the driving task, such as operational performance and tactical behaviour, and may be regarded as good candidates for the unitary measures that serve as an organi­zing principle in the model presented here.

 

The general idea underlying the adaptation model is that any factor that affects operational performance may result in adaptation of behaviour on the tactical level, if the driving task is self-paced and if the driver is able to perceive these effects on operational performance. In this, feedback of the effects on operational performance may have a direct effect on tactical behaviour. For example, windgusts affect operational performance which, if detected by the driver, result in the choice of a lower speed. Alternatively, feedback effects may result in learned associations of the effects of various factors on operational performance, resulting in anticipatory adaptation responses. If, for example, the driver detects a fog bank in the direction of the vehicle path, he may already decrease speed, although the effect of fog on operational performance has not been experienced yet. The speed reduction then is an anticipatory adaptation response resulting from associations learned in the past between fog and effects on operational performance. If the driving task is not fully self-paced, the driver may elect to increase effort in order to improve operational performance. Time-based safety margins are proposed as the regulating mechanisms of behavioural adaptation. The strategies involved in this are a matter of further experimentation.

In the experimental section of this thesis two driving tasks, curve negotiation and car-following, are analyzed in more detail. Curve negotiation is essentially a lateral control task, while car-following is predominantly a longitudinal control task. Figure 9 presents a model for lateral control tasks. Figure 10 does the same for the longitudinal control tasks of car-following. Both models are almost identical, although they differ in the kinds of safety margins and behavioural adaptations.

 

 

Figure 9. Model for the lateral control task.

 

 

 

 

Figure 10. Model for the longitudinal control task of car-following.

 

 

2.6 Experimental validation of the model: research questions

 

Two different car driving tasks, negotiating curves and car-following, are studied in detail in the chapters that follow. The goal of the six experiments discussed in the chapters 4 to 9 is to examine one aspect of the present model: the prediction that individual differences in operational performance affect behaviour on the tactical level. The experiments were performed in a car driving simulator.

In experiment 1 the driver task of curve negotiation is analyzed. It focuses on the relation between steering performance and speed choice in curves with different radii. Drivers differ in steering performance in that some drivers consistently commit larger steering errors than others. Curve radius is manipulated as an situational factor that affects operational performance. In general, steering errors are larger in curves with smaller radii. It is then investigated how speed is affected by curve radius and by individual differences in steering competence as an adaptative response to steering performance. Drivers already decrease speed before the curve is entered and, thus, before the effect of radius on steering performance is experienced. The adaptation of speed then is assumed to be an anticipatory adaptation response that has been learned by experience in curve negotiation. Time-to-line-crossing (TLC) is used as a safety margin and it is explored whether this safety margin is affected by curve radius and steering competence.

In the experiments 2 to 6 the longitudinal control task of car-following is analyzed. It is examined whether choice of time-headway (THW), as behaviour on the tactical level, is affected by operational braking performance. During car-following, the driver has to take account of the possibility that the driver of the lead vehicle might brake. However, the driver never knows when the lead vehicle will brake, and if it does, how hard it will brake and for how long. It is then assumed that the driver has learned the quality of his or her own braking performance from previous experiences and that this results in a preference for a specific THW. THW is the time available to the driver to reach the same level of deceleration as the lead vehicle in case it brakes, without becoming involved in a collision. Braking performance is assumed to affect the time required to reach the same level of deceleration as the lead vehicle. Adaptation of THW may then be regarded as a tuning of available time to required time that is determined by braking performance.

In experiment 2 it is investigated whether choice of THW is related to the ability to brake as fast as possible in situations where the driver knows that the lead vehicle will brake and the level of deceleration at which it will brake. In this experiment the locus of effect of differences in braking performance is examined as well.

In experiments 3 and 4 it is examined whether choice of THW is constant over different speeds and whether individual differences in choice of THW are consistent. In experiment 3 the role of time-to-collision (TTC) on the moment the lead vehicle starts to brake is examined in detail. More specifically, it is tested whether the sensitivity of the braking response to TTC information differs as a function of preferred THW. In experiment 4 the process of braking itself is examined in more detail and a model of braking is presented starting from modern theories of perceptual-motor performance. The process of braking is separated into three sequential phases: a reaction time (RT) phase, an open-loop phase covering the initial motor response, and a closed-loop phase during which visual feedback is used to control the process of braking. It is tested whether TTC on the moment the driver detects the braking of the lead vehicle affects the early motor phase (open-loop component) of the braking process and whether the motor response differs as a function of preferred THW in unexpected emergency braking situations.

In the experiments 5 and 6 it is tested explicitly whether short followers differ from long followers in the open- and closed loop phases of the braking response by manipulating both phases. However, in experiment 5 specific task-related factors induced startle responses and vigilance effects requiring some methodological changes in the final experiment. In experiment 6 the level of deceleration of the lead vehicle is manipulated. This affects the TTC on the moment the driver detects the deceleration of the lead vehicle and this procedure aims to manipulate the duration of the open-loop response. It is tested whether the open-loop response of short followers is more strongly affected by this manipulation compared to long followers. This would support the idea that the sensitivity of the motor response to TTC information differs as a function of preferred THW, and thus, that short followers differ in operational performance from long followers. It is also examined whether preferred THW is related to differences in performance in other tasks that require a fast dynamic perception-response coupling as a test of the hypothesis that preferred THW is  related to perceptual-motor abilities that are more general than braking performance.

 


« Previous Entries