LOGO Imperfect in- vehicle collision avoidance warning systems can aid distracted drivers Masha Maltz, David Shinar Transportation Research Part F 10 (2007)

LOGO

Imperfect in-vehicle collision avoidance warning systemscan aid distracted drivers

Imperfect in-vehicle collision avoidance warning systemscan aid distracted driversMasha Maltz, David Shinar

Transportation Research Part F 10 (2007) 345–357

Introduction

Automated systems designed to aid drivers are increasingly popular among car manufacturers, and exist in a myriad of forms, from the drawing board to instrumented vehicles.

It may introduce two types of problems: (1) failure by the driver to take appropriate action on the rare occasion when the system malfunctions and the driver has relaxed his/her vigilance, and (2) false alarms, when the driver responds to an erroneous signal.

The purpose of this study was to examine the potential pitfalls of two critical aspects of automation: information acquisition and information analysis.

An effective IVCAWS that reliably monitors headway and warns an inattentive driver when his or her vehicle approaches a lead vehicle too closely.

Study Goals

In this study, we addressed the issue of driver mental workload.

The addition of a secondary task to the driving environment can be a realistic approximation of driving on the real road where the driver is often distracted by other tasks.

Method-participants

Forty-three undergraduate students participated in the experiment as part of an Ergonomics course requirement.

The participants, 20 males and 23 females, ages 22–43 (average age 26) were all licensed drivers with 2–27 years’ experience.

The experiment was conducted on a STISIM DriveTM simulator, a fixed-base driving simulator developed by Systems Technology, Inc.

The simulator continually sent digital output, consisting of temporal headway measurement, to a second computer via coaxial cable.

Method-physical setup

A second computer generated the auditory alarms, based on the input from the simulator and on internal calculations of warning system reliability.

The alarms were medium pitch tones emitted by a sound blaster through speakers at a sound level of 84 dBA for the regular alarms and 82 dBA for the graded alarms.

All performed the loading visual task that consisted of responding to changes in two peripheral targets.

The loading task (DA) appeared at random intervals of between 5 and 10 seconds.

The driver’s task was to respond to the arrow by activating the turn signal, signaling ‘left’ for a left or down arrow, and signaling ‘right’ for a right or up arrow.

If at the time an error was to be generated the headway was too short (less than one second), the system would fail to sound the alarm for a 2 s period. If at the time an error was to be generated the headway was one second or more, the system would sound a false alarm for 1 s.

The three experimental groups had high, medium, and low reliability, with 1, 4, and 8 errors per minute, respectively.

Before the start of the experimental trials, the drivers were told that the visual loading task was of primary importance.

They were also told to maintain 1 s (i.e., safe) headway and to try and reach the destination as quickly as possible, while keeping a safe distance.

Experimental procedure

The first lead vehicle appeared once the driver had traveled 100 feet. If the driver approached the lead vehicle to within .5 s, the lead vehicle sped ahead and disappeared.

During the course of the trial, nine additional vehicles entered the driver’s lane in order to ensure that a lead vehicle would always be present.

Participants were told that the IVCAWS system was designed to help them keep the proper headway, but that it was imperfect.

Each participant received two 4.5 mile trials that took about seven minutes to complete, one with a single alarm and one with a graded alarm.

During each trial, 85 visual loading task events occurred at variable intervals.

Results an discussion

If TH increased by at least 3% within two seconds from the event, it was recorded as a deceleration response since lead cars were programmed to travel at fixed speeds.

TH was consolidated into five levels for the analysis: <0.5 s, 0.5 to <1.0 s, 1.0 to <1.5 s, 1.5 to 62.0 s, and >2.0 s.

No significant effect or difference in the headways maintained by drivers in the experimental and the control groups was found (F(4,164) = .25; p = .91).

While the participants actually increased their short headways – from 13.6% in the pretrial segment to 35.2% in the experimental trials – the experimental groups retained the percentage of time in short headways – averaging 11.8% in the pretrial and 10.1% in the experimental trials, respectively.

Loading task

They responded correctly to between 73% and 100% of the events, and missed between 0% and 18% of the events.

They also incorrectly responded to some events, which, apparently, was the result of some directional confusion.

We found that the performance of the loading task remained stable across all groups.

The control group and the experimental groups as a whole both missed an average of 3% of the visual motor task events, and the high, medium, and low system reliability groups missed an average of 3%, 2%, and 4% of the events.

Responses to alarms

There were no significant differences among the experimental groups in the percentage of the time that the drivers responded (slowed down in response) to graded alarms (F(2,26) = .52; p < .5983).

With the low system reliability group decelerating the over 80% of the time compared to the other groups who decelerated an average of 50%of the time.

Therefore, they were probably most likely to notice that they were in the danger zone independently of the IVCAWS alarms.

A 2-way ANOVA [system reliability group (3) X alarm state (sounded or not sounded)] was computed on the fraction of the time in the danger zone that the drivers slowed down, with alarm treated as a repeated measure.

Drivers slowed down 28% of the time in response to false alarms compared to 21% of the time when there were no alarms while in the safe zone.

Subjective measures

73 percent of the participants rated the single alarm system as helpful or very helpful and 60% of the participants rated the graded alarm system as helpful or very helpful.

The source of the difference between the system types was the high system reliability group: 90% of the participants in this group found the IVCAWS helpful with a single alarm, but only 50% of them considered the graded alarm helpful.

Post hoc comparison showed a marginally significant difference between the control group and the medium system reliability group.

The control group found the task more difficult than the medium system reliability.

Summary and conclusions

The general findings of this study are consistent with our previous findings that show that IVCAWS can be effective in helping drivers recognize and maintain optimum headways.

We consider that distracted drivers will maintain longer headways with an IVCAWS of this type.

We found that higher system reliability actually had a negative effect on performance. Surprisingly, the IVCAWS appeared to be most effective when it was least reliable.

This finding is consistent with the demonstrated phenomenon of ’complacency’ that automation introduces (Desmond et al., 1998).

The drivers in this study who were given the least reliable system learned from the error-prone system to discern one-second headway more slowly.

The drivers with the high-reliability system, on the other hand, were led astray when the system erred.

Given a demanding loading task, the drivers without an IVCAWS (the control condition) suffered the expected drop in driving performance, as measured by their headways.

The drivers with the IVCAWS maintained their driving performance at a reasonable level.

At higher reliabilities, the IVCAWS led to over reliance and complacency which ultimately led to more mistakes than with the less reliable system.