Auto Safety and Human Adaptation
The effectiveness of new high-tech auto safety devices will depend on how drivers use them.
Vehicle manufacturers around the world are spending large sums of money to develop sophisticated new safety devices. Anti-lock brakes were one of the first. Within the next 5 to 10 years, adaptive cruise control, collision warning, and vision enhancement systems are expected to become standard features on new cars, all in the name of safety. Governments, especially in Japan, Europe, and the United States, are contributing research funds for the development of these devices. Safety is the stated goal, but clearly economics is a major driving force. These devices will intrigue and attract car buyers.
Expectations for improved safety are high. But they may not be met if the human penchant to adapt is ignored. Take vision enhancement systems as an example. These systems use thermal imaging and a heads-up display to enhance the driver's view of the central part of the road scene, allowing drivers to more easily spot pedestrians and animals on the road at night. Such a system was first available in the United States on the 2000 model year Cadillac deVille. Such systems ought to improve safety but may not if they prompt people to drive more frequently in low-visibility conditions. There is already good reason for skepticism. Anti-lock brakes were expected to significantly reduce crashes. These devices work by sensing lockup and releasing the brake before applying it again: the same thing a driver does when pumping brakes but far more rapidly. By these means skidding is prevented and steering control is maintained. But as large studies have shown, they have not had a demonstrable effect on overall crash rates. Drivers appear to have changed their behavior in ways that reduced or eliminated the safety cushion provided by the improved braking.
Because we expect so much of these devices and because of their cost, it is time for governments and vehicle manufacturers to examine more fully the nature of driver adaptation and its effects. If policymakers truly want to improve safety, they must ensure that the research involved in developing and implementing these devices is comprehensive in its analysis of the human element. It is not enough to develop a better device; one has to know how humans interact with it.
There is also the issue of whether drivers will grasp how these devices function. Some of the limitations of the new systems are subtle. For example, adaptive cruise control, which is being introduced as an option on this year's European Jaguars and Mercedes, maintains a selected speed but also senses slower-moving vehicles ahead and responds to them by slowing the vehicle. However, because of technical limitations, it will not respond to stopped vehicles, which may result in an unpleasant surprise for the uninformed driver. A safety system that is poorly understood by the user isn't an improvement; it's a liability. Governments and vehicle manufacturers should consider the need for some form of education for drivers of these increasingly sophisticated vehicles.
Pervasiveness of adaptation
Although the aim of high-tech devices is to improve safety, the human proclivity for adaptation makes this a challenge. Adaptation, defined as the process of modifying to suit new conditions, is an everyday occurrence in driving and happens on many levels. Short-term adaptations occur when we are pressed for time and take a chance on running a red light. Long-term adaptations occur as we age. Older drivers reduce their speed by a few miles per hour on average and allow longer headways to vehicles in front.
We adapt our focus of attention to the specific driving task. Eye-movement studies of drivers show a dramatic narrowing of eye fixations when drivers are closely following another vehicle. Drivers in heavy traffic reduce by 20 percent the length of time that they spend glancing at the car radio while operating it, as compared to when they drive in light traffic. Adaptations also occur in response to the roadway environment. A change in traffic signals to provide an all-directions red light clearance interval will increase the numbers of drivers who enter the intersection during the caution period. Increasing the lane width, widening the shoulder, and resurfacing the roadway all result in higher speeds.
Adaptations also occur in response to vehicle changes. Changes that occurred before the advent of high-tech devices probably resulted in various adaptations. For example, the installation of turn signals inside the vehicle may have increased the likelihood of drivers signaling, especially in inclement weather. Automatic transmissions have accelerated the learning process for novice drivers, who no longer have to deal with shifting gears while controlling vehicle speed and lane position. In Canada, a standard driving course is 18 lessons for those learning on standard and 13 for those learning on automatic transmissions. Power-assisted brakes must have allowed drivers to approach situations requiring a stop at higher speeds. Improved car handling is thought to be one of the elements behind continual increases in average speed during the past 20 years.
Adaptation is intrinsically human. It is one of our most valuable characteristics and the reason why a human presence is desirable in monitoring even the most highly automated systems: to adapt to and therefore deal with the unexpected. Adaptation is a manifestation of intelligent behavior.
However, engineers who develop new devices to assist drivers frequently assume that drivers will not change their behavior. For example, when anti-lock brakes were introduced, predictions about their impact on safety were based on the assumption that only stopping distance and directional control during braking would change; speed and headways would not be affected. But that has proved incorrect.
Why do engineers make such assumptions? According to Ezra Hauer, a civil engineering professor at the University of Toronto, engineers are trained to deal with the characteristics of inanimate matter such as loads, flows, stress, strain, and so forth. Once the physics of the situation and the properties of the materials are understood, engineers can predict fairly well what will happen and make the corresponding design choices. But drivers adapt, and speed and headway choices and reaction times cannot be considered to be invariant quantities that remain the same once the roadway or the vehicle has changed. That adaptation will occur is predictable. We should be more surprised by its absence.
A prime example of unfulfilled predictions because of adaptation is anti-lock braking. In early proof-of-concept studies, test drivers drove at a designated speed and then braked. Not surprisingly, braking distances were found to decrease on wet surfaces. Moreover, directional control was maintained during braking on wet or dry surfaces. Based on such studies, optimistic predictions were made. For example, one German engineer concluded that the universal adoption of anti-lock brakes in Germany would result in a 10 to 15 percent reduction in accidents involving heavy damages and/or injuries.
Later studies considered the possibility of adaptation. A test track study showed that when drivers could choose their speed, they traveled slightly faster after practicing with anti-lock brakes on wet surfaces, with the result that emergency stopping distance was no different than with standard brakes. Other researchers observed 213 taxi drivers en route to an airport and likely to be pressed for time. Drivers whose vehicles were equipped with anti-lock brakes were found to allow significantly shorter headways to the vehicles in front of them.
How was safety affected? In an extensive study, the Highway Loss Data Institute compared claim frequency and size of 1991 models without anti-lock brakes to those of 1992 models with the system. No significant differences were found in either claim frequency (8 per 100 vehicles) or size (an average of $2,215 per 1991 model claim versus $2,293 per 1992 claim). Researchers then examined a subsample from the northern states in the winter and still found no significant differences. Based on the performance studies and on this crash rate study, it appears that drivers with anti-lock brakes adapted by trading off safety for mobility to the extent that there was no safety benefit--a far cry from the predicted 10 to 15 percent improvement.
The fact that adaptation frequently leads to less safety and more mobility should not surprise us. Unfortunately, safety and mobility are frequently though not always inversely correlated. An improvement in mobility, higher speeds, or easier lane changing may result in a decrease in safety. Mobility improvements provide an immediate payoff: Drivers reach their destinations faster. Safety improvements are far more intangible; for example, a change in the risk of a certain type of accident from 1 every 100 years to 1 every 150 years.
Another potentially important tradeoff influencing driver strategy has only recently been recognized. New devices such as navigational aids make it possible for drivers to devote less attention to the road and more to other activities. We live in an age in which people are trying to accomplish more in less time. The proliferation of cell phone use in vehicles is evidence of the desire to be more productive while driving. How this tradeoff might affect safety is only beginning to be studied.
As in-vehicle systems change the nature of driving, they will also affect the choices made by drivers. In particular, they are likely to affect the decision to drive. Vision enhancement systems may make drivers feel more comfortable about driving in poor visibility conditions. Collision warning systems may encourage a fatigued driver to keep going when he or she might otherwise have stopped. A navigation system may encourage tourists to explore more widely than they might have otherwise.
In-vehicle systems will also affect the choices made while driving. Anyone who has driven a car with brakes in need of maintenance knows that one becomes more cautious and drives more slowly and allows greater headway to the vehicle in front. It is hardly surprising that drivers equipped with anti-lock brakes do the reverse.
Adapting to change
One of the up-and-coming driver aids is adaptive cruise control. In one study, adaptive cruise control was compared to standard cruise control and manual driving in an on-road test. Not surprisingly, the results indicated that the adaptive cruise control system conferred a substantial margin of safety as compared to the two other modes. But how will the availability of adaptive cruise control affect behavior? Will people drive more or for longer periods? Will they be more inclined to drive in high-density traffic, to drive when tired, and to spend more time on nondriving tasks? More study is needed.
A hazard of particular concern with adaptive cruise control is a stopped vehicle ahead. Currently, these systems respond only to moving vehicles, in order to avoid other contingencies, such as a vehicle that slows inappropriately on a curve because it assumes that a stationary object on the side of the road is actually on the road. This means that the system will not respond to a stopped vehicle, such as one at the end of a stopped line of cars. Unfortunately, drivers are slow to perceive rapid closing of their vehicle with another, especially at night. This makes stopped vehicles particularly hazardous, especially if the driver has become dependent on the system to detect and respond to unsafe headways. Lack of attention to the road ahead because of dependence on adaptive cruise control may contribute to crashes into such stopped hazards. In fact, a simulator study has demonstrated this to be the case.
Vision enhancement systems are designed to assist drivers in detecting hazards, particularly pedestrians and animals, under low-visibility conditions. As noted earlier, they do so by providing an enhanced view of the central portion of the road ahead. Unfortunately, according to studies, these systems also appear to reduce the likelihood that peripheral objects will be detected and identified. Better vision of the central portion of the road may prompt drivers to drive faster. Studies in Finland found that improving roadway guidance by using post-mounted reflectors on winding, substandard roads resulted in inappropriate increases in speeds and higher rates of nighttime collisions. Vision enhancement systems may well result in a decreased likelihood of crashes involving hazards on or near the road but an increase in the number of crashes involving hazards entering the road. In addition, as with adaptive cruise control, vision enhancement systems may also lead to more driving in poor visibility, particularly by older drivers.
Navigation systems have received much attention from researchers, with most studies focused on how their use affects driver attention to the road ahead. Navigation systems are intended to help drivers find their way in unfamiliar areas by presenting visual and/or auditory directions in response to the entry of a destination address. These have been available for almost 10 years as an option on a few car models and are now becoming more widely available. An early system, the ETAK navigator, used a screen on the dashboard to show drivers a map indicating where they were relative to their destination. It allowed the driver to chose the map scale by using a zoom feature. One study used video cameras to observe how many times drivers using an ETAK navigator glanced away from the road compared with drivers using a map or following a memorized route. Study results showed that with ETAK, 43 percent of glances were away from the road ahead, compared to 22 percent with the map and 15 percent on the memorized route. Other studies indicate that older drivers are particularly affected by the use of in-vehicle navigational aids. They glance away from the road more frequently and for longer periods of time.
Although these studies raise safety concerns, we cannot be sure of the effect without knowing much more about where drivers are looking. Studies of driver eye movements done 30 years ago suggest that drivers have a fair amount of spare capacity; they can glance at objects other than road signs and other vehicles without diminishing safety. However, today's roads are much busier, and spare capacity is likely to be considerably less.
To date, visual demand associated with navigation systems has been measured with video cameras that allow researchers to separate glances at the navigation display from those at the mirrors and at the road scene ahead. However, we need to measure eye glances more precisely to really understand how using a navigation system affects safety. Specifically, we need to know how far ahead the driver is looking and how appropriately he or she monitors nearby traffic, with and without a navigation system. A particular concern is vulnerable road users. One study showed that drivers at a T junction turning right spent much more time looking left toward oncoming vehicles than right toward pedestrians or bicyclists who were about to cross the driver's path.
Although the amount of time that drivers spend looking at navigation displays raises safety concerns, there is also reason to believe that drivers adapt appropriately to increases in traffic. The time drivers spend glancing at signs in high-density traffic is about half that found in low-density traffic. Similarly, drivers using a map-based navigation system in an on-road study had glance durations 30 percent less than those in a simulator study where the traffic demands were lower.
An on-road study using a map display navigator examined the influence of traffic density on attention to the display. Subjects used the system to drive in unfamiliar areas that varied greatly with respect to traffic density. The driving difficulty of various road sections was rated and compared to driver eye scan patterns. As driving difficulty increased, the probability of a glance to the roadway center increased, whereas the probability of a glance to the navigational display decreased. In addition, the length of glances to the roadway center increased for high-density as compared to low-density traffic and even more so when critical incidents occurred.
These data suggest that most drivers will tailor their glances at in-vehicle displays or tasks to the driving workload. However, it is necessary to examine changes in the detection of on-road hazards to be sure that safety is not compromised. Such an approach was taken in a study using the Federal Highway Administration driving simulator to compare driver detection ability, as well as car control, for various types of navigation systems, including maps, auditory messages, and visual displays. The detection task involved watching dashboard instrument gauges for out-of-range indications. Various driving scenarios were used to vary the difficulty of driving and the difficulty of the detection task. Drivers appeared to cope with greater display complexity and greater task difficulty by dropping their speed and by reducing the attention paid to the detection task. The detection task was performed most poorly for the paper map group and next most poorly for the complex map display. Overall, subjects missed 16 percent of the signals presented. Older subjects using the complex map display or the paper map missed large numbers of signals (approximately 40 and 50 percent, respectively). Other types of visual and auditory devices (with the exception of the paper map) were associated with much lower miss rates.
Although this study did examine changes in attention, the task used was one of watching gauges inside the car. The more critical task in driving is watching the road for hazards such as pedestrians, bicyclists, or debris. The effect of navigation systems on such detection remains to be studied.
There has been little research addressing how any high-tech device changes the extent of driving. One Japanese experiment demonstrates some interesting adaptive effects of a navigation system. The results showed that lost drivers benefited from car navigation information and revised their route more easily than those who used maps. Users of car navigation systems appeared to worry less about the consequences of becoming lost and therefore intentionally traveled more on neighborhood streets to avoid congested arterial streets. Widespread use of such systems and traffic congestion information may increase neighborhood congestion unless countermeasures are taken.
Further research is required on other potential changes resulting from the use of navigation systems. There may be more driving by drivers unfamiliar with routes. There may be less attention to the road ahead, resulting in poorer detection of hazards. The overall safety effect will depend on the tradeoff between fewer lost and distracted drivers relative to greater exposure of unfamiliar drivers. It will also depend on the tradeoff between reduced attention required to the road ahead because of the navigation task being aided and greater demands required inside the vehicle.
If the task is changed, drivers will modify their behavior. The task of designers and researchers is to ensure that the design encourages optimal modification. This is done by considering the likely changes in strategy and by modifying the design to ensure that the resulting behavior is appropriate to the design goal of increased safety.
A good example of this approach is a 1997 study by Weil Janssen of TNO Human Factors Research Institute, Soesterberg, the Netherlands, and Hugh Thomas of Bristol Aerospace, Bristol, United Kingdom. Performance was measured for three types of collision avoidance systems: driver's braking distance shown by a horizontal red line projected onto the windshield with a heads-up display; drivers warned through accelerator pedal resistance when the time to collision to another vehicle was less than 4 seconds; and drivers warned, as above, either when time to collision was less than 4 seconds or when the time headway to the car in front was less than 1 second
These three systems were compared with the use of a driving simulator. Vehicles ahead were presented with an initial headway of seven seconds. A variety of closing speeds were used, ranging from 10 to 40 kilometers per hour. In a quarter of the scenarios, the vehicle ahead of the driver braked, creating an emergency situation. Frequent but irregular oncoming traffic made passing difficult. The results showed that only the second collision avoidance system, which warned the driver of less than four seconds time to collision, provided a safety benefit. It reduced the percentage of time that the headway was less than one second, without increasing average speed. In simulated fog conditions, the heads-up display that showed braking distance significantly decreased driver safety by increasing short headways relative to when drivers had no collision warning system.
Based on a purely mechanistic analysis, one would expect the third system to be better than the second. However, the results showed that adding a simple one-second headway trigger criterion to four seconds time-to-collision criterion significantly worsened driver safety by increasing the proportion of short (less than one second) headways and the average speed. Because there were two distinct criteria, drivers may have found it more difficult to understand how the system was operating. It is sobering to remember that, in one of the first accidents involving an anti-lock braking system, a police officer in a high-speed chase responded to the unfamiliar vibration of the anti-lock brake by releasing his foot. In short, the driver's understanding of how the device operates is an important issue that has received little attention to date.
Where public funds are spent on high-technology development, it is incumbent on governments to ensure that adaptive effects are considered. This means observing whether and how driving strategy changes as a result of using a new device before making unfounded predictions on likely improvements in safety. Better predictions require a variety of evaluations. Initially, simple mockups can be used to evaluate driver understanding of how to operate the interface and driver expectations of how such a device would function (for example, the expectation that adaptive cruise control will detect stopped cars ahead). This will give insight into likely driver errors or misunderstandings when using the device. At the next stage, the device can be studied in a driving simulator. Behavior--such as speed, following distances, length of glances required to operate the device, removal of the foot in response to the vibration associated with anti-lock brakes--can be compared with and without the device being used, and devices with different functionality can be compared, such as adaptive cruise control that merely reduces acceleration versus adaptive cruise control that applies some braking. These results can be used to optimize the functional design.
The final stage involves testing the device on the road to observe how drivers use it in real traffic initially and over time as they adapt to it. For example, do drivers equipped with adaptive cruise control turn more and more of their attention to nondriving tasks such as using a cell phone? Over time, do elderly drivers with vision enhancement systems drive more at night than those not so equipped? In other words, what tradeoffs are made and is the net result likely to improve safety?
Responsible vehicle manufacturers must concern themselves with optimizing these devices to achieve the greatest safety possible. It may not matter if a high-tech VCR or microwave oven confuses its owner. But injury and death can result from a driver who does not understand the functioning of his or her brakes, vision enhancement system, adaptive cruise control, and so on.
It is also becoming clear that some form of education is needed for drivers using vehicles with sophisticated systems. A recent study in Quebec showed widespread misunderstanding by drivers of whether their own vehicles were equipped with anti-lock brakes (27 percent did not realize they had them) and how braking was affected (47 percent thought anti-lock brakes improved braking on dry surfaces). Anti-lock brakes are just the beginning. Much more complex devices are coming soon. Drivers with adaptive cruise control and vision enhancement systems will have to understand their specific limitations. Manufacturers can and should provide well-designed manuals and educational videotapes. This is already being done for adaptive cruise control systems. However, manufacturers are not in a position to verify through testing that drivers have understood the new technology. It is up to departments of motor vehicles to consider modifications to licensing tests to assess driver understanding of these new technologies. There may even need to be retesting requirements for drivers who buy vehicles equipped with several devices.
There is an enormous need to improve road safety; worldwide, about half a million people are killed annually in traffic crashes. In the United States alone in 1999 more than 40,000 people were killed and more than 3 million were injured. The economic cost in 1994 was more than $150 billion. Road safety can be improved through high technology, but to do so, the complexities of human adaptation must be addressed and drivers must be informed about how these devices function.
T. A. Dingus, J. F. Antin, M. C. Hulse, and W. W. Wierwille, "Attentional Demand Requirements of an Automobile Moving-Map Navigation System," Transportation Research Record 23A, no. 4 (1989): 301315.
Highway Loss Data Institute (HLDI), "Collision and Property Damage Liability Losses of Passenger Cars with and Without Antilock Brakes," Highway Loss Data Institute Report A-41 (Arlington, Va.: HLDI, 1994).
W. Janssen and H. Thomas, "In-Vehicle Collision Avoidance Support under Adverse Visibility Conditions," in Ergonomics and Safety of Intelligent Driver Interfaces, ed. Y. I. Noy, Mahwah, N.J.: Lawrence Erlbaum Associates, 1997), 221229.
F. Sagberg, S. Fosser, and I.-A. F. Saetermo, "An Investigation of Behavioral Adaptation to Airbags and Antilock Brakes among Taxi Drivers," Accident Analysis & Prevention 29, no. 3 (1997), pp. 293-302.
J. Walker, E. Alicandri, C. Sedney, and K. Roberts, In-Vehicle Navigation Devices: Effects on the Safety of Driver Performance [Warrendale, Pa.: Society of Automotive Engineers (SAE), Publication No. P-253, SAE Paper No. 912793, 1991].
Alison Smiley (email@example.com) is president of Human Factors North Inc. in Toronto, Canada, and chair of the Transportation Research Board's Group 3: Operations, Safety and Maintenance of Transportation Facilities. This article is adapted from "Behavioral Adaptation, Safety and ITS" (Journal of the Transportation Research Record, in press).