Interaction Weaknesses of Personal Navigation Devices [original]

Interaction Weaknesses of Personal Navigation Devices

Markus Hipp*, Florian Schaub*, Frank Kargl†, Michael Weber*

*Institute of Media Informatics

Ulm University

89069 Ulm, Germany

{firstname.lastname}@uni-

ulm.de

†DIES Research Group

University of Twente

P.O. Box 217

7500 AE Enschede, NL

f.kargl@utwente.nl

ABSTRACT

Automotive navigation systems, especially portable navi-

gation devices (PNDs), are gaining popularity worldwide.

Drivers increasingly rely on these devices to guide them to

their destination. Some follow them almost blindly, with

devastating consequences if the routing goes wrong. Wrong

messages as well as superfluous and unnecessary messages

can potentially reduce the credibility of those devices. We

performed a comparative study with current PNDs from dif-

ferent vendors and market segments, in order to assess the

extent of this problem and how it is related to the inter-

action between device and driver. In this paper, we report

the corresponding results and identify multiple interaction

weaknesses that are prevalent throughout all tested device

classes.

Categories and Subject Descriptors

H.1.2 [User/Machine Systems]: Human factors; H.5.2

[User Interfaces]: Evaluation/Methodology; H.5.2 [User

Interfaces]: Ergonomics

General Terms

Human Factors, Experimentation

Keywords

HCI, personal navigation devices (PND), study

1. INTRODUCTION

Automotive navigation systems have become an assistant

technology many drivers rely on. Navigation systems are ei-

ther directly integrated in the vehicle or come as add-on solu-

tions, like dedicated portable navigation devices (PNDs), or

smartphone software solutions. By navigation system we re-

fer to a device or component that performs routing decisions

locally, in contrast to online routing services where some or

all routing calculations are performed by servers online. In

this work, we focus on PNDs, due to their widespread adop-

tion and global market share [5].

AutomotiveUI’10, November 11-12, 2010, Pittsburgh, Pennsylvania.

ACM 978-1-4503-0437-5

Frequent drivers, business travelers, vocational drivers, and

families entrust their navigation systems with finding a suit-

able route to their destinations. Nowadays, some drivers fol-

low them almost blindly, i.e., by reacting instantaneously to

driving directions announced by the device. Such blind trust

can lead to dangerous situations as underlined by anecdotal

reports of people driving their cars into a river because they

followed the commands of their navigation system too liter-

ally [7]. However, the case of a German driver who caused an

accident by turning around on a highway because his satel-

lite navigation systems told him so [6] underlines the serious-

ness of the problem. Hanowski et al. [4] were able to show

that the confidence placed in the commands of a navigation

system and its credibility increases with the driver’s unfa-

miliarity with the environment. They also showed that the

driver’s self-confidence decreases in such situations. Thus,

the further away drivers are from their known environment,

the more they rely on their navigation systems and trust

their commands. In such situations drivers may also be-

come susceptible to gullibility errors [3], which could result

in them acting upon erroneous commands that they would

most likely reject under normal conditions.

1.1 The Credibility Issue

Apparently, credibility is an important aspect of the problem

at hand. Fogg and Tseng [3] distinguish between different

types of credibility of software systems. Presumed credibility

is based on general assumptions about the functionality of

a system, surface credibility is attributed based on the im-

pression made by the hardware and interface design, reputed

credibility stems from the recommendations and opinions of

third parties, while experienced credibility is based on earlier

personal experiences. The first three aspects are prevalent

in forming the initial credibility a driver attributes to the

navigation system. However, experienced credibility evolves

over time. While the first three credibility aspects in com-

bination with unfamiliar environments may facilitate gulli-

bility errors, the consequence of such an error will almost

certainly destroy experienced credibility of the navigation

system. Mistrust towards future commands of the device

will be the result.

In familiar environments, a slightly different situation un-

folds. Drivers feel more confident and due to their back-

ground knowledge can better judge the correctness of nav-

igation commands, at least to a certain extent. There-

fore, the danger of drivers acting upon erroneous commands

should decrease. But in a familiar environment drivers also

form their own (not necessarily correct) belief what the best

route would be in a given situation. A mismatch between

the planned route of the navigation system and the driver’s

belief of the best route is possible. Drivers may deliberately

deviate from the suggested route to evade traffic jams or

to make a purposeful detour, e.g., to stop at the grocery

store, or because they think they know a shortcut. Typi-

cally, navigation systems react by issuing commands to lead

the driver back onto the planned route or a newly calcu-

lated alternative route. But because the driver deliberately

deviated from the route such messages will be considered

annoying especially if they occur repeatedly. It is likely that

such annoying or superfluous commands also have a negative

effect on the experienced credibility of the device.

Another feature that may cause similar effects is the inte-

gration of dynamic traffic information in the routing pro-

cess. Some navigation systems can receive dynamic traffic

messages from different sources to inform the driver about

traffic obstructions ahead and to suggest a detour. While

potentially a useful service, the presentation of such infor-

mation can impact the experienced credibility of the system.

For example, displaying the age of a traffic message would

be a useful indicator to judge its relevance and accuracy,

while its absence may foster gullibility errors, if the naviga-

tion system suggests a detour to avoid a traffic jam based on

information that is hours old. Furthermore, the benefit of

a detour needs to be apparent to drivers and should match

their understanding of an appropriate alternative route.

Thus, the credibility of navigation systems can be reduced

by gullibility errors, but superfluous or annoying messages

can also impact it negatively. The experienced credibility

can potentially degrade to such a level that incredulity er-

rors occur [3]. Drivers would ignore navigation commands

because they doubt their accuracy, even if following these

commands would actually benefit them. This of course can

lead to a further decrease of experienced credibility. In ad-

dition, the exchange of negative experiences between cus-

tomers can also decrease the reputed credibility of a prod-

uct. Therefore it is in the best interest of vendors to ensure

that not only presumed and surface credibility remain high

but also experienced and reputed credibility.

1.2 Contribution & Outline

In this work, we analyze the extent of interaction issues of

PNDs, which may cause credibility deterioration. In partic-

ular, we focus on the frequency of superfluous and erroneous

messages and the integration of dynamic traffic warnings.

We performed a comparative study with PNDs of different

vendors under real world conditions to assess their inter-

action behavior with the driver. Based on the results, we

identified weaknesses in the interaction process and particu-

larly problematic scenarios. As part of a structured research

approach, these results will be used in future work to assess

credibility deterioration in navigation systems and to de-

velop optimized interaction processes for such devices.

Section 2 provides background information on navigation

systems with a focus on information sources that impact

routing decisions and navigation commands. Section 3 out-

lines the setup of our study, and Section 4 introduces the

applied metrics. Results are presented in Section 5. A dis-

cussion of these results and the identified weaknesses follows

in Section 6. Section 7 concludes the paper.

2. BACKGROUND

Navigation systems draw information form multiple sources

to make routing decisions. Besides the current position,

which is usually determined with GPS, static map data

stored in the device constitutes a main information source.

Streets are represented by directed graphs, where nodes rep-

resent intersections or fixed points on the map and edges

represent streets connecting nodes. Directed multigraphs

are used to represent multiple lanes, i.e., two nodes can be

connected with multiple edges. Direction of a lane is repre-

sented by edge direction. Additional attributes can express

the type of streets, speed limits, etc. Edges are weighted. In

the easiest implementation, a weight represents the length

of a street segment, but weights may also be time-dependent

to indicate different traffic situations over the day. The lat-

ter enables devices to optimize routes for the current time of

day. To identify the optimal route, the cheapest path from

start to finish has to be found. Different graph search algo-

rithms exist to calculate these paths, for example, Djikstra’s

SPF [2] or A* [1].

Another information source for PNDs are dynamic traffic

messages. Service providers as well as some PND vendors,

like TomTom1or Garmin2, collect driving information from

their users to generate time-dependent traffic information.

TomTom devices with iQ-Routes come preloaded with a

database filled with the information from over 10 billion

driven kilometers. This additional information is used to

calculate the most effective route for the driver depending

on the current time of day, which might not necessarily be

the shortest. Garmin offers n¨

uLink! online services, which

provides traffic information and Internet search functional-

ity to models equipped with an integrated GSM module.

In addition to proprietary solutions, different services ex-

ist to broadcast information about traffic jams, accidents,

or other incidents to vehicles. The most common service

is the Traffic Message Channel (TMC), which is a partic-

ular service of the Radio Data System (RDS), now under

control of the Traveller Information Services Association3

(TISA). TMC provides travel information via signals broad-

cast over the FM-Data-Channel. TMCpro is an improved

derivative of the TMC service, operated as a commercial

service by Navteq.4While TMC relies mainly on reports by

drivers, TMCpro utilizes real-time traffic data automatically

generated by data sensors, which are installed on highways.

The process of extracting traffic warning from the gathered

data is automated and not subject to editorial control. As

the extension possibilities of TMC and TMCpro are limited,

the TPEG5standard, also maintained by TISA, is able to

provide more precise information about traffic jams or acci-

dents. TPEG will replace TMC as the standard service to

transmit dynamic traffic information.

1TomTom website: http://www.tomtom.com/

2Garmin website: http://www.garmin.com/

3TISA website: http://www.tisa.org/

4Navteq website: http://www.navteq.com/

5Transport Protocol Experts Group

Navigation system manufacturers also provide other features

to improve user interaction. For example, in order to pro-

vide more precise announcements, some devices use text-

to-speech technology to convert natural language text into

speech. This enables the inclusion of street names and other

dynamic attributes into announcements, which helps the

driver to relate the provided information to the real world

environment.

3. STUDY DESIGN

The goal of our study is the comparison of the interaction

behavior of different navigation systems under real world

driving conditions. We focus on the frequency of naviga-

tion commands and messages, as well as the quality of the

provided messages. See Section 4 for the employed metrics.

But first, we describe the analyzed devices and the driving

scenarios of the study.

3.1 Devices

For the study, only portable navigation devices (PNDs) have

been considered as they constitute the most commonly used

type of navigation systems [5]. Due to their portability they

also facilitate comparison under the exact same conditions,

as multiple PNDs can be installed in one car.

We want to gain a reasonable overview of the interaction

concepts currently prevalent in the market and, therefore,

selected devices covering bottom-range ($), mid-range ($$),

and top-range ($$$) segments of the German PND market.

The analyzed devices are listed in Table 3.1. In the re-

mainder of this paper, we will refer to the devices by their

designated labels.

Three of the five test devices support the reception of dy-

namic traffic information via TMC or TMCpro. Tomtom

further supports the iQ routes system, while Garmin only

supports the Garmin operated n¨

uLink! service. Pearl, as

a bottom-range device, is the only navigation system that

does not come directly with a receiver for dynamic traffic

information services. Note that text-to-speech is only sup-

ported by the mid-range and top-range devices.

3.2 Driving Scenarios

The performed test drives were split into predefined driving

scenarios to analyze the behavior of the navigation systems

in different situations. During the test drives, the devices

were mounted on a common board inside the vehicle. A

setup of two cameras monitored the audiovisual output of

the navigation systems as well as the current road situation.

Additionally, two persons transcribed navigation commands

and audiovisual messages issued by the devices. Fig. 1 de-

picts the in-car setup. In preparation for each scenario, all

tested devices were set to the same destination, and it was

verified that the devices calculated a similar route (Rc). De-

pending on the driving scenario (see below) deliberate devi-

ations from the route were planned and performed, resulting

in a different driven route (Rd). The driven route Rdis sup-

posed to reflect the intentions of a driver and the driver’s

belief of the correct route.

The driving scenarios have been designed to specifically an-

alyze how the devices react in common driving situations.

Figure 1: In-car setup of the navigation systems.

Thereby, a balanced mixture of urban, suburban, rural, and

highway road usage was factored in.

Detour (city). The calculated route Rcis left for a de-

liberate urban detour. A typical reason for such driving

behavior would be the need to run quick errands before em-

barking onto the actual journey, e.g. to stop by a grocery

store or drop children off at school. In this scenario, the

driven Rdand the calculated route Rcdo not match. The

main purpose is to analyze how devices react to deviations

from the route. Road usage in this scenario is mainly urban

and suburban.

Highway. On the highway, commands of the navigation

systems are followed to analyze normal operation. In ad-

dition, a stop at a petrol station is simulated to see how

navigation systems react.

Inner city. In this scenario, the routing behavior of the

devices close to a destination is analyzed. Devices should

support the driver in finding the destination, yet provide

freedom to search for a parking spot in proximity of the

destination. Urban roads dominate this scenario.

Detour (rural). Similar to the detour (city) scenario, the

driver deviates from the route on purpose. However, the

detour is chosen over rural roads in such a way that there is

no obvious alternative route the PNDs could switch over to.

A typical scenario for this case would be the spontaneous

decision to visit some touristic attraction or friends who live

close to the original route.

Dynamic traffic warning. Here, we deliberately travel a

chosen route during rush hour with many traffic jams along

the route. The aim of this scenario is the analysis of the

integration of dynamic traffic warnings. Of special inter-

est is the presentation of the warning message, associated

information, and potential alternatives to the driver.

The first four scenarios were combined in one route from

Ulm University, just outside the town of Ulm, to a loca-

tion in the inner city of Munich, and back. Fig. 2 gives an

Table 1: Analyzed Devices

Label Manufacturer and model Text-to-speech Dynamic traffic services Price segment

Becker Becker Traffic Assist Z205 yes TMC, TMCpro $$$

Garmin Garmin n¨

uvi 1690 yes n¨

uLink! $$$

Falk Falk F6 yes TMC, TMCpro $$

Tomtom TomTom XL Europe yes TMC, iQ Routes $$

Pearl Pearl V35-1 no - $

detour

(city)

detour

(rural)

highway

inner

city

Figure 2: The calculated route Rc(blue) and the different scenarios (red ). Map data: OSM.org

overview of the planned route Rc(blue), the different sce-

narios, and the driven route Rd(red). The dynamic traffic

warning scenario was performed and tested separately, as

one had to dynamically look for suitable traffic jams.

4. METRICS

To allow an objective comparison of the different devices

under test, we define specific metrics for measuring their

behavior. First of all, we distinguish between acoustic and

visual messages and navigation commands. Visual messages

contain all information that is shown on the display. Acous-

tic messages are voice instructions from the system, which

either support visual messages or provide distinct informa-

tion. A navigation command may consist of an acoustic or

visual message, or both. For example, a typical navigation

command for turning right in 300 meters would consist of

the acoustic message “turn right in 300 meters” and a visual

message on the display composed of an arrow pointing to

the right and the distance below or next to it. Animations

that convey device activity are also counted as visual mes-

sages, e.g. two animated turning arrows when the route is

recalculated.

In the analysis, we distinguish observed navigation com-

mands according to message correctness and driver antic-

ipation. We define message correctness to reflect the cor-

rectness of the issued navigation command in terms of the

driver’s intention. This means a message is correct, if its

content conforms with and supports the driver’s belief of

the correct route. All other messages are labeled false. In

our study, all events that correspond to the route intended

by the driver Rdare labeled as correct.

 





 

 

Figure 3: Definition of four message categories.

Driver anticipation measures if the driver can expect the

message or if the message comes as a surprise, i.e., is unex-

pected. Expected messages are predictable in the sense that

they fit the current situation and align with the behavior

of the driver, but not necessarily with the driver’s intention.

They constitute the majority of all messages. All other mes-

sages appear unexpected. In our study, a message is labeled

expected if it supports the calculated route Rcin the current

situation.

Fig. 3 depicts the two axes and the four possible message cat-

egories. Correct/expected messages constitute desired nav-

igation commands. Correct/unexpected messages provide

correct information but are not anticipated in the current

situation, e.g., the message “follow the road for 60 kilome-

false correct

expected

unexpected

41% 56%

2%1%

Figure 4: Average distribution of all observed mes-

sages.

ters” is correct but unexpected if the driver has been follow-

ing this road for several kilometers already. False/expected

messages occur when the driver deviates from the route. The

message content does not conform to the driver’s intention,

but due to explicitly leaving the route such messages can

be anticipated by the driver, e.g., repeating “turn around”

messages. The instruction to turn around while driving on a

highway falls in the category of false/unexpected messages.

5. RESULTS

In the following, we present the results of the performed

study and analyze the behavior of the tested devices for

each driving scenario.

In total, 162 kilometers where driven in 2:06 hours. In this

time, 189 acoustic messages and 191 visual messages have

been observed. Due to GPS radio reception problems, the

Tomtom device had to be replaced to the side window, where

it could not be captured by the camera anymore. However,

the behavior of the Tomtom device was still observed and

recorded manually.

The cumulative observation results over all devices and the

first four driving scenarios are given in Fig. 4. Unexpected

messages account only for 3% of all messages. The rea-

sons for their occurrence are GPS radio reception problems.

Thus, these messages are indeed dangerous but relatively

rare. False/expected messages account for 41% of all ob-

served messages, while merely 56% of all messages were cor-

rect/expected. While the extreme nature of the chosen driv-

ing scenarios degraded these values compared to a normal

drive, it shows that PNDs react inflexible to changing driver

intentions. Interestingly, all devices behaved quite similar

irrespective of their price segment.

A detailed analysis of each driving scenario is provided in

the following, highlighting typical reactions and behavior.

The cumulative results for each scenario are summarized in

Fig. 5.

5.1 Detour (city)

The 16 km long scenario was completed in 24 minutes. As

expected, all devices called on the driver to turn around,

as soon as the driver deviated from the planned route. The

Falk called on the driver to turn around seven times (Becker

five times, Pearl three times). As acoustic messages, Pearl

and Becker used phrases like“turn left, then turn left”. Only

Falk used the words “please turn around”.

With 41 acoustic messages, the Pearl was most active in

this scenario, while the Becker acted very passive, with only

six acoustic messages. Falk and Pearl produced nine false

messages each, Becker only produced seven. Messages were

false in the sense that they did not correspond to the driver’s

intention. Falk and Pearl recalculated their route nine times

(indicated by visual messages). In case of the Pearl two of

the recalculations resulted from GPS radio reception prob-

lems. The Becker performed six recalculations.

Fig. 5(a) gives the cumulative results for this scenario. The

high percentage of correct/expected messages (57%) results

from the fact that all devices stopped issuing turn around

messages and reverted to an alternative route R0

cas soon as

that route became shorter than he orignal route Rc. Subse-

quently, mainly correct/expected messages were issued, be-

cause R0

ccorresponded to Rd. Thus, the 40% false/expected

messages occurred almost completely in the first minutes of

this scenario.

5.2 Highway

In this scenario, we analysed the devices’ behaviour during a

highway trip, where the driver followed the calculated route,

i.e., the driven route Rdmatched the calculated route Rc.

As expected, all navigation systems remained passive and

calm during the whole trip of 123 kilometers (76 minutes).

The average number of visual messages was 10.6, from which

74.4% have been correct (see Fig. 5(b)). There have been

an average of 4.3 acoustic messages during the scenario. At

one point, shortly before the end of the scenario, the calcu-

lated route Rcof two PNDs differed from the others and the

intended route Rd.

To simulate the refueling at a petrol station, the highway

was left once. Due to the fact that the gas station was lo-

cated near the shoulder of the highway, none of the devices

corrected the displayed position but continued showing the

car on the highway. When the trip was resumed, one de-

vice temporarily displayed an erroneous position and orien-

tation for the car. Because the device believed the vehicle

to be driving in the opposite direction, the driver was in-

structed to leave the highway at the next exit to turn around

(false/unexpected message).

5.3 Inner city

During the inner city scenario (17 km, 19 minutes), reli-

able routing by all devices was observed. 92.5% of visual

messages and all of the acoustic messages were correct (see

Fig. 5(c)). The precise acoustic messages provided by the

Falk device near intersections are notable. Announcements

included the indication of direction and street names. For

example, “In 500 meters, leave the highway towards B13

Munich, therefore keep right” in contrast to instructions like

“After 800 meters, keep right, then keep left” issued by other

devices. After reaching the destination, which all devices

announced acoustically, all devices suspended routing and

switched into free drive mode to support the driver in the

search for parking.

false correct

expected

unexpected

40% 57%

1%2%

(a) Detour (city)

false correct

expected

unexpected

13% 83%

1%1%

(b) Highway

false correct

expected

unexpected

95%

0% 1%

false correct

expected

unexpected 0% 0%

93% 7%

(d) Detour (rural)

Figure 5: Message Distribution for the different driving scenarios.

5.4 Detour (rural)

In this scenario, the driver deviated from the calculated

route Rconto rural roads. Roads with few intersections

were chosen to reduce potential alternative routes the de-

vices could change to. The detour extended over 6 km (7

minutes) before the driver turned around eventually. In this

time, 6.3 route recalculations were performed on average.

This corresponds to 1.16 recalculations per kilometer by

Becker and Pearl, and 0.83 recalculations by Falk. None

of the devices explicitly informed the driver about the rea-

son of recalculation. Instead, turn around commands were

repeatedly issued.

In combination with the large number of messages (one

acoustic message every 37 seconds, one visual message ev-

ery 57 seconds, on average) the behavior suggests, that the

driver is rather stressed than supported by the devices.

5.5 Dynamic traffic warning

The integration of dynamic traffic warnings by the PNDs

has been analyzed independently of the other four scenar-

ios. We analyzed the behavior of devices in situations where

warnings about traffic obstructions on the planned route Rc

appeared. Hereby, the focus was placed on the information

the devices provide about the obstruction and how it is con-

veyed, as well as the decision support provided to the driver.

Due to different information sources for the dynamic traf-

fic warnings, the listing of traffic information varied between

devices, e.g., between Becker (TMC) and Garmin (n¨

uLink!).

Additionally, TMC messages can also differ strongly between

devices, because, in Germany, radio broadcasting corpora-

tions are responsible for creating and updating traffic infor-

mation. Each broadcasting corporation relies on it’s own

mix of different information sources, e.g., listeners, automo-

bile associations like the German ADAC6, or police notifi-

cations.

If traffic warnings existed, the driver could call up a list of

received warnings and their information on all devices. A

detailed view for each message provides the reason for the

obstruction (traffic jam,accident, etc.) and displays the ap-

proximate position of the obstruction. None of the tested

devices provided information about a message’s time of ori-

gin. When a dynamic traffic warning for an obstruction on

the planned route Rcis received, the tested devices call on

the the driver to decide if the traffic jam should be circum-

vented.

As an example, Fig. 6 show the dialog of the Becker device.

A traffic jam is depicted as a purple area on the planned

route (red). A symbol on the right depicts the type of ob-

struction. The navigation system provides only one alterna-

tive to avoid the obstruction. Only approximations of the

6ADAC – Allgemeiner Deutscher Automobil-Club e. V.:

http://www.adac.de

Figure 6: Recommended route to circumvent the

traffic jam.

Figure 7: Display of an obstruction of traffic and the

possibility to avoid this route.

saved time and the distance difference of the detour are pro-

vided as additional information. How the value of, in this

case, 28 minutes has been determined is not obvious to the

driver and not explained by the system. If additional in-

formation is available for the calculation of the saved time,

e.g., characteristics of the traffic jam, it is not displayed to

the driver. In the example, it is also not obvious and not

explained, why this exact route (yellow) is recommended

to avoid the traffic jam, while several other (presumably

shorter) routes are discernible on the map (see Fig. 6).

However, other devices, e.g., the Garmin, do not even dis-

play the alternative route to the driver. Fig. 7 shows that

the only information available to support the driver’s deci-

sion is the length of the traffic jam (0.1 km) and the cause

of the obstruction (construction site). Thereby, the degree

of obstructions is displayed in different colors, e.g., green

indicates a slight obstruction and red indicates a heavy ob-

struction.

In general, the tested devices provide not enough informa-

tion about the obstruction and possible alternatives to en-

able the driver to make an informed decision about staying

on the route or avoiding the obstruction. Unfortunately, the

tested devices did also not provide a recommendation for

the best choice.

6. DISCUSSION

In this Section, the results from the study are discussed and

interaction weaknesses are summarized.

Currently, a driver can only influence the route prior to a

trip by choosing between different route characteristics, e.g.,

shortest, fastest, or most economic route. Furthermore, the

driver can choose to avoid toll roads or highways. Based

on these adjustments, the navigation system calculates the

route. At this point no reasons are given why this particular

route is chosen, and the driver has no opportunity to com-

pare this route to other alternatives. A comparison maybe

performed internally by the system, but it is invisible to

the driver. Almost all devices instantly started the route

guiding. Only the Pearl displayed a general overview of

the planned route before commencing navigation. While in-

stantly switching to route guiding might save time, display-

ing a route overview can prevent mistakes due to wrongly

selected destinations.

The study showed that PNDs are reliable and perform well

in normal situations, when the calculated route Rcmatches

the intended route Rd. However, while all PNDs issued cor-

rect and expected messages for navigation commands, sev-

eral devices produced cryptic messages. Here, the top range

devices performed notably better.

When the driver deviates from the calculated route Rcon

purpose, the devices are not able to dynamically adapt to the

driver’s decision. The driver is not properly informed about

the deviation detected by the device, route recalculations

occur mostly silently and are only accompanied by short

visual messages (e.g. turning progress arrows). The navi-

gation systems instruct the driver to turn around but give

no indication why, i.e., to return to the calculated route Rc.

Interestingly, only one device actually used the vocal instruc-

tion “turn around”, while the others resorted to sequences of

turn commands (“turn left, then left”). The latter behavior

not only increases the number of messages but also does not

convey the intention of the device to the driver. Further-

more, the driver has no possibility to inform the navigation

system about its intention (deviation from the route) with-

out performing multiple clicks in touch menus to abort the

routing process.

The study showed further that in both scenarios where the

driver left the route intentionally the navigation systems

produced 56% false acoustic and 65% false visual messages

in total, respectively. While those messages can be expected

by the driver, because they intentionally left the route, the

observed frequency and persistence of such messages sug-

gests increased stress for the driver in already stressful sit-

uations. In contrast, the numbers of correct messages were

considerable higher in scenarios, where the driver followed

the calculated route (93% correct acoustic and 88% correct

visual messages).

Another critical point is the interaction concerning dynamic

traffic warnings. While tested devices that support the re-

ception of dynamic traffic warnings notified the driver about

obstructions on the planned route, the provided information

is often not sufficient for the driver to make an informed deci-

sion if the obstruction should be avoided or not. The driver’s

options for interaction are also very limited. The driver can

either accept the calculated detour (which is not displayed

by all devices) or continue on the current route. It remains

to be analyzed what information needs to be displayed to

actually support the driver in this decision and to improve

system credibility in such scenarios.

In summary, major interaction weaknesses could be identi-

fied for the route selection, the purposeful deviation from

routes, and the integration of dynamic traffic warnings. Al-

though the study included devices of different price ranges,

the issues were prevalent throughout all devices with minor

variations.

We conclude that in order to improve credibility, the inter-

action between the driver and the system must be improved.

The driver should be able to comprehend system decisions,

e.g., why a specific alternative route has been selected and

not another. At the same time, devices should enable the

driver to interact with them in order to convey changes in

driving intention. With improved bidirectional information

interchange, the amount of false messages could be poten-

tially reduced and the devices could better adapt to drivers.

7. CONCLUSION

In this work, we studied the interaction behavior of several

current personal navigation devices. We designed driving

scenarios to analyze the behavior of PNDs from different

categories in specific situations. All devices have been tested

in parallel to ensure the same conditions.

The results of the study support our hypothesis that PNDs

perform well as long as drivers follow their directions, but fail

in deviating situations. As soon as the driver leaves the cal-

culated route, all devices show interaction weaknesses. They

issue cryptic commands (“turn left, then left”) repeatedly,

without communicating the actual problem to the driver,

e.g., that the driver left the calculated route. While this be-

havior is desired to some extent when the driver leaves the

route by accident, it becomes an annoyance in situations

where the driver acts on purpose. Current PNDs are not

flexible enough to cater for both situations. We argue that

the root cause of these problems is the lack of interactivity

with the driver en route. While vendors constantly work to

improve routing functionality, as well as output mechanisms

(e.g., text to speech), what is missing in today’s devices

are information exchange loops between driver and device.

The navigation system does not expose enough information

about its internal decisions. Instead, they should offer mean-

ingful information about commands and route alternatives

to empower drivers to understand why a certain route has

been recommended. Otherwise navigation systems risk the

loss of credibility with the driver.

At the same time, drivers have neither explicit nor implicit

means to communicate their dynamic intentions to the nav-

igation device.

They can only influence routing beforehand by specifying a

vague route preference, e.g. fastest route. We claim that

such static settings are not sufficient to capture a person’s

driving habits. Instead, navigation systems should utilize

available sensors to implicitly infer the driving behavior over

time and adapt accordingly. Furthermore, by offering mean-

ingful interaction opportunities and alternatives, navigation

systems could engage in a dialog with the driver, better

adapt to the driver, and maintain high credibility. However,

great care is required in interaction design to ensure that

the driver can concentrate on the primary task – operating

the vehicle.

We are currently developing and evaluating an interaction

model for navigation systems. Our model facilitates dy-

namic adaptation to driver intentions and addresses the is-

sues uncovered in this work by improving the credibility of

interaction processes.

8. ACKNOWLEDGMENTS

The authors would like to thank Bj¨

orn Wiedersheim for his

support during the test drive, the manufacturers and retail-

ers that kindly provided the tested devices, and the anony-

mous reviewers for their valuable comments. This work was

supported by Transregional Collaborative Research Centre

SFB/TRR 62 (“Companion-Technology for Cognitive Tech-

nical Systems”) funded by the German Research Foundation

(DFG).

9. REFERENCES

[1] R. Dechter and J. Pearl. Generalized best-first search

strategies and the optimality of A*. Journal of the

ACM, 32(3):505–536, 1985.

[2] E. W. Dijkstra. A note on two problems in connexion

with graphs. Numerische Mathematik, 1:269–271, 1959.

[3] B. J. Fogg and H. Tseng. The elements of computer

credibility. In CHI ’99: Proceedings of the SIGCHI

conference on Human factors in computing systems,

pages 80–87, New York, NY, USA, 1999. ACM.

[4] R. J. Hanowski, S. C. Kantowitz, and B. H. Kantowitz.

Driver acceptance of unreliable route guidance

information. In Proceedings of the Human Factors and

Ergonomics Society 38th Annual Meeting, volume 2 of

System Development, pages 1062–1066, 1994.

[5] A. Malm. Personal navigation devices (3rd edt). Market

report, Berg Insight, November 2009.

http://www.berginsight.com/ShowReport.aspx?id=92.

[6] The Local. Driver obeys car navigation literally,

promptly crashes.

http://www.thelocal.de/society/20100423-26745.html,

23 April 2010.

[7] The Times. Sat-nav dunks dozy drivers in deep water.

http://www.timesonline.co.uk/tol/news/article707216.ece,

20 April 2006.