A Study of the Effect of Looming Intensity Rumble Strip Warnings in Lane Departure Scenarios
DAVID SANDBERG
KTH ROYAL INSTITUTE OF TECHNOLOGY
A Study of the Effect of Looming Intensity Rumble Strip Warnings in Lane Departure
Scenarios
DAVID SANDBERG dsandb@kth.se
Master’s Thesis in Computer Science
School of Computer Science and Communication (CSC) Royal Institute of Technology, Stockholm
Supervisor at CSC: Sten Ternström Supervisor at Osaka University: Takao Onoye
Examiner: Jens Lagergren
November 12, 2015
Abstract
In lane departure warning systems (LDWS) it is impor- tant that the auditory warning triggers a fast and appro- priate reaction from the driver. The rumble strip noise is a suitable warning to alert the driver of an imminent lane departure. A short reaction time is important in lane de- parture scenarios, where a late response may have fatal con- sequences. For abstract sounds an increase in intensity can influence the perceived urgency level of the warning, which may also trigger a faster reaction from the listener. In this thesis, the effect of a rumble strip warning with looming (increasing) intensity was analyzed by letting test persons drive a driving simulator and measuring how quickly they reacted to the auditory warning. These results were com- pared with those for a rumble strip warning with a constant intensity, and two versions of an abstract warning; constant intensity and looming intensity. A survey regarding the per- ceived urgency, annoyance and acceptance of the warnings was also carried out.
The results show no differences in reaction time be- tween the four warning signals. This may be because the test persons expected the warnings, or because of their lim- ited experience. The survey suggests that adding a looming intensity to the rumble strip warning results in a higher ur- gency, while keeping the annoyance low, which could be of importance to avoid unwanted reactions from the driver.
En studie av effekten av bullerräffleljud med ökande intensitet vid ofrivilligt lämnande av
körfältet
I varningssystem för personbilar används ofta ett system som signalerar ett stundande ofrivilligt lämnande av körfäl- tet, s.k. lane departure warning systems (LDWS), genom att en varningssignal ljuder. Det är viktigt att en sådan akustisk varningssignal frammanar en snabb och lämplig reaktion från föraren. Ljudet av en bullerräffla är en lämp- lig varningssignal för detta ändamål. En kort reaktionstid är viktig när fordon är på väg att ofrivilligt lämna körfäl- tet, då en långsam reaktion kan ha förödande konsekven- ser. Studier på abstrakta akustiska varningssignaler har vi- sat att en ökande intensitet kan få en varning att verka mer brådskande, vilket i sin tur kan leda till att lyssna- ren reagerar snabbare. I denna rapport analyseras hur ett bullerräffleljuds ökande intensitet påverkar förarens reak- tionstid. Analysen gjordes genom att mäta reaktionstiden hos testpersoner som körde en bilsimulator med fyra oli- ka varningssignaler; en bullerräffleljudsvarning och en ab- strakt varning, båda med konstant intensitet och ökande intensitet. Reaktionstiderna för de olika signalerna jämför- des, varpå en enkät utfärdades där testpersonerna uppgav hur brådskande och irriterande de uppfattade varningarna, samt till vilken grad de skulle acceptera varningarna i ett verkligt körscenario.
Resultaten visar inga skillnader i reaktionstid mellan varningarna, vilket kan bero på att testpersonerna förut- såg när varningarna skulle komma, eller på grund av deras begränsade erfarenhet av bullerräffleljud. Enkätens utfall antyder att bullerräffleljudsvarningen med ökande intensi- tet är mer brådskande än versionen med konstant intensi- tet, men att irritationsnivån inte påverkas när intensiteten ökar, vilket kan vara viktigt för att inte framkalla oönskade reaktioner hos föraren.
Acknowledgements
Thank you to my supervisor at KTH, Prof. Sten Ternström, for continuous support and feedback throughout the research process.
Thank you to my supervisor at Osaka University, Prof. Takao Onoye, for giving me the opportunity to do the research in his lab and providing the necessary equipment and guidance.
Thank you Dr. Wataru Kobayashi of Arnis Sound Technologies for suggesting the field of this research and providing suggestions for the warning signal designs.
Thank you Prof. Jens Lagergren for examining the thesis.
Thank you to fellow students Andreas Wedenborn, Dennis Johansson, Christoffer Carlsson and Jens Eriksson for reading and commenting on the thesis throughout the research process.
Thank you Ann Bengtsson for helping me find a supervisor and examiner on short notice.
Thank you Dr. Johan Fagerlönn for answering my questions and providing feedback regarding the auditory warning signal design.
Thank you to all the test persons who participated in the evaluation process.
Thank you to the Scandinavia-Japan Sasakawa Foundation for providing funding to help make this research possible.
Thank you to my family for always supporting me.
1 Introduction 1
1.1 Background . . . 1
1.2 What makes an auditory warning? . . . 2
1.3 Non-speech auditory warnings . . . 3
1.3.1 Auditory icons . . . 4
1.3.2 Earcons . . . 6
1.4 Speech warnings . . . 6
1.5 Urgency and annoyance . . . 7
1.6 Related work . . . 7
1.6.1 Ziegler et al. (1995) . . . 7
1.6.2 Haas & Edworthy (1996) . . . 7
1.6.3 Suzuki & Jansson (2003) . . . 8
1.6.4 Gray (2011) . . . 8
1.7 This thesis . . . 10
2 Method 11 2.1 Driving simulator design . . . 11
2.1.1 Game engine . . . 12
2.1.2 Car model . . . 12
2.1.3 Road and surroundings . . . 12
2.1.4 Triggers . . . 15
2.2 Warning signal design . . . 15
2.2.1 Rumble strip warning . . . 15
2.2.2 Abstract warning . . . 16
2.2.3 Choosing intensity levels . . . 17
2.2.4 Implementing the looming . . . 18
2.3 Evaluation . . . 18
2.3.1 Test persons . . . 19
2.3.2 Setup . . . 19
2.3.3 Secondary task . . . 20
2.3.4 Driving test . . . 21
3 Results 23
3.1 Driving test . . . 23
3.1.1 Reaction times . . . 23
3.2 Survey . . . 24
3.3 Test persons’ opinions . . . 26
4 Conclusion 27 4.1 Discussion . . . 27
4.1.1 Reaction times . . . 27
4.1.2 Survey . . . 28
4.2 Limitations . . . 28
4.3 Future work . . . 29
Bibliography 31
Chapter 1
Introduction
In the following section the background to this thesis will be presented and impor- tant theories in the field of auditory warnings will be explained. Lastly, related work that has been the foundation for, or in other ways inspired, the current research will be discussed.
1.1 Background
In 2009, 34,500 people were killed in road traffic accidents in the EU. Although the number of road casualties was twice as high ten years earlier, it was still an indication that a greater effort was needed in order to increase road safety. Therefore, in 2011 the European Commission set a goal to reduce the number of casualties by half by 2020, and to have zero fatalities by 2050 (European Commission, 2011). Measures of reaching this goal include educating drivers on the importance of using the provided safety equipment, for example seat-belts, as well as developing a safer infrastructure.
Furthermore, it is also important to make use of current road safety technology to battle this problem. Implementing Advanced Driver Assistance Systems (ADAS) in cars is one way of making use of such technology. ADASs are systems that assist the driver in various ways, and can involve systems handling features such as cruise-control, collision detection and lane departure warnings. At present, there are no strict guidelines as to what should be included in an ADAS, and whether or not to implement an ADAS in a vehicle is up to the manufacturer to decide.
Consequently, the warning signals for collision detection and lane departure can also differ between different ADASs. However, it is common to use some sort of auditory warning signals to catch the driver’s attention. The question is what kind of auditory warning signal is the most appropriate for a certain situation.
According to a report published by The American Association of State Highway and Transportation Officials (2008), close to 60% of all fatal car accidents in the U.S. were due to the car leaving its lane (i.e. a lane departure). Another analysis of data from the Swedish Traffic Accident Data Acquisition (STRADA) system on heavy trucks in Sweden between 2003 and 2008 revealed that lane departure was
the cause in 40% of the cases where the driver was killed or seriously injured (Volvo Trucks, 2013). These figures suggest that a large amount of lives can be saved if the amount of lane departure accidents is reduced. One common way of trying to solve this problem is through implementing Lane Departure Warning Systems (LDWS) into cars. The LDWS will assist the driver by triggering a warning signal if the car gets too close to the lane boundary, or crosses it. Most research has focused on auditory warning signals, but there are also studies comparing auditory and haptic warning signals (e.g. Ziegler et al., 1995; Stanley, 2006). One reason that auditory warning signals are widely used is that they catch our attention immediately even if we are not looking at the source of the sound, as compared with visual warnings that require us to already be focusing on the source in order to alert us (Patterson, 1989).
One commonly used auditory warning signal is the “rumble strip noise”, which imitates the sound that is produced as the wheels of a car hits the rumble strip that is often present at the edge of a road. Research has shown that the rumble strip noise can be effective in alerting a driver of an imminent lane departure (Ziegler et al., 1995; Fagerlönn, 2011a). These studies were performed on professional truck drivers with years of driving experience. However, the perceived urgency of a sound does not only depend on its acoustical characteristics, but also on the listener’s previous experience of that sound (Västfjäll et al., 2006). It can therefore be interesting to study if less experienced drivers perceive the rumble strip noise in the same way as experienced drivers. Furthermore, most research evaluating auditory warning signals in cars have used signals that have constant acoustical parameters. In other words, parameters such as pitch and intensity do not change over time. However, a study on auditory warning signals for frontal collision detection showed that a signal with a looming (increasing) intensity can be more effective in reducing a driver’s reaction time (Gray, 2011). Applying this idea to lane departure warnings could therefore be of interest.
This thesis presents an implementation of a looming intensity rumble strip warn- ing signal. The signal will be evaluated in lane departure situations by letting test persons, with limited driving experience, drive a driving simulator and measuring their reaction time (RT) once a warning signal has been triggered. A survey will also be conducted to find out how urgent, annoying and accepted the signal is in a lane departure scenario.
1.2 What makes an auditory warning?
When designing auditory warnings there are many things to keep in mind. If the warning system is to be used in an environment which requires multiple warnings, such as hospitals or flight decks, one of the challenges lies in designing signals that have different acoustic characteristics, in order to avoid confusion. Another problem is masking. Masking happens when one sound interferes with another sound of the same, or a very similar, frequency (Ulfvengren, 2003b). For example, if you are talking to a friend at a sound level of 65 dB and someone turns on the vacuum
1.3. NON-SPEECH AUDITORY WARNINGS
cleaner at 75 dB, then the speech will be masked by the vacuum cleaner, and you would have to raise your voices above 75 dB to hear each other. The level difference between the two signals is known as the masked threshold. If the sound levels are the same, the signals will merge and be difficult to tell apart. Furthermore, if the warning signal is too loud it may cause a startle reaction which can result in the concentration level dropping. It may also make any necessary communication more difficult (Edworthy et al., 1994). Another problem that is often found in warning systems is that the perceived (psychoacoustic) urgency of a warning is not properly matched with the urgency of the situation (Edworthy et al., 1991). Consider two warnings with different perceived urgency; high and low. Let us assume that there is a mismatch between the two warnings’ urgency levels and the situations they portray; the “high” warning is mapped to a low urgency situation, and vice versa.
If the two warnings are being sounded at the same time, the less urgent situation may be attended to first, which may have unwanted consequences.
Edworthy et al. (1994) observed two main requirements that need to be fulfilled in order to create an auditory warning signal. Firstly, the warning has to have an appropriate intensity level; not too loud, but still loud enough to catch the listeners attention. If the intensity of a warning rises too quickly, it will startle the listener who in turn might respond instantaneously, and out of reflex. These types of responses are often not the most appropriate in a given situation (Patterson, 1989).
Secondly, the warning should in some way be properly related to the situation that it is portraying. For example, the psychoacoustic urgency of a warning should match the urgency of the situational urgency, or have a natural connection to the desired action (e.g. screeching tires to tell a driver to hit the brakes). Furthermore, a warning must also invoke a quick and accurate response (Graham, 1999).
Västfjäll et al. (2006) claimed that the bond between a sound and a person’s previous experiences of that sound is important. When we first hear a sound, the acoustical characteristics (pitch, intensity etc.) are analyzed to see if it should be interpreted as a warning. If a certain arousal threshold is exceeded, then we respond to it. However, if the “arousal potential” is too low we try to find a connection to the sound in our bank of memories. If a match is found that has a dangerous or negative connotation, then we respond. Otherwise, the sound will not be be regarded as a warning, and no response is necessary. This suggests that warnings may be more effective if they not only trigger an arousal, but that they also have a connection to a person’s past experiences.
1.3 Non-speech auditory warnings
A method to solve the problems mentioned in the previous section was found in Patterson’s guidelines to designing auditory warnings in 1982, and consisted of four main steps; deciding the appropriate loudness level of the warning (15-25 dB over the masked threshold), creating a short sound pulse (100-300 ms long), creating a burst consisting of several pulses of various frequencies and pauses, and finally
adding several bursts together with a silence in between to form the final warning.
It was also suggested that each pulse should have an onset and offset of 20 ms to avoid startling the listener. (Patterson (1982) cited in Edworthy et al., 1991). These guidelines proposed a new way of designing auditory warnings, where the perceived urgency level could be altered to better fit the situational urgency. The masking problem has also been addressed by Patterson (1989), who claims that if the sound consists of four or more harmonics it is less likely to be masked than one that has all the energy focused on just one harmonic.
Using Patterson’s proposed guidelines, Edworthy et al. (1991) showed in more detail how pulse and burst parameters should be altered to convey different levels of urgency. The acoustic parameters studied were fundamental frequency, harmonic regularity, amplitude envelope and delayed harmonics. These were part of the design of the pulse. In a regular harmonic series, every harmonic is an even integer multiple of the fundamental frequency. A delayed harmonic is a harmonic whose onset is delayed compared with the fundamental frequency. It was discovered that fundamental frequency had a smaller impact on urgency than expected, and that irregular harmonics and a standard onset (20 ms) were the two acoustic parameters that were most important to convey a high level of urgency. Among the burst parameters, it was found that the single most effective parameter was speed; the faster the burst, the more urgent the warning. By altering the pulse and burst parameters accordingly, it was possible to create auditory warnings with the desired urgency levels.
A few years later, research showed that the signals yielding the highest perceived urgency were those with a high speed, a high frequency and a high level of loudness (Haas and Edworthy, 1996). Furthermore, increasing pitch and loudness showed a decrease in reaction time.
The research mentioned above has focused only on variations of tones, which is part of the abstract sounds family. Abstract sounds have no obvious connection to what they are portraying, and consist of sounds such as bells, buzzers and sirens.
Even though they can be learnt over time, “they may not be well suited to emergency warning situations, which in certain applications occur only very rarely” (Graham, 1999, p. 1234). It has also been shown that abstract sounds may produce slower reaction times than other types of sound (McKeown, 2005). A different type of sounds that has a more natural connection to the situation it is portraying, is known as auditory icons.
1.3.1 Auditory icons
The concept of auditory icons was introduced by Gaver in 1989 who defined them as
“everyday sounds meant to convey information about computer events by analogy with everyday events” (p. 67). The main idea of auditory icons is to take everyday sounds and apply them to related events within another field, for example by using the sound of a piece of paper being crumpled when emptying the trashcan in a computer. By using sounds that have connections with our everyday lives, it is
1.3. NON-SPEECH AUDITORY WARNINGS
easier to convey the meaning of the sounds in the new context. This may be because we tend to hear the source of the sound rather than the sound itself. For example, as a door shuts, we hear its size and with what force it shuts, and not the pitch, loudness or other acoustical parameters of the sound.
Ulfvengren (2003a, cited in Ulfvengren, 2003b) introduced the concept of “asso- ciability” as “the required effort to associate sounds to their assigned alert function meaning” (p. 53). Furthermore, she states that the more “associable” a sound is, the fewer cognitive resources are needed, which would make it appropriate for au- ditory warnings. Other benefits with sounds with a high associability are that they are easy to remember and discern from other sounds. Ulfvengren carried out tests with various types of auditory warnings to find out which sounds were the easiest to remember. It was shown that auditory icons were better than all other sounds, including animal sounds and abstract sounds. These results suggests that audi- tory icons has the highest associability, and may therefore be suitable for auditory warnings.
Graham (1999) compared two auditory icons (tyre-skid sound and car-horn sound), a verbal warning (’ahead!’) and a beep sound in a car driving environ- ment. The warnings were triggered when a collision was imminent. Examining the reaction times of the drivers showed that the auditory icons produced significantly faster reaction times than the other warnings. However, it was also shown that the auditory icons produced a higher rate of inappropriate responses, such as making the driver hit the brakes even in a non-collision situation (false-positives). A sur- vey carried out among the test participants revealed that the car-horn sound was considered to be very appropriate for all collision situations (more than the beep sound), while the tyre-skid sound was much less appropriate. This suggests that although auditory icons are related to everyday events, they may have several possi- ble interpretations, some being less appropriate for a certain situation than others.
Consequently, they may also be misinterpreted.
Similar to Graham’s results, research by McKeown (2005) showed that auditory icons resulted in faster reaction times than abstract sounds when used in within- vehicle scenarios. This strengthens the claim that auditory icons are suitable as warnings when a quick response is needed. In the tests, warnings were mapped to driving scenarios of various urgency. It was observed that all warnings except speech were perceived as less pleasant the more urgent the scenario was, revealing a connection between annoyance and urgency. This suggests that the annoyance level of a warning may depend on its learnt meaning, more than its acoustic charac- teristics. Furthermore, McKeown points out that it may be better to use auditory icons with different learned urgency, than to manipulate one sound acoustically to make it sound more or less urgent. This is because such manipulation may affect the recognizability of the auditory icon.
1.3.2 Earcons
The concept of earcons was introduced by Blattner, Sumikawa and Greenberg (1989), who described them as “nonverbal audio messages used in the user-computer interface to provide information to the user about some computer object, operation, or interaction” (p. 13). The earcons were divided into two types; representational earcons and abstract earcons. The former is just another name for the auditory icons described in the previous section, while abstract earcons are “abstract, syn- thetic tones that can be used in structured combinations to create sound messages to represent parts of an interface” (Brewster, 1994, p. 7). Parameters such as pitch, timbre, and dynamics may be altered to form a specific earcon. While auditory icons sound like what they represent, the meaning of an abstract earcon is not in- tuitive, and has to be learned. From here on, the word earcon will be used for abstract earcons. An example of a usage of an earcon may be to indicate where on the computer screen a message is displayed. An earcon with a low pitch could indicate a message appearing at the bottom of the screen, while an earcon with a higher pitch could indicate a message appearing at the top of the screen.
Lucas (1995) studied how well auditory icons and earcons described actions and objects in a human-computer interface. Test persons were asked to connect each sound to the action or object that they thought the sound best represented. It was shown that it was more difficult to connect accurately the earcons to the correct action or object, than for the auditory icons. However, Lucas points out that the limited number of sounds used in the test may have impacted the result to a higher degree than the audio cue design method.
1.4 Speech warnings
In the research by McKeown (2005) mentioned in section 1.3.1, it was shown that speech sounds could accurately be mapped to the situations they were supposed to portray. However, the urgency of the speech sounds were constantly rated as inter- mediate, regardless of the situational urgency. On the other hand, earlier research had shown that speech warnings could in fact produce various levels of urgency, although the range of potential urgencies were shown to be wider for nonspeech warnings (Edworthy, et al., 2000).
Furthermore, other research suggests that the reaction times for speech can be slower than those for both abstract sounds and auditory icons (Graham, 1999).
Graham mentions that the reason is that even a short spoken warning will take a longer time to interpret, and that it is not likely that the reaction time would be improved by changing the warning’s parameters. However, recent research has shown that the accuracy and reaction time for short speech warnings and auditory icons may be similar (Fagerlönn & Alm, 2010). Still, it is mentioned that speech can be affected by background noise or other people talking, which may cause problems in a real driving scenario.
1.5. URGENCY AND ANNOYANCE
1.5 Urgency and annoyance
Although the urgency of a warning signal is an important parameter when it comes to catching a person’s attention, the perceived annoyance level must also be con- sidered. The urgency may have an effect on a driver’s ability to quickly recognize and react to a warning, but the annoyance can have a negative influence on the driver, causing the warning to be disabled or ignored (Marshall et al., 2007). In other words, if the warning is too annoying it may be turned off, which defeats the whole purpose of using a warning system. Furthermore, if a driver gets angry, it may affect the driving performance and result in more traffic violations (King and Parker, 2008). Tan and Lerner (1995) showed that there is a correlation between a warning’s perceived urgency level and its annoyance level, while other research has suggested that certain acoustical parameters have a stronger effect on urgency than others (Marshall et al., 2007). It is therefore of importance to evaluate both the urgency and annoyance when designing auditory warnings.
1.6 Related work
Several important sources have already been described in the previous sections.
However, below is some of the research that has been most significant to the decisions made regarding the evaluation methods and warning signal designs in this thesis.
1.6.1 Ziegler et al. (1995)
Ziegler et al. let 18 professional truck drivers drive a driving simulator for 30 minutes. Upon an imminent lane departure a warning signal was triggered. The warning signals consisted of one haptic signal with a slightly oscillating steering wheel, one haptic warning with a torque that turned the steering wheel towards the middle of the lane, and one auditory icon (rumble strip noise). After the test drive, interviews showed that the highest acceptance was achieved by the rumble strip warning, and that the majority of drivers said that it was well suited for the lane departure scenario. Further tests measuring reaction times showed that the rumble strip warning had a mean reaction time of 0.5 seconds, which was considered as being “very fast”.
1.6.2 Haas & Edworthy (1996)
Haas and Edworthy carried out research to find out what acoustic parameters make a sound more urgent. They also investigated how the perceived urgency was related to reaction time. The fundamental frequency of a pulse, the inter-pulse interval (time between pulses), and the pulse level (intensity) were the independent variables.
The evaluation was made by letting 30 college students listen to 27 auditory signals.
They were asked to subjectively rate each signal’s urgency, and were then asked to
press a button as soon as they heared the signal, in order to measure the reaction time for each signal. It was concluded that the most urgent signals also had the shortest reaction times. These signals shared the same characteristics; they had a high frequency, a fast speed, and were loud.
Haas and Edworthy suggested that an auditory signal with a fundamental fre- quency of at least 500 Hz, a loudness of between 15 and 30 dB above ambient, and with a 0 ms inter-pulse interval was a suitable signal for designers who wish to achieve the highest level of perceived urgency and the shortest reaction times.
In this thesis, these guidelines were followed when designing the abstract warning signal.
1.6.3 Suzuki & Jansson (2003)
Suzuki and Jansson investigated the effect of four different warning methods on drivers’ reaction times in lane departure scenarios. The warning methods used were monaural beeps (sound played from both speakers), stereo beeps (sound played from the speaker on the side of the lane departure), steering vibration, and pulse- like steering torque. The tests were carried out with 24 experienced drivers, reported to drive at least 5000 km/year. In the tests where the subjects were aware that the warning signified a lane departure, both the monaural and stereo beeps reduced the reaction times. However, there were no pairwise differences between the monaural and stereo beeps, and it was observed that the drivers looked at the road before reacting in both cases. In the tests where the drivers were unaware of the meaning of the warning, the steering vibration showed a significant decrease in reaction times.
In order to trigger a lane departure, the subjects were asked to perform a sec- ondary task shown on a separate display mounted on the passenger seat. On the separate display, five random numbers were shown, and the subjects had to read the numbers out loud. While the subjects were engaged in the secondary task, the yaw angle of the vehicle was changed randomly to the left or right by two degrees.
In this thesis, a similar secondary task and yaw angle change were implemented, as described in more detail in Chapter 2: Method. The subjects were asked if they drove more than 5000 km/year, which is the same threshold that Suzuki and Jansson used. The road and shoulder widths used in this thesis also correspond to those used by Suzuki and Jansson, at 3.5 m and 1.0 m respectively.
1.6.4 Gray (2011)
Gray investigated what auditory warning signals produced the fastest reaction times in a collision scenario. 20 test persons drove behind a lead car in a driving simulator and were asked to react to an auditory warning signal when the lead car slowed down. In total seven signals were tested; four nonlooming signals and three looming signals. The nonlooming signals were constant intensity, ramped, pulsed and a car horn sound. The looming signals, which had rising intensities to simulate that the lead car was getting closer, were veridical (actual), early or late. The early signal
1.6. RELATED WORK
had a steep time-intensity curve to suggest that the time to collision (TTC) with the lead car was shorter than it actually was, while the late signal had a flatter time- intensity curve which suggested that the lead car was further away than it was. The veridical signal had a time-intensity curve that matched the TTC, meaning that the maximum intensity would be reached at the same time as a collision with the lead car would occur. All signals except the car horn had a frequency of 2000 Hz. The constant intensity signal had an intensity of 75 dB, the ramped signal increased linearly from 60 dB to 85 dB, while the pulsed signal started at 0 dB and reached a maximum of 75 dB for each pulse. The intensity of the looming warnings varied as shown in Figure 1.1.
Figure 1.1: Time-intensity profiles for veridical looming warnings triggered at different speeds (from Gray, 2011, used with permission).
The tests showed that the veridical looming warning signal produced, on aver- age, 77 to 115 ms faster brake reaction times (BRT) than the abstract warnings.
Although the car horn had BRTs similar to the veridical looming warning, it also produced more brake responses in the situations when the warning was triggered as a false alarm. It was also observed that the BRT was faster for the early looming warning than for the late looming warning. The results suggest that it is possible to influence how fast a driver reacts in a collision situation by altering the TTC of a looming warning.
The time-intensity curve used for the looming warnings in this thesis was created to be an approximation of the curves used by Gray, but with different onset and offset intensities, and with a shorter duration. Furthermore, Gray’s idea of telling the driver how far away a potential danger is by altering the intensity has been an
inspiration for this thesis.
1.7 This thesis
The primary goal of this thesis is to see if a looming rumble strip warning will produce faster reaction times than warning signals with constant intensity in a lane departure scenario. This is done by first building a simple driving simulator used in the driving tests.
Two types of auditory warning signals will be used; one rumble strip warning and one abstract warning. Each warning will have two variations; one version with constant intensity, and one version with looming intensity. The warnings will be designed in Matlab and automatically triggered from the driving simulator.
Test persons will participate in the evaluation which involves driving the car along the road. At certain points where the car is driving on a straight section, a secondary task will be triggered in order to force a lane departure. While the test persons are busy with the secondary task, a slight shift in direction of heading will be introduced. This will prevent the drivers from knowing on which side of the road the lane departure will occur. As a lane departure occurs, a warning signal will be triggered, and the test persons have to respond to the warning by steering back onto the road. The elapsed time from the warning onset to the time of the reaction (when the driver turns the steering wheel) will be recorded. This experiment will be performed for each of the four different warning signals, and the results will then be evaluated for statistical significance.
A second goal is to determine the perceived urgency, annoyance and acceptance for each warning signal. This will be done through a survey where the test persons rate the warnings on a scale from 1 to 5 for each category. It is important to know if a driver would accept a certain warning in the lane departure scenario, since the warning may otherwise be ignored or have unwanted effects on the driver.
The hypothesis is that both variations of the rumble strip warning (constant intensity and looming) will result in faster reaction times than their abstract warning counterparts. Furthermore, the looming rumble strip warning is expected to show faster reaction times than the constant intensity variant, due to the increasing intensity. Since previous research has shown a high acceptance for the rumble strip warning, similar results can be expected from the survey in this thesis.
Chapter 2
Method
The method can be divided into four main parts; designing the driving simulator, designing the auditory warning signals, evaluating the reaction times, and evaluat- ing the perceived urgency, annoyance and acceptance. Each part will be described in detail in the following sections.
2.1 Driving simulator design
In order to get the best results from tests in the field of collision and lane departure warnings, it is important to perform the tests in an environment that is as close to a real driving experience as possible. Since performing the tests in a real driving scenario would be dangerous and unethical, most research has used top of the line driving simulators consisting of a vehicle cabinet that the test person enters, where several displays present the driving environment to the driver. The driving physics are also designed to properly simulate the behavior of a real vehicle. Although a similar simulator environment would have been the most suitable for the current research, due to time and budget constraints it was not a feasible solution.
The modification of an existing open-source driving game such as Speed Dreams (2014) was also considered. However, this idea was abandoned since it was not certain that triggering warning signals, saving reaction time data to file, and other necessary functionality would be available without extensive modifications. Instead, in order to fit the purpose of this research and to have total control over the test en- vironment, it was decided to build the the driving simulator from scratch. Although the main goal of this research was to evaluate the various warning signals, since a large portion of the research was spent on developing the driving simulator, some details of the development process and how the driving simulator works should also be mentioned.
2.1.1 Game engine
The driving simulator was built in the game engine Unity (Unity Technologies, 2014), which is one of the most popular game engines used by game developers today. The decision to use Unity was mainly due to the author’s previous experience with the game engine, and also because Unity has a large community of users which has provided assets such as road building tools. Furthermore, the size of the community suggested that many of the potential problems involved in building a driving simulator had most likely already been faced and discussed by other users in the community online forums. The API is well documented and most functionality required for designing a simple driving simulator exists. The programming in Unity is done through writing scripts in either Javascript or C# and attaching them to the different game objects. In the current driving simulator the choice was to use C#.
2.1.2 Car model
The car model used in the driving simulator was called “Peugeot” and was included in the official Unity Car Tutorial package available from the Unity Asset Store.
The package included scripts for the car’s behavior. The scripts included variables such as damping, grip, friction and braking torque, all of whose values were kept as defaults. The size of the car was altered to fit the dimensions of a Peugeot 205 as follows; 3.705 m (length) x 1.572 m (width) x 1.4 m (height). The script controlling the car only supported digital steering, acceleration and braking. Therefore, in order to fit the steering wheel input that was going to be used during the tests, the script was edited to support analog input as well. The speed of the steering also had to be altered in order to make the car controllable. Although these alterations were made to the best of the author’s abilities, and thoroughly tested throughout the development process, it should be noted that they may not perfectly represent the properties of a real vehicle. However, it was estimated that the settings were appropriate enough to give the drivers good control over the vehicle, and also to measure their reaction times when a warning was triggered. Handling the car only involved acceleration and braking - no gear shift was implemented.
2.1.3 Road and surroundings
The road was built using the road building tool EasyRoads3D Free (AndaSoft, 2015), available from the Unity Asset Store. This tool provides the user with a graphical interface and the ability to create roads by placing markers on the desired points on the map. Once all the markers have been placed, the road will be drawn following the path of the markers. The road used in the current research was 9.85 km long, connected, and included five longer straights (see Figure 2.1). The straights would be suitable places to introduce the secondary task to the driver and thereby forcing a lane departure. The road consisted of a single 3.5 meter wide lane, with one 1.0 meter wide shoulder on each side. The 3.5 meter width of the lane is the
2.1. DRIVING SIMULATOR DESIGN
Figure 2.1: Overhead view of the course used in the driving tests
standard for Swedish motorways. Each side of the lane was marked with white lane markings. An overhead view of the road can be seen in Figure 2.2. The choice of using only a one-lane road was due to limitations in the EasyRoads3D Free tool.
At first, the plan was to use the car’s position relative to the road in order to determine when a lane departure had occurred. However, these coordinates could not be acquired since the position of a game object is relative to the whole game world. Consequently, the idea to trigger a lane departure warning based on the car coordinates was not feasible. However, the road building tool had a function to alter the width of the road. Therefore, once the road (R1) had been drawn (with a width of 5.5 m), an invisible copy of the road (R2) with a narrower width (3.5 m) was added on top of R1. Changing the width to 3.5 m meant that R2 became 1.0 m narrower on each side, compared with R1. A lane departure could therefore be registered as soon as the car had started to leave R2 on either side. This was done by using triggers, as described in the next subsection.
The surroundings were kept simple, without any buildings, trees, vehicles or other objects. This was mainly due to the limited time available, but also in order to prevent any latency that may appear due to heavy rendering. The first-person view that the test persons experienced while driving can be seen in Figure 2.3.
Figure 2.2: Close up of the car and the road
Figure 2.3: First-person view of the road and surroundings
2.2. WARNING SIGNAL DESIGN
2.1.4 Triggers
In Unity, a game object (x) can be set to be a trigger. This means that if another game object (y) enters the space where x is currently residing, instead of the two objects colliding with each other, y can enter x. When this happens, the OnTrig- gerEnter method in the script assigned to x is called. Any code that should be executed when another game object enters x can be put in this method. A similar method, OnTriggerExit, is called when an object leaves a trigger.
In the driving simulator, one trigger was attached to each side of the car, close to the front wheels. Both triggers were in contact with the road. Whenever one of the triggers left the R2 road, the OnTriggerExit method was called, in which the code to playback the warning signal was executed.
2.2 Warning signal design
The main purpose of this research was to study the effect of a looming intensity rumble strip warning on a driver’s reaction time. In order to evaluate the effect of the looming intensity, it was necessary to compare it with a rumble strip warning with constant intensity. Furthermore, it could be interesting to examine whether the looming intensity had an effect also on abstract warning signals. Therefore, two abstract warning signals were also developed; one with looming intensity and one with constant intensity. All the warning signals were developed in Matlab, and the looming was introduced using the sound editor software Audacity (Audacity Team, 2015). In the following sections, the warning signal design will be discussed in more detail.
2.2.1 Rumble strip warning
In order to create a synthetic warning signal resembling a real rumble strip noise as much as possible, a recording of a rumble strip noise from inside a car was used as a reference (U.S. Department of Transportation, 2011). Since the rumble strip noise is a low-frequency sound, the fundamental frequency of the warning signal was set to 90 Hz. The fundamental frequency is the lowest frequency present in a sound wave. Four additional harmonics were used to complete the sine wave in the rumble strip noise. Each harmonic is an integer multiple of the fundamental frequency.
The harmonics chosen were 180, 270, 360 and 450 Hz, and the amplitude ratios relative to the fundamental frequency were 1.14, 1.43, 1.14 and 0.71 respectively.
In order to create a rumbling sound, the Matlab Audio Toolkit (Smith, 2012) was downloaded, and the noisefilter function was used to create a white noise in the frequency range 150 to 500 Hz, with an amplitude ratio of 2.86 relative to the fundamental frequency. The frequency of a rumble strip noise is not constant.
Instead, the sound displays acoustic properties similar to that of a vibrato or a beating. A vibrato is a pulsating pitch change in a sound wave, revolving around the fundamental frequency. To simulate a vibrato, the vibrato function in the Audio
Figure 2.4: Frequency spectrum of the rumble strip warning signal
Toolkit was used. The sine wave was assigned a vibrato frequency of 10 Hz, and a depth of 0.01. The noise was assigned a vibrato frequency of 15 Hz, and a depth of 0.05. The depth is the size of the frequency deviation around the fundamental frequency and it determines between which frequencies the vibrato should vary. A depth of 0.01 means that the vibrato will vary with +/- 1% from the fundamental frequency. The vibrato frequency is the speed at which the vibrato should move.
Combining the sine wave and the noise completed the design of the rumble strip warning signal. The frequency spectrum can be seen in Figure 2.4. In order to confirm that the created signal was similar to a real rumble strip noise, Dr. Wataru Kobayashi of Arnis Sound Technologies was consulted because of his expertise in the field of sound engineering. He confirmed that the acoustic characteristics of the signal were similar to those of a real rumble strip noise.
2.2.2 Abstract warning
A signal with a fundamental frequency of 500 Hz or higher has been found to be effective when a fast reaction time and a high perceived urgency is desired (Haas
& Edworthy, 1996). An abstract warning with a 500 Hz fundamental frequency was therefore created. In order to avoid any masking, three additional harmonics were added to the signal. These were 1000, 1500 and 2000 Hz, with amplitude ratios relative to the fundamental frequency of 0.5, 0.2 and 0.1 respectively. The frequency spectrum can be seen in Figure 2.5.
2.2. WARNING SIGNAL DESIGN
Figure 2.5: Frequency spectrum of the abstract 500 Hz warning signal
2.2.3 Choosing intensity levels
As mentioned in the introduction, an appropriate level of intensity for an auditory warning signal is around 15-25 dB above the ambient noise (Patterson (1982) cited in Edworthy et al., 1991). The ambient and engine noise in this research was a recording from inside a driving car, downloaded from the internet (SoundEffects- Factory, 2014). A part of this recording was looped and its intensity level was adjusted to approximately 55 dB. The intensity level was measured with a Thanko Rama 11O08 sound level meter, at the ears of a head and torso manikin seated in the driver’s seat. The same method was used to measure the intensity of the warning signals.
The looming warning signals had an initial intensity of approximately 66 dB, which was considered audible even with the ambient noise present. The intensity increased by 12 dB to reach a peak intensity of 78 dB (23 dB above ambient), which was within the suggested intensity range for warning signals as mentioned above.
A small onset and offset (less than 20 ms) was added to the signal in order to avoid any clipping.
The constant intensity warning signals were given an intensity of 72 dB (17 dB above ambient), which was the median value for the looming warning signal, and an onset and offset of 20 ms to avoid any startling reactions from the drivers.
2.2.4 Implementing the looming
The looming intensity was added to the rumble strip noise and the 500 Hz abstract warnings by using the Adjustable Fade function in Audacity. While a standard fade function creates a linear fade, the Adjustable Fade function can be used to shape the curve of the fade, by pushing the middle of the fade up or down.
In order to create a looming intensity that rises slowly at first, and then gradually rises more rapidly, the fade type was set to Fade Up, and the Mid-Fade Adjust to -70% (the lower the value, the more curved the fade will be). The intensity was set to increase by 12 dB. The time-intensity curve can be seen in Figure 2.6.
Figure 2.6: Time-intensity curve for the looming warning signals
2.3 Evaluation
There are many parameters that can be evaluated in order to decide whether an auditory warning signal is appropriate in a lane departure scenario. Parameters such as lateral deviation, time to return to the lane and correct response percentage (is the steering wheel turned in the correct direction?) are often considered. However, one of the most studied parameters is the reaction time since a fast response in a dangerous traffic situation can be the difference between life and death. Due to the limited scope of this thesis, the focus has therefore been to examine if there is a difference in reaction time between the four warning signals. The reaction time will be evaluated through a driving test session.
Although the reaction time is an important factor to consider when designing a warning signal, how the drivers perceive the signal is also of interest, since an annoying warning may cause unwanted behavior (Marshall et al., 2007). It has also
2.3. EVALUATION
been shown that angry drivers are more prone to violate traffic rules (King & Parker, 2008), which is why it is important that the drivers accept the warning signal. In order to assess how the test persons perceived the warning signals, a survey was conducted after the driving tests had been completed. The survey was designed to evaluate the perceived urgency, annoyance and acceptance of the warning signals in the lane departure scenario. The test persons were asked to listen to each warning signal as many times as they wanted and then rate each warning signal from 1 to 5. For the urgency and annoyance, 1 meant that the signal was “not urgent” or
“not annoying”, while 5 meant that it was “very urgent” or “very annoying”. To evaluate the acceptance, the test persons were asked; “Would you accept the sound as a warning signal if it were used in a lane departure scenario?”. They would then rate each signal from 1 to 5, 1 meaning “would not accept it at all”, and 5 meaning
“would accept it very much”.
One test session including instructions, driving test and completing the survey, took about 1 hour and 20 minutes to complete.
2.3.1 Test persons
14 test persons participated in the evaluation. All the test persons were students at Osaka University in Japan, between 22 and 26 years of age (M = 23.43, SD = 1.29), and of varying nationalities. The majority were Japanese (8 persons). All of the participants were licensed drivers with between 1 to 7 years of experience (M = 4.25, SD = 1.35). However, only three test persons reported an annual driving distance of 5000 km or more, which is the lower threshold for experienced drivers (Suzuki
& Jansson, 2003). Several participants stated that they had not been driving since they received their driver’s licenses. One person reported a lot of driving game experience, but only using a standard game controller. No test persons had any significant experience playing a driving game using steering wheel and pedals. All test persons reported normal hearing and normal or corrected-to-normal vision. Five test persons were members of the laboratory where this research was conducted, and had some prior knowledge of the purpose of the research. However, none of the test persons knew any details regarding the driving test procedures before participating.
The rest of the test persons had no knowledge of the purpose of the research. All test persons were told that the driving tests main purpose was to measure their ability to concentrate on a secondary task while driving.
2.3.2 Setup
The driving tests were carried out in a regular office room. The driver’s seat was a swivel chair with freely adjustable height. The driving simulator ran on a 2,7 GHz Intel Core i5 Macbook Pro with 8GB of RAM and an Intel Iris Graphics 6100 1536 MB graphics card. The Macbook was connected to a 37 inch Toshiba Regza 37Z2000 TV via HDMI with a resolution of 1920x1080 pixels. The ambient noise was played back from a music device connected to two JBL Creature 3 cm
diameter loudspeakers, placed on each side of the TV, directed towards the driver at approximately 1.20 m from the driver’s ears.. The car was controlled using a Logicool (Logitech’s brand name in Japan) G27 steering wheel and pedals. The driving simulator warning signals were played back through a Panasonic RX-ED57 portable stereo system, with two 8 cm diameter loudspeakers. The stereo system was aligned with the center of the steering wheel, facing the driver at a distance of approximately 0.80 m from the driver’s ears. The initial idea was to put loud- speakers on each side of the driver, and play back the warning signal from only one loudspeaker, depending on the side of the lane departure. However, since research (Suzuki & Jansson, 2003) suggests that the reaction times are the same regardless if the driver is exposed to the warning signal from both speakers at the same time, or only from one side, this idea was abandoned. The secondary task (described below) was triggered on a Samsung Galaxy S2 smartphone.
2.3.3 Secondary task
In real driving scenarios, a lane departure may occur if the driver is occupied with a secondary task while driving, such as using a smartphone, and not focusing on the driving. A lane departure can also be the result of the driver nodding off while driving. When testing warning signals in a driving simulator, it is common to simulate one of these two scenarios. The response to a warning signal collected from drowsy drivers may be a more natural response than if the warning signal is expected, as in a secondary task scenario. However, more effort may be required to make the test persons drowsy, and more equipment such as eye-tracking devices and physiological sensors is needed. The research by Ziegler et al. (1995), as well as Rimini-Doering et al. (2005) used this method. Forcing lane departures by using a secondary task has the advantage that a similar number of lane departures can be collected for each test person, and that no external equipment is needed. This has been the method of choice in other research (Suzuki & Jansson, 2003; Stanley, 2006), and is also the method used in this thesis.
An Android smartphone app was developed and used as a secondary task. When triggered from the driving simulator, the app sounded a short alarm signal. After a brief pause (0.9 s) to let the driver react to the alarm, the display showed five arrows in succession, pointing either to the left or the right of the screen. Each arrow was displayed for 0.5 s. As soon as an arrow was displayed, the driver was told to keep the steering wheel fixed in a neutral position while pressing the corresponding button on the steering wheel; either on the right-hand side or the left-hand side. All test persons were told to focus on the secondary task unless they heard a warning signal, in which case they were told to immediately react and steer back into the lane. The smartphone was placed on a chair to the left of the driver’s seat, in line with the test person’s head. This forced the test person to look down to the left in order to view the smartphone display, thereby removing the focus from the road, as can be seen in Figure 2.7.
2.3. EVALUATION
Figure 2.7: The secondary task triggered while driving
2.3.4 Driving test
The test persons started by driving one lap around the 9.85 km long course as a practice run, in order to become familiar with the car handling and the road. They were asked to drive at the max speed of 80 km/h as often as possible, and to perform lane departures on purpose in order to get a feel for the width of the road. The lane departure warning signal during the practice run was a short musical sequence (G4, H4, G4, H4) designed in Logic Pro X (Apple Inc., 2015), played with the 80s FM Piano instrument for a duration of 1.0 s. The secondary task was triggered manually by the author several times during the practice run to teach the test persons how they were supposed to behave. All test persons were told that the warning signals signified a lane departure.
When the test persons had familiarized themselves with the controls, the road and the secondary task, four test runs were conducted; one for each warning signal.
Each test run lasted approximately eight minutes (one lap around the course), and between each run there was a short five minute break. The order of the warning signals was randomized in order to avoid any bias. The course had five longer straight sections. For each straight, the secondary task was triggered at random places, resulting in a total of approximately five secondary tasks per run (on a few occasions the test persons were asked to drive longer if enough lane departure data had not been collected by the completion of one lap). In order to force a lane departure while the test persons were busy performing the secondary task, the yaw angle of the car was changed (+/- 4 deg) manually by the author as soon as the test persons started looking at the smartphone display. The direction of the yaw
angle change was randomized. This yaw angle change resulted in a lane departure almost every time. Once a lane departure was committed, the warning signal was played back and the current angle of the steering wheel and the timestamp (t1) was recorded. As soon as the test person turned the steering wheel more than the threshold (2 deg), the new timestamp (t2) was recorded. The time difference t2 - t1 was saved as the reaction time. The two degree threshold was added because it was considered to be too difficult to hold the steering wheel completely still.
Without this threshold the slightest movement of the steering wheel would be falsely recorded as a response to the warning signal. The data was recorded at a rate of approximately 70 Hz.
Chapter 3
Results
3.1 Driving test
All test persons tried the secondary task at least five times per warning signal.
Although the secondary task almost always forced a lane departure, some lane departures were judged as being invalid, and were consequently excluded from the test data. The main reasons were that the test persons had been looking at the road, or that the reaction time was too fast. All reaction times below 0.16 s were removed because this is considered to be the minimum reaction time threshold for human beings reacting to sounds (Triggs & Harris, 1982). The total number of excluded lane departures was 37.
In total, 248 valid lane departures were registered and the reaction times recorded.
The abstract warning with constant intensity (AC) accounted for 63 lane departures, the abstract warning with looming intensity (AL) for 59, the rumble strip warning with constant intensity (RC) for 62, and the rumble strip warning with looming intensity (RL) for 64 lane departures.
On average, each test person left the lane 18 times (SD = 1.62). Except for one test person who only had two valid lane departures for the RC warning, all test persons had at least three valid lane departures per warning.
3.1.1 Reaction times
Contrary to the expected results, the mean reaction times showed little variance between the warning signals, as can be seen in Table 3.1. In order to decide what method to use for determining the statistical significance of the data, a normal prob- ability test was performed. This was done by creating a normal probability plot (see Figure 3.1), which showed that the reaction times were positively skewed (skew- ness = 2.07), and consequently not normally distributed. Since the data was not normally distributed, a non-parametric Kruskal-Wallis test was conducted, which showed that there was no significant difference in reaction time between the four warning signals (H = 2.397).
Table 3.1: Mean reaction time per warning signal
Warning signal Count Mean reaction time (s) Standard deviation (s)
AC 63 0.58 0.28
AL 59 0.66 0.33
RC 62 0.64 0.33
RL 64 0.60 0.23
Figure 3.1: Normal probability plot for distribution of reaction times
3.2 Survey
All 14 test persons participated in the survey after completing the driving test. The mean ratings and standard deviations are found in Table 3.2 below. The data was tested for normal distribution in the same way as for the reaction times. The normal probability plots showed that the data was normally distributed, which allowed for a one-way Analysis of Variance (ANOVA) to be conducted. The ANOVA is a method to determine whether any changes between various means are statistically significant. The p-value is evaluated to determine the significance of the results.
In this case, a p-value lower than .05 would indicate that any differences between the groups (the warnings) were most likely significant. The ANOVA showed that there were significant differences at the p<.05 level between the four warning signals for each of the three categories (urgency [F(3, 52) = 13.76, p < .001], annoyance [F(3, 52) = 3.62, p = .02] and acceptance [F(3, 52) = 4.28, p = .01]). However, the ANOVA can only indicate that there are differences among all the warnings, but not between which warnings. Therefore, two-sided paired-samples t-tests were
3.2. SURVEY
conducted for each category to evaluate pairwise differences between the warning signals.
Analyzing the perceived urgency, it was found that the rumble strip warnings were significantly less urgent than their abstract warning counterparts (Table 3.2);
RL was less urgent than AL; t(13)=5.83, p<.001, and RC was less urgent than AC;
t(13)=4.22, p=.001. It can be seen that AL displays a high level of urgency, while RC is perceived as not very urgent. Furthermore, for the rumble strip warnings, the looming warning was perceived as significantly more urgent than the constant warning; t(13)=2.86, p=0.01. No similar difference was found among the abstract warnings; t(13)=2.09, p=.06.
Looking at the Annoyance column in Table 3.2, it is found that the rumble strip warnings were significantly less annoying than their abstract warning counterparts;
RL was less annoying than AL; t(13)=2.33, p=.04, and RC was less annoying than AC; t(13)=2.31, p=.04. Both versions of the rumble strip noise were equally an- noying; t(13)=1.33, p=.21, and the same relation was found between the abstract warnings; t(13)=0.74, p=.47. No significant difference in perceived annoyance was found between RL and AC; t(13)=1.13, p=.28. It can be noted that no warning was rated as being not annoying; all of the warnings have a medium high or higher perceived annoyance.
Lastly, the degrees to which the test persons would accept the warning signals in a lane departure scenario are shown in the Acceptance column (Table 3.2). It can be observed that the two rumble strip warnings have the same means, and were hence equally accepted. Comparing the two abstract warnings with each other, it can be seen that these two warnings were also accepted to the same degree; t(13)=0.18, p=.86. Looking at the rumble strip warnings compared with the abstract warnings, the paired-samples t-tests found that RC was significantly less accepted than AC;
t=(13)=2.83, p=.01, but that there was no significant difference compared with AL;
t(13)=2.11, p=.05. However, RL was found to be significantly less accepted than both AC; t(13)=2.45, p=.03, and AL; t(13)=2.38, p=.03.
Table 3.2: Mean perceived urgency, annoyance and acceptance per warning signal
Warning signal Urgency Annoyance Acceptance
mean sd mean sd mean sd
AC 3.71 0.96 3.64 1.23 3.79 0.86
AL 4.36 0.81 3.93 0.70 3.71 1.22
RC 2.43 0.73 2.64 1.11 2.64 1.04
RL 3.07 0.70 3.00 1.31 2.64 1.29
*Ratings are on a scale from 1 (lowest) to 5 (highest)
3.3 Test persons’ opinions
During the driving test session and while answering the survey, some test persons voiced their opinions on the warning signals. The first person said that the rumble strip warning sounded too much like an engine sound, and that he had never heard a rumble strip noise before. He also stated that a warning was supposed to be urgent and annoying to catch a driver’s attention.
The second person claimed that the rumble strip warning sounded too much like a real rumble strip noise. He then explained that he had become used to the noise when driving in real life, and sometimes drove on the rumble strips just for fun. In other words, the warning signal did not convey any sense of urgency to him.
The third person mentioned that the rumble strip warning sounded “more real- istic” than the practice warning signal, and that it sounded urgent and “felt like in reality”. It was also mentioned that the abstract warnings conveyed the feeling of a car approaching.
Chapter 4
Conclusion
4.1 Discussion
4.1.1 Reaction times
Contrary to the hypothesis, all warning signals displayed similar reaction times;
neither the rumble strip noise or the looming intensity reduced the reaction times.
Auditory icons, with a sound related to the real world, have displayed faster reac- tion times than other sounds when used in driving scenarios (Graham, 1999; McKe- own 2005). Although they are easier to remember than abstract sounds (Ulfvengren, 2003a, cited in Ulfvengren, 2003b), they nevertheless have to be learnt through pre- vious experience. Since 11 of the 14 test persons were inexperienced drivers, likely with limited experience of the rumble strip noise, they may not have related the rumble strip noise to a dangerous situation, which in turn could have caused a reaction similar to when responding to an abstract warning signal.
On the other hand, since the test persons knew that all warning signals signified a lane departure, they understood how to react regardless of what signal they heard, which may have caused the similar reaction times. This may also explain why the warning signals with a looming intensity showed no faster reactions than the signals with constant intensity.
Another reason for the lack of difference between the warning signals may have been that the test persons learnt that almost each time the secondary task was triggered, there would be a following lane departure. They may therefore have expected the warning signal, and were ready to react once they heard it. This may have been avoided if, for some secondary tasks, the yaw angle of the car had not been altered. By altering the yaw angle less frequently, there would have been several occasions where the secondary task was triggered without a following lane departure.
4.1.2 Survey
It was not surprising that the abstract warnings were perceived as more urgent than the rumble strip warnings, since their frequency was higher and followed the guidelines suggested by Haas and Edworthy (1996) for creating a high urgency warning. That the high urgency abstract warnings displayed a higher perceived annoyance than the low urgency rumble strip warnings, was also in line with previous research that suggests that as the urgency increases, so does the annoyance (Tan
& Lerner, 1995; McKeown, 2005). It may also be argued that because of the participants’ limited driving experience, they may not relate the rumble strip noise to a dangerous situation, and thereby perceiving the warning as less urgent. That RL showed a significantly higher level of urgency than RC suggests that a looming intensity may be used to portray how urgent a certain situation is. This could for example be used to tell the driver how close the car is to the lane edge; a low intensity meaning that there is still plenty of space left, while increasing the intensity means that the lane edge is getting closer.
Interestingly enough, although RL is perceived as more urgent than RC, there is no difference between them when it comes to the perceived annoyance. This result suggests that a rumble strip warning with a looming intensity could effectively be used to increase the urgency of a warning while maintaining a low level of annoyance, which is important in a driving scenario in order to avoid any unwanted responses from the driver.
That the rumble strip warnings were not accepted to a high degree was contrary to what previous research has found (Fagerlönn, 2011b). However, these results were based on a survey among professional truck drivers, with much more experience than the participants in this research. The lack of experience may be a reason why the acceptance rate was not higher. It could also be because this research solely focused on lane departure warnings; no other warnings, such as collision warnings, were present, which minimized the confusion regarding how to react. In this case the urgency of a signal may be more important than how easy it is to interpret the meaning of the signal. However, in a real driving scenario the ability to discern the meaning of a warning from another is important, which is why the rumble strip noise may be preferred to an abstract sound.
4.2 Limitations
It should be noted once again that the driving simulator was built from scratch, with no time to properly measure any latency issues, which may have affected the reaction times to some degree. Furthermore, since the road did not have any other traffic, and there was no risk of collision, the perceived danger of a lane departure may have been low, which would result in few incentives for staying on the road.
However, the lack of danger is apparent even in standard driving simulators (Blana, 1996). Naturally, since the driving tests were conducted in a regular office room, and not in a car cabinet, the feeling of being in a real world driving situation was
4.3. FUTURE WORK
limited. No studies have been found comparing how research results from using a high-cost driving simulator differ from those of a low-cost driving simulator.
4.3 Future work
If the same driving simulator was to be used again, it would be interesting to carry out longer driving tests, where only some of the secondary tasks are followed by a lane departure, in order to make the drivers less anticipating of the warning signals. Since this research only focused on lane departure warnings, it would also be interesting to study how the rumble strip warning is perceived if other warning systems, such as collision warnings, are present as well. In these situations it will be important to tell the various warnings apart, potentially making the abstract warning less accepted, and reducing the reaction time for the rumble strip warning.
Furthermore, it could be relevant to study what effect altering the intensity curve for the looming would have. Using the looming intensity to alter the urgency of a warning, and thereby telling the driver how close the vehicle is to the lane boundary, could also be interesting. Such an implementation could potentially be used to guide the driver when the weather is bad or when other factors prevents the driver from accurately determining how much space is left until the lane boundary has been reached.
