Integrated Vehicle-Based Safety Systems (IVBSS): Human Factors And Driver-Vehicle Interface (DVI)

Summary

PDF Version 4.92MB

February 2008
DOT HS 810 905

Technical Report Documentation Page

1. Report No. DOT HS 810 905	2. Government Accession No.		3. Recipient’s Catalog No.
4. Title and Subtitle Integrated Vehicle-Based Safety Systems (IVBSS): Human Factors and Driver-Vehicle Interface (DVI) Summary Report			5. Report Date February 2008
			6. Performing Organization Code 052004
7. Author(s) Green, P., Sullivan, J., Tsimhoni, O., Oberholtzer, J., Buonarosa, M.L., Devonshire, J., Schweitzer, J., Baragar, E., & Sayer, J.			8. Performing Organization Report No. UMTRI-2007-43
9. Performing Organization Name and Address The University of Michigan Transportation Research Institute 2901 Baxter Road Ann Arbor, Michigan 48109-2150			10. Work Unit no. (TRAIS)
			11. Contract or Grant No. Cooperative Agreement DTNH22-05-H-01232
12. Sponsoring Agency Name and Address The University of Michigan Industry Affiliation Program for Human Factors in Transportation Safety			13. Type of Report and Period Covered November 2005 – November 2007
			14. Sponsoring Agency Code Office of Human Vehicle Performance Research – Intelligent Technologies Research Division, NVS-332
15. Supplementary Notes
16. Abstract The IVBSS program is a four-year, two-phase project to design and evaluate an integrated crash warning system for forward collision, lateral drift, lane-change merge, and curve speed warnings for both light vehicles and heavy trucks. This report, covering human factors research and DVI development in the first two years of the program, describes five laboratory studies, four driving simulator studies, and two on-road pilot tests conducted to assess a variety of driver-interface concepts related to the development of integrated warning systems. Selected major findings are as follows: 1) For the vehicles selected, warning sounds should be at least 80 dB(A) in the 1 to 5 KHz range. 2) Auditory warning durations should be less than the expected mean response time. 3) No approaches to warning combination (single, dual-simple, dual-hybrid, or multiple warnings) led to noticeably better driver responses, though drivers favored the multiple warning approach least, and for a variety of reasons a dual-warning approach is recommended for IVBSS. 4) Delays between 150 and 300 ms are acceptable for the LDW algorithm. 5) No single prioritization scheme for warnings (simultaneous, priority interrupt, or delayed presentation) is recommended based on the findings from a simulator study. Extended pilot testing is likely to suggest minor refinements to the DVIs developed here. In the pilot tests that have been conducted, all of the warning systems operated as planned, with some changes required to reduce false alarm rates. Overall, drivers reported IVBSS to be intuitive and easy to use. Most drivers stated warnings were received with about the right frequency, and in general the warnings were not distracting . Results from the laboratory and simulator experiments, in particular, are likely to assist future developers of driver-vehicle interfaces for integrated crash warning systems.
17. Key Words Integrated Vehicle-Based Safety System, IVBSS, DVI, driver vehicle interface, human factors, ergonomics, warnings				18. Distribution Statement Document is available to the public through the National Technical Information Service, Springfield, VA 22161
19. Security Classification (of this report) Unclassified		20. Security Classification (of this page) Unclassified		21. No. of Pages 396	22. Price

List of Acronyms

AASHTO	American Association of State Highway and Transportation Officials
ACAS	Automotive Collision Avoidance System
ANOVA	Analysis of Variance
BSD	Blind Spot Detection
CDL	Commercial Drivers License
CSW	Curve Speed Warning
CWS	Crash Warning System
DIU	Driver Interface Unit
DVI	Driver-Vehicle Interface
FAD	Light-vehicle module for FCW, Arbitration, and DVI
FCW	Forward Collision Warning
FOT	Field Operational Test
HT	Heavy Truck
ISO	International Organization for Standardization
IVBSS	Integrated Vehicle-Based Safety Systems
LAM	Look Ahead Module
LCM	Lane Change-Merge Warning
LDW	Lateral Drift Warning
LFAD	Light-vehicle module for LCM, FCW, Arbitration, and DVI
LV	Light Vehicle
MT	Masking Threshold
MUTCD	Manual on Uniform Traffic Control Devices
NHTSA	National Highway Traffic Safety Administration
RDCW	Road Departure Crash Warning
TCP/IP	Transmission Control Protocol/Internet Protocol
TLX	Task Load Index
TTC	Time To Collision
U.S. DOT	United States Department of Transportation
UM	University of Michigan
UMTRI	University of Michigan Transportation Research Institute
VOC	Voice Of the Customer
VORAD	Vehicle Onboard RADar

1     Executive Summary

1.1    Goal of the Human Factors and Driver-Vehicle Interface Effort

The goal of this project is to contribute to the design and implementation of safe and effective driver-vehicle interfaces for integrated vehicle-based crash warning systems installed in light vehicles and commercial trucks by studying warning characteristics and their effect on driver responses to crash warnings.

1.2    Overview

In November 2005, the U.S. Department of Transportation entered into a cooperative research agreement with a team led by the University of Michigan Transportation Research Institute (UMTRI) to develop and test an integrated, vehicle-based, crash warning system to help reduce rear-end, lane change-merge, and road departure crashes for light vehicles and heavy commercial trucks. The first two years of the Integrated Vehicle-Based Safety Systems (IVBSS) Program included a series of human factors tests and various driver-vehicle interface (DVI) engineering tasks to support the development and testing of the integrated warning systems. Specifically, and as outlined in this report, the human factors and DVI development is intended to help ensure that the DVI is safe and effective at accurately communicating crash threats to drivers.

Table 1 shows the crash warning subsystems integrated into the IVBSS program.

Table 1 . IVBSS warning subsystems

Warning	Abbrev.	Platform	Description
Forward collision warning	FCW	Light vehicle Heavy truck	Warns drivers of the potential for a rear-end collision
Lateral drift warning	LDW	Light vehicle Heavy truck	Warns drivers that they may be leaving their lane of travel, or possibly the roadway
Lane change-merge	LCM	Light vehicle Heavy truck	Warns drivers that it might not be safe to perform a lane change or merge maneuver due to the presence of other vehicles
Curve speed warning	CSW	Light vehicle	Warns drivers that they may be traveling too fast to successfully navigate an upcoming curve

The IVBSS program is different from previous field tests of crash warning systems in the number of subsystems that are being integrated and the number of vehicle platforms on which the integration is taking place. While the human factors and DVI efforts have focused on safe and effective communication with the driver, even greater effort has gone into engineering, integrating the subsystems into vehicles, fusing sensor data, and prioritizing warnings. The human factors studies took place in parallel with the development of the vehicles themselves. In anticipation of this, DVI hardware was designed from the beginning of the program to remain flexible in order to accommodate the findings of the human factors studies.

Early in the IVBSS program, the IVBSS team considered a wide variety of approaches to DVIs for both light vehicles and heavy commercial trucks, including the use of several warning modalities (visual, audio, and haptic). In doing so, it was recognized that light vehicles and commercial trucks are distinctly different both in the characteristics of their drivers and the environment in which the DVI has to function. Therefore, a singular DVI design and approach for both platforms was ruled out. Nonetheless, the human factors studies that were performed in the laboratory and a driving simulator sought to answer basic questions specific to developing DVIs for integrated crash warning systems. Many of the experiments dealt with general human performance issues (e.g., responses to tone sequences in experiment 1, subtask 5 combined warnings in experiment 3, delays in experiment 4, and interference in experiment 5) and the results should therefore apply to both platforms.

What follows in this executive summary, and the remainder of the report in much greater detail, is information on how the experiments, evaluations, and pilot tests were carried out and reports of the associated findings. Wherever applicable, the report highlights how findings from the human factors testing contributed to DVI development for vehicles to be used in the IVBSS field operational test.

1.3    Initial Human Factors and DVI Development Efforts

After a review of existing literature on warning design (in general and specific to crash warnings for drivers), expert human factors judgment, and discussions with those overseeing the engineering efforts of subsystem development and integration, seven research questions that warranted study were identified by the human factors team. Given the available time and resources allocated for human factors testing, a team of human factors experts designed a series of five experiments that would attempt to answer these seven research questions with particular interest towards recommending how DVIs should be implemented in the IVBSS program. A work plan was developed, and several upgrades to the UMTRI driving simulator were made to support the conduct of the experiments.

A series of experiments was carried out to help the team members select among DVI options. In experiment 1, subjects rated warnings on various dimensions and responded to them by pressing buttons (to indicate which was presented), and researchers collected physical measurements of cab environments. The human performance experiments were conducted in the laboratory and in vehicle cabs.

In experiments 2 through 5, subjects drove in the UMTRI simulator while various events occurred (vehicles cut in, lead vehicle braked, lead vehicle changed lanes to reveal parked vehicles, etc.). When these events occurred, one or more of the four warnings (FCW, CSW, LDW, and LCM) would trigger, and sometimes the subject would respond by slowing down, braking, or returning to the lane. Within and across experiments, the warning modalities and, in particular, sound characteristics varied in a systematic manner to examine issues pertaining to simultaneous warnings, warning processing delays, etc. In addition to collecting driving performance data (speed, lane position, throttle position, brake on or off, etc.), warnings were rated on various characteristics (loudness, frequency of occurrence, ease of understanding, usefulness, etc.) at various points in time.

1.4    Available DVI Options

In the IVBSS program the integration of the IVBSS system, including the DVI, had to occur in post-production vehicles. Therefore, certain constraints associated with existing vehicle designs existed a priori. The current DVI designs are based upon the options available to the team, while retaining the goal of safe and effective communication of warnings to drivers. The design approaches may not, however, be completely representative of those that might be selected by a vehicle manufacturer if an IVBSS-like system was planned for early in the vehicle development stage.

On both vehicle platforms, the primary modality used to warn drivers is auditory, with the addition of some haptic warnings on the light-vehicle platform. The auditory warnings are directional, as they are presented from the location where the threat resides (forward, left, or right of the vehicle). Visual displays are not a source of presenting warnings per se, but they indicate the presence of vehicles in blind spots so that drivers might see those indicators, located in or near the left and right rearview mirrors, prior to initiating a maneuver that would otherwise result in an auditory warning. Visual displays are largely used to convey system status information on both platforms. Haptic warnings that are implemented on the light-vehicle platform consist of a directional-vibrating seat pan to convey cautionary lateral drift warnings (LDWs), and a brake pulse that is used in conjunction with an auditory warning to convey imminent forward collision warnings (FCWs) and curve speed warnings (CSWs).

With the emphasis on the use of auditory warnings for both platforms, the human factors experiments conducted to support DVI design focused largely on how auditory warning characteristics, and methods of implementing auditory warnings, affect both objective and subjective driver response.

1.5    Experiment 1: Auditory Warnings

The first experiment consisted of five subtasks:

• Characterize the light-vehicle and truck cabin sound environments;
• Systematically study the acoustic properties of sounds having listeners rate the sounds for perceived urgency, annoyance, noticeability, and loudness, and use the gained understanding in developing crash warnings;
• Construct two suites of sounds comprised of auditory icons and abstract sounds that could serve as crash warnings, and test them for the relative speed at which naïve subjects could learned and respond to them;
• Study how the addition of noise and spatial enhancement modifications made to sounds affect their ability to be localized to determine if such modification might enhance response speed and accuracy; and
• Examine how the characteristics of pulsed tones (the number of bursts, number of beeps in a burst, and time between bursts) affect reaction time to directional warnings.

1.5.1    Summary Findings

• Based upon measurements and analyses of sound environments in the Honda Accord and International 8600 tractor using some standard assumptions, IVBSS warning sounds should be a minimum of 80 dB(A) included frequency content in the 1 to 5 KHz range.
• The empirical results from subtask 3 suggest that both urgency and annoyance increase as frequency increases, with annoyance increasing more rapidly. Use of high frequencies risks increasing annoyance levels. 1,000 Hz is two octaves above middle C and may be considered high with respect to fundamental pitch (only the best soprano singers can reach this note). The recommendation, however, is based on considerations beyond just perceived urgency and annoyance, and must also consider the need for the warning to not be masked by background noise levels.
• In comparison to abstract sounds, one set of auditory icons required substantially fewer trials to learn and resulted in substantially shorter reaction times. However, another set of sounds that were modest variants of the auditory icons (and assigned to the same scenarios) were more difficult to learn and produced longer reaction times. Thus, even minor alterations to the sound characteristics of auditory icons can result in markedly different performance.
• Neither the addition of noise nor spatial enhancement improved the accuracy or the speed of responding to a warning sound.
• Response times increased slightly when warnings have increasing numbers of bursts, numbers of beeps in a burst, or delays between bursts. To minimize response time, pulsed warnings should consider two bursts of three beeps each, and short gaps between bursts.

1.6    Experiment 2: Driver Response to Warnings

This experiment addressed three questions:

• How do drivers respond to warnings; especially, where do they look?
• Do drivers respond differently to warnings when they are distracted?
• How well do drivers understand a candidate set of IVBSS warnings, and are any confused or misunderstood?

1.6.1    Summary Findings

• For all threat types, in most cases, a driver’s vision was likely to fixate on the area forward of the vehicle (about 65% of the time during the baseline periods and 75% of the time immediately after a warning), even for lane change-merge (LCM) warnings where the hazard was on the side of the vehicle.
• There were no statistically significant effects associated with distraction; however, the complexity of the basic task meant that many participants were not diligent in completing the distraction task—so that they were not in fact truly distracted.
• All of the warnings were rated as somewhat easy to understand with only small differences among them; however, none of the warnings was initially well understood when drivers were uninformed.

1.7    Experiment 3: Combined Warnings for IVBSS

• How does the combining of warnings, using a single cue to represent more than one threat scenario, affect (a) driver performance when responding to them, and (b) driver ratings of them? Of interest were single warnings, dual warnings (simple and hybrid), and multiple warnings.

1.7.1    Summary Findings

• No combination of threat scenarios into one or more combined cues (single [master warning], dual-simple [lateral warning and longitudinal warning], dual-hybrid, or multiple [each treat type had a distinctly different warning]) led to substantially better driver responses than any other combination.
• Subjects least preferred multiple warnings for the four subsystems.
• IVBSS should use one of the dual-warning approaches, or a variation of them.

1.8    Experiment 4: Warning Time-Accuracy Trade

• How does the tradeoff between warning system processing time (to start to inform the driver) and warning accuracy affect driver responses to warnings?
• How well do drivers respond to a candidate set of IVBSS warnings? Are any confused or misunderstood?

1.8.1    Summary Findings

• Subjects did not perceive the difference in warning delays.
• Delays (the time between the onset of a threat and the presentation of a warning to a driver) between 150 and 300 ms are acceptable for the LDW algorithm implemented in this experiment.

1.9    Experiment 5: Driver Response to Simultaneous Warnings

• How does responding to one warning differ from responding to two warnings that are co-occurring (simultaneous or nearly simultaneous)?
• Should co-occurring warnings be presented:
• Whenever they occur, even if simultaneously?
• One at a time with higher priority warnings interrupting those of lower priority?
• In sequence of occurrence, with the first warning playing to completion before the second (delayed warning) starts?
• In the simulator, how well do drivers respond to the set of candidate IVBSS warnings? Are any confused or misunderstood?

1.9.1    Summary Findings

• No single prioritization rule can be recommended based on the data collected.
• BSD (blind spot detection) should be used as is. All other warnings could use enhancements.

1.10    Light-Vehicle Stage 2 Pilot Test

• How well did the warning hardware or software work on the road and what could be improved ?
• How often did warnings occur?

The light-vehicle stage 2 pilot testing sought to gain feedback and first impressions from 18 laypeople while driving a vehicle equipped with a developmental version of IVBSS. This evaluation was performed along a 90-mile, prescribed route with a researcher present. Objective measures of warning type and frequency were collected, as was subjective data on preliminary acceptance.

1.10.1    Summary Findings

• The number of warnings triggered was sufficient for the purpose of the pilot test.
• The IVBSS warnings were rated as easy to use.
• The IVBSS hardware and software worked fairly well, but some changes are needed to reduce the false alarm rate (13 warnings per 100 miles), particularly for LDW. Those changes are readily achieved.

In over 1,528 miles of driving, a total of 379 warnings were received in the pilot test. The average number of warnings per driver was 21, with one driver receiving only five warnings and another receiving 38. There were a total of 263 LDW warnings, which were dominated by false warnings when drivers drifted toward a lane boundary and IVBSS mistakenly identified an adjacent threat. Based upon a sampling of the LDW imminent alerts, the false alarm rate for LDW is approximately 12.8 warnings per 100 miles, as compared to an overall false alarm rate of 13 warnings per 100 miles. This was a known problem with the LDW subsystem. Scheduled changes to the LCM subsystem that provides AMR data to the LDW subsystem will significantly reduce false warnings. The second most common warning was cautionary LDW, which was largely associated with lane changes in which the turn signal was not used (98). The frequency of FCW and CSW warnings was quite low.

Drivers subjectively reported IVBSS to be intuitive and easy to use. Most drivers stated they received warnings with about the right frequency, and on average were not distracted by IVBSS warnings. The warnings were also deemed to be helpful in identifying potential conflicts. However, the high false warning rate associated with the known LDW problems did lead to driver uncertainty about what each warning was intended to represent. Overall, when compared to the subjective results from previous field evaluations of crash warning systems (RDCW and ACAS), the results are on par despite the recognized need to correct the LDW subsystem.

1.11    Heavy-Truck Stage 2 Pilot Test

The heavy-truck stage 2 pilot testing sought to gain feedback and first impressions from commercial truck drivers operating a vehicle equipped with a developmental version of IVBSS. This evaluation was performed along a prescribed route with a researcher present. Objective measures of warning type and frequency were collected, as was subjective data on preliminary acceptance.

1.11.1    Summary Findings

The heavy-truck stage 2 pilot test will be run in mid-November 2007. Findings will be reported when the test is complete.

1.12    Conclusions

The human factors testing described in this report provided guidance in developing the driver interfaces for the prototype vehicles used in the IVBSS program. If the program is approved to move forward with the planned field operational test in Phase II, then additional testing of the interface designs will provide data for further improvement.

Both vehicle platform teams worked with the human factors staff to determine the constraints imposed by vehicle hardware and software, as well as driver characteristics. The team identified the research questions most critical to integrated crash warning system implementation. The integrated system was developed for existing production vehicles. The specifics of the driver interface implementation for this project, however, are likely to be different from how a vehicle manufacturer might elect to implement a suite of integrated warnings, having the benefit of designing the system from the onset of vehicle planning. This is particularly true with regard to the possible use of advanced visual and haptic displays that could not readily be retrofitted into a production vehicle. The results of the experiments presented in this report will provide information for such future systems, as well as the driver-vehicle interfaces implemented in this research program.

The integrated system driver interface design is centered on auditory warnings, which is consistent with both current and accepted automotive human factors practice; auditory signals are appropriate for time-critical events, are less likely to interfere with the visual aspects of driving, and capture driver attention even when the driver is distracted. The experimental results given in this report support the particular implementation of auditory warnings for the IVBSS program. Supplemental warnings are provided by visual and haptic alerts to reinforce the effectiveness of the auditory warnings. These supplemental warnings were developed by each platform’s design team apart from the human factors experiments discussed in this report or incorporated into the DVI design, in consultation with the all IVBSS partners.

At the conclusion of this stage of the IVBSS program, the DVI design for the light vehicle platform will have one warning for longitudinal hazards (FCW and CSW), and a second for lateral hazards (LDW and LCM), supplemented by directional cues. Based on the findings of driver preference surveys, this approach will provide a safe, effective and usable driver-vehicle interface for the IVBSS field operational test. For the heavy truck platform, a similar approach for all subsystems was used. The heavy-truck implementation differs from the light-vehicle approach due to the constraints imposed by the Eaton crash warning system, which was used as the basis for the production of the heavy-truck warning system.

The results of the human factors experiments described in this report will not only provide the basis for the current driver interface implementations, they will also substantially extend the knowledge of how the design of auditory warnings for a suite of integrated warnings impact driver response to those warnings.

2    Overview

In November 2005, the U.S. Department of Transportation entered into a cooperative research agreement with an industry team led by the University of Michigan Transportation Research Institute (UMTRI) to develop and test an integrated, vehicle-based, crash warning system to reduce rear-end, lane change-merge, and roadway departure crashes for light vehicles and heavy commercial trucks. The work being carried out under this agreement is known as the Integrated Vehicle-Based Safety System (IVBSS) program.

The IVBSS program is a four-year effort divided into two consecutive, non-overlapping phases of 24 months each. The UMTRI-led team is responsible for designing, building, and field-testing the prototype integrated crash warning systems. This report summarizes the initial human factors testing and driver-vehicle interface development performed during the first phase of the program in support of an overall integration effort, all prior to the field test. This first phase includes (1) the development and specification of the driver-vehicle interfaces (visual, audio, and haptic information provided to the driver), (2) the development of prototype hardware, and (3) the design and conduct of a series of laboratory and driving simulator studies to assess and enhance the ease of use and usefulness of the evolving driver interface. Subsequent research in Phase II will assess the safety benefits and driver acceptance associated with the prototype integrated crash warning systems.

Preliminary analyses conducted by the U.S. DOT indicate that the number of crashes can be reduced significantly by the widespread deployment of integrated crash warning systems that address rear-end, lateral drift, and lane change-merge crashes (NHTSA, 1996; Pomerleau & Everson, 1999; Talmadge, Chu, Eberhard, Jordan, & Moffa, 2001). Such integrated warning systems have the potential to provide comprehensive, coordinated information, from which the individual crash warning subsystems can determine the existence of a threat and, thus, provide the appropriate warning to drivers.

Three crash warning subsystems are being integrated into each platform of the IVBSS program: forward collision warning (FCW), lateral drift warning (LDW), and lane change-merge (LCM) warning. A fourth, curve speed warning (CSW), is being integrated into the light-vehicle platform.

• Forward collision warning provides warnings to drivers to assist them in avoiding or mitigating rear-end crashes with other vehicles.
• Lateral drift warning consists of a system that warns drivers that they may be drifting inadvertently from their lane or departing the roadway.
• Lane change-merge warning warns drivers of possible unsafe maneuvers based on adjacent or approaching vehicles in adjacent lanes, and includes full-time side-object-presence indicators.
• Curve-speed warning warns drivers that they may be driving too quickly into an upcoming curve and as a result might lose control and depart the roadway.

What differentiates the IVBSS program from previous programs supported by the U.S. DOT is that these subsystem are being evaluated as part of an integrated crash warning system, rather than independently. To realize the maximum potential benefits, the integration in the IVBSS

program is greater than that undertaken in any prior program of its kind. The integration should dramatically improve the IVBSS performance relative to the standalone subsystems by increasing system reliability and reducing false warnings. As a result, consumer acceptance of crash warning systems, in general, might be expected to improve. However, the scope of integration effort on the IVBSS program is not limited to sensor data, but includes the arbitration of warnings based upon threat severity and the development of an integrated driver-vehicle interface. Arbitration and a well-designed driver-vehicle interface are critical to ensuring driver comprehension of warnings, reduction of driver workload, and reduction of driver reaction times.

The IVBSS team at the Department of Transportation includes representatives from the National Highway Traffic Safety Administration, the Research and Innovative Technology Administration (specifically, its Intelligent Transportation Systems Joint Program Office and the Volpe National Transportation Systems Center), the Federal Motor Carrier Safety Administration, and the National Institute of Standards and Technology.

The team led by UMTRI working on the light-vehicle platform includes Visteon Corporation (a major supplier), Honda R&D Americas (a manufacturer), and Cognex Corporation (a supplier of crash warning systems). On the heavy-truck platform the partners are Eaton Corporation (a supplier of sensors), International Truck (a truck manufacturer), and Cognex Corporation. In addition, Con-Way Freight (a commercial trucking company) is working on the program. The involvement of industrial partners on the IVBSS program is seen to be critical, given the partners’ technical knowledge of and ultimate ability to deploy actual systems into the Nation’s vehicle fleet. Additional members of the team include Battelle Memorial Institute, which is assisting in the development of the heavy-truck driver-vehicle interface, and the Michigan Department of Transportation, which is providing technical support as it relates to the acquisition of crash and roadway geometry data.

Additional information detailing the development of the integrated crash warning systems during the first year can be found in the first annual report of the IVBSS program (UMTRI, 2007).

2.1    The Need to Conduct Studies on Integrated DVIs

The design of an integrated crash warning system differs significantly from that of a single, stand-alone system. A stand-alone system does not need a warning that is readily distinguishable from any other warnings presented to the driver. A single system simply has to present a warning that is readily detected and acted upon. In the implementation of a stand-alone system, effective warnings can be designed that do not convey much in the way of meaning or intent. In other words, if a vehicle is only equipped with a forward crash warning system, the driver can readily learn that the presentation of a warning, almost independent of its characteristics, is associated with a forward crash scenario. There is less cognitive processing required by the driver to determine what the warning means, or how to respond, in a stand-alone system relative to an integrated crash warning system. For an integrated crash warning system, the driver must determine (1) what warning was presented, (2) what the warning means (i.e., what crash type is detected (forward, curve speed, lateral drift, or lane change-merge) and (3) how best to respond.

Furthermore, a warning stimulus that works well in a stand-alone system (e.g., FCW) may not work as well in a multiple warning system, even when warnings do not occur concurrently. If the stimulus does not work well in a multiple warning system , it could be a result of the extra time required for the driver to compare this stimulus to the possible stimuli from other warnings to verify its identification. The implication for design is that the identification of several stimuli, all of which appear to be optimal for stand-alone systems, may require further empirical comparison to select the most effective of those stimuli for a multiple warning system. This is a significant issue because it appears likely that the majority of warnings issued from a multiple warning system will be issued for single conflicts.

Ideally, the best possible warnings for an integrated crash warning system would result in reaction times, and the types of responses, that one would observe in a well-designed stand-alone crash warnings system. However, most of the human factors testing and DVI development work to date that is published in the open literature has concentrated on stand-alone systems (e.g., what constitutes a good auditory warning regardless of the application); in addition, it has not taken into consideration the potential for confusion or uncertainty by a driver when multiple warnings are present and what is needed to mitigate uncertainty. Therein lies the challenge facing the design of an integrated warning system, and hence the need for DVI testing and development on the IVBSS program, which is the central theme to the simulator testing in particular.

2.2    Prior Warning Studies and What Do They Say About How Drivers Should Be Warned?

A vast body of literature exists for the design and evaluation of warning systems, in particular integrated warning systems. Most notably, this body of literature has recently been surveyed and discussed in the context of IVBSS in Campbell, Richard, Brown & McCallum (2007). This section identifies the literature that is of specific importance to this project in several categories: warning timing, warning reliability, multi-collision warnings, modality of warning, multi-modal warnings, auditory warnings, warning design, warning urgency, auditory icons, and sound localization. Notice that the emphasis of this review and the literature in general is on the characteristics of individual warnings, not on the integration of warnings, which is the focus of IVBSS. For a complete listing of the literature reviewed, see Appendix D.

2.2.1    Warning Timing

The timing of warnings has a critical effect on the safety benefit of any warning system. Early warnings have been shown to reduce the number and severity of crashes (McGehee, Brown, Lee, & Wilson, 2002) and to help drivers to react more quickly to avoid collisions, even when drivers are not distracted (Lee, Ries, McGehee, Brown, & Perel, 2000). The extent to which appropriate timing predictions can be made is straightforward, as predictions must be based on assumptions about the driver’s expectations and typical responses (Kiefer, LeBlanc, & Flannagan, 2005). The timing of warnings not only affects the driver’s immediate responses, but also the trust they develop in the warning system. A series of simulator studies has shown that in imminent crash situations, response and trust for early warnings (e.g., .05 seconds after a lead vehicle braking) is better than for late warnings (e.g., 0.99 seconds) (Abe & Richardson, 2004, 2005, 2006). Although there is clear evidence to support use of early warnings, there also appears to be a price—overall warning reliability may be reduced if warnings are produced too early. The timing issue is particularly relevant to experiment 5.

2.2.2    Warning Reliability

Response frequency to alarms has been shown to decline with decreasing reliability (Bliss & Acton, 2003). Reliability, in this context, can be defined as the extent to which a system yields the same results on repeated trials. According to signal detection theory, in assessing the reliability of a warning, there are four cases to consider: (1) a signal and a response (hit); (2) a signal and no response (miss); (3) no signal and a response (false warning); and (4) no signal and no response (correct rejection). What confuses the matter is that responding to an event is a two-stage process. In the first stage, the warning system responds to the event; in the second stage, the driver responds either to the warning system alone or to the warning system and the event, depending on the situation. So, from the perspective of the warning system, if an event leads to a warning based on the system rules, it is a hit. However, if from the driver’s perspective the warning is considered unnecessary, it would be considered a nuisance warning relative to the original event (even though is was correctly triggered based on the warning system rules). Thus, what the driver considers a nuisance warning depends on what their assessment is with regard to the warning system and the triggering event. Nuisance warnings are important because of their effect on driver acceptance and sometimes the extent to which they can be reduced through the application of technology. Yamada and Kuchar (2006) found that the mean driving speed decreased as the missed detection rate of a collision warning system increased, demonstrating a decrease in a driver’s reliance on warnings when the system has low reliability. Furthermore, they found that both the acceleration pedal and brake pedal reaction time increased when the system became more prone to false alarms. The inherent problem of a collision warning system, however, is that the base rate of crash events is extremely small. As a result, even excellent warning systems, which provide correct detection 99 percent of the time and generate false alarms only 1 percent of the time, will be prone to generate many more false alarms than hits because the opportunity to provide a true positive alarm is scarce when the baseline is low, while the opportunity to provide false positives is immense (Parasuraman, Hancock, & Olofinboba, 1997). Drivers need to accept this probability imbalance and expect to hear many false alarms if they want a system with the potential to save lives. Another effect of reliability on system effectiveness is annoyance, which is likely to increase as the number of inappropriate warnings increases. For example, a rate of four inappropriate warnings per hour has been rated in a naturalistic driving study as most annoying (Lerner, Dekker, Steinberg, & Huey, 1996). Recognizing the importance of reliability, a significant number of false alarms were included in experiments 2 through 5.

2.2.3    Warnings Systems with Multiple Warnings

These systems are now becoming available but have not been studied extensively in the literature. Chiang, Brooks, & Llaneras (2004) were particularly interested in the rare event of warnings that occur at the same time. They found that drivers were not confused when they received different warnings for different collision systems, as opposed to receiving a single combined warning for all different systems. They found that drivers receiving a multiple warnings looked in the direction of the threat more often than drivers receiving a combined warning, and they were sometimes able to avoid a collision as they were quicker to realize that the second warning was distinct from the first. In another simulator study, no difference in reaction time and response accuracy was found between a single collision warning and individual alerts (Ho, Cummings, Wang, Tijerina, & Kochhar, 2006) . Subjects did show a preference, however, for the multiple warning condition. The topic of multiple warnings is

central to all of the major experiments. The mapping of threats onto warnings was specifically examined in experiment 3.

2.2.4    Modality of Warning and Multi-Modal Warnings

The literature on multi-modal warnings offers mixed recommendations. The redundancy effect and the overall increased magnitude of combined signals contribute to a general preference for multi-modal warnings. Manual reaction time responses have been found to be faster with tri-modal stimuli (visual, auditory, and tactile) than with bi-modal stimuli, which were faster than uni-modal stimuli (Diederich & Colonius, 2004). Similarly, bi-modal interfaces (vision and tone or vision and voice) for forward collision warning were judged more helpful than uni-modal interfaces (Maltz & Shinar, 2004). In contrast to these and other findings, Lee, McGehee, Brown, and Marshall (2006) found that combining all four redundant warning modes resulted in a driver reaction that was 400 ms slower than just an auditory and visual alert. They suggest that redundant multi-modal warnings are not universally beneficial, and that in some circumstances might introduce complex sensory interplay that counteracts the benefits of redundancy. An additional point of view has to do with the magnitude of the signals and their spatial separation. In general, for strong multi-sensory facilitation, stimuli must occur simultaneously and in the same spatial location. Further, a strong facilitation effect is found only if the individual signals are relatively weak (Schnupp, Dawe, & Pollack, 2005). These findings imply that some intervening factors (such as the magnitude of the signals and their spatial location) may overwhelm the redundancy effect. To support that conclusion, a study that compared driver attentional prompting to a spatial location found that there was a facilitatory effect of cross-modal auditory prompting of the spatial direction, but not for vibrotactile prompting (Ho, Tan, & Spence, 2006).

2.2.5    Auditory Warnings

The use of sound to signal an imminent warning condition is ubiquitous in many warning contexts, and virtually indispensable for collision avoidance. There are few other ways to inform a driver quickly (without involving a redirection of gaze) that something is, or may be going very wrong. In comparing the suitability of an auditory to a visual warning, Deatheridge (1972) suggested that an auditory signal is best suited to convey a simple and short message requiring immediate action, while the visual system is overburdened, and the person’s position may not be fixed. It is difficult to imagine a better fit to an in-vehicle collision avoidance system.

Indeed, the use of sound to warn of danger is ubiquitous. This has generated a variety of secondary problems, not the least of which is the need to ensure that a driver interprets the sound as appropriately urgent, distinctive, and recognizable in order to make a timely response. In the context of collision avoidance, a fast and appropriate response is indispensable. In the context of an automotive product it is also desirable that the sound should not annoy those in the vehicle.

The literature on auditory warnings covers several broad themes: general discussions of methods for constructing warning sounds (e.g., Casali, 2003; Deatheridge, 1972; Edworthy, Stanton, & Hellier, 1995; Patterson & Mayfield, 1990) to ensure they are heard; investigations of the relationship between the acoustic attributes of a sound and a listener’s perception of urgency (e.g., Arrabito, Mondor, & Kent, 2004; Edworthy & Stanton, 1995; Guillaume, Drake, Rivenez, Pellieux, & Chastres, 2002; Haas & Edworthy, 1996; Hellier, Edworthy, & Dennis, 1993; Hellier, Edworthy, Weedon, Walters, & Adams, 2002; Marshall, Lee, & Austria, 2007) ; the

association of meaning with a sound through the use of natural sounds or auditory icons (Belz, Robinson, & Casali, 1999; Graham, 1999; Hellier et al., 2002; Stephan, Smith, Martin, Parker, & McAnally, 2006); and the use of spatial information in sounds to enhance localization (Catchpole, McKeown, & Withington, 2004; Tan & Lerner, 1996). Each of these topics is discussed below. As noted elsewhere, this literature is sufficiently supportive of the use of sound for imminent warnings that the auditory modality was chosen as the primary modality for IVBSS warnings that were assessed in this project.

2.2.5.1    Design of Warnings

Existing guidelines for sound construction include prescriptions about the contexts in which auditory warnings are best used (Deatheridge, 1972) and algorithms to effectively design a sound that is not excessively invasive. For example, Patterson and Mayfield (1990) discuss the use of spectral analysis of background noise to design warnings that reduce spectral overlap with the noise. Rather than designing a sound that exceeds the overall decibel level of the background noise, the warning only needs to exceed the noise levels in a few spectral bands. This allows the design of a sound that can be heard without being excessively loud. These authors also put forward a common method and vocabulary of warning construction in which pulses are combined to form sound bursts. Concern for background sound levels led to experiment 1, subtask 1.

2.2.5.2    Urgency

There has been a good deal of focus on the perceived urgency of warning sounds especially for circumstances requiring immediate response. In part, this is due to results that find response time to warnings perceived as urgent is shorter than to those perceived as less urgent. Much of the research on urgency has been directed toward establishing which acoustic properties of a sound are associated with urgency (e.g., Edworthy, Loxley, & Dennis, 1991). Several acoustic attributes have been associated with urgency—high frequency, rapid pulse rate, high or rising volume (to name a few), short onset time. Other attributes appear to be more equivocal (e.g., timbre, rhythmic variation).

Although urgency has been recognized as a particularly important component for crash warnings, it is also recognized that sounds produce other subjective impressions in listeners as well. Tan and Lerner (1995) made a comprehensive study of 28 warning sounds (including speech-based warnings) using a multiple attribute evaluation (MAE) method that quantified the utility of each sound in the context of crash warnings. This report includes a correlation matrix that suggests many attributes are related to each other—in particular, annoyance seems to be highly correlated with urgency. This makes the design of an urgent warning particularly challenging for a safety system embedded in a consumer product. If the warning is insufficiently urgent, the warning may not be effective. If the warning is sufficiently urgent, it may annoy the prospective customer so that it is avoided or somehow subverted. Recent research has been directed toward finding the combination of acoustic attributes that increase urgency without also raising annoyance levels, or at least finding attributes that raise urgency more than annoyance (e.g., Marshall et al., 2007).

There are two important issues that this focus on acoustic properties associated with urgency does not address. Warning sounds do not occur in a vacuum; they are often presented in the

context of other warning sounds and thus must be easily distinguishable. If all sounds were designed to be the most urgent, they would likely all sound the same. A second problem is that occasionally, the acoustic properties of a sound suggest it should be perceived as urgent, but instead it is perceived as silly or comical. For example, Guillaume et al. (2002) found that a learned semantic association may override the effects predicted by an acoustic analysis alone. The need to relate sound physical characteristics to the perception of urgency led to experiment 1, subtask 2.

2.2.5.3    Auditory Icons

An auditory icon is a natural sound that has a semantic association with a warning condition. In the context of collision warnings, horn honks, squealing tires, and rumble-strip sounds have been employed to represent side-collision, forward collision, and lane departure warnings.

Several studies have shown that auditory icons can be learned easily and responded to quickly (Belz, 1997; Graham, 1999; Stephan et al., 2006). To obtain the best results, a preexisting association between the sound and the warning condition is important. In some situations the warning condition may not lend itself to an obvious sound, or the semantic relationship may be odd. For example, the sound of squealing brakes may not immediately be interpreted by a driver as the need to brake, only that someone nearby is braking. Contrast this with a rumble strip sound for a lane departure warning—the sound mimics what is heard if a vehicle wanders onto the shoulder.

The naturalness of the sound of an auditory icon is relevant only in as much as such sounds usually have built-in semantic associations. Unnatural sounds with semantic associations can work equally as well. For example, if a warning sound happened to mimic the sound of a familiar radio clock, a frequently heard artificial sound, that warning would carry semantic associations that would likely influence the perception of urgency. Thus, warning sounds that resemble cartoon pratfalls or video games are often reported by listeners as silly, inappropriate, or unsuitable as warning sounds. These associations have been learned through repeated exposure. Thus warning urgency may also be learned through repeated exposure, much as one becomes highly sensitized to the sound of one’s cell phone. Issues pertaining to auditory icons and their naturalness were examined in experiment 1, subtask 3.

2.2.5.4    Sound Localization

Directional warnings should coincide with the location of lateral warning (such as for lane departure and lane change-merge warnings). This should lead to faster recognition of a warning and reduce the effort to locate the radial direction the conflict. Research on this issue is scant and has generally shown that, inside a vehicle, localization is dependent on the kind of sounds presented (speech and complex sounds are more easily localized than simple sounds) and speaker placement (localization performance is best with speakers directed toward the listener’s ears) (Tan & Lerner, 1996).

It remains to be demonstrated that a lateralized sound indeed enhances a driver’s response to a lateralized warning. In most cases, lateralized warnings lack sufficient radial resolution to pinpoint the direction of concern. At best, lateralized warnings distinguish only left and right. Moreover, sound direction cues are easily overshadowed by visual cues—even if the voice of a speaker is displaced from the speaker’s radial direction, the voice is still likely to be heard as

originating from the speaker’s direction. If drivers have even a modest level of situational awareness, they are likely to be aware of the direction of their merge or lane drift without being told. Issues pertaining to sound localization were examined in experiment 1, subtask 4.

2.3    Research Questions Identified and Addressed

As extensive as the literature on warnings is, there were still many questions that needed to be addressed in order to implement an easy-to-use, understandable, and useful driver interface for IVBSS. Based on an expert review of the literature and review of the design issues for the option space being considered for IVBSS, the seven most prominent questions regarding human factors considerations in the design of an integrated crash warning system were identified. Although there were many other possible questions that could have been addressed, those listed in Table 2 were considered most pertinent to the IVBSS program, and could be addressed within the constraints of the program.

Table 2 . Seven research questions examined

Issue	Comment
Q1. Shared warnings (When and how should warnings be shared or differentiated, e.g., FCW and CSW, LDW and LCM?)	In response to warnings, drivers can return to their lane, steer out of it, or slow down. Warnings can indicate what is wrong (so each warning is unique), what to do (which suggest common warnings based on desired actions), or both.
Q2. Sequencing co-occurring warnings (Should warnings occurring at the same time be presented together or with a delay between them?)	Presenting two warnings at the same time (e.g., forward collision warning and lateral drift) could confuse drivers as they will not be able to determine what each warning is.
Q3. Warning set/confusion (Are warnings in the IVBSS sets confused with each other?)	Warnings that sound, look, or feel alike, could be confused. But, what constitutes “alike”?
Q4. Time course of driver actions (When responding to warnings, what is the process by which drivers respond?)	To design warnings, the sequence of how drivers respond to warnings needs to be known—in particular where and when they look, when they release the throttle, and when they brake or steer.
Q5. Warning processing time/accuracy tradeoff (How does the tradeoff between warning system processing time [to start to inform the driver] and warning accuracy affect driver responses to warnings?)	For some systems, waiting to respond improves warning accuracy, for example allowing a radar unit to make more sweeps and increase threat identification accuracy. However, that delay gives the driver less time to respond.
Q6. Auditory characteristics of warnings (How does auditory warning effectiveness vary with warning sound characteristics [loudness, pitch, speed] in sound environments of each vehicle platform?)	Although there are basic data on auditory discrimination, their application to multidimensional variations found in real warnings is difficult. In real systems, due to signal generator limitations and the desire for warning sounds to resemble particular real-world sounds, there are constraints on which sounds can be used.
Q7. Influence of pauses and repetitions (For sounds that involve periods of silence (or pauses), are responses deferred to coincide with silence? What is the optimal number of repetitions?)	For lateral drift, sounds resembling a rumble strip are sometimes used. That sequence takes time to play, potentially delaying a driver response. Can the sequence be sped up?

2.4    Work Plan

Table 3 shows the mapping of the seven issues onto five planned experiments, with experiment 1 actually being comprised of five smaller experiments (subtasks). Experiments 2 to 4 involve use of the driving simulator.

Table 3 . Sequence of experiments and mapping to research questions

Experiment	Question/ Topic	Central Theme	Procedure	Subsystem
Exp 1 jury selection	Auditory warning characteristics (Q6)	Characterize sound environment of light vehicle and heavy truck; select sounds best suited to environment, five sub-tasks	Jury evaluations: (sub-task 1) of masking of warnings, (2) of sound appropriateness, (4) localization of candidates sounds RT evaluations: (sub-task 3) [MJ2] confusability of ensemble, (5) repeating sounds	All
Exp 2	Time course, method (Q3, Q4)	How people respond (where and when they look) suggests warning presentation modality and content	Collect eye fixations, steering and brake data, etc. to initial warnings (includes uninformed warnings)	FCW, LDW, CSW, maybe LCM
Exp 3	Shared warnings (Q1, Q3)	If two warnings (FCW, CSW) lead to the same response, should the warning be the same?	Collect steering and brake data, etc. for shared warnings and unique warnings	All
Exp 4	System time/accuracy tradeoff (Q3, Q5)	Warnings that are delayed may be more accurate? What tradeoff is “best?”	Use full set of candidate warnings, vary accuracy and delay of each warning, collect steering and brake data, etc.	All
Exp 5	Co-occurring warnings (Q2)	When two warnings occur at the same time, should one be delayed and by how much?	Create situations to trigger two warnings at the same time. Sometimes present both, sometimes present in priority order with delays. Collect steering and brake data, etc.	All

2.4.1    Impact of Human Factors Testing on IVBSS Design

2.4.1.1    Laboratory and Driving Simulator Testing

The goals of human factors testing in the laboratory and driving simulator were to provide information useful in making IVBSS driver interface design decisions and to provide more general knowledge that will assist in the development of driver interfaces for collision warning systems. Additionally, the driving simulator experiments served to test elements of the DVI before they were implemented on prototype vehicles. For example, some of the messages that would appear on the center console display of the light-vehicle platform were tested and modified in experiment 3. Based on feedback from experimenters and subjects, the messages were modified to make them easier to understand and less distracting.

Experiment 1 guided the selection of warning tones primarily for the light-vehicle platform. It provided the experimenters with tools for the selection of suites of auditory warnings based on predicted levels of parameters such as perceived urgency, discriminability from other tones, and annoyance. Using the CSound software tool (www.csounds.com), the experimenters were able to change sounds in a systematic manner. After one of the original warning tones was not well received, an alternative tone was quickly developed for the jury drives using CSound, with predicted levels of urgency and discriminability developed from testing of other tones. In addition, subtask 5 led to a heightened awareness of the effect of auditory warning on response time and an effort to keep auditory warnings short.

Experiments 2 through 4 were conducted in the UMTRI driving simulator. Experiment 2 examined where drivers looked when responding to warnings and their reactions to the initial warnings set. In fact, the primary class of measures of interest, those relating to eye fixations, seemed to be relatively unaffected by warnings, but that may be because the analysis was at too gross a level. However, this experiment did point out that several of the warnings needed improvement and provided useful data on how often planned and unplanned triggering of each warning actually occurred.

Experiment 3 was designed to determine how much, if at all, warnings should be combined. Results from the study ruled out the use of four individual warnings (one each for LDW, CSW, FCW, and LCM). Furthermore, an analysis that led up to the experiment discouraged the use of a single warning to represent all subsystem warnings collectively. As a result, a solution using two warnings, one for longitudinal and one for lateral threats, was selected. Objective results from the simulator experiment did not strongly favor any one of the warning approaches examined, so a subsequent design decision was made based on other considerations (such as the preferences of experts in the jury drives and laypeople in the initial pilot test).

Experiment 4 examined the effect of warning delays (time to allow drivers to gather additional information). For LDW, there was no difference between 0 and 150 ms delays in terms of the time to return to the lane, but there was a difference between 150 and 300 ms. For FCW, there were differences between the three delays, but that difference may be due to other confounded factors, not delay.

Experiment 5 addressed the issue of simultaneous warnings. The results of this experiment were important in restricting the length of auditory warnings and in confirming their adequacy. The experiment also provided useful perspective concerning the relatively rare occurrence of near-simultaneous warnings, even when there was a deliberate effort to force them to occur.

2.4.1.2    Light-Vehicle Jury Drives

The light-vehicle jury drives involved human factors experts and IVBSS team engineers driving a fixed route on public roads using a prototype system. One significant issue identified in the light-vehicle jury drives was an inter-vehicle variation in the haptic brake pulse used for FCW and CSW warnings. There were several cases in which the brake pulse was not noticed. This finding led to additional consideration of the technical aspects of the brake pulse cue.

Also as a result of comments received during the jury drives, the possibility of adding icons to the central console display to signify the occurrence of alerts, simultaneous warnings, and subsystem availability was excluded. A textual display was deemed sufficient and more straightforward.

Lastly, two tones that could serve to indicate a lateral threat were examined in the jury drives and one was selected as the preferred choice. The length of tones was deemed adequate, so it was decided to continue with the 7,000 ms warnings to drivers. The timing of the LDW/LCM warnings that were experienced during the jury drives was generally perceived as being late. As a result, the onset timing was adjusted inward for both subsystems.

2.4.1.3    Impact of Human Factors Testing on the Heavy-Truck Platform

The heavy-truck jury drives involved truck drivers, both on a test track and public roads. The threshold for changes to the heavy-truck DVI was set very high, given that the VORAD system, which serves as the primary element in the heavy-truck DVI, has previously been subjected to several evaluations, and the associated hardware was not as flexible as that of the light-vehicle platform. The greatest impact on the current design of the heavy-truck DVI was from the first experiment, which reexamined of auditory cues used in the heavy truck. That experiment identified a particular concern with the location of the visual display.

2.4.2     Timeline

The timeline for the overall IVBSS DVI effort is illustrated in the Gantt charts below. Figure 1 provides a high-level overview of all DVI efforts, while Figure 2 provides details specific to the laboratory- and simulator-based experiments.

Note that for reasons of scheduling, the data collection and analysis for experiment 1, subtask 5 was completed after experiment 5.

Figure 1. IVBSS DVI timeline

Figure 2. Timeline for experiments

2.5    Report Structure

The remainder of this report is organized as follows:

• Section 3 provides a summary of the research conducted in this study, including performance-based hardware results, the available option spaces on each platform, research questions addressed, and overviews of the driving simulator and test scenarios.
• Section 4 provides an overview of experiment 1 (subtasks 1 through 5) on auditory warnings (the complete details of which are provided in Appendix E).
• Section 5 provides an overview of experiment 2 on driver response to warnings (the complete details of which are provided in Appendix F).
• Section 6 provides an overview of experiment 3 on integrating IVBSS warnings (the complete details of which are provided in Appendix G).
• Section 7 provides an overview of experiment 4 on warning-delay/accuracy tradeoffs (the complete details of which are provided in Appendix H).
• Section 8 provides an overview of experiment 5 on driver response to simultaneous warnings (the complete details of which are provided in Appendix I).
• Section 9 provides an overview of the light-vehicle stage 2 pilot test (the complete details of which are provided in Appendix J).
• Section 10 provides an overview of the heavy-truck stage 2 pilot test (the complete details of which are provided in Appendix K).
• Section 11 provides next steps regarding the development of the DVIs and conclusions.
• Section 12 contains the list of references.
• Appendix A provides details of the driving simulator used in the experiments.
• Appendix B discusses the various driving scenarios used in this study.
• Appendix C details the prototyping tools for scenario development and warning interface creation.
• Appendix D provides an additional literature review, in greater detail than could be provided in Section 3.
• Appendix E details the five subtasks of experiment 1, auditory warnings.
• Appendix F details experiment 2, driver response to warnings.
• Appendix G details experiment 3, integrating warnings for IVBSS.
• Appendix H details experiment 4, warning delay-accuracy tradeoff.
• Appendix I details experiment 5, driver response to simultaneous warnings.
• Appendix J details the light-vehicle stage 2 pilot test.
• Appendix K details the heavy-truck stage 2 pilot test.

3    Research Summary

3.1    Platform-Based Hardware Constraints

This section identifies any existing DVI design constraints, by platform, which had to be taken into consideration when planning the DVI option spaces and experiments. This includes, but is not limited to, issues related to working with vehicles that are post-production.

3.2    Available Option Space

The IVBSS program recognized that the integration of the warning system, including the DVI, would have to occur in post-production vehicles and be consistent with the constraints imposed by real products. Thus, working within the range of available modifications that could be made to the production vehicles, option spaces (outlines of the DVI alternatives available for implementation) were determined for both light-vehicle and heavy-truck platforms. For production vehicles, the development of the DVI for an integrated crash warning system would most likely occur early in vehicle design. However, the goal of developing DVIs for the IVBSS program was to design an interface that was effective in communicating warnings to drivers, but not necessarily to develop the optimal DVI for an integrated crash warning system. Despite the initial constraints, early feedback and evaluation suggest that the approaches taken in the IVBSS DVI development are useful, effective, and acceptable to drivers.

3.2.1    Light-Vehicle Option Space

The DVI option space for the light-vehicle platform was explored during weekly meetings that included representatives from UMTRI, Visteon, and Honda. Many of the decisions to include or exclude certain options were the result of fruitful discussions among those who attended these meetings. Much of the justification for decisions was based on the literature review and on other human factors considerations. The design process started with an option space that included as many options as possible and then narrowed the design down by eliminating options that were not technologically feasible for installation in a production vehicle, did not meet human factors guidelines, or were inconsistent with the overall goals of IVBSS.

Among the constraints for the light-vehicle option space, the primary concern was that large-scale structural changes could not be made to the production vehicles. For example, the installation of a head-up display (HUD) would require excessive modifications to the structure of the production vehicle. Similarly, in considering a display for advisories, there was an attempt to use hardware that already exists in other Honda Accord implementations. An additional constraint of major impact was that of safety. Some ideas that may have been explored further in a design phase were dropped upfront for the production vehicle. For example, there was discussion of vibrating the steering wheel as a means of warning the driver. This was not pursued because of concerns that the vibration from a post-production subsystem would input vibration into the steering system and might adversely affect the safety of the system. Nevertheless, structural changes were made to the car seat to allow for seat vibration, and an implementation of a brake pulse was added after thorough discussions between Visteon and Honda representatives on implementation and on the safety aspects of such a modification.

Auditory warnings were selected as the primary modality for alerting drivers on the light-vehicle platform. A strategic decision had been made early in the IVBSS design process to center the

DVI for light vehicles on imminent collision warnings. This decision was primarily driven by the need to keep the number of warnings to a minimum and the recognition that several multistage warnings would be confusing to lay drivers. The focus on imminent collision warnings led to the choice of warning drivers with the auditory modality. Justification for the use of auditory warnings can be found in a summary of design guidelines for auditory warnings (Campbell, Richard, Brown, & McCallum, 2007, and COMSIS, 1996). Campbell et al. concluded that auditory warnings were appropriate when: (1) a high-priority warning is needed (i.e., imminent collision warnings), (2) drivers may be distracted, and (3) attention needs to be drawn directly to the location of a potential crash threat.

In the light vehicle, auditory warnings are played through an MP3-capable sound card and then amplified and played through headrest speakers. Headrest speakers were chosen because they are close to the driver’s ears (making them more likely to be heard and less likely to be masked by other sounds), separate from other speakers in the vehicle (making the discrimination task easier because of spatial separation), not so intrusive to other passengers, and likely to be perceived as louder than they actually are. The volume of the headrest speakers can be set to one of three preset levels (80, 85, and 90 dB). The auditory tones that were selected were based on a review of the literature and on findings from experiment 1. The headrest speakers were used in the simulator experiments and were well accepted by subjects.

Haptic warnings were considered as additional or redundant cues for auditory warnings. There was a preference for warnings that would intuitively draw the driver’s attention to the appropriate location of the threat. For longitudinal warnings, the driver’s attention should be drawn to the forward scene, and for lateral warnings the driver’s attention should be drawn to the appropriate side. Although a few haptic solutions involving the steering wheel were initially considered for longitudinal warnings, they were not explored further because of the potential for confusion between the physical location of the steering wheel (front) and the implied connotation of steering (lateral control). Consequently, a haptic warning in the form of a brake pulse was explored and later implemented for longitudinal warnings. Currently, the brake pulse consists of two phases. The purpose of the first phase is to get the brake pump ready for the second phase so that brake pulses appear consistent to drivers. The magnitude of the first phase is 75 psi and its duration is 160 ms. The magnitude of the second phase is 325 psi and it lasts for 240 ms. The resulting vehicle deceleration is 0.20 plus or minus 0.02 g. Although a rough simulation of a haptic brake pulse was built into the driving simulator, most of the testing and design of the brake pulse was done by Visteon on the road in development vehicles, with limited input from UMTRI during several test drives, and then with additional input throughout the jury drives.

Given the success of the haptic transducers located in the pan of the driver’s seat for lateral drift warning (LDW) in the road departure crash warning (RDCW) field test (LeBlanc et al., 2006), they were again chosen as the haptic cue conveying certain lateral threats in the current project. As an exception to the decision to focus on the use of auditory cues for imminent warnings, for the relatively benign case of crossing a lane boundary into an unoccupied adjacent lane, providing a haptic warning, without an accompanying auditory warning, was selected. The magnitude of the signals was tested at Visteon, with UMTRI team members present to approve and comment on the design. Early on, an analysis of the preferred position of the seat shakers was provided by the biosciences division at UMTRI. Later in the process, possible confusion between right and left shakers was identified by UMTRI in the prototype seat, and addressed by Visteon for all development vehicles.

The visual modality was considered for warnings, advisories, and supplemental information. Because of the inherent directionality of visual warnings, visual warnings were decided to accompany, but not replace, an auditory or haptic warning. Visual warnings were selected as a means to inform the driver of threats in the surrounding area (a top-view display) and as a way to draw the attention to the front of the vehicle (a light bar reflected off of the windshield). The former was excluded from the final design so that drivers would not have to shift attention away from the road. This decision followed a thorough examination of the concept and a unanimous decision by UMTRI and Visteon human factors researchers not to include it in the final design of IVBSS. The light bar was excluded because its added value, especially when the driver is not already looking ahead, was deemed insufficient to justify its intrusive nature. Initial testing of the visual display was done in experiment 3 and the display continued to be used in later experiments. Testing of the wording of messages and their timing were a byproduct of these experiments.

A visual advisory for lane change-merge (LCM) warnings was included in the final design. Specifically, LCM indicators appear in the side-view mirrors to convey the presence of a potential threat in an adjacent lane should the driver decide to attempt a lane change ( Figure 3). If the driver chooses to look at the mirror before making a lane change, an indicator would serve as a means to advise against initiating an unsafe maneuver because of a vehicle in the blind zone (red indicator) or a vehicle that is approaching the blind zone (yellow indicator). There is consideration to combine both warnings to a single red LED indicator. Lane changes that are initiated without checking the side-view mirror, when a threat was present in the adjacent lane, would result in an auditory warning once the maneuver begins. Some testing with prototypes of icons was done at UMTRI before the decision to keep the information in the side mirrors to a minimum. UMTRI researchers used the literature review and their knowledge of human factors literature to support the implementation of the LCM indicator as a part of the LCM system rather than an extension of a blind spot detection icon.

Figure  3. LCM indicator (LED in right mirror).

The visual modality is also used for supplemental information by providing text messages to the driver via a reconfigurable display located in the top of the center stack or cluster, above the radio and just below the hood line ( Figure 4). Because of the potential conflict between looking at the road and looking at a visual display, the visual demand of the display was reduced in the following ways: (1) any information provided would remain on the display for an extended amount of time (ten seconds), and (2) information would not appear immediately so that drivers would not be trained to always glance immediately at the display after an warning has sounded. (For technical reasons, this delay is not currently implemented in the prototype vehicle.) Although these measures are not necessarily in strict agreement with a driver’s intuitive desires, they are deemed as safety measures to reduce glancing away from the road at undesired times.

Figure  4. Center stack visual display for advisory information

There are some road conditions under which false warnings are likely to occur repetitively, for example, when repeatedly crossing a lane marker near a Jersey barrier in a construction zone. To avoid being inundated with warnings, a mute button is provided. Pressing the button temporarily disables all warnings for two minutes. Repeated presses increase the mute time to four and six minutes. An additional press resets the mute and returns to normal operation.

Sensitivity adjustments of the subsystems for several threshold levels will not be implemented. An analysis of the use of sensitivity adjustments during the RDCW field operation test (LeBlanc et al., 2006) was done to support this decision. Figures 5 and 6 illustrate the frequency of changing the LDW and CSW sensitivity settings by exposure week for all 78 drivers. As can be seen in these graphs, drivers rarely changed the sensitivity setting, particularly with increased exposure to the systems. In fact, one-third of the drivers never adjusted the sensitivity switches in the RDCW test. When RDCW FOT subjects were asked to rate their level of agreement with the statement, “I frequently adjusted the LDW sensitivity setting during my drive,” 41 of 78 respondents (53%) indicated some level of disagreement with the statement (i.e., they rated it a 1, 2, or 3 on a 7-point Likert scale with 1 = strongly disagree and 7 = strongly agree).  For CSW sensitivity adjustment (“I frequently adjusted the CSW sensitivity setting during my drive”), 50 of 78 respondents (64%) indicated some level of disagreement with that statement. Lastly, the team has received feedback from several system developers and vehicle manufacturers stating that they are very unlikely to allow drivers to adjust the level of imminent warning thresholds for several reasons: a) expense and design challenges associated with limited locations to place the controls; b) concern over increasing the complexity of the driver’s mental model on how the system operates (particularly a driver’s uncertainty about exactly when a system will warn); c) the added complexity associated with the integration of multiple warnings systems; and d) a belief that drivers are very unlikely to perform adjustments even when given the opportunity to do so.

Figure 5. Frequency of LDW sensitivity adjustment from the RDCW FOT

Figure 6. Frequency of CSW sensitivity adjustment from the RDCW FOT

A few additional pieces of information were considered and dropped from the current implementation. There is a good possibility that they will be bundled together and shown on the text display to drivers who are interested in additional information and are comfortable with changing the system settings (e.g., young drivers). Those elements of information include: lane boundary detection and availability, distance to detected object for FCW, and reference speed for CSW.

Table 4 shows the resulting light-vehicle option space with the associated warning subsystems (FCW, CSW, LDW, and LCM).

Table 4. Light-vehicle DVI

Warning	Auditory	Haptic	Visual	Driver Adjustments
FCW You are approaching a hazard ahead	Tone from headrest speakers.	Brake pulse	Inform driver which alert occurred, about two seconds after it is over	Drivers can select from three predefined headrest volume levels (low, medium, and high). Temporary warning mute button: a key-press can disable all warnings for two, four, or six minutes.
CSW You are entering a curve too fast
LDW You are unintentionally drifting across a lane boundary (out of lane or off road). Possible object identified as crash threat.	When object identified as crash threat, directional LDW tone from headrest speakers from side where threat exists	Directional haptic vibration in seat pan
LCM The lane you are intentionally entering is hazardous	LCM tone, from side of threat, through headrest speakers		Inform driver which alert occurred, about two seconds after it is over. LCM icon in left and right side view mirror appears in advance of the LCM warning. Blind zone: red. Closing zone: yellow.

3.2.2    Heavy-Truck Option Space

Drivers of light vehicles are generally both the vehicle owner and the vehicle operator; the situation is different in heavy trucks where the driver is often not the owner—instead, a commercial operation is often the purchaser. One consequence of this difference is that in heavy trucks, styling and comfort may play a secondary role to concerns about safety and efficiency. Thus, a warning system is conceived as both a means of implementing a carrier’s safety policy and as a driver support system. When a conflict arises between the two roles, the carrier policy usually prevails. Consequently, heavy-truck warning systems limit the degree to which a driver can control warning characteristics. Notably, limits are typically placed on control over warning sensitivity and sound volume, and strict policies regulating minimum following distances are incorporated into carrier safety policies. Indeed, one reason a fleet would obtain Eaton VORAD systems is to enforce such a following-distance policy.

Another difference between light-vehicle and heavy-truck production involves the degree to which component customization is feasible. Heavy-truck components are typically produced by several different independent suppliers—the engine, transmission, seating, and suspension may come from different suppliers. Moreover, the purchaser is given a great deal of control over the final vehicle configuration. This degree of configuration flexibility imposes some constraints on customization. For example, one cannot integrate a collision warning system employing haptic actuators into a generic seating system without significant customization of the seating. Customization may make little business sense for a seating manufacturer or a collision warning system manufacturer if only a small volume of the combination of components is projected.

As a group, truck drivers are professional drivers—they have significantly more training and spend significantly more time behind the wheel than an average light-vehicle operator. Presumably, heavy-truck drivers may be more capable of managing somewhat more complicated vehicle-based systems than average light-vehicle drivers.

Finally, because of the greater mass of heavy trucks, they are generally less maneuverable than light vehicles—they take significantly more time to stop, their turn radius is large, and their roll threshold is small. Consequently, truck drivers need generally longer lead times for warnings, especially in forward collisions. Unfortunately, extension of warning lead times also increases the likelihood that nuisance warnings will be generated. The dilemma is that if a warning is withheld until a problem is certainly imminent, it may be too late for a driver to effectively respond. If the warning is delivered too early, before a collision is certainly imminent, but in a time window that permits an effective response, the number of nuisance warnings may become problematic. Given these constraints for forward collision warnings, the heavy-truck driver interface has graded forward warnings. Progressively urgent auditory warnings are produced at progressively shorter headways.

3.2.2.1    Warning Presentation Modalities

The heavy-truck platform warning presentation is restricted to auditory presentation via one of three audio channels, and visual presentation via a central display unit ( Figure 7) and two lateral warning displays located near the driver- and passenger-side rearview mirrors ( Figure 8). Haptic warnings (e.g., seat rumble, steering wheel shake, and brake-pulse) were determined to fall outside of the feasible scope of the heavy-truck implementation and have not be implemented as part of IVBSS on the heavy-truck platform.

Figure 7. VORAD driver interface unit

Figure 8. Side sensor display unit

Sound reproduction in the previous generation of the Eaton VORAD system limited the maximum output sound frequency to 2.2 KHz, with an eight-bit dynamic range. This is arguably below the fidelity required to easily distinguish natural sound sources and thus precludes use of auditory icons, or extensive use of sound timbre for auditory warnings. The system speaker size of the VORAD unit (as well as environmental noise in the truck cabin) places a practical lower limit of 500 Hz on the usable auditory frequency.

The resulting heavy-truck option space is shown in Table 5 along with the associated warning condition.

Table 5. Heavy-truck DVI

Warning	Auditory	Visual	Driver Adjustments
FCW You are approaching a hazard ahead	Forward sound source from driver interface unit (DIU)	Red and yellow warning LEDs on DIU, collision warning LCD display on DIU DIU contains indicator LEDs and a monochromatic LCD display	Drivers control sound volume (70 to 90 dB), display brightness Temporary warning mute function: A key-press can disable all warnings for up to six minutes in 120-second increments No sensitivity adjustment is provided to drivers for warnings
LDW You are un-intentionally drifting across a lane boundary	Directional, from side of threat, using lateral speaker channels controlled by DIU	Informational only – e.g., status, availability; drift diagram	Adjustment same as above
LCM The lane you are intentionally entering is hazardous	Directional, from side of threat, using new speakers controlled by DIU	Lateral indicators mounted to A-pillar area. Always visible, directional indicator near each side view mirror	Adjustment same as above

3.3    Research Questions

The goals of the simulator studies on the IVBSS program were to (1) provide information to guide the design of individual IVBSS warnings; (2) determine how warnings should be combined to maximize effectiveness, safety, and acceptability of the system; and (3) provide a better understanding of how warnings should be presented in general. To satisfy these goals, test methods were developed. The application of the test methods and results are largely independent of whether one is working on the heavy-truck or light-vehicle platforms, except as noted.

Admittedly, a significant body of literature on warning design provides a good basis for the development of DVIs for integrated crash warning systems. (See Appendix D.) However, the literature is far from complete, especially from an engineering perspective. Many of the conclusions in the literature are qualitative, for example, indicating that drivers respond more rapidly to louder sounds. Often, however, absolute or percentage differences do not appear in the literature (e.g., specifying by how much brake response time varies between the least meaningful and most meaningful speech message).

In addition, there is very little framework presented in the literature about how drivers respond to warnings other than open-loop or closed-loop concepts (e.g., Abe & Richardson, 2004). A greater understanding of the process of responding to warnings, especially where drivers look and when, should help considerably with selecting warning modalities and content. Do drivers just hear something and respond, or do they look towards a target area, confirm, and respond? Decisions about where and when to provide visual, haptic, and auditory warnings, alone and in combination, will be much better informed as a result of addressing this issue.

When and which multiple warnings might be confused is a core issue of the project and is addressed in several experiments. However, the definitive study cannot be carried out until the end of the sequence of simulator studies when individual warnings have been finalized. Otherwise, one would not know if confusability problems are generic or due to the specific (and potentially suboptimal) set of warnings examined.

Nuisance warnings are one of the problems that plague warning systems. For example, in the ACAS FOT “36 percent of all alerts were of the nuisance type that became triggered by non-threatening, stationary targets” (Ervin, Sayer, LeBlanc, Bogard, Mefford, Hagan, Bareket, & Winkler, 2005). With multiple warnings systems, drivers are likely to experience more nuisance warnings per unit of time, potentially leading to greater annoyance and other consequences. Hence, for most, if not all, of the simulator experiments, subjects experienced a mixture of real, nuisance, and false alarms.

Finally, during the discussions of how IVBSS should be designed, many design and engineering issues could not be resolved based on the existing literature or logic, and as a consequence, led to some of experiments proposed in this project. Balancing the need to support the design of a real system with pursuing basic research questions is always difficult. The resources proposed here are sufficient to answer the major questions and develop a warning set that will satisfy the objective of fielding an effective, safe, and practical integrated crash warning system. In many ways, the questions to be addressed are a mixture of specific implementation questions for IVBSS and more overarching questions about how drivers respond to multiple warnings, individually, collectively, and in close temporal proximity.

The seven questions listed in Table 6 were identified as requiring further attention. Each question is discussed in the following subsections.

Table 6. Research questions mapped to experiments

#	Question	Topic	Exp.	Subsystem
Q1	When and how should warnings be shared or differentiated?	Shared warnings	3	All
Q2	Should warnings occurring at the same time be presented together or with a delay between them?	Sequencing co-occurring warnings	5	All
Q3	Are warnings in the IVBSS sets confused with each other?	Warning set confusion	2, 3, 4, 5	All
Q4	When responding to warnings, what is the process by which drivers respond?	Time course of driver actions	2	FCW, LDW, CSW, maybe LCM
Q5	How does the tradeoff between warning system processing time and warning accuracy affect driver responses to warnings?	Warning processing time-accuracy tradeoff	4	All
Q6	How does auditory warning effectiveness vary with warning sound characteristics (loudness, pitch, speed) in sound environments of each vehicle platform?	Auditory warning characteristics	1	All
Q7	For sounds that involve periods of silence or pauses, are responses deferred to coincide with silence? What is the optimal number of repetitions?	Influence of pauses and repetition	1- subtask 5	LDW

3.3.1    Q1. Shared Warnings

When should warnings be shared or differentiated (e.g., FCW and CSW, LDW, and LCM) and if so, how? How does that depend on factors such as having a common action in response to the warning (brake or slow down, stay in your lane), the collision potential or severity of the outcome (crash target present or absent), and, possibly, the warning reliability and nuisance warning frequency?

Rationale. Should multiple situations, such as approaching a curve too fast or a potential forward collision, have the same warning if the same driver response is required, or should there be multiple (unique) warnings? Using common warnings could shorten response time by removing steps in the process of sensing a warning, deciding how to respond, and executing that response. If drivers simply do what the warning suggests, then a warning should only indicate the response desired. However, if the driver assesses the situation, then indicating what is wrong as part of the warning could shorten the duration of the assessment and response time. The shared warnings question arose both in discussions of the DVI and in the U.S. DOT’s review of the initial proposed experimental plan, and is central to IVBSS. The answer could be specific to a particular warning set or context. This question is distinct from the master caution warning concept explored by Chiang, Llaneras, and Foley (2006) because warnings are linked to specific driver actions, either brake or slow down, or stay in or return to one’s lane.

3.3.2    Q2. Sequencing Co-Occurring Warnings

When sequencing co-occurring warnings:

1. Should only one warning be presented because the second will delay the driver’s response? or
2. Should the second warning be presented with a delay (and what should that delay/lockout be)? or
3. If the second warning is of higher priority, should it preempt the first and, if so, how (fade out the first, immediately start the second, provide delay or lockout and then start, etc.)?

Rationale. A central integration issue is how to sequence multiple warnings and whether one or multiple warnings should be presented. The relevant fundamental concept, the psychological refractory period, asserts that presenting two signals in close temporal proximity (for simple lights and tones, within about 500 ms) can interfere with responding to either of them (Karlin & Kestenbaum, 1968; Wu & Liu, 2004). The impression from Ho, Cummings, Wang, Tijerina, and Kochhar (2006) is that a single master warning leads to performance equivalent to multiple tailored warnings (for FCW, LDW), but drivers prefer tailored warnings. (See also Chiang, Brooks, & Llaneras, 2004.) What clearly emerges from that research is the need for a more detailed look at the specific warnings and the process of how drivers respond to them (e.g., where they look and when). Additional thought is needed concerning the hypothetical situations in which this could occur. Much of that thinking will be guided by an ongoing review of crash statistics.

3.3.3    Q3. Warning Set Confusion

How well do drivers respond to candidate sets of IVBSS warnings? Are any confused or misunderstood?

Rationale. One of the consequences of integrating warnings from independently developed systems is that the warnings can be confused or misunderstood. That concern needs to be addressed before warnings are implemented in a test vehicle.

3.3.4    Q4. Time Course of Driver Actions

What is the time course of driver actions in responding to single and multiple warnings, both when the warnings are unique to the situation and when multiple situations lead to the same warning (such as a common warning for LDW and LCM)? Of particular interest is where drivers look.

Rationale . From previous research, in particular the road departure crash warning (RDCW) simulator experiments (LeBlanc, Sayer, Winkler, Ervin, Bogard, Devonshire, Mefford, Hagan, Bareket, Goodsell, & Gordon, 2006), the project team has some sense of the time course response for steering wheel movements and brake actuations of drivers responding to warnings. Responding to warnings always involves some visual element, either searching for hazards when alerted by warnings or confirming their existence while responding. Thus, an important element of the task is getting drivers to look in the appropriate place and then to execute the desired response. However, where and when drivers look, both pre- and post-warning, for each of the four subsystems being implemented in the IVBSS program, has yet to be fully explored.

Understanding this process should provide insights into the use of shared warning signals (where the driver action is the same), desired warning durations, the need for repetition, and the use of orientation cues.

3.3.5    Q5. Warning Processing Time-Accuracy Tradeoff

How does the tradeoff between warning system processing time (to start to inform the driver) and warning accuracy affect driver responses to warnings?

Rationale. This question has not been addressed in the published literature. In brief, the system integrators wanted to know how long a system could take to process information and warn the driver. Ideally this would be zero, but in reality systems need to sample over time, computers need to process the data, and information needs to be sent over a network to other devices for presentation to the driver. For example, delaying a warning will allow for additional sweeps of the radar system, resulting in higher confidence levels for target detection and fewer false and nuisance warnings. The current time limit, from the beginning of signal processing to the initiation of a warning, is probably on the order of 300 ms (estimations range from 100 to 500 ms). Driver response times are on the order of one to two seconds, and accordingly, this is a 30 percent difference—a nontrivial effect.

3.3.6    Q6. Auditory Characteristics of Warnings

How does auditory warning effectiveness vary with warning sound characteristics (loudness, pitch, and timbre) in sound environments representative of each vehicle platform?

Rationale. To be effective, an auditory warning must be heard and recognized above the din of other sounds that naturally occur during driving. This includes background cabin noise, sounds from the vehicle audio system, and other in-vehicle warning sounds like low-fuel indicators and safety belt reminders. Moreover, warning sounds designed for IVBSS must also be easily learned and unlikely to be confused for each other, and for some warning conditions, they must also be localized. The ultimate goal is to produce warnings that are easily detected and quickly learned, and which produce quick and accurate avoidance responses.

3.3.7    Q7. Influence of Pauses and Repetition

For sounds that involve periods of silence (or pauses), are responses deferred to coincide with silence? What is the optimal number of repetitions?

Rationale. In the RDCW study (LeBlanc et al., 2006) response times were often on the order of a second for some warnings. However, the durations of the auditory warnings were much longer—some lasted as long three seconds. One question, for example, is that for a simulated rumble strip, can the number of simulated strips or the silent period between them be reduced? The silent period is important because the continuing sound of a warning may serve to inhibit a person from responding. In a previous study (Nowakowski, Friedman, & Green, 2001) there was a strong bias toward answering the phone in the silent period of the ringing sequence.

3.4    Driving Simulator Overview

The simulator-based experiments took place after the first major upgrade of the third-generation UMTRI driving simulator (www.umich.edu/~driving/sim.html). The simulator consists of a full-size cab, ten computers, six video projectors, seven cameras, audio equipment, and other electronic devices. The main functions (generation of scene graphics; processing of steering wheel, throttle, and brake inputs; provision of torque feedback; and data collection) were controlled by hardware and software provided by DriveSafety (Vection and HyperDrive Authoring Suite, version