Traffic Injury Prevention
ISSN: 1538-9588 (Print) 1538-957X (Online) Journal homepage: http://www.tandfonline.com/loi/gcpi20
The Real-World Safety Potential of Connected Vehicle Technology Sam Doecke, Alex Grant & Robert W. G. Anderson To cite this article: Sam Doecke, Alex Grant & Robert W. G. Anderson (2015) The Real-World Safety Potential of Connected Vehicle Technology, Traffic Injury Prevention, 16:sup1, S31-S35, DOI: 10.1080/15389588.2015.1014551 To link to this article: http://dx.doi.org/10.1080/15389588.2015.1014551
© Sam Doecke, Alex Grant, and Robert W. G. Anderson. Published with license by Taylor & Francis© Sam Doecke, Alex Grant, and Robert W.G. Anderson. Published online: 01 Jun 2015.
Submit your article to this journal
Article views: 404
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=gcpi20 Download by: [109.161.192.181]
Date: 05 November 2015, At: 23:02
Traffic Injury Prevention (2015) 16, S31–S35 Published with license by Taylor & Francis ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2015.1014551
The Real-World Safety Potential of Connected Vehicle Technology SAM DOECKE1, ALEX GRANT2, and ROBERT W. G. ANDERSON1 1 2
University of Adelaide, Centre for Automotive Safety Research, Adelaide, Australia Cohda Wireless, Adelaide, Australia
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Received 14 November 2014, Accepted 29 January 2015
Objective: This article estimates the safety potential of a current commercially available connected vehicle technology in real-world crashes. Method: Data from the Centre for Automotive Safety Research’s at-scene in-depth crash investigations in South Australia were used to simulate the circumstances of real-world crashes. A total of 89 crashes were selected for inclusion in the study. The crashes were selected as representative of the most prevalent crash types for injury or fatal crashes and had potential to be mitigated by connected vehicle technology. The trajectory, speeds, braking, and impact configuration of the selected in-depth cases were replicated in a software package and converted to a file format allowing “replay” of the scenario in real time as input to 2 Cohda Wireless MK2 onboard units. The Cohda Wireless onboard units are a mature connected vehicle technology that has been used in both the German simTD field trial and the U.S. Department of Transport’s Safety Pilot project and have been tuned for low false alarm rates when used in the real world. The crash replay was achieved by replacing each of the onboard unit Global Positioning System (GPS) inputs with the simulated data of each of the involved vehicles. The time at which the Cohda Wireless threat detection software issued an elevated warning was used to calculate a new impact speed using 3 different reaction scenarios and 2 levels of braking. Results: It was found that between 37 and 86% of the simulated crashes could be avoided, with highest percentage due a fully autonomous system braking at 0.7 g. The same system also reduced the impact speed relative to the actual crash in all cases. Even when a human reaction time of 1.2 s and moderate braking of 0.4 g was assumed, the impact speed was reduced in 78% of the crashes. Crash types that proved difficult for the threat detection engine were head-on crashes where the approach angle was low and right turn–opposite crashes. Conclusions: These results indicate that connected vehicle technology can be greatly beneficial in real-world crash scenarios and that this benefit would be maximized by having the vehicle intervene autonomously with heavy braking. The crash types that proved difficult for the connected vehicle technology could be better addressed if controller area network (CAN) information is available, such as steering wheel angle, so that driver intent can be inferred sooner. More accurate positioning in the real world (e.g., combining satellite positioning and accelerometer data) would allow the technology to be more effective for near-collinear head-on and rear-end crashes, because the low approach angles that are common in such crashes are currently ignored in order to minimize false alarms due to positioning uncertainty. Keywords: accident avoidance, intelligent transport system (ITS), vehicle to vehicle communications (V2V)
Introduction For well over a decade, technology has been under development to allow vehicles to send and receive information to and from one another, other road users, and infrastructure. Many uses for such technology have been conceptualized, including primary safety applications. The information sent and received for safety applications, referred to as the basic safety message, includes position, positional accuracy, speed, heading, © Sam Doecke, Alex Grant, and Robert W. G. Anderson Address correspondence to Sam Doecke, University of Adelaide, Centre for Automotive Safety Research, Adelaide 5005 Australia. E-mail: [email protected]
accelerations, and yaw rate (Ahmed-Zaid et al. 2011). This emerging technology is known by several names, including vehicle-to-vehicle communication, vehicle-to-infrastructure communication, V2X (when generally considering communication between vehicles and another entity), car-to-car, and the associated C2I and C2X acronyms, intervehicle communication, cooperative driving, and the term used in this article, connected vehicles. The exchange of information between connected vehicles can be used to detect risks or potential collisions and could be used to trigger a vehicle response such as providing a warning to the driver and/or to autonomously intervene using the vehicle’s braking or steering systems. Previous research by 2 of the authors found that connected vehicles could reduce injury crashes by 43 to 55% and fatal crashes by 31 to 37% (Doecke
32
Doecke et al.
Table 1. Number of crashes simulated by crash type and speed zone group Speed zone group Crash type
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Rear end Right angle Head on Right turn–adjacent Right turn–opposite Total
50 and 60 km/h
70, 80, and 90 km/h
100 and 110 km/h
Total
10 10 6 9 9 44
1 4 3 2 4 14
2 9 10 10 NA 31
13 23 19 21 13 89
and Anderson 2014); however, this was based on basic threat detection code developed by the researchers that had not been developed to avoid false alarms in the real world. The U.S. Department of Transport found similarly high crash reductions for 2 specific connected vehicle applications. Intersection movement assist was found to help drivers avoid 41 to 55% of target intersection crashes and left turn assist (equivalent to right turns in countries like Australia where vehicles drive on the left) would have prevented 36 to 62% of crashes (Harding et al. 2014). These estimates were based on simulation models and driving simulations using input from real-world situations. The purpose of the research detailed in this article was to estimate the safety potential of a current, commercially available connected vehicle technology in real-world crashes using threat detection software developed for real-world applications.
Methods At-scene in-depth crash investigations can provide a much more complete picture of the circumstances of a crash than is available in mass crash data derived from police reports. The Centre for Automotive Safety Research has been conducting such investigations for over 4 decades. The data used in this article were limited to investigations taking place between 1995 and 2011. During this time, 364 crashes had been reconstructed so that travel and impact speeds were known, information that is required to simulate the crash. Crash types that were most prevalent in injury or fatal crashes and can be addressed by connected vehicle technology were selected for inclusion. These crash types were rear end, right angle, head-on, right turn–adjacent (vehicles originally traveling on different roads), and right turn–opposite crashes (vehicles originally traveling on the same road, in opposing directions). Pedestrian crashes may be able to be addressed by connected vehicles in the future but were not included in this analysis. The objective was to simulate 10 crashes from each crash type within a speed zone group. This proved particularly difficult in the less common speed zones of 70, 80, and 90 km/h. A total of 89 crashes were chosen for simulation. The number of cases in each crash type can be seen in Table 1.
The trajectory, speeds, braking distance, and impact configuration determined in the original reconstructions of the selected in-depth cases were replicated in software called PreScan, developed by TASS (TASS International, Helmond, The Netherlands). PreScan is a simulation environment for primary safety technologies, and many types of sensor can be simulated, as can the response of a vehicle to sensor signals in many types of environmental conditions. Though the PreScan software is capable of performing very detailed simulations of advanced driver assistance systems, these capabilities were not used in this study. Rather, PreScan was used to generate time-based position data for both vehicles based on the original reconstruction conducted around the time the crash was investigated. An example of how an in-depth crash investigation case was replicated in the PreScan software is shown in Figure 1. The site diagram from the original crash is shown on the left and same scenario replicated in PreScan is shown on the right. The colored lines in PreScan diagram represent the trajectories of the vehicles and the spacing of the colored symbols represents the speed of the vehicle. The onboard units (OBUs) used for the experiment were provided by Cohda Wireless (Cohda Wireless, North Adelaide, Australia). This was the technology used for the German simTD trial and for approximately half of the fleet in the SafetyPilot deployment. These OBUs operate in the 5.9 GHz band and are compliant with the IEEE802.11p wireless standard and the IEEE 1609 networking standards. Position determination for these units is provided by the Global Positioning System (GPS). The OBU software suite for the units under test included the complete IEEE 802.11p/1609 communications stack and the Cohda Wireless v2xlib facilities layer. The facilities layer provides software APIs for local vehicle (vehicle where the OBU is installed) position and remote vehicle (other vehicles) position. The OBUs were running the Cohda Wireless Aftermarket Safety Device (ASD) software application (as used in SafetyPilot). The ASD software includes a threat detection engine that determines different levels of driver warning based on tracking and predicting the trajectory of the local vehicle and all nearby remote vehicles. Three threat levels are generated by the threat detection engine. A level 1 warning is issued when a collision is predicted such that “normal” driver behavior or modest braking is sufficient for avoidance. In real-world operation, level 1 warnings are not passed on to the driver. Level 2 warnings are issued to the driver when more urgent braking would be required. Level 3 driver warnings indicate the need to brake at vehicle limits (appropriate to the class of vehicle). The output of the PreScan software trajectory reconstructions was used to produce synthetic pcap files for input to the OBUs. The pcap file is a format normally used to capture detailed output logs from an onboard unit for analysis purposes. For example, pcap files were used in SafetyPilot to collect data from the trial. A pcap replay application allows a pcap file to be used as input for an OBU, replacing the GPS input with the contents of the pcap file. The pcap replay application (with appropriate synchronization) was used to independently provide inputs to 2 OBUs.
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Connected Vehicle Technology
33
Fig. 1. Original site diagram (left) and PreScan simulation (right).
Other than the replaced GPS data, the OBUs operated normally. The position fix was “perfect”; no noise or other disturbances were introduced to the simulated GPS data. No attempt was made to simulate the actual physical radio environment from the crash scene, and the units were connected with an RF cable. Previous extensive on-road testing of the radio system performance of these OBUs was reported by Alexander et al. (2011). The ASD application generated standard SAE J2735 Basic Safety Messages at 10 messages per second. These were transmitted over the full communications stack (1609.1 resource manager, 1609.3 short message protocol, 802.2 link layer control, 1609.4 multichannel control, and 802.11p multiple access control and physical layer). The experiments were recorded by capturing pcap log files. These files were postprocessed to output the distance from the original collision point at which a level 2 threat would have been issued to the driver. This distance was used to calculate a new impact speed for the striking vehicle using 3 different reaction scenarios and 2 levels of braking. It was assumed that only the striking vehicle had a full collision avoidance system and the struck vehicle was simply sending a basic safety message. In a head on crash the nonerrant vehicle was considered the striking vehicle. The reaction times used were a 1.2 s reaction time to the warning, a 0.7 s reaction time to the warning, and an autonomous system that begins braking as soon as the warning is issued. The reaction times were based on the best reaction times found by Green (2000) and Mohebbi et al. (2009) and the reaction time used in the European New Car Assessment Programme Test Protocol for autonomous emergency braking (European New Car Assessment Programme 2013). The levels of braking used were 0.4 and 0.7 g. The lower braking level of 0.4 g was chosen as a moderate braking reaction, above gentle braking but below full braking. The higher level of braking was chosen for 2 reasons. The main reason was
to match the level of braking used in the original reconstructions. In the original crash reconstructions, braking could only be known from skid marks and the general braking level for full skidding was found to be 0.7 g for dry roads in the region that crashes occurred. The second reason for choosing 0.7 g was to represent a level of braking that was high but not fully optimized, such as a human might achieve without the aid of emergency brake assist.
Results The number of crashes where the impact speed was reduced relative to the actual crash, the number of crashes avoided completely, and the average impact speed by reaction time and braking level are shown in Table 2. A fully autonomous system braking at 0.7 g was found to always reduce the impact speed compared to the driver’s response in the actual crash. The scenario with a reaction time of 1.2 s and a braking level of 0.4 g still reduced the impact speed in 69 of the 89 crashes (78%) compared to the drivers’ responses. If the striking vehicle was found to have been able to come to a complete stop before the impact point the crash was considered avoided. Between 37 and 77% of the crashes were avoided with a connected vehicle system. The average initial speed of the vehicles in the crashes was 72.3 km/h. The drivers in the actual crashes were able to reduce their impact speed to an average of 59.3 km/h. The simulations found that with a connected vehicle system fitted the impact speed was reduced much further to between 4.4 and 35.2 km/h. Figures 2, 3, and 4 show the results from the individual cases for the 3 different speed zone groups by reaction time and crash type for a braking level of 0.7 g. For 50 and 60 km/h zones crash types that proved most challenging were head-on and right turn–opposite crashes. Such crashes were often not able to be avoided by even the autonomously intervening system. In 70, 80, and 90 km/h zones right turn–opposite crashes proved
34
Doecke et al.
Table 2. Impact speed reduced, crashes avoided, and average impact speed with connected vehiclesa 1.2 s reaction
Impact speed reduced Crash avoided Average impact speed (km/h)
0.7 s reaction
Autonomous
Braking level (g)
Actual crash
Number
Percentage
Number
Percentage
Number
Percentage
0.4 0.7 0.4 0.7 0.4 0.7
0 0 0 0 59.3 59.3
69 77 33 51 35.2 21.6
77.5 86.5 37.1 57.3 — —
73 83 43 68 27.5 12.4
82.0 93.3 48.3 76.4 — —
81 89 53 77 17.0 4.4
91.0 100.0 59.6 86.5 — —
aThe
Downloaded by [109.161.192.181] at 23:02 05 November 2015
average initial speed in the crashes was 72.3 km/h, meaning that drivers without any connected vehicles system were able to reduce their speed by an average of 13 km/h before impact.
challenging, as in the lower speed zones, but head-on crashes did not pose a challenge at all, though the low number of crashes simulated from the 70, 80, and 90 km/h zones should be kept in mind. In the high-speed zones (100 and 110 km/h) many of the right turn–adjacent crashes were challenging for the systems that warned the driver rather than intervening autonomously. However, in these cases the speed of the vehicle was still reduced considerably from the initial speed. Some head-on crashes in high-speed zones were challenging to mitigate, though many were also completely avoided.
Fig. 2. Impact speed of crashes in 50 and 60 km/h zones with 0.7 g braking by reaction time.
Fig. 3. Impact speed of crashes in 70, 80, and 90 km/h zones with 0.7 g braking by reaction time.
Discussion The experiments reveal some real-world crash types, which are particularly challenging. These consist of head-on approaches with late, low-angle deviations and similar for approaches from the rear. The current threat detection engine is tuned to provide a very low false alarm rate from real-world noisy GPS data. As a result, some heading “wobble” is filtered out. The results of our experiments reveal that some real-world crashes begin with parameters within these false alarm thresholds. This can, however, be addressed by the availability of more accurate sources of position data via a combination of multi-global navigation satellite system (GNSS), accelerometer-based dead reckoning, and other data available from the CAN bus, such as steering wheel angle and rate of change. The sample of 89 crashes was chosen to examine the effect of connected vehicle technology by crash type and speed zone, not to be representative. This should be borne in mind when considering the results across the whole sample. The actual braking level used by the driver will depend on his reaction to the warning or the design of the autonomous intervention. It is also limited by the peak coefficient of friction developed between the tire and road surface. Unpublished tests conducted by the Centre for Automotive Safety Research found that a sensor-based autonomous emergency braking system used a staged braking level. The initial braking level of about 0.3 g was increased to 0.95 g, the full capacity of the antilock braking system on the dry test road. It may be desirable for manufacturers to implement such a staged response in autonomous systems to reduce the annoyance of any false alarms. If full antilock braking at around 0.95 g (dry roads of average quality) is used in an autonomously intervening system, even greater benefits than found in this research could be realized. Drivers’ actual reaction times to warnings will vary and could depend on factors such as age, level of alertness, or intoxication. The driver will also have the ability to swerve as well as, or instead of, braking. This could have both positive and negative effects on crash avoidance. The best results in this study were found with fully autonomous systems. Such systems should be the ultimate goal, but user acceptance of them will be dependent on achieving a very low level of false alarms. No attempt was made to simulate the actual physical radio environment from the crash scene. Real-world, scenario-
Connected Vehicle Technology
35 or the vehicles missing each other entirely. However, it should be noted that the assumption of a crash being avoided if the striking vehicle’s speed reached zero might not be true for head-on crashes. Further, if the struck vehicle had a connected vehicle collision warning system as well as the striking vehicle it is likely that more crashes could have been avoided. The results of this study are very encouraging, especially because they are most likely conservative. They further reinforce the need to have connected vehicle technology in the vehicle fleet as soon as possible.
Acknowledgments
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Fig. 4. Impact speed of crashes in 100 and 110 km/h zones with 0.7 g braking by reaction time.
based radio system performance results have been previously reported by Alexander et al. (2011). Other than replacing the GPS signal with the simulated trajectories, the onboard units functioned as they would in a typical installation, employing industry standard communications protocols and messaging formats. PreScan trajectory reconstruction was only run for approximately 100 m prior to the original crash point. However, as reported in Alexander et al. (2011), the Cohda Wireless OBUs routinely establish reliable radio connection at longer ranges than this. When running the experiments we noticed that in some scenarios the threat detection engine went immediately into level 2 or 3 warning. This indicates that future experiments should use a greater “lead-in” distance. Accordingly, the results presented in this article may be considered conservative. The speed of the striking car was the only result examined. Changes in the relative positions of the vehicles were not taken into account. Earlier braking caused by the connected vehicle system could have resulted in a different impact configuration
The authors gratefully acknowledge the assistance of Dr. Paul Alexander and Mark Yan from Cohda Wireless.
References Ahmed-Zaid F, Bai F, Bai S, et al. Vehicle Safety Communications—Applications (VSC-A) Final Report. Washington, DC: NHTSA; 2011. DOT HS 811 492A. Alexander P, Haley D, Grant A. Cooperative intelligent transport systems: 5.9-GHz field trials. Proc IEEE. 2011;99:1213–1235. Doecke SD, Anderson RW G. The safety potential of connected vehicles. Paper presented at: Australasian Road Safety Research, Policing and Education Conference; November 12–14, 2014; Melbourne, Australia. European New Car Assessment Programme. Test Protocol—AEB systems. Version 1.0. Brussels, Belgium: Author; 2013. Green M. “How long does it take to stop?” Methodological analysis of driver perception–brake times. Transp Hum Factors. 2000;2(3):195–216. Harding J, Powell GR, Yoon R, et al. Vehicle-to-Vehicle Communications: Readiness of V2V Technology for Application. Washington, DC: NHTSA; 2014. Report No. DOT HS 812 014. Mohebbi R, Gray R, Tan H. Driver reaction time to tactile and auditory rear-end collision warnings while talking on a cell phone. Hum Factors. 2009;51:102–110.
ISSN: 1538-9588 (Print) 1538-957X (Online) Journal homepage: http://www.tandfonline.com/loi/gcpi20
The Real-World Safety Potential of Connected Vehicle Technology Sam Doecke, Alex Grant & Robert W. G. Anderson To cite this article: Sam Doecke, Alex Grant & Robert W. G. Anderson (2015) The Real-World Safety Potential of Connected Vehicle Technology, Traffic Injury Prevention, 16:sup1, S31-S35, DOI: 10.1080/15389588.2015.1014551 To link to this article: http://dx.doi.org/10.1080/15389588.2015.1014551
© Sam Doecke, Alex Grant, and Robert W. G. Anderson. Published with license by Taylor & Francis© Sam Doecke, Alex Grant, and Robert W.G. Anderson. Published online: 01 Jun 2015.
Submit your article to this journal
Article views: 404
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=gcpi20 Download by: [109.161.192.181]
Date: 05 November 2015, At: 23:02
Traffic Injury Prevention (2015) 16, S31–S35 Published with license by Taylor & Francis ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2015.1014551
The Real-World Safety Potential of Connected Vehicle Technology SAM DOECKE1, ALEX GRANT2, and ROBERT W. G. ANDERSON1 1 2
University of Adelaide, Centre for Automotive Safety Research, Adelaide, Australia Cohda Wireless, Adelaide, Australia
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Received 14 November 2014, Accepted 29 January 2015
Objective: This article estimates the safety potential of a current commercially available connected vehicle technology in real-world crashes. Method: Data from the Centre for Automotive Safety Research’s at-scene in-depth crash investigations in South Australia were used to simulate the circumstances of real-world crashes. A total of 89 crashes were selected for inclusion in the study. The crashes were selected as representative of the most prevalent crash types for injury or fatal crashes and had potential to be mitigated by connected vehicle technology. The trajectory, speeds, braking, and impact configuration of the selected in-depth cases were replicated in a software package and converted to a file format allowing “replay” of the scenario in real time as input to 2 Cohda Wireless MK2 onboard units. The Cohda Wireless onboard units are a mature connected vehicle technology that has been used in both the German simTD field trial and the U.S. Department of Transport’s Safety Pilot project and have been tuned for low false alarm rates when used in the real world. The crash replay was achieved by replacing each of the onboard unit Global Positioning System (GPS) inputs with the simulated data of each of the involved vehicles. The time at which the Cohda Wireless threat detection software issued an elevated warning was used to calculate a new impact speed using 3 different reaction scenarios and 2 levels of braking. Results: It was found that between 37 and 86% of the simulated crashes could be avoided, with highest percentage due a fully autonomous system braking at 0.7 g. The same system also reduced the impact speed relative to the actual crash in all cases. Even when a human reaction time of 1.2 s and moderate braking of 0.4 g was assumed, the impact speed was reduced in 78% of the crashes. Crash types that proved difficult for the threat detection engine were head-on crashes where the approach angle was low and right turn–opposite crashes. Conclusions: These results indicate that connected vehicle technology can be greatly beneficial in real-world crash scenarios and that this benefit would be maximized by having the vehicle intervene autonomously with heavy braking. The crash types that proved difficult for the connected vehicle technology could be better addressed if controller area network (CAN) information is available, such as steering wheel angle, so that driver intent can be inferred sooner. More accurate positioning in the real world (e.g., combining satellite positioning and accelerometer data) would allow the technology to be more effective for near-collinear head-on and rear-end crashes, because the low approach angles that are common in such crashes are currently ignored in order to minimize false alarms due to positioning uncertainty. Keywords: accident avoidance, intelligent transport system (ITS), vehicle to vehicle communications (V2V)
Introduction For well over a decade, technology has been under development to allow vehicles to send and receive information to and from one another, other road users, and infrastructure. Many uses for such technology have been conceptualized, including primary safety applications. The information sent and received for safety applications, referred to as the basic safety message, includes position, positional accuracy, speed, heading, © Sam Doecke, Alex Grant, and Robert W. G. Anderson Address correspondence to Sam Doecke, University of Adelaide, Centre for Automotive Safety Research, Adelaide 5005 Australia. E-mail: [email protected]
accelerations, and yaw rate (Ahmed-Zaid et al. 2011). This emerging technology is known by several names, including vehicle-to-vehicle communication, vehicle-to-infrastructure communication, V2X (when generally considering communication between vehicles and another entity), car-to-car, and the associated C2I and C2X acronyms, intervehicle communication, cooperative driving, and the term used in this article, connected vehicles. The exchange of information between connected vehicles can be used to detect risks or potential collisions and could be used to trigger a vehicle response such as providing a warning to the driver and/or to autonomously intervene using the vehicle’s braking or steering systems. Previous research by 2 of the authors found that connected vehicles could reduce injury crashes by 43 to 55% and fatal crashes by 31 to 37% (Doecke
32
Doecke et al.
Table 1. Number of crashes simulated by crash type and speed zone group Speed zone group Crash type
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Rear end Right angle Head on Right turn–adjacent Right turn–opposite Total
50 and 60 km/h
70, 80, and 90 km/h
100 and 110 km/h
Total
10 10 6 9 9 44
1 4 3 2 4 14
2 9 10 10 NA 31
13 23 19 21 13 89
and Anderson 2014); however, this was based on basic threat detection code developed by the researchers that had not been developed to avoid false alarms in the real world. The U.S. Department of Transport found similarly high crash reductions for 2 specific connected vehicle applications. Intersection movement assist was found to help drivers avoid 41 to 55% of target intersection crashes and left turn assist (equivalent to right turns in countries like Australia where vehicles drive on the left) would have prevented 36 to 62% of crashes (Harding et al. 2014). These estimates were based on simulation models and driving simulations using input from real-world situations. The purpose of the research detailed in this article was to estimate the safety potential of a current, commercially available connected vehicle technology in real-world crashes using threat detection software developed for real-world applications.
Methods At-scene in-depth crash investigations can provide a much more complete picture of the circumstances of a crash than is available in mass crash data derived from police reports. The Centre for Automotive Safety Research has been conducting such investigations for over 4 decades. The data used in this article were limited to investigations taking place between 1995 and 2011. During this time, 364 crashes had been reconstructed so that travel and impact speeds were known, information that is required to simulate the crash. Crash types that were most prevalent in injury or fatal crashes and can be addressed by connected vehicle technology were selected for inclusion. These crash types were rear end, right angle, head-on, right turn–adjacent (vehicles originally traveling on different roads), and right turn–opposite crashes (vehicles originally traveling on the same road, in opposing directions). Pedestrian crashes may be able to be addressed by connected vehicles in the future but were not included in this analysis. The objective was to simulate 10 crashes from each crash type within a speed zone group. This proved particularly difficult in the less common speed zones of 70, 80, and 90 km/h. A total of 89 crashes were chosen for simulation. The number of cases in each crash type can be seen in Table 1.
The trajectory, speeds, braking distance, and impact configuration determined in the original reconstructions of the selected in-depth cases were replicated in software called PreScan, developed by TASS (TASS International, Helmond, The Netherlands). PreScan is a simulation environment for primary safety technologies, and many types of sensor can be simulated, as can the response of a vehicle to sensor signals in many types of environmental conditions. Though the PreScan software is capable of performing very detailed simulations of advanced driver assistance systems, these capabilities were not used in this study. Rather, PreScan was used to generate time-based position data for both vehicles based on the original reconstruction conducted around the time the crash was investigated. An example of how an in-depth crash investigation case was replicated in the PreScan software is shown in Figure 1. The site diagram from the original crash is shown on the left and same scenario replicated in PreScan is shown on the right. The colored lines in PreScan diagram represent the trajectories of the vehicles and the spacing of the colored symbols represents the speed of the vehicle. The onboard units (OBUs) used for the experiment were provided by Cohda Wireless (Cohda Wireless, North Adelaide, Australia). This was the technology used for the German simTD trial and for approximately half of the fleet in the SafetyPilot deployment. These OBUs operate in the 5.9 GHz band and are compliant with the IEEE802.11p wireless standard and the IEEE 1609 networking standards. Position determination for these units is provided by the Global Positioning System (GPS). The OBU software suite for the units under test included the complete IEEE 802.11p/1609 communications stack and the Cohda Wireless v2xlib facilities layer. The facilities layer provides software APIs for local vehicle (vehicle where the OBU is installed) position and remote vehicle (other vehicles) position. The OBUs were running the Cohda Wireless Aftermarket Safety Device (ASD) software application (as used in SafetyPilot). The ASD software includes a threat detection engine that determines different levels of driver warning based on tracking and predicting the trajectory of the local vehicle and all nearby remote vehicles. Three threat levels are generated by the threat detection engine. A level 1 warning is issued when a collision is predicted such that “normal” driver behavior or modest braking is sufficient for avoidance. In real-world operation, level 1 warnings are not passed on to the driver. Level 2 warnings are issued to the driver when more urgent braking would be required. Level 3 driver warnings indicate the need to brake at vehicle limits (appropriate to the class of vehicle). The output of the PreScan software trajectory reconstructions was used to produce synthetic pcap files for input to the OBUs. The pcap file is a format normally used to capture detailed output logs from an onboard unit for analysis purposes. For example, pcap files were used in SafetyPilot to collect data from the trial. A pcap replay application allows a pcap file to be used as input for an OBU, replacing the GPS input with the contents of the pcap file. The pcap replay application (with appropriate synchronization) was used to independently provide inputs to 2 OBUs.
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Connected Vehicle Technology
33
Fig. 1. Original site diagram (left) and PreScan simulation (right).
Other than the replaced GPS data, the OBUs operated normally. The position fix was “perfect”; no noise or other disturbances were introduced to the simulated GPS data. No attempt was made to simulate the actual physical radio environment from the crash scene, and the units were connected with an RF cable. Previous extensive on-road testing of the radio system performance of these OBUs was reported by Alexander et al. (2011). The ASD application generated standard SAE J2735 Basic Safety Messages at 10 messages per second. These were transmitted over the full communications stack (1609.1 resource manager, 1609.3 short message protocol, 802.2 link layer control, 1609.4 multichannel control, and 802.11p multiple access control and physical layer). The experiments were recorded by capturing pcap log files. These files were postprocessed to output the distance from the original collision point at which a level 2 threat would have been issued to the driver. This distance was used to calculate a new impact speed for the striking vehicle using 3 different reaction scenarios and 2 levels of braking. It was assumed that only the striking vehicle had a full collision avoidance system and the struck vehicle was simply sending a basic safety message. In a head on crash the nonerrant vehicle was considered the striking vehicle. The reaction times used were a 1.2 s reaction time to the warning, a 0.7 s reaction time to the warning, and an autonomous system that begins braking as soon as the warning is issued. The reaction times were based on the best reaction times found by Green (2000) and Mohebbi et al. (2009) and the reaction time used in the European New Car Assessment Programme Test Protocol for autonomous emergency braking (European New Car Assessment Programme 2013). The levels of braking used were 0.4 and 0.7 g. The lower braking level of 0.4 g was chosen as a moderate braking reaction, above gentle braking but below full braking. The higher level of braking was chosen for 2 reasons. The main reason was
to match the level of braking used in the original reconstructions. In the original crash reconstructions, braking could only be known from skid marks and the general braking level for full skidding was found to be 0.7 g for dry roads in the region that crashes occurred. The second reason for choosing 0.7 g was to represent a level of braking that was high but not fully optimized, such as a human might achieve without the aid of emergency brake assist.
Results The number of crashes where the impact speed was reduced relative to the actual crash, the number of crashes avoided completely, and the average impact speed by reaction time and braking level are shown in Table 2. A fully autonomous system braking at 0.7 g was found to always reduce the impact speed compared to the driver’s response in the actual crash. The scenario with a reaction time of 1.2 s and a braking level of 0.4 g still reduced the impact speed in 69 of the 89 crashes (78%) compared to the drivers’ responses. If the striking vehicle was found to have been able to come to a complete stop before the impact point the crash was considered avoided. Between 37 and 77% of the crashes were avoided with a connected vehicle system. The average initial speed of the vehicles in the crashes was 72.3 km/h. The drivers in the actual crashes were able to reduce their impact speed to an average of 59.3 km/h. The simulations found that with a connected vehicle system fitted the impact speed was reduced much further to between 4.4 and 35.2 km/h. Figures 2, 3, and 4 show the results from the individual cases for the 3 different speed zone groups by reaction time and crash type for a braking level of 0.7 g. For 50 and 60 km/h zones crash types that proved most challenging were head-on and right turn–opposite crashes. Such crashes were often not able to be avoided by even the autonomously intervening system. In 70, 80, and 90 km/h zones right turn–opposite crashes proved
34
Doecke et al.
Table 2. Impact speed reduced, crashes avoided, and average impact speed with connected vehiclesa 1.2 s reaction
Impact speed reduced Crash avoided Average impact speed (km/h)
0.7 s reaction
Autonomous
Braking level (g)
Actual crash
Number
Percentage
Number
Percentage
Number
Percentage
0.4 0.7 0.4 0.7 0.4 0.7
0 0 0 0 59.3 59.3
69 77 33 51 35.2 21.6
77.5 86.5 37.1 57.3 — —
73 83 43 68 27.5 12.4
82.0 93.3 48.3 76.4 — —
81 89 53 77 17.0 4.4
91.0 100.0 59.6 86.5 — —
aThe
Downloaded by [109.161.192.181] at 23:02 05 November 2015
average initial speed in the crashes was 72.3 km/h, meaning that drivers without any connected vehicles system were able to reduce their speed by an average of 13 km/h before impact.
challenging, as in the lower speed zones, but head-on crashes did not pose a challenge at all, though the low number of crashes simulated from the 70, 80, and 90 km/h zones should be kept in mind. In the high-speed zones (100 and 110 km/h) many of the right turn–adjacent crashes were challenging for the systems that warned the driver rather than intervening autonomously. However, in these cases the speed of the vehicle was still reduced considerably from the initial speed. Some head-on crashes in high-speed zones were challenging to mitigate, though many were also completely avoided.
Fig. 2. Impact speed of crashes in 50 and 60 km/h zones with 0.7 g braking by reaction time.
Fig. 3. Impact speed of crashes in 70, 80, and 90 km/h zones with 0.7 g braking by reaction time.
Discussion The experiments reveal some real-world crash types, which are particularly challenging. These consist of head-on approaches with late, low-angle deviations and similar for approaches from the rear. The current threat detection engine is tuned to provide a very low false alarm rate from real-world noisy GPS data. As a result, some heading “wobble” is filtered out. The results of our experiments reveal that some real-world crashes begin with parameters within these false alarm thresholds. This can, however, be addressed by the availability of more accurate sources of position data via a combination of multi-global navigation satellite system (GNSS), accelerometer-based dead reckoning, and other data available from the CAN bus, such as steering wheel angle and rate of change. The sample of 89 crashes was chosen to examine the effect of connected vehicle technology by crash type and speed zone, not to be representative. This should be borne in mind when considering the results across the whole sample. The actual braking level used by the driver will depend on his reaction to the warning or the design of the autonomous intervention. It is also limited by the peak coefficient of friction developed between the tire and road surface. Unpublished tests conducted by the Centre for Automotive Safety Research found that a sensor-based autonomous emergency braking system used a staged braking level. The initial braking level of about 0.3 g was increased to 0.95 g, the full capacity of the antilock braking system on the dry test road. It may be desirable for manufacturers to implement such a staged response in autonomous systems to reduce the annoyance of any false alarms. If full antilock braking at around 0.95 g (dry roads of average quality) is used in an autonomously intervening system, even greater benefits than found in this research could be realized. Drivers’ actual reaction times to warnings will vary and could depend on factors such as age, level of alertness, or intoxication. The driver will also have the ability to swerve as well as, or instead of, braking. This could have both positive and negative effects on crash avoidance. The best results in this study were found with fully autonomous systems. Such systems should be the ultimate goal, but user acceptance of them will be dependent on achieving a very low level of false alarms. No attempt was made to simulate the actual physical radio environment from the crash scene. Real-world, scenario-
Connected Vehicle Technology
35 or the vehicles missing each other entirely. However, it should be noted that the assumption of a crash being avoided if the striking vehicle’s speed reached zero might not be true for head-on crashes. Further, if the struck vehicle had a connected vehicle collision warning system as well as the striking vehicle it is likely that more crashes could have been avoided. The results of this study are very encouraging, especially because they are most likely conservative. They further reinforce the need to have connected vehicle technology in the vehicle fleet as soon as possible.
Acknowledgments
Downloaded by [109.161.192.181] at 23:02 05 November 2015
Fig. 4. Impact speed of crashes in 100 and 110 km/h zones with 0.7 g braking by reaction time.
based radio system performance results have been previously reported by Alexander et al. (2011). Other than replacing the GPS signal with the simulated trajectories, the onboard units functioned as they would in a typical installation, employing industry standard communications protocols and messaging formats. PreScan trajectory reconstruction was only run for approximately 100 m prior to the original crash point. However, as reported in Alexander et al. (2011), the Cohda Wireless OBUs routinely establish reliable radio connection at longer ranges than this. When running the experiments we noticed that in some scenarios the threat detection engine went immediately into level 2 or 3 warning. This indicates that future experiments should use a greater “lead-in” distance. Accordingly, the results presented in this article may be considered conservative. The speed of the striking car was the only result examined. Changes in the relative positions of the vehicles were not taken into account. Earlier braking caused by the connected vehicle system could have resulted in a different impact configuration
The authors gratefully acknowledge the assistance of Dr. Paul Alexander and Mark Yan from Cohda Wireless.
References Ahmed-Zaid F, Bai F, Bai S, et al. Vehicle Safety Communications—Applications (VSC-A) Final Report. Washington, DC: NHTSA; 2011. DOT HS 811 492A. Alexander P, Haley D, Grant A. Cooperative intelligent transport systems: 5.9-GHz field trials. Proc IEEE. 2011;99:1213–1235. Doecke SD, Anderson RW G. The safety potential of connected vehicles. Paper presented at: Australasian Road Safety Research, Policing and Education Conference; November 12–14, 2014; Melbourne, Australia. European New Car Assessment Programme. Test Protocol—AEB systems. Version 1.0. Brussels, Belgium: Author; 2013. Green M. “How long does it take to stop?” Methodological analysis of driver perception–brake times. Transp Hum Factors. 2000;2(3):195–216. Harding J, Powell GR, Yoon R, et al. Vehicle-to-Vehicle Communications: Readiness of V2V Technology for Application. Washington, DC: NHTSA; 2014. Report No. DOT HS 812 014. Mohebbi R, Gray R, Tan H. Driver reaction time to tactile and auditory rear-end collision warnings while talking on a cell phone. Hum Factors. 2009;51:102–110.
