EUROPEAN COMMISSION DG RTD ASPECSS

EUROPEAN COMMISSION DG RTD SEVENTH FRAMEWORK PROGRAMME THEME 7 TRANSPORT - SST SST.2011.RTD-1 GA No. 285106 ASPECSS Assessment methodologies for forward looking Integrated Pedestrian and further extension to Cyclists Safety Deliverable No. ASPECSS D2.5 Deliverable Title Dissemination level Written By Test procedure including test target and misuse tests, based on test results Patrick Seiniger (BASt), Oliver Bartels (BASt), Martin Kunert (Bosch), Thomas Schaller (BMW) 21.5.14 Checked by Thomas Schaller (BMW) 12.6.14 Approved by Monica Pla (IDIADA) 12.6.14

Executive summary Test conducted within the ASPECSS project framework withseveral series and developmentvehicles showed that many of the accident scenarios, being reflected in the AsPeCSS test scenario catalogue, are already addressed by some state-of-the-art AEB systems.test tools and methodologies established so far do have the potential to achieve the necessary repeatability. However, the original test scenarios, as derived from accident data, do not seem to show significant differences between the tested vehicles since they are either too easy or too difficult to deal with. For a better differentiation of the AEB criteria for the individual vehicles the challenges are still within the proper test suite definition (high dummy speed - dummy obscuration - different dummy impact points) and an adequate distribution on the continuously evolving test scenarios. In chapter 2 of this deliverable two kind of test scenarios are distinguished, the base and the advanced test scenarios. The prevailing boundary conditions are described, sensor limitation effects are discussed and robustness of assessment methodologies is addressed. A comparative survey of existing test protocols is presented in chapter 3. To date several interest groups have worked in parallel on different levels, regions and constellations on the definition of AEB test scenarios. The German vfss group mainly focused on forward facing safety systems related to accidents with pedestrians and cars. The AEB group, under the lead of insurance companies, had similar targets. The European Commission funded project ASPESS concentrated on VRUs and especially on pedestrians and finally came up with proposals on this topic for EuroNCAP test protocol. All interest groups exchanged views and opinions and aligned and adapted their proposal in several joint meetings and workshops. Chapter 4 reconsiders the case of justified and unjustified system responses by first analyzing the potential operating points and system reaction patterns, i.e. true positive, true negative, false positive, false negative and near miss situations. The crucial point to find a good trade-off in system reaction between these situative states is further influenced by the uncertainty in traffic scenario evolution and the always existing variance of the most important system parameters (vehicle deceleration, pedestrian deceleration, vehicle width and TTC). Nine trade-off scenarios are proposed that bear the risk to produce false alarms by unjustified reaction or responses from the AEB system. These trade-off scenarios may be used to check robustness of the AEB functions or to identify too highforecast times of future traffic situations that can be achieved by simple linear prediction over more than a second. Such kind of generalization checks are not yet part of the EuroNCAP test protocol in force since 2014. In the case of a later adoption, such trade-off tests should be in line with the setup and methodology of the base tests due to economical and test time reasons. Finally, in this chapter a rationale for a typical parameter set that can be used when conducting the proposed trade-off tests is given. The typical lateral safeguard distance when passing a pedestrian with a car is 1m and the typical deceleration of a stopping pedestrian is 3 m/s². With these values different TTC zones for rating the trade-off tests can be defined which give much more room and flexibility for a good AEB system design (i.e. more conservative or a little bit more pro-active) compared to only unique, fixed values. In chapter 5 the available test tools are listed as well as the test target requirements. However, extended information regarding test tools can be found in public project deliverables 2.1 and 2.3. Chapter 6 provides a survey of initial tests results conducted during the project runtime. And the last chapter 7 deals withvalidation of test scenarios and test setup conditions. 2/74

Contents 1 Introduction... 4 2 Accidentology and boundary conditions... 5 2.1 Accidents involving passenger cars and pedestrians... 5 2.1.1 Impact point... 7 2.1.2 Definition oftestscenarios... 8 2.1.3 Base test scenarios and advanced test scenarios... 9 2.2 Boundary conditions... 9 2.3 Technical limitations due to sensor capability... 11 2.4 Assessment methodology for performance testing... 12 2.5 Robustness of assessment methodology against scatter in results... 14 3 Comparative Survey of Existing Test Protocols... 17 4 Assessment of justification of system response... 20 4.1 Potential operatingpoints of pedestrian protection systems... 20 4.2 Uncertainty due to traffic scenario evolution... 21 4.3 Trade-off scenarios with potential risk of false alarms... 24 4.4 Rationale for selected parameters in pedestrian crossing scenarios... 28 4.4.1 Typicalaveragepedestrian deceleration... 28 4.4.2 Typical lateral safe-guard distance when passing by a pedestrian... 30 4.4.3 Conclusion... 31 5 Available Test tools... 32 6 Initial Experiments... 34 6.1 Detailed performance investigation for vehicle A... 34 6.2 Additional test data from other sources... 35 6.3 Derivation of final test scenarios... 36 6.4 Performance investigation for vehicle C... 37 7 Validation of Test Scenarios and Test Setup... 42 7.1 Repeatability of Test Setup... 42 7.1.1 Impact position... 42 7.1.2 Speed accuracy... 43 7.1.3 Repeatability of pedestrian speed... 44 7.2 Performance test results per vehicle... 45 7.2.1 Speed reductions per scenario and vehicle... 45 7.3 Influence of reproducibility on test results... 47 7.4 Conclusion on the test scenario and setup validity... 49 8 Literature... 50 9 Risk Register... 51 10 Conclusion... 52 11 Acknowledgment... 53 12 Test protocol... 54 3/74

1 Introduction Automatic emergency braking systems for vehicle-to-vehicle accidents are already state of the art since 2003 and will be considered in consumer testing by Euro NCAP from 2014 on. These systems use a sensor system to observe the area in front of a vehicle, assess the risk of collisions with another vehicle and brake automatically if needed to mitigate or even avoid an accident. Technology has made great progress in the last decade, and today also systems for avoiding or mitigating vehicle-to-pedestrian-accidents are within reach with first systems already on the market. Systems to address pedestrian accidents are more challenging from a technology point of view due to three main reasons: pedestrians are able to change their direction of travel almost immediately making it difficult to correctly predict their movement; pedestriansare relatively small with at mosta little amount of metal, making it hard to detect and classify them with radar sensors, and last but not least, the majority of pedestrian accidents happen in cross-traffic situations. In longitudinal-traffic situations, directions of travel for both partners are parallel, while in cross-traffic they are perpendicular. Therefore, longitudinal accidents do happen as a function of velocity difference only.for cross-traffic accidents speeds and starting positions of the accident opponents have to match within small boundaries. As of March 2014, pedestrian AEB systems are available in four different vehicle type series 1, more vehicles certainly are currently being developed. The technology is relatively new and still these systems will be tested in the Euro NCAP rating as early as 2016. For a valid assessment methodology that also reflects real world accidentology it is of high importance to involve all stakeholders in an early state. Proposed requirements for pedestrian AEB systems (and corresponding test procedures) are being developed mainly by three initiatives: the seventh-framework-program-founded research consortium AsPeCSS 2, the initiative vfss 3 with German, Japanese and US-American manufacturers as well as independent research institutes) and the AEB 4 -group (Thatcham Research, various manufacturers and suppliers, amongst others). The proposals then are considered by Euro NCAP, so that pedestrian AEB systems can be rated from 2016 on. This deliverable summarizes the proposals and concepts of the AsPeCSS-project. An overview towards other proposals as well as an outlook towards test scenarios that can be expected is also given. 1 Lexus LS, Mercedes E-Klasse/S-Klasse, Volvo, BMW 3er GT/i3/Mini 2 Amongst other with BMW, Toyota, PSA, TRW, Bosch, Autoliv, IDIADA, TNO, BASt 3 Advanced forward-looking safety systems 4 Automatic Emergency Braking 4/74

2 Accidentology and boundary conditions 2.1 Accidents involving passenger cars and pedestrians Within the AsPeCSS project different European accident data sources (especially high level national data and indepth accident data from Germany and Great Britain) were used to investigate the causations and backgrounds of road traffic accidents with pedestrians [Wisch 2013]. Results of the initiatives vfss und AEB were reviewed and further extensive analysis was performed within AsPeCSS, focusing on the following topics: Aggregation of traffic accident data involving one passenger car and one pedestrian by accident type and impact configuration to Accident Scenarios Proportions of pedestrians by injury severity Darkness (i.e.frequency; accompanied visibility constraints due to weather conditions) Investigation of considerable accident parameter (i.e.pedestrian age;pedestrian impact points; crash reconstruction details such as driving and collision speeds of passenger cars, braking behavior) As result of the accident data analysis seven accident scenarios have been identified which are shown in Table 1. Table 1: AsPeCSS Accident Scenarios (A and B classify day (includingtwilight) and night times) It became clear that most frequently crossing conflict situations occurred which have led to collisions between one pedestrian and one passenger car, see also Table 2 and Table 3. Smaller proportions could be assigned to crash situations with obstructed view as well as to collisions where the pedestrian went along the road. Crashes while reversing or parking have been excluded from the dataset due to their minor relevance regarding current forward-looking pedestrian protection systems. The relevance of the different accident scenarios varied significantly by the categorization into the injury severity groups slightly injured, seriously injured, seriously injured or killed (KSI), killed or all injury severities. 5/74

Table 2shows the proportions of the AsPeCSS accident scenarios (crashes between one passenger car and one pedestrian) for seriously injured or killed pedestrians based on national accident data from Great Britain and Germany. Hereby, highest average shares were established for scenario 1 (25%; crossing straight road, nearside, no obstruction), followed by scenario 2 (20%; crossing straight road, offside, no obstruction), other scenarios (16%) and scenario 7 (15%, along straight road, no obstruction). Table 2: AsPeCSS Accident Scenarios in car-to-pedestrian collisions in day/night times (national accident data from Great Britain and Germany of years 2008-2010 regarding killed and seriously injured pedestrians) Table 3 shows the proportions of the AsPeCSS accident scenarios (crashes between one passenger car and one pedestrian) for killed pedestrians only again based on national accident data from Great Britain and Germany. Hereby, highest average shares were established for scenario 2 (30%; crossing straight road, offside, no obstruction), followed by scenario 1 (23%; crossing straight road, nearside, no obstruction), scenario 7 (19%, along straight road, no obstruction) and other scenarios (17%). 6/74

Table 3: AsPeCSS Accident Scenarios in car-to-pedestrian collisions in day/night times (national accident data from Great Britain and Germany of years 2008-2010 regarding killed pedestrians) In summary, accident scenarios 1, 2 and 7 were found as the three most important scenarios for car-topedestrian crash configurations (sum of weights concerning KSI is 60% and concerning fatalities is 72%) that may potentially be addressed by forward-looking integrated pedestrian safety systems. However, accident scenarios 3&4, 5 and 6 (KSI: 24%, Fatalities: 11%) also have a significant potential regarding future active pedestrian protection systems. Further, it was seen that more than half of the crashes which led to seriously injured pedestrianswere assigned to dark light conditions. This proportion increased to around three-quarters in case of killed pedestrians. In addition, the following conclusion could be drawn from the analysis of the accident data: Higher proportion of car-to-pedestrian crashes in urban areas, but higher injury severity on rural roads, Elderly (65+ years) record the highest percentage of casualties killed or seriously injured, Winter months November, December and January show higher number of car-to-pedestrian crashes compared to other months, Proportion of accident scenarios varied sometimes conspicuously between the investigated countries Great Britain and Germany (i.e. accident scenario 7). 2.1.1 Impact point The impact point of the pedestrian on the front of the impacting vehicle is a key factor for the expected AEB system performance. This will be highlighted in detail in section2.2. Anevaluation of in-depth data from GIDAS 5 shows that the impact point is actually distributed on the vehicle front, rather than being only in the vehicle center.in fact the occurrence of impacts at the outer regions of the vehicle front is higher than in the center. Impact point as function of vehicle velocity is shown in Figure 1 for 5 German In-Depth Accident Study, see www.gidas.org 7/74

running pedestrians that come from behind an obstruction. The lateral impact point on the vehicle front is given in m, with 0 m corresponding to the vehicle centerline. It can be seen that the impact is more likely to occur near the corner on that side where the pedestrian is coming from. Concluding from these observations the impact point is a parameter that needs to be considered in the definition of test scenarios. impact point in m relative to vehicle center 1.5 1 0.5 0-0.5 pedestrian from far side pedestrian from near side -1 0 10 20 30 40 50 60 impact speed in km/h Figure 1: Impact point as function of vehicle travelling speed for running pedestrians appearingfrom behind an obstruction 2.1.2 Definition oftestscenarios Test scenarios need to reflect real accident situations as good as possible. Since the accurate simulation of real accident situations in lab-testing is difficult and complex, characteristic parameters that have a significant impact on system performance have to be identified. Test scenarios then are defined using these characteristics. For scenario definition, an accident kinematics model is developed using physical parameters like velocities, positions and starting times. In a first step, velocities are assumed to be constant, and in a second step, pedestrian detection timing and a brake logic where integrated. Using this model, the following parameters have been identified as most important for scenario definition: initial velocity of pedestrian and passenger car, impact point of pedestrian on the car front, start of pedestrian movement (timing) and total travel distance to impact point distance of obscuration (if applicable) to the passenger car path. In addition, the following aspects need to be taken into account: test scenarios should reflect realistic accident situations as good as possible, test scenarios should be able to simulate various accident scenarios and not only those occurring most frequently, test scenarios should take estimated abilities of current and near-future AEB systems as well as current test tools (e.g. repeatability, contrast, light conditions) into account. 8/74

2.1.3 Base test scenarios and advanced test scenarios The developed AsPeCSS test scenarios (see Table 4) were defined first for daylight conditions and were categorized with regard to view obstructions and pedestrian speed (Base Test Scenarios): Test scenario 1: Crossing Child, which runs out from behind an obstruction (distance 1 m between vehicle path and obstruction). Test scenarios 2 and 3: Crossing elderly (walking slowly) and adult (walking) without view obstruction. The test scenarios 2 and 3 are supposed to be also tested in dark light conditions as soon as itis feasible to conduct these tests reproducibly. The classification of child, adult and elderly to the test scenarios was made based on the related assignment of personal data to the accident scenarios and reflects the size and walking speed of pedestrians. Advanced test scenarios are considered to be developed in a later phase. Hereby, turning maneuvers as well as the scenario where the pedestrian is walking along a road on the nearside should be covered. This latter scenario is technically an unobstructed scenario with a lateral pedestrian speed of zero. However, the longitudinal speed and movement of the pedestrian is relevant for the detectability for some sensors, especially radar.it is currently not yet possible to simulate this satisfyingly with available test tools (because the dummy can only travel between two bearings, and cannot be overrun). Table 4: AsPeCSS Test Scenarios 2.2 Boundary conditions Pedestrian travel speeds are generally much lower than car speeds, but the achieved decelerations are comparable [Tiemann, 2010]. Consequently pedestrians can still comfortably avoid collisions with passenger cars just fractions of a second prior to contact. This is particularly true for accidents with impact on the nearside edge of the vehicle front and is less critical for impact points near the far-side edge. Determining for the achieved speed reductions is the time when full braking commences (which corresponds to a lateral distance of the pedestrian to the vehicle path). This brake timing limits the achievable speed reduction (which is a function of initial speed and deceleration). The ability of the brake system to increase the brake pressure and deceleration (maximum brake jerk) is another factor, as well as is the delay time between brake command and brake actuation. If the pedestrian is hidden behind an obscuration and therefore is not freely visible for the vehicle's sensor, the detection and classification time also becomes relevant for system performance. The derivation of performance boundaries in this paper is done based on an ideal braking system and a pointshaped pedestrian. A full investigation of the effects of real brake systems (e.g. increase of brake force) and finite size of pedestrians is published in [Seiniger, 2013]. Error! No se encuentra el origen de la referencia. shows the scenario configuration for the frequent case of a pedestrian crossing perpendicular to the vehicle. For full avoidance, the stopping distance s Stop of an ideal vehicle is a function of initial velocity v Vehicle and deceleration a x 9/74

s 2 v = 2 Vehicle a x Stop. (1) The quantity "Time-To-Collision"[Winner, 2011] is more reasonable for describing these situations than the stopping distance since it depends on the brake timing rather than a fixed distance and especially is independent from the travelling speed. For the crossing pedestrian, is can be simplified to: x Vehicle TTC =. (2) v Vehicle TTC's unit is the second. TTC is the time until the collision occurs when all speeds are held constant and in this case it is used to describe discrete points in time. The TTC (when braking needs to start) for full avoidance is the combination of (1) and (2): TTC v Vehicle Stop = (3) 2 ax,vehicle This equation is only valid for avoidance, meaning for situations when TTC Available (calculated from the actual speed and distance) is greater than TTC Stop : TTC v Vehicle Available (4) 2 ax,vehicle The residual speed is then of course zero. If avoidance is not possible, the residual speed becomes v 2 Residual = v0 2 a x, Vehicle TTC Available v (5). Vehicle The essential quantity for speed reduction in pedestrian AEB systems is the time available for braking - which is in general the TTC at which the pedestrian is considered "critical". Real AEB systems will consider a pedestrian critical when it is near to the vehicle path (e.g. within a second) and on collision course. The pedestrian's impact point on the vehicle front is the second essential quantity determining possible speed reduction. Pedestrians with a high probability for an impact on the near-side of the vehicle will become critical much later than those with an impact on the far-side of the vehicle, because those latter ones do travel within the path of the vehicle for a significant amount of time. Balancing these criteria with the acceptable rate of false-positives then is the task of the vehicle manufacturer's philosophy, considering also product liability, customer acceptance and the likes, see for instance [Lübbe, 2014]. A challenge for any AEB system is pedestrians coming from behind an obscuration near the side of the road. In this case, the total time available for detection, classification, decision, and braking is limited by the time the pedestrian is visible. 10/74

FAR SIDE Vehicle Path IP=100% IP=0% NEAR SIDE y Ped y x v Ped w Ped v Veh Obstruction x Veh IP Obstruction Distance w Veh Figure 2: Definition of the test scenario 2.3 Technical limitations due to sensor capability In Table 5 the key performance parameters for the five mostly used environmental perception technologies (Video, RADAR, CV-LIDAR, LIDAR and PMD) are listed. Each of these technologies has advantages and disadvantages depending on the physical measurement principle and other environmental influences. When using only one of these sensor technologies the AEB system performance may be impaired by the individual sensor deficiencies and should be carefully taken into account. When applying sensor fusion among of at least two different sensor technologies the robustness and dependability of the AEB system with respect to the environmental perception can be drastically improved (assuming that both sensors are fully overlapping with their observation zones and detection capability, thus forming a completely redundant sensor cluster). 11/74

Table 5: Comparison of sensor performance for different technologies Sensorytpe CV / IR / Short Range Challenge/Problem: Video Radar Lidar Lidar PMD Separability of objects lat. + - - + + Separability of objects long. o ++ -- ++ ++ Measurement uncertainty lat. + o - ++ + Measurement uncertainty long. - ++ ++ ++ ++ Typical updatarate >10 Hz >10 Hz >10 Hz <50Hz <50Hz Detection capability @ night - ++ + + ++ Detection capability @ bad weather conditions - + - - - Detection capability against low standing sun / light spots -- + o o o Detection range (Warning functionality possible?) o ++ - + - Detection capability influenced by light reflecting material + ++?? - Multi-Target capability (separation of each person in groups ) + o - + + Typical opening angle / field of view + o o + + Object classification as pedestrian / distinguish btw. standing pedestrians and other standing objects (relevant for warning systems) ++ -o - o + Detection dependance on IR reflectivity ++ ++ o + - 2.4 Assessment methodology for performance testing The assessment methodology for pedestrian AEBs needs to be in line with that established already for vehiclevehicle AEB systems [Euro NCAP 2013]. For each scenario, tests with increasing test speeds are proposed. In principle a linear assessment of speed reductions as measured in the individual test cases seems to be reasonable: 0% of "assessment units" for no speed reduction, 100% assessment units for full avoidance. The unified result of a test suite then depends on the weighting of the individual test speeds (within a test scenario) and the weighting in-between the scenarios. The dose-response-model [Wisch 2013] is an established methodology for the weighting of speed reduction within a test scenario (= for different tested speeds). This methodology defines weighting factors by multiplying injury probability and accident occurrence frequency for all speeds. The factors then correspond to the "risk" and measure the relevance of a given test speed. Due to practicality reasons, the risk curve is approximated with 0 to 3 points, see Figure 3. 12/74

Figure 3: Derivation of in-scenario weighting from accident velocity and injury risk Maximum score for a test scenario according to this approach amounts to 20 points. The achieved score (consisting of full points for full avoidance and a point ratio for tests with mitigation) can then be normalized with the maximum score to generate a percentage value. Since avoidance is not possible for all speeds, AsPeCSS proposes to consider a maximum possible speed reduction with a boundary speed reduction achieved by an ideal brake when braking 0.5 seconds before the accident (as defined with the equations (2) and (5)), see Figure 4 13/74

Figure 4: Scoring as function of assumed maximum speed reduction and exemplary measurement results Weighting of the individual scenarios needs to reflect the frequency of these scenarios in real-world accidents. AsPeCSS is currently investigating a proposal for this. 2.5 Robustness of assessment methodology against scatter in results All results gained from experiments can only be interpreted with respect to the test conditions. Of course, all standardized test scenarios will require to maintain test conditions (e.g. initial speeds, positional deviations of test objects) within small boundaries. On the other hand, the smaller the boundaries for test conditions, the higher becomes the test effort. In general, there will be an optimum point of acceptable test repeatability with acceptable effort. Goal of this chapter is to identify this optimum point, based on an accepted tolerance for the overall test result, a simulation model for an exemplary AEB pedestrian system and the assessment methodology. The evaluation of test scenarios regarding realistic AEB performance on one hand and regarding the necessary repeatability on the other hand requires a description of the sequence of events in an AEB system. An AEB system will constantly monitor driving situation, surroundings and objects in the front of the vehicle and continuously judge the probability of an accident. When a dangerous situation is detected, the system's algorithms will take the decision for an intervention and command the braking. Brakeline pressure increases with a gradient that in general is limited by the brake pump performance. Brake deceleration will also increase proportionally until the maximum possible deceleration is reached. This deceleration then is maintained either to a standstill, to the moment of collision or until the risk of collision is over. 14/74

For the definition of necessary repeatability, the speed reduction of an exemplary AEB system working as described in simple test scenarios is simulated. The exemplary system has the following characteristics: it brakes as soon as the pedestrian enters the vehicle path, maximum deceleration shall be 9 m/s², brake swell time shall be 500 ms (with linear increase of deceleration during the swell time), treatment of the pedestrian as a point. In general, the criticality of a pedestrian will increase with decreasing lateral distance to the vehicle path. The lateral relative position between pedestrian and vehicle throughout the experiment is assumed to cumulate in the quantity impact point. Deviations of the speed reduction are expected for deviations of the impact point. Deviations of the impact point result from lateral shift of the path of the vehicle under test (independent from test speed), deviations in the travel speed of the vehicle under test (larger influence for smaller travel speed), deviations in dummy travel speed (larger influence for smaller vehicle travel speed), variations of pedestrian start of movement (dummy trigger timing). Figure 5 shows consequences of typical impact point variations on the speed reduction for a simple simulation of a crossing pedestrian, walking at 5 km/h, with different nominal impact points. Speed reduction in km/h 60 50 40 30 20 Repeatability ± 5 cm (fully robot-controlled) 25% 50% 75% 2.5% below des. 2.5% above des. Speed reduktion in km/h 70 60 50 40 30 20 Repeatability ± 15 cm (driver) 25% 50% 75% 7.5% below des. 7.5% above des. 10 10 0 0 10 20 30 40 50 60 70 80 Test speed in km/h 0 0 10 20 30 40 50 60 70 80 Test speed in km/h Figure 5: Simulation of speed reduction for several test speeds, based on the variation of the impact point of an unobscured walking adult (5 km/h). Solid lines correspond to the nominal point of impact, dashed lines correspond to the worst-case variations of the impact point in both directions (as given). These results suggest that primary concern in a test setup must be to keep variation of impact points (in general all lateral deviations) limited within boundaries that allow still acceptable variations in measured speed reduction. The required repeatability depends on the overall assessment of the system. The overall assessment for the simulated AEB system performance is given in Table 6 (without considering maximum speed reduction). The results show that the expected rating variations are relevant especially for mitigation cases rather than for avoidance cases for the middle and high speeds. 15/74

Table 6: Variation of overall assessment for the simulated speed reductions according to Figure 5, left case (2.5% deviation of impact point ±5 cm). 50% Walking Adult 75% Walking Adult 25% Walking Adult Test speed Available Pts SR Pts SR Pts SR Pts SR Pts SR Pts SR Pts SR Pts SR Pts SR Pts 10 1 10 1,00 10 1,00 10 1,00 10 1,00 10 1,00 10 1,00 10 1,00 10 1,00 7 0,70 15 1 15 1,00 15 1,00 15 1,00 15 1,00 15 1,00 15 1,00 9 0,60 7,5 0,50 6 0,40 20 1 20 1,00 20 1,00 20 1,00 20 1,00 20 1,00 20 1,00 8 0,40 7 0,35 6 0,30 25 2 25 2,00 25 2,00 25 2,00 25 2,00 25 2,00 25 2,00 8 0,64 7 0,56 6 0,48 30 2 30 2,00 30 2,00 30 2,00 30 2,00 30 2,00 30 2,00 8 0,53 7 0,47 6 0,40 35 3 35 3,00 26 2,23 23 1,97 35 3,00 35 3,00 35 3,00 8 0,69 7 0,60 6 0,51 40 3 26 1,95 23 1,73 21 1,58 40 3,00 40 3,00 40 3,00 8 0,60 7 0,53 6 0,45 45 3 24 1,60 22 1,47 19 1,27 45 3,00 45 3,00 45 3,00 8 0,53 7 0,47 6 0,40 50 2 23 0,92 21 0,84 18 0,72 50 2,00 50 2,00 50 2,00 8 0,32 7 0,28 6 0,24 55 1 22 0,40 20 0,36 18 0,33 55 1,00 55 1,00 46 0,84 8 0,15 7 0,13 6 0,11 60 1 21 0,35 20 0,33 18 0,30 50 0,83 44 0,73 40 0,67 7,5 0,13 7 0,12 6 0,10 Sum 20 15,22 13,96 13,16 19,83 19,73 19,50 5,58 4,99 4,09 Percent 76,10% 69,79% 65,80% 99,17% 98,67% 97,52% 27,91% 24,96% 20,47% The influence of variations on the overall assessment is in the region of ±5%, depending on the nominal impact point. These are the worst cases. In reality, not all deviations have the same sign, and thus variations tend to compensate each other, which reduce the sensitivity significantly. The identified variation of ±5% for the overall results seems to be acceptable. This implicates a repeatability of the impact point in the range ±5 cm. 16/74

3 Comparative Survey of Existing Test Protocols Because of their potential in crash avoidance and injury mitigation Euro NCAP intends to include assessment of AEBs in future protocols. Procedures will be defined by the PNCAP group using information from a number of projects: 1. Advanced Forward-Looking Safety Systems (vfss) Cooperation between OEMs, research and insurance groups worldwide developing test and assessment methods for forward facing safety systems related to accidents with pedestrians and cars. vfss also develops and applies methods on system effectiveness. 2. Advanced Emergency Brake Group (AEB Group) Cooperation between insurance organisations Thatcham and IIHS with support from research groups, a supplier and two OEMs. Aims and goals are identical to vfss. 3. Assessment methods for Integrated Pedestrian Safety Systems (ASPECSS) EU FP7 project consortium of OEM s, suppliers, testing houses, research organisations and universities.research on test methods considering driver behavioural aspects (warning), pre-crash performance evaluation, crash performance evaluation and system effectiveness. AsPeCSS, vfss and AEB Group have all proposed similar accident scenarios based upon the current system limitations and the frequencies of pedestrian collisions. To increase the feasibility in testing a wide range of scenarios appropriate accident scenarios are mapped to specific test scenarios that assess the most critical elements and factors in a controlled manner. The proposed scenarios from each organization (see Table 7) all focus on pedestrians crossing from both obstructed and unobstructed positions. Only vfss has included a test involving a pedestrian entering the road from the far side, AsPeCSS suggests that roads are typically wide enough so that any far side obstruction becomes irrelevant and so the near side test is sufficient to test both cases. Additionally, vfss has included entrance tests designed as prerequisites that demonstrate the systems basic functionality in both day and night time conditions whereas AEB Group suggests a mandatory 100% clearance for each of the standard tests at 20km/h. Another point of difference is the age of the pedestrian in each scenario. Given that older pedestrians (65+ years) are more likely to suffer serious or fatal injuries, AsPeCSS suggests the use of a slow moving dummy to account for this specific risk. Similarly, the likely case of children running out from behind an obstacle is represented by all agencies in some capacity with the introduction of a smaller child dummy. The clear focus is on scenarios whereby the pedestrian crosses the road from either an obstructed or unobstructed position. Slight disagreement between the different groups is found in the importance of children or elderly pedestrians w.r.t. the defined scenarios. Table 7Comparison of proposed test scenarios from AsPeCSS, vfss and the AEB Group AsPeCSS vfss AEB Group Test Scenario Entrance Test Test Scenario Test Scenario Unobstructed Pedestrian - Crossing near side Ped Type : Adult Ped Vel : 5km/h Veh Vel : 10-50km/h Imp Pos : 25% Ped Type: Adult Ped Type : Adult Ped Vel : 5km/h Veh Vel : 40km/h TTC: 2700ms Dist car : 30m Dist ped : 3.8m Ped Type : Adult Ped Vel : 5km/h Veh Vel : 10-50km/h Imp Pos: 25% Ped Type : Adult 17/74

Ped Vel : 5km/h Veh Vel : 10-50km/h Impact Pos : 75% Ped Vel : 5km/h Veh Vel : 10-50km/h Impact Pos : 75% Unobstructed Pedestrian - Crossing far side Ped Type : Elderly Ped Vel : 3km/h Veh Vel : 20-60km/h Impact Pos : 50% Ped Type : Adult Ped Vel : 8km/h Veh Vel : 20-60km/h Impact Pos : 50% Ped Type : Child Ped Vel : 10km/h Veh Vel : 40km/h TTC: 2700ms Dist car : 30m Dist ped : 7.6m Ped Type : Adult Ped Vel : 3km/h Veh Vel : 20-60km/h Imp Pos: 50% Ped Type : Adult Ped Vel : 8km/h Veh Vel : 20-60km/h Impact Pos : 50% Obstructed Pedestrian - Crossing near side Ped Type : Child Ped Vel : 5km/h Veh Vel : 20-60km/h Impact Pos : 50% Ped Type : Adult Ped Vel : 5km/h Veh Vel : 20km/h TTC: 1300ms Dist car: 8.3m Dist ped:2.1m Lum Min : 5000 lx Ped Type : Adult Ped Vel : 5km/h Veh Vel : 40km/h TTC: 1300ms Dist car :14.5m Dist ped : 1.8m Ped Type : Child Ped Vel : 5km/h Veh Vel : 20-60km/h Impact Pos : 50% Ped Type : Adult Ped Vel : 5km/h Veh Vel : 50km/h TTC: 2700ms Dist car: 20.8m Dist ped:2.1m Lum Min : 5000 lx Night Test (Fig. 6) Ped Type : Adult Ped Vel : 5km/h Veh Vel : 20km/h TTC: 1300ms Dist car: 8.3m Dist ped:2.1m Lum: 3 lx (at Ped.) Obstructed Pedestrian - Crossing far side Ped Type : Child Ped Vel : 10km/h Veh Vel : 40km/h TTC: 1300ms Dist car : 14.5m Dist ped : 3.6 Vehicle Straight - walking along carriageway Ped Type : Adult Ped Vel : 5km/h Veh Vel : 40km/h Distance Long : 30m Distance Lat : 0.8m TTC: 2700ms Lum Min : 5000 lx Other Group Bystanders. Ped Type : Adult & Child Ped Vel : 0km/h 18/74

Veh Vel : 40km/h Distance Long : 30m Distance Lat : 2m TTC: 2700ms Lum Min : 5000 lx Ped Type - Pedestrian Type Ped Vel Pedestrian Velocity Veh Vel Vehicle Velocity Impact Pos Impact Position taken from the curb side and as a percentage of the vehicle width Dist car Distance from Initial car position to collision point Dist ped Distance from Initial Pedestrian position to collision point Lum min Minimum Lighting conditions given in lumens Distance Long Initial longitudinal distance from the vehicle to the impact point Distance Lat - Initial latitudinal distance from the pedestrian to the impact point In Figure 6 a proposal of vfss group for a test set-up for VRU city test scenario at night with active illumination by streetlamps is sketched. Figure 6: vfss City Night Test Scenario 19/74

4 Assessment of justification of system response The differentiation whether an AEB system should automatically brake or not in a specific situation is difficult to grasp because no exact physical or mathematical formulation can be found for the typical and expected decision and brake activation process. Furthermore, traffic situations and their accompanying risk levels are perceived rather dissimilar by different individuals. While one person would judge a situation as harmless it could be seen already very critical for another one. This mainly depends on the experience and background everyone has made so far in his driving -life and the personal risk sensation that may also vary over a large and rather unspecific range within the population. There are, however, two crucial factors that can be well considered when classifying the AEB system behavior: i. A good trade-off between the desiredsystem performance and the still acceptable unjustified system ii. response occurrences; A good generalization of the system reaction to avoid direct tweaking or locking on simple test situation or linear movement scenarios (that may easily extend the prediction horizon capability to a few seconds but never really happen in everyday traffic life). The first bullet point is directly related to the user acceptance criterion. By earlier braking the speed reduction potential is definitely improved, resulting consequently in a better system performance. However, the risk of unjustified braking maneuvers is increased on the other side because the prediction horizon becomes much more uncertain when forecast time increases. With a large number of unjustified braking maneuvers (i.e. false positives) the customer acceptance will drop dramatically, often resulting in a deactivation of the whole safety system. The second bullet point addresses the unconscious or intended adaption of the AEB system reaction to the prevailing test constellations, as e.g. realized with respective EuroNCAP test procedures. While the unintended tweaking on specific test characteristics and motion patterns can be avoided with proper system design and manifoldness, the voluntary abuse of a specific system behavior can be only counteracted with additional and absolutely unpredictable cross-check test sequences (i.e. the so-called trade-off scenarios). One exemplary case of voluntary test system behavior abuse is the standard pedestrian crossing test case with constant speed both for the pedestrian and the target vehicle. Here the forecast of the expected impact point can be calculated many seconds in advance by simple linear prediction of the constant movements. With this knowledge a vehicle can be easily come to a standstill in front of a crossing pedestrian even for vehicle speeds above 80 km/h and more. This early braking already starts when the pedestrian is still far away from the virtual driving corridor of the vehicle. A cross-check (or trade-off) test case to inhibit this voluntary abuse of wellknown test characteristics is the random stopping of the pedestrian in his course towards the vehicles driving corridor, which in that specific case should not provoke any premature system braking reaction. 4.1 Potential operatingpoints of pedestrian protection systems In Figure 6 the different reaction patterns of an AEB system response are sketched in a Venn diagram. The true positive (TP) and true negative (TN) zones are that what is expected from the system and denotes the typical reaction. The false negative (FN) and false positive (FP) are that what is undesirable and should be suppressed. Often these cases are unavoidable due to the complexity of the traffic situation and the impairments of the sensorial perception system and central processing unit reasoning capabilities. 20/74

Figure 7: Venn diagram of possible AEB system behavior The Near Miss (NM) case is the grey-zone between wanted and unwanted system reaction that strongly depends on the situation and risk awareness of the individual drivers. No intervention is the typical fall-back solution for detected system malfunction or sensor perception fault. The number of false negatives (FN) describes the deficiency of the system to detect a critical situation (i.e. to overlook a crossing pedestrian) that results in a smaller benefit ratio and has the same effect as when a system is not at all activated (i.e. no safety protection activated). False positives (FP) have the most noticeable impact and outcome as they deteriorate unjustifiably the elsewise normal situation by sudden and unexpected braking maneuvers that are not at all needed and very annoying for the driver and the cars following from behind. The big challenge is to find an acceptable trade-off between good system effectivity (True Positive [TP]) and unjustified system responses (False Positives [FP] and False Negatives 6 [FN]). In addition, test specific tuning and tweaking of the system parameters shall be avoided under all circumstances to guarantee the system performance permanently in typical real world scenarios and not only on the test track. 4.2 Uncertainty due to traffic scenario evolution Uncertainty due to sensors perception shortage was already introduced in Section 2.3.Besides the sensor impairness there exists a further source of uncertainty that stems from the unknown evolution of the traffic scenario over time, especially when non-linear dynamics in pedestrian and vehicle movements prevail. 6 For False Negatives the unjustified system response is that the system doesn t react at all (as breaking is expected to happen in this case) 21/74

Figure 8: Uncertainty of traffic scenario evolution due to driver/pedestrian (re)actions The different kinds of interaction possibilities between the driver of a vehicle and a crossing pedestrian are shown in Figure 7. Each action of one traffic participant may induce a (re)action of the other traffic partner that can either improve or deteriorate the overall situation. Risk prediction and possible evasive maneuvers are therefore more difficult to determine compared to e.g. a fixed obstacle on the road. During the evolution of a given scenario the collision risk may vary several times to higher or lower values in a non-linear and rather unpredictable way. While the possible movements of the vehicle are still limited to a quite manageable range the dynamics of a pedestrian are almost unlimited. Pedestrians are the traffic participants with the highest dynamical change rates both in moving direction and speed. From literature [Fuer2005] acceleration and deceleration of a pedestrian can easily reach 1 g in a very short fraction of time. Depending on the pedestrian speed and max. deceleration different breaking distances, as shown in Table 6, can be calculated. Table 8: Breaking distance of a pedestrian for different speeds and decelerations v Ped 3 kph 5 kph 8 kph s BrakePed [m] in case of a dec = 3m/s² 0.12 m 0.32 m 0.82 m s BrakePed [m] in case of a dec = 9m/s² 0.04 m 0.11 m 0.27 m Equivalent TTC for 50% impact point in case of a dec = 3m/s² and vehicle width of 2m 1.34 s 0.95 s 0.82 s Equivalent TTC for 50% impact point in case of a dec = 9m/s² and vehicle width of 2m 1.25 s 0.8 s 0.57 s AEB systems with a low false positive (FP) alarm rate will only react if the distance of the pedestrian to the vehicle s driving corridor boundary is smaller than the stopping distance of the pedestrian with a maximum 22/74

deceleration of around 1g (i.e. 9 m/s 2 ). Other, even more pro-active systems will already start breaking if quite comfortable (i.e. 3 m/s 2 ) stopping of the pedestrian is no longer possible 7. Based on the following parameters: vehicledeceleration pedestriandeceleration vehiclewidth TTC the earliest breaking point can be determined. Note: If both braking and evasive steering is allowed or conducted, then the timing for breaking may largely differ between these two cases. To this day in Euro NCAP steering maneuvers are not foreseen and thus AEB systems with breaking only are further considered. Another straightforward breaking concept is oriented towards the so-called driver comfort zone. From driving simulator studies and road tests it was elaborated that a typical driver already starts breaking even when the pedestrian is still capable to comfortably stop before the driving corridor borderline. This is owed to the fact that nobody would drive with 30 or 40 km/h aside a pedestrian that is standing only a few centimeters away from the driving corridor borderline. Results from a study conducted by the project partners Toyota and Autoliv with 60 candidates (seefigure 8) showed that all drivers already started breaking 2 seconds before impact (and still 90% of all drivers 2,5 seconds before TTC). Figure 9: Test study in Carson City (Sweden) to determine brake onset in pedestrian collision scenario [Lübbe, 2014] The different assumptions w.r.t.motion dynamics together with the different braking concepts and reaction philosophies finally results to different TTCs for system activation and speed reduction. In Figure 9the diagrams for system activation and speed reduction are shown for two different vehicle parameter sets: While car 1 represents a very early reacting system; the timings for car 2 are much more conservative. This leads directly to much lower speed reduction potential due to the later system activation time. 7 Rationale: When the pedestrian stops with a lower deceleration, the stopping distance and time becomes longer and the AEB system has to break earlier w.r.t. the distance of the pedestrian to the vehicle s driving corridor. 23/74

While at lower vehicle test speeds the difference in speed reduction potential and TTC of system activation are still quite marginal, at higher driving speed above 20 km/h the distinction becomes much more significant. With lower TTC intervention less speed reduction is possible, but this is done with a higher certainty and good knowledge of the traffic situation, thus reducing the FP ratio. TTC of System Activation [s] 1,5 1 0,5 0 20 30 40 50 Test Speed [kph] Car1 Car2 Speed Reduction [kph] 60 50 40 30 20 10 0 10 20 30 40 Test Speed [kph] Car1 Car2 Figure 10: System reaction differences of two cars (car 1 early&pro-active vs. car 2 rather conservative ) With higher TTC intervention more speed reduction is possible but also the false activation rate grows because of the much higher situational uncertainty on the traffic scenario motion parameters. The challenge for system designers is to find the ideal trade-off between a good braking performance versus a tolerable customer acceptance of false positives (FP) in real world scenarios. 4.3 Trade-off scenarios with potential risk of false alarms The ASPECSS consortium members elaborated and discussed in several meetings and telecon calls possible traffic scenarios with the risk to produce false positive (FP) activations or with the aim to counteract and unveil too much test-oriented system tweaking. There are many uncritical situations where pedestrians are located within or next to the predicted driving path with no need of any system activation, but a high risk of a false positive (FP) activation due to unexpected and often non-linear moving patterns. Such kind of test scenarios are very well apt to check system response generalization and the degree a system is tuned to the often very simple and linear predictable test conditions. In Figure 10 some typical FP trade-off scenarios are shown. Trade-off scenarionumber Description Picture 1 Pedestrian walks and comes to a comfortable still stand next to driving corridor 2 Pedestrian walks into corridor and steps back again 3 Pedestrian stands next to driving corridor and accelerates to run over the street 24/74

4 Running child comes out behind obstruction and stops next to driving corridor 5 Pedestrian stands at the end of an acceleration strip 6 Turning maneuver - pedestrian stands on the opposite side of the road 7 Multi-target scenario in S-Curve parcours 8 Pedestrian stands on a traffic island 9 Multiple target scenario - for example group of people standing close to the vehicle path and one person starts crossing Figure 11: Traffic scenarios with a high risk of FP activation due to specific situations and dynamics The question whether trade-off test scenarios should be introduced or added to the already existing standard EuroNCAP tests is not finally answered. In 2013 EuroNCAP staff members were of the opinion that it is within the liability of the OEMs to guarantee a good generalization of the safety system performance to real traffic scenes and not only tweaked to the current test scenarios in force. However, if it becomes obvious that the system performance obtained in the EuroNCAP tests cannot be (completely) transferred to a suitability everyday use, further appropriate measures and additional generalization capability tests may be introduced in the future. In the following paragraphs a first draft proposal of such generalization check and tradeoff tests is shown. From an economical viewpoint it is much better to extend the existing test scenarios and use the same test equipment and methodology instead of defining completely new test setups from the scratch (see Figure 11). 25/74

Figure 12: Development of the so-called quality tests by independent or extended design The AsPeCSS consortium finally came to the conclusion that an extended design approach forthe so-called quality or trade-off tests is sufficient both on a technical and economical basis. For the performance assessment and rating of these trade-off tests different areas in front of the vehicle were defined in which system reaction like a braking intervention can be still tolerated or not, respectively. These areas are defined by using the TTC zones, as marked in Figure 12. TTC yellow TTC green TTC driving corridor d PedSafety s Ped Braking Figure 13: TTC zones used as assessment and rating criterion for the trade-off test scenarios The measurement variable TTC was already introduced and described in chapter 2.2 and is the perfect indicator for rating the quality of the trade-off test scenarios. With the introduction of different TTC zones the system design can be much better accommodated with respect to the desired behavior that can go from very conservative to highly pro-active. Indeed, there is no unique value or reaction scheme that is appropriate for a given situation while any parametric change would be totally off target. The TTC driving corridor is the lower bound where system reaction is mandatory to happen, the TTC green zone opens variations in timing to act earlier with the help of anticipation and the end of the TTC yellow area marks the region where prediction already starts to become rather unsure and intervention strategies are often too early in time. The TTC zones are derived by the motion capability of a pedestrian and classified in Figure 13 accordingly. TTC Intervention <TTC green TTC green <TTC Intervention <TTC yellow TTC Intervention >TTC yellow Pedestrian is not able to come to a complete stop before entering the driving corridor Pedestrian is able to stop between the beginning of the driving corridor and an additional safety distance to the Safety system is prematurely triggered and the unsure intervention is still unsubstantiated and typically 26/74

driving corridor not tolerated by the user Figure 14: The different TTC zones for rating the trade-off tests For the calculation of the boundaries between the different TTC zones the following parameters were defined: - w veh : width of the vehicle - x overlap : ratio of the distance the pedestrian needs to walk into the driving corridor to reach the impact point w.r.t. the vehicle width w veh - s PedBrake : distance the pedestrian needs to brake - a decped = 3m/s² (mean value of typical deceleration for a pedestrian) - d latsafety = 1m (distance of the pedestrian to the driving corridor after stop that a driver still feels comfortable when passing by; very subjective parameter that varies from individual to individual) with the above parameters, the following formulas apply for TTC calculation: TTC drivingcorridor w = veh x v Ped Overlap ( ped. enters driving corridor) TTC green = w veh x Overlap v Ped + s Ped brake = w veh x v Ped Overlap v + 2 Ped Ped adec wveh xoverlap v d TTC yellow = + + v 2 v Ped Ped Ped adec safety lat Ped In Table 7 the TTC values for driving corridor, green and yellow are listed for some pedestrian speeds and impact point overlaps for the test scenarios used in AsPeCSS. Table 9: Different TTC values for some of the AsPeCSS test scenarios AsPeCSS scenario parameter TTC drivingcorridor [s] TTC green [s] TTC yellow [s] V ped =3kph, x overlap = 50 % 1,2 1,34 2,53 V ped =5kph, x overlap = 50% 0,72 0,95 1,67 V ped =8kph, x overlap = 50% 0,45 0,82 1,27 V ped =5kph, x overlap = 25% 0,36 0,59 1,31 V ped =5kph, x overlap = 75% 1,08 1,31 2,03 27/74

The main difference between the time values in column two to four of Table 7 stems from the different terms applied in the formula for the respective TTC zones. While TTC driving corridor is the time the pedestrian needs to travel from the border of the driving corridor to the impact point with constant speed, for TTC green and TTC yellow further time portions are added due to the deceleration of the pedestrian and the lateral safety distance of 1m, respectively. Both the pedestrian deceleration of 3m/s 2 and the lateral safety distance of 1m are values typically met in real life traffic scenarios. A rationale for these two parameters is given in chapter 3.4. 4.4 Rationale for selected parameters in pedestrian crossing scenarios A more detailed justification of the two parameters, average pedestrian deceleration and lateral safety distance, is given in the following sections of this deliverable. These two values, a decped = 3m/s² and d latsafety = 1m directly influence the earliest possible activation time for automatic intervention that is still accepted by the driver and constitutes a good compromise between system performance and false alarm rate. 4.4.1 Typicalaveragepedestrian deceleration Within the German funded project AKTIV pedestrian movements were recorded and representative motion models were developed. In a paper presented at the Airbag2010 conference in Karlsruhe, Germany [Tiem2010] the motion patterns of a moving pedestrian were analyzed in depth. With a GPS-base motion box from Racelogic placed in the backpack of a pedestrian (see Figure 14) the velocity and acceleration were recorded with a very high precision and accuracy. Results of a short trip including walking, jogging and running are drafted in Figure 15. For better evaluation of the acceleration values the speed signal was low-pass filtered to remove excessive over- and undershoots due to noisy speed signals. Physically a pedestrian can easily achieve any acceleration or deceleration up to 1g or 9,81 m/s 2. What is more interesting to know is the mean uniform acceleration process (MUA), i.e. the acceleration with which the test person could accelerate uniformly intime [0; t]. Figure 15: Pedestrian motion pattern analysis with VBOX [Tiemann, 2010] 28/74

Figure 16: Speed and acceleration profile of a moving pedestrian [Tiemann, 2010] When least-square averaging the MUA of several test persons over the three motion types (walking, jogging and running) the following distribution, as depicted in Figure 16, can be achieved. Figure 17: Mean uniform acceleration process (MUA) averaged over several individuals Both acceleration and deceleration can achieve values between 1 to 3 m/s 2 within 1 second of motion time. In the case of an emergency situation it is expected that the pedestrian will brake or accelerate with the maximum peak values, i.e. with ± 3 m/s 2. 29/74

This value is therefore used in the calculation of the TTC zones while always having the caveat mentioned in 3.4.3 in mind. 4.4.2 Typical lateral safe-guard distance when passing by a pedestrian In two studies conducted by the AsPeCSS consortium partners Autoliv and Toyota the typical average lateral safe-guard distance when passing a pedestrian standing at the road border was determined. While Autoliv conducted real world tests with 62 test drivers at an intersection the typical lateral safe-guard distance was determined by Toyota in a driving simulator. In Figure 14 the test setup for the Autoliv real world study, conducted at Carson City in Sweden, is sketched. Figure 18: Carson City test setup configuration for lateral safety distance determination [Lübbe, 2014] Figure 15 shows the cumulative occurrence for longitudinal and lateral brake onset for the two test speeds of 30 km/h and 50 km/h. 30/74

Figure 19: Results for longitudinal and lateral brake onset in the Carson city study [Lübbe, 2014] It can be seen that the longitudinal distance to the dummy at brake onset is different for 30 km/h and 50 km/h. This is of course self-evident because the TTC for both speed values was more or less identical and thus directly converts to different brake onset distances. More important and interesting, however, was the result from the lateral distance at brake onset. At that moment in time all the pedestrian were still more than 1m away from the driving corridor; more than halve of all the pedestrians even 2m and more. 4.4.3 Conclusion It can be concluded that the chosen values for the average pedestrian deceleration a decped = 3m/s² and the lateral safety distance d latsafety = 1m are appropriate numbers that can be used in the equations to calculate the different TTC zones, as defined in chapter 3.3.With these two parameters that are very naturalistic and wellrepresenting the human being behavior and risk perception, sufficient space is opened to incorporate some flexibility in the determination of the automatic brake onset triggering time that best represents the human expectation and feelings. When well respecting the before mentioned different TTC zones and their inherent time limits the AEB system performance can be optimized without jeopardizing user acceptance by too high false alarm rates. AEB systems that are acting with much earlier and thus too premature trigger times, either by exaggerated evaluation of linear motion prediction or by directly locking on test case sequences, can be simply detected and eliminated from the further test procedures. As final caveat it should be mentioned that the results and numbers presented in this document are of preliminary draft level and reflect the current opinion of the AsPeCSS partners. Further modifications and changes towards a final procedure for the trade-off test scenarios are still possible and most likely to happen. Any changes after the submission date of this deliverables will not be tracked or amended later-on. 31/74

ASPECSS D2.5 Test procedure including test target and misuse tests, based on test results 5 Available Test tools Testing pedestrian AEB systems requires a pedestrian dummy as well as an apparatus that is able to move the dummy with a necessary accuracy. Requirements for the dummy are as follows: appears like a human for the vehicle sensor system, impactable without significant damage to the vehicle under test (especially damage to the AEB system's sensors), durability: the possibility to conduct a high number of tests with impacts without significant change in the physical properties. AsPeCSS uses a dummy as initially specified by the vfss group. A detailed description of the dummy characteristics can be found in Deliverable 2.1 [Lemmen, 2013], a detailed summary of the dummy improvements achieved throughout the AsPeCSS project can be found in Deliverable 2.3 [Lemmen, 2014]. The finaldummy used in the project proved to appear close to a human in extensive verification testing with various test vehicles, and the current version is able to take over 100 impacts and does not damage the vehicle under test. Dummy movement is done with two different system concepts: a transportable platform system that has relatively good position accuracy due to its tooth belt drive (Figure 20) that carries the dummy by means of magnets, as well as a gantry-type unit that carries a hanging dummy (Figure 21). Driving robots for accurate longitudinal and lateral vehicle control were used in all experiments in this project. Figure 20: Test setup with child dummy, obscuration, platform, belt drive and light switch system 32/74

Figure 21: gantry-type system with child dummy and obscuration (source: Applus IDIADA) The platform propulsion system used by BASt is a commercial solution by the Austrian company 4a engineering, developed in parallel to the progress of the AsPeCSS project (and substantially influenced by its' progress). The final version of that propulsion system receives position data from the Vehicle under Test and triggers the dummy movement appropriately. If a radio connection is not available, light switches are used to determine the Vehicle under Test's position. 33/74

6 Initial Experiments As mentioned in the introduction, pedestrian AEB is currently only implemented in four series vehicles of different brands. Of each of those four series cars, one representative specimen has been tested by ADAC. AsPeCSS did test one of those vehicles in more detail. An overview of the vehicles and their sensor equipment is available in Table 10. Data from these tests helps to estimate performance and limitations of current AEB pedestrian systems and to define future test procedures. Table 10: Tested vehicles and their respective sensor equipment Vehicle Vehicle A Vehicle B Vehicle C Vehicle D Sensors used for pedestrian AEB Stereo camera, near infrared, RADAR Stereo camera, near infrared, RADAR Mono camera, RADAR Mono camera 6.1 Detailed performance investigation for vehicle A Vehicle A has been tested by BASt and IDIADA in detail, according to the initial AsPeCSS test scenarios. BASt used a platform-type dummy propulsion system, IDIADA used a gantry-type system. Some of the tests were done during night condition. Speed reduction and TTC of start of braking are available for evaluation. All vehicles had been conducted speed reduction measurements with a comparable platform-type dummy propulsion system. A detailed description of test setup and results is available in [Rigling, 2013]. Figure 1 shows speed reductions and TTCs for start of braking for vehicle A, distinguished between dummy type (child/adult), dummy speed, obstruction, propulsion system and desired impact point. Speed reduction in km/h 90 80 70 60 50 40 30 90 80 70 60 50 40 30 Elderly 75% Adult 25% Adult 50% Adult 75% Child 50% Child 50% Obstruction Child 75% Obstruction Static 50% 20 20 10 10 0 0 1 2 TTC AEB in s 0 0 10 20 30 40 50 60 Test speed in km/h Figure 22: Speed reduction (right) and start of braking TTC (left) as function of test speed for vehicle A The test results for vehicle A show that a large portion of the proposed test suite can be addressed already today. However, especially test cases with a speed ratio between dummy speed and vehicle speed approaching equality and the resulting large angles between pedestrian and passenger car throughout the test lead to 34/74

relatively low speed reductions. This is particularly true for tests with 10 km/h vehicle speed and 5 km/h dummy speed, as well as for 20 km/h vehicle speed and 8 km/h dummy speed. A reason for this may be that the field of view of the vehicle s sensor system is not sufficient to detect the pedestrian dummy right from the start conditions - the dummy will appear in the sensor s observation range only during the experiment. Another observation is the start of braking TTC and therefore also the achieved speed reduction do stay on the same level in running child experiments above 30 km/h. Route cause by sensor field of view is no explanation here, since the needed observation sector decreases with increasing test speed. The time needed for decision making might exceed the reaction time available in these cases. The test scenarios with running child from behind an obscuration clearly show the limits for active safety: the child appears approximately 1.3 seconds before the impact (75%). This time is not sufficient for more than a few km/h speed reduction. For a 25% situation, the appearance time is approximately 0.7 seconds, which is even worse for AEB performance. The test results suggest some potential for the running child scenario with 75% impact, but it seems that running child situations with 25% impact point might not be solvable in the mid and far future, and even effective mitigation does not seem to be possible. Vehicle A does not seem to have problems with dark light conditions (night, dusk), and there seems to be no significant difference in results with regard to the propulsion system type. The brake timing seems to stay constant from 40 km/h upwards. In fact, the remaining distance of the vehicle after standstill to the dummy line of movement decreases for these higher speeds. This suggests that the intended avoidance speed is 40 km/h, and all speed reductions above this speed occur because of the relatively high friction on one of the test tracks. Lower speed reductions for the gantry-type propulsion system therefore do not contradict the assumption that there is no influence of the type of the propulsion system; the friction of the test track might make the difference. 6.2 Additional test data from other sources Figure 31 summarizes the ADAC test results for vehicles A to D, however the tested scenarios do differ significantly from AsPeCSS scenarios in terms of impact point and dummy speed in some cases and are not directly comparable. Test results for test vehicle A for the scenario "walking adult" (center in Figure 23) are comparable; especially problems with test speeds of 10 km/h are observed as well, and the obscured pedestrian (even when moving with a slower speed) is still a tough challenge to be solved. Another significant observation is the relatively robust speed reduction of 10 km/h for test vehicle D in almost all cases and test speeds. A comparison between all four tested vehicles shows the obscured scenario (right) is most challenging. Any collision avoidance in this test scenario only occurs up to 20 km/h, and that only for two of the four vehicles. Above 20 km/h, only accident mitigation can be observed, and the higher the test speed, the lower is the achieved speed reduction. The static adult (not an AsPeCSS scenario since this is not found anywhere in the accident databases) does pose no problems on any of the vehicles. Walking adult scenarios show avoidance up to highest speed ranges for two of the vehicles, but only weak performance with the other two. There are significant differences between the vehicles and the observed speed reductions are higher than those found when applying the AsPeCSS test scenarios (very likely due to the modified dummy moving speed of the child dummy). 35/74

60 50 Static adult 50% A B C D 60 50 5 km/h adult 50% 60 50 5 km/h child 50% (obsc.) Speed reduction in km/h 40 30 20 40 30 20 40 30 20 10 10 10 0 0 20 40 60 Test speed in km/h 0 0 20 40 60 Test speed in km/h 0 0 20 40 60 Test speed in km/h Figure 23: Speed reduction as function of test speed and test scenario for vehicles A to D, without using the final dummy (for details see [Rigling, 2013]) 6.3 Derivation of final test scenarios First of all, test results within this paper have been gained only with high-end vehicles, with the exception of vehicle D. A holistic view with regard to possible AEB pedestrian performance for all vehicle segments might be possible in a few years. For the time being, it seems that today's vehicles do not have problems with slow and readily visible adults, and there are observable differences between different vehicles. The AsPeCSS child scenarios however do not allow a real differentiation in performance and benefit assessments between the different vehicles or AEB systems used. In this aspect, ADAC scenarios with decreased dummy speed provide more meaningful results. Preliminary AsPeCSS scenarios with the obscured and running child did challenge all vehicles, only slight speed reductions were possible. This type of scenario combines the three main challenges - obscuration, high dummy speed, small dummy - in a single scenario. In collaboration with other European initiatives, AsPeCSS developed the idea of distributing these difficulties into more than one scenario in order to be able to better assess and compare system performances. The test results have been improved based on the results shown in this deliverable (AsPeCSS, ADAC) and on the test results generated within the AEB group and together with Euro NCAP for better comparability of vehicles, see Table 11. The dummy speed in the child scenario has been decreased to 5 km/h, and the different impact points (a fact that is quite important according to accident figures) will be tested with a walking, unobscured adult dummy instead. Higher as well as lower pedestrian walking speeds are transferred to a scenario with an adult dummy coming from the far side - the dummy will be readily visible long before the impact in this case. 36/74

Table 11: current scenario definitions from AsPeCSS Scenario Walking adult Running adult Walking adult Walking adult Walking child Pedestrianspeed 3 km/h 8 km/h 5 km/h 5 km/h 5 km/h Dummy type Adult Adult Adult Adult Child Dummy initialposition Far side Far side Nearside Nearside Nearside Vehiclespeeds 20-60 km/h 20-60 km/h 10-50 km/h 10-50 km/h 20-60 km/h Obscuration No No No No Yes Impact point 50 % (Center) 50 % (Center) 6.4 Performance investigation for vehicle C 25% (Nearside) 75% (Nearside) 50 % (Center) Vehicle C has been tested by Autoliv in detail after the AsPeCSS scenario update, but with a different dummy type. Due to time constraints, only a subgroup of the AsPeCSS scenarios was tested. Autoliv used a gantry-type system. Speed reduction is available for further evaluation in Figure 25 to Figure 30. Some variations in results were obtained and some test scenarios gave results that were not in line with the manufacturer s expectations. E.g., brakes were sometimes released before impact and AEB was sometimes activated later than expected. The car manufacturer believes that this was due to contrast issues between the dummy and the environment as well as dummy posture and distance between dummy feet and ground. Contrast was particularly important in this testing, since the used dummies are static and thus lacked important pedestrian features. Figure 24: Dummies used by Autoliv To still be able to compute a total test score, the measured test results and estimated point of brake onset were combined with the manufacturer s comments to predict likely results in all AsPeCSS tests: Vehicle C always applies a similar brake pulse. It has an approximately linear ramp-up phase of 500 ms, during which >1g deceleration is reached. The deceleration then fluctuates due to ABS. During the ABS phase, the deceleration first decreases slightly and then increases again. The brake pulse may be approximated by a linear ramp-up phase of t ramp = 500 ms followed by a phase of constant deceleration with a = 0.95g. Theoretical collision speeds as function of the test speed v can then be calculated as v Coll = v a t ramp 2 2 2 a v ( TTC t ramp ) a 2 2 t ramp / 3, where TTC is time to collision at brake decision, and v is test speed in m/s. 37/74

A comparison between the measured speed reduction and calculated results is given in Figure 25 to Figure 29, and an overview over all determined speed reductions is shown in Figure 30. 40 35 Elderly walking visible Speed reduction (km/h) 30 25 20 15 10 5 Theoretical Measured 0 0 10 20 30 40 50 60 70 Test speed (km/h) Figure 25: Speed reductions, vehicle C, scenario walking elderly 25 Adult running visible 20 Speed reduction (km/h) 15 10 5 Theoretical Measured 0 0 10 20 30 40 50 60 70 Test speed (km/h) Figure 26: Speed reductions, vehicle C, scenario running adult 38/74

Adult walking visible, 75% offset 35 30 Speed reduction (km/h) 25 20 15 10 Theoretical Measured 5 0 0 10 20 30 40 50 60 70 Test speed (km/h) Figure 27: Speed reductions, vehicle C, scenario walking adult, 75% impact 35 Adult walking visible, 25% offset 30 Speed reduction (km/h) 25 20 15 10 Theoretical Measured 5 0 0 10 20 30 40 50 60 70 Test speed (km/h) Figure 28: Speed reductions, vehicle C, scenario walking adult, 25% offset 39/74

Child walking obstructed 25 Speed reduction (km/h) 20 15 10 5 Theoretical Measured 0 0 10 20 30 40 50 60 70 Test speed (km/h) Figure 29: Speed reductions, vehicle C, scenario walking child 50%, obstructed Summary of theoretical results Speed reduction (km/h) 40 35 30 25 20 15 10 5 0 0 20 40 60 80 Test speed (km/h) EWV AWV 75% AWV 25% ARV CWO Figure 30:Speed reductions, vehicle C, scenario walking adult, 25% offset The overall results show a significant improvement over the test results gained from testing the initial AsPeCSS scenarios: there is a significant speed reduction observed in all tested scenarios, while in the original running child scenario this was not the case. Without any speed reduction, a comparison between different vehicles is not possible. There is also a significant difference in performance between the offset walking adult scenarios and the centered walking elderly scenario. These results suggest that the refined AsPeCSS scenarios will fulfill the requirements from the scenario definition. 40/74

Assessment of justification of system responses, as described in chapter 4, was done as far as possible during these initial test experiments. The different TTC zones, as defined in chapter 4.3, for rating the trade-off tests were used by the four tested vehicles in an either more conservative or more pro-active way. Further discussion and evaluation with the relevant stakeholders is needed to come up with a first proposal for additional trade-off or entry tests, based on the initial work done in chapter 4. Up to now, EuroNCAP staff is of the opinion that trade-off test procedures are not needed for the current NCAP AEB VRU test suite but fully fall within the responsibility of the OEMs to design and tune their systems with the necessary precaution to deliver the expected customer benefit in everyday operation on public roads and not (only) on the proving grounds. 41/74

7 Validation of Test Scenarios and Test Setup Test vehicles A, B and D were available for additional validation of the refined test scenarios. Due to timing and capacity problems, a prioritization of test cases was necessary. The walking elderly could not be tested with all vehicles. Vehicle speeds below 20 km/h andvehicle speeds above 45 km/h with a speed reduction below 20 km/h have also not been tested. This prioritization is in compliance with Euro NCAP s upcoming test methodology. In order to identify the repeatability of the test setup and the robustness against scatter, priority was also given to repeated conduction of those tests that generated inconsistent results. This chapter is based on tests conducted mainly by BASt (for one car, the test series was run by BMW). Since the number of test labs therefore is low, a detailed reproducibility analysis cannot be carried out. 7.1 Repeatability of Test Setup The sensitivity analysis in section 2.5identified the impact point repeatability as key contributing factor towards the test setup repeatability. The vehicles brake in most of the cases. This braking action gives the pedestrian dummy more time to move, so the impact point is shifted in the pedestrian s direction of travel. In order to assess the repeatability of the test setup with regard to the impact point, it is therefore necessary to calculate the impact point as if the vehicle had not braked. The impact point depends on vehicle speed, pedestrian dummy speed and pedestrian starting time. Pedestrian starting time in general can be adjusted according to the actual vehicle speed impact point repeatability is independent from vehicle speed repeatability, as long as the vehicle speed is maintained constant within certain boundaries. The repeatability of vehicle speed is investigated as well it is mainly in line with the upcoming Euro NCAP requirements (±0.25 km/h), as is the pedestrian dummy speed (±0.2 km/h from 3 m distance to the impact position). 7.1.1 Impact position The impact position observed during a test in general does not match the desired impact point as long as the vehicle brakes before the impact (which of course is the desired action). The impact point for an unbraked test run needs to be calculated. Known properties are vehicle speed v x,vut and vehicle position x VuT, pedestrian position y Ped and the time the vehicle deceleration exceeds 0.3 m/s² for the first time t AEB (identified according to the Euro NCAP AEB test protocol, see [Euro NCAP 2013]). The concept to identify the projected impact point is to calculate at what time the vehicle would have crossed the pedestrian line of travel if the braking would not have happened and check at what position the pedestrian had been at that (earlier) time: x& x t unbraked unbraked impact, unbraked y = x& ( t < t = x& impact, projected unbraked = t( x = y AEB ) ( t t unbraked Ped ( t AEB = x ) + x( t = t Pedestrian impact, unbraked ) ) AEB ) Vehicle speed is held constant before t AEB, however there is a small drop of speed during the last fractions of a second before t AEB. This might influence the projected impact point, sot AEB has been shifted backwards 0.3 seconds during the data processing. Figure 31 shows an example of the calculations. 42/74

0 x VuT Distance in m -1-2 -3-4 y Dummy x VuT,theoretical y Dummy -y VuT Projected Impact Position Real Impact Position 3.5 3.6 3.7 3.8 3.9 4 4.1 Figure 31: Example for calculation of projected impact point: positions of vehicle and dummy over time. The dotted line is the calculated vehicle position if the braking hadn't commenced, the solid black line is the vehicle's measured position. The projected impact position is 0 (vehicle center) which is also the desired impact position for the shown test run. Taking all valid test runs into account, the requirement of achieving a repeatability of ± 5 cm was met in the vast majority of cases (all but 7), see Figure 32. 1 0.9 0.8 Dummy Impact Position Deviation, Relative to Vehicle Reference CPAN75 CPAN25 CPAF CPCN 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0-0.1-0.05 0 0.05 0.1 0.15 0.2 0.25 Projected Impact Point - y VuT,end - Desired Projected Impact Point [m] Figure 32: Accuracy of pedestrian impact on vehicle front (note that deviations of the vehicle path are included in this value). 7.1.2 Speed accuracy As discussed above, the initial (unbraked) speed of the vehicle is another important parameter for the interpretation of the test results. Figure 33 shows the difference between VuT speed and desired test speed for all valid test runs as a function of dummy position, from TTC=4 seconds until end of test (braking is included). In general, a good repeatability in the range lower than ± 0.25 km/h is achieved, but some singular test runs show significant deviations especially during the beginning of the tests. A further analysis of the absolute speed suggests that this problem occurs only during tests with relatively high initial speeds (50, 55, 60 km/h), see Figure 34. This is caused by the small available distances on the test track in that particular test setup: the 43/74

stabilized test speed had to be reached after as little as 70 m distance for the higher speeds, see Figure 35 for an overview of the BASt test site. 1 VuT Speed deviation over x position 70 VuT Speed over x position 0.5 0 60-0.5 50 v x,vut - v Test [km/h] -1-1.5-2 -2.5 v x,vut [km/h] 40 30 20-3 -3.5 10-4 -70-60 -50-40 -30-20 -10 0 10 x [m] 0-70 -60-50 -40-30 -20-10 0 10 x [m] Figure 33: Deviation of VuT speed from TTC=4s until end of test Figure 34: VuT speed from TTC=4s until end of test Test runs relative to test track 140 120 Test trackboundaries 100 80 60 startingpoints TTC=4s@60km/h Impact Position 40 20 0-20 -140-120 -100-80 -60-40 -20 0 20 40 60 Figure 35: Test setup on BASt's Vehicle Dynamics Area. Due to insufficient acceleration distance, higher speeds (50, 55, and 60) could not be achieved within the required boundaries. 7.1.3 Repeatability of pedestrian speed The pedestrian dummy s speed is an important parameter within the test scenario definition since it determines the amount of time available for automatic braking actuations. The observed pedestrian dummy speeds in all test runs are shown in Figure 36. 44/74

8 Pedestrian Dummy speed 7 6 Dummy speed [km/h] 5 4 3 2 1 Figure 36: Dummy speeds over dummy position 7.2 Performance test results per vehicle The only performance indicator during the pedestrian active safety tests is the speed reduction. The brake timing can also be identified it is used to determine the initial speed of the vehicle. The AsPeCSS rating however is a filter that converts the speed reduction into a rating result per test speed. The result for a test scenario is the sum of all individual tests, and the result of all test scenarios is the mean value of all sums of all individual tests (no weighting of single scenarios is performed). Rating is done in a two-step approach: for all speeds up to and including 40 km/h, the quotient of speed reduction and test speed is multiplied with the points defined in Table 12to generate the individual rating result (sliding scale). This means that the total number of points can be obtained only if collision avoidance is achieved in all tests. For speeds above 45 km/h, collision avoidance in general is very demanding and cannot be expected. Full points are given for a speed reduction of at least 20 km/h. No partial points are awarded, and the consecutive test speed will only be tested for those test cases with a speed reduction above 20 km/h. Table 12: Rating method Test speed 0-4 -2 0 2 4 6 8 Dummy position [m] 20 25 30 35 40 45 50 55 60 Points 1 2 2 3 3 3 2 2 1 Method Sliding Scale 7.2.1 Speed reductions per scenario and vehicle Pass for Speed Reduction > 20 km/h The speed reductions are shown as functions of test speed, for all scenarios, and in one plot per vehicle, see Figure 37 to Figure 41. 45/74

60 Speed reduction for all scenarios Brake timing for all scenarios 1.2 50 1 Speed reduction [km/h] 40 30 20 TTC AEB [s] 0.8 0.6 0.4 CPAN75 10 CPAN25 CPAF CPCN 0 0 10 20 30 40 50 60 Test speed [km/h] 0.2 CPAN75 CPAN25 CPAF 0 0 10 20 30 40 50 60 Test speed [km/h] Figure 37: Speed reductions for vehicle A Figure 38: Brake timing per scenario and test speed for vehicle A 60 Speed reduction for all scenarios Brake timing for all scenarios 1.2 50 1 Speed reduction [km/h] 40 30 20 TTC AEB [s] 0.8 0.6 0.4 CPAN75 10 CPAN25 CPAF CPCN 0 0 10 20 30 40 50 60 Test speed [km/h] 0.2 CPAN75 CPAN25 CPAF 0 0 10 20 30 40 50 60 Test speed [km/h] Figure 39: Speed reductions for vehicle B Figure 40: Brake timing per scenario and test speed for vehicle B Speed Reduction in km/h 60 50 40 30 20 CPAN75 A B D Speed Reduction in km/h 60 50 40 30 20 CPAN25 Speed Reduction in km/h 60 50 40 30 20 CPAF Speed Reduction in km/h 60 50 40 30 20 CPCN 10 10 10 10 0 0 20 40 60 Test Speed in km/h 0 0 20 40 60 Test Speed in km/h 0 0 20 40 60 Test Speed in km/h 0 0 20 40 60 Test Speed in km/h Figure 41: Speed reductions per scenario and vehicle for vehicles A,B and D 46/74

Vehicles A and B achieve full score in the scenario Adult-near side-75% impact point. Avoidance is observed for speeds up to 45 km/h (A) and 50 km/h (B). Speed reductions above this test speed always stay above 20 km/h. Vehicle A manages to achieve almost full score for the scenario running adult far side as well, despite the scatter that is observed in almost all scenarios without impact avoidance. The main reason for this is the pedestrian becomes critical relatively short before the impact (running adult farside, walking adult 25% near side) or even becomes visible relatively late (walking child). The scatter is most significant in the walking child scenario. Brake timing values show a similar trend (see Figure 38, Figure 40). Vehicle D manages to reduce the impact speed by around 10 km/h in all scenarios but the running adult farside. 7.3 Influence of reproducibility on test results While there is significant scatter within the performance test data (especially for the child and adult running scenarios), this test data is "filtered" with the rating scheme. This filtering significantly improves the scatter. Figure 42 shows all possible test results per scenario and in total under the assumption that only one of the repeated tests would have been selected for the performance assessment. In other words: the data shown is a systematic combination of all available test results. Since there is only little scatter in the walking adult 75% scenario and full points would have been awarded for both vehicles, no matter which of the repeated test results would have been selected, there is no scatter in the results for this scenario as well (blue curve). For vehicle B, surprisingly, there is also only very little scatter in the results for scenarios walking adult 25% and running adult, despite the fact that the observed speed reductions showed substantial scatter (red curve, green curve). There is, however, significant scatter in the results for scenario walking child (black curve). For this vehicle, possible results range from around 10% to around 35% in this scenario. Since there is no weighting in the overall (total) results and the result scatter for all but the walking child scenarios is low, the total result for this particular vehicle does not show significant scatter. Total results for this vehicle could have been from 43% to 53%, with the highest probability around 48%. It is anticipated that this scatter should be tolerable. Since the test results for vehicle A are generally better, the possible results are higher as well. According to the available test data, the median test result for vehicle A is 80%, with a range from 76 %to 83%, which is the same range as observed for vehicle B. Since no repetitions had been carried out for vehicle D, there is no distribution of the test result. Based on the conducted tests, the result is 9.5%. 47/74

1 Possible test result per scenario 0.9 0.8 0.7 0.6 0.5 0.4 0.3 CPAN75 0.2 CPAN25 CPAF 0.1 CPCN Total 0 0 10 20 30 40 50 60 70 80 90 100 Results in % Figure 42: Possible results from the recorded test data. Statistics for total result: min 43.16%, max 52.49%, mean 48.51%, median 48.63%, std. deviation 1.6% 1 Possible test result per scenario 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 CPAN75 CPAN25 0.1 CPAF CPCN 0 20 30 40 50 60 70 80 90 100 Results in % 48/74

Figure 43: Possible results from the recorded test data per scenarios, vehicle A. Total results distribution not computable due to high number of variations. Median total result is 80%, max total result is 83%, min total result is 76%. 7.4 Conclusion on the test scenario and setup validity The validation tests show that the test repeatability in general stays in the required boundaries according to the calculations from section 2.5. There is significant scatter in the observed speed reductions for scenarios adult near side 25% and adult running 50%, and substantial scatter in the speed reductions for scenario walking child. The resulting overall result for this particular vehicle, determined as a systematic combination of only one speed reduction for all the test speeds however stays in the predicted boundaries from section 2.5, page 14(note that in total 113.246.208 different total results would have been possible, depending on the selected speed reductions). Since these requirements from section 2.5 have been reached, the test procedure is considered appropriate to robustly judge a vehicle s automatic emergency braking performance for pedestrian accidents. 49/74

8 Literature [Euro NCAP, 2013]:Euro NCAP: Assessment Protocol - Adult Occupant Protection, Version 6.0. July 2013, S.30. Download at http://www.euroncap.com/ [Fuer2005]: Fuerstenberg, K. Pedestrian protection using laser scanners. Proceedings of the 8th International IEEE Conference on Intelligent Transportation Systems Vienna, Austria, 2005 September: 13-16 [Lemmen, 2013]:Lemmen P et al: Evaluation of Pedestrian Targets for Use in Autonomous Emergency Brake System Testing - A Report from the Harmonisation Platform 2 Dealing With Test Equipment. Paper presented at: 23nd Conference on the Enhancement of the Safety of Vehicles (ESV). May 27-30, 2013; Seoul, South Korea. [Lemmen, 2014]: Lemmen P et al: Deliverable 2.1b. [Lübbe, 2014]: Lübbe N, Rosén E. Pedestrian crossing situations: Quantification of comfort boundaries to guide intervention timing. Accident Analysis and Prevention. In Press. [Rigling, 2013]: ADAC press release, published November 14, 2013 [Seiniger, 2013]: Seiniger P, Bartels O, Wisch M, Pastor C. An Open Simulation Approach to Identify Chances and Limitations for Pedestrian Active Safety. Paper presented at: 23nd Conference on the Enhancement of the Safety of Vehicles (ESV). May 27-30, 2013; Seoul, South Korea. [Tiemann, 2010]:Tiemann N, Branz W, Schramm D. Predictive Pedestrian Protection Situation Analysis with a Pedestrian Motion Model. Paper presented at: 10th International Symposium on Vehicle Control. August 22-26, 2010; Loughborough, UK. [Winner, 2011]: Winner H. Frontalkollisionsschutzsysteme. In: Winner H, HakuliS, Wolf G (Hrsg).: Handbuch Fahrerassistenzsysteme. Wiesbaden: Vieweg + Teubner, 2012:S.522. [Wisch, 2013]: Wisch M., Seiniger P., Pastor C., Edwards M., Visvikis C.: Scenarios and weighting factors for pre- crash assessment of integrated pedestrian safety systems. AsPeCSS D1.1, download available at www.aspecss -project.eu, 2013. 50/74

9 Risk Register Risk No. What is the risk WP2 Since this deliverable contains the final results of WP2, there is no risk. Level of risk 8 - - Solutions to overcome the risk 8 Risk level: 1 = high risk, 2 = medium risk, 3 = Low risk 51/74

10 Conclusion Based on work from AsPeCSS work package 1, accident scenarios had been defined in Deliverable 1.1. This work has been summarized in this Deliverable. In an initial test series, various vehicles had been tested in these initial test scenarios.the results (shown in section 6, page 34 of this Deliverable) do show some difficulties: one of the scenarios is so demanding that only neglectable AEB system performance is observed. One the other hand, the other two scenarios do show relatively good performance. It is expected that this good performance would be achieved by other vehicles as well. The problem with this situation is: there is no means to distinguish two vehicles from each other with this mixture of very demanding and very easy test scenarios. Based on work from other parties (e.g. ADAC, see section 6.2, page 35) and in collaboration with other Euro NCAP members, a refined set of test procedures had been defined. Test data from other sources / test data gained with these scenarios but different dummy setup suggests that the refined test scenarios reflect the differences between different vehicles much better. A test procedure requires thresholds for the necessary repeatability of important variables during the test. These thresholds were defined in section 2.5, page14, based on a model of a generic AEB system and based on assumptions for a tolerated scatter in overall test results. Final test data shows the repeatability requirements are met quite well. Also the scatter in overall results matches the expected values. Therefore the test setup itself is verified to comply with the specifications. 52/74

11 Acknowledgment This project is co-funded by the 7th FP (Seventh Framework Programme) of the EC - European Commission DG Research http://cordis.europa.eu/fp7/cooperation/home_en.html http://ec.europa.eu Disclaimer The FP7 project has been made possible by a financial contribution by the European Commission under Framework Programme 7. The ation as provided reflects only the authors view. Every effort has been made to ensure complete and accurate information concerning this document. However, the author(s) and members of the consortium cannot be held legally responsible for any mistake in printing or faulty instructions. The authors and consortium members retrieve the right not to be responsible for the topicality, correctness, completeness or quality of the information provided. Liability claims regarding damage caused by the use of any information provided, including any kind of information that is incomplete or incorrect, will therefore be rejected. The information contained on this website is based on author s experience and on information received from the project partners. 53/74

12 Test protocol The definition of a test protocol for pedestrian AEB is part of WP2. However, during the conduction of AsPeCSS, Euro NCAP in coorperation with AsPeCSS members such as BASt, TNO and IDIADA defined a test protocol for AEB pedestrian testing as well The selected Euro NCAP scenarios are a subgroup of the updated AsPeCSS scenarios - in fact, everything except the walking elderly scenario is covered. For this reason, AsPeCSS did not define a test protocol of its' own. The current draft of Euro NCAP's test procedure is as follows. Please note this document is not for the public yet. 54/74

55/74