Counterfactual simulations applied to SHRP2 crashes: The effect of driver behavior models on safety benefit estimations of intelligent safety systems

How would autonomous vehicles behave in real-world crash scenarios?

Autonomous Vehicles (AVs) have the potential to revolutionize transportation systems by enhancing traffic safety. Safety testing is undoubtedly a critical step for enabling large-scale deployment of AVs. High-risk scenarios are particularly important as they pose significant challenges and provide valuable insights into the driving capabilities of AVs. This study presents a novel approach to assess the safety of AVs using in-depth crash data, with a particular focus on real-world crash scenarios. First, based on the high-definition video recording of the whole process prior to the crash occurrences, 453 real-world crashes involving 596 passenger cars from China In-depth Mobility Safety Study-Traffic Accident (CIMSS-TA) database were reconstructed. Pertinent static and dynamic elements needed for the construction of the testing scenarios were extracted. Subsequently, 596 testing scenarios were created via each passenger car's perspective within the simulation platform. Following this, each of the crash-involved passenger cars was replaced with Baidu Apollo, a famous automated driving system (ADS), for counterfactual simulation. Lastly, the safety performance of the AV was assessed using the simulation results. A logit model was utilized to identify the fifteen crucial scenario elements that have significant impacts on the test results. The findings demonstrated that the AV could avoid 363 real-world crashes, accounting for approximately 60.91% of the total, and effectively mitigated injuries in the remaining 233 unavoidable scenarios compared to a human driver. Moreover, the AV maintain a smoother speed in most of the scenarios. The common feature of these unavoidable scenarios is that the AV is in a passive state, and the crashes are not caused by the AV violating traffic rules, but rather caused by abnormal behavior exhibited by the human drivers. Additionally, seven specific scenarios have been identified wherein AVs are unable to avoid a crash. These findings demonstrate that, compared to human drivers, AVs can avoid crashes that are difficult for humans to avoid, thereby enhancing traffic safety.

Comparison of Experienced and Novice Drivers' Visual and Driving Behaviors during Warned or Unwarned Near-Forward Collisions

Forward collision warning systems (FCWSs) monitor the road ahead and warn drivers when the time to collision reaches a certain threshold. Using a driving simulator, this study compared the effects of FCWSs between novice drivers (unlicensed drivers) and experienced drivers (holding a driving license for at least four years) on near-collision events, as well as visual and driving behaviors. The experimental drives lasted about six hours spread over six consecutive weeks. Visual behaviors (e.g., mean number of fixations) and driving behaviors (e.g., braking reaction times) were collected during unprovoked near-collision events occurring during a car-following task, with (FCWS group) or without FCWS (No Automation group). FCWS presence reduced the number of near-collision events drastically and enhanced visual behaviors during those events. Unexpectedly, brake reaction times were observed to be significantly longer with FCWS, suggesting a cognitive cost associated with the warning process. Still, the FCWS showed a slight safety benefit for novice drivers attributed to the assistance provided for the situation analysis. Outside the warning events, FCWS presence also impacted car-following behaviors. Drivers took an extra safety margin, possibly to prevent incidental triggering of warnings. The data enlighten the nature of the cognitive processes associated with FCWSs. Altogether, the findings support the general efficiency of FCWSs observed through a massive reduction in the number of near-collision events and point toward the need for further investigations.

Characteristics of automatic emergency braking responses in passenger vehicles evaluated in the IIHS front crash prevention program

Researchers can estimate the potential safety benefits of front crash prevention (FCP) systems by simulating system performance in rear-end crash scenarios reported to police or captured during naturalistic driving. Data to support assumptions about FCP systems in production vehicles, particularly automatic emergency braking (AEB), are limited. This study used detailed information from the Insurance Institute for Highway Safety's (IIHS's) FCP evaluation to characterize interventions in vehicles that performed well (superior-rated vehicles) and those that did not perform as well (basic/advanced-rated vehicles) when approaching a stationary surrogate vehicle on a test track at 20 and 40 km/h, and estimated performance in similar conditions at higher speeds. Vehicle and video data from 3,231 IIHS FCP tests conducted at 20 and 40 km/h and 51 IIHS FCP research tests conducted at 50, 60, and 70 km/h with AEB responses were analyzed. Forward collision warning (FCW) and AEB time-to-collision (TTC), mean deceleration, maximum deceleration, and maximum jerk from the beginning of automatic braking to the end of braking or impact were computed for each test. Each dependent measure was modeled with test speed (20 km/h, 40 km/h), IIHS FCP test rating (superior, basic/advanced), and the interaction between test speed and rating. The models were used to estimate each dependent measure at 50, 60, and 70 km/h, and model predictions were compared with the observed performance of six vehicles in IIHS research test data. Vehicles with superior-rated systems warned and began braking earlier, had a greater average rate of deceleration, reached a higher peak deceleration, and had greater jerk than vehicles with basic/advanced-rated systems, on average. The interaction between test speed and vehicle rating was significant in each linear mixed-effects model, indicating that these differences changed with test speed. FCW and AEB in superior-rated vehicles occurred 0.05 and 0.10 s earlier, respectively, per 10-km/h increase in test speed compared with basic/advanced-rated vehicles. Mean deceleration and maximum deceleration for FCP systems in superior-rated vehicles increased 0.65 m/s² and 0.60 m/s² more, respectively, per 10-km/h increase in test speed than for systems in basic/advanced-rated vehicles. Maximum jerk increased 2.78 m/s³ per 10-km/h increase in test speed for basic/advanced-rated vehicles but decreased 0.25 m/s³ for systems in superior-rated vehicles. The root mean square error between the observed performance and estimated values at 50, 60, and 70 km/h indicated that the linear mixed-effects model had reasonable prediction accuracy for every measure except jerk at these out-of-sample data points. The findings from this study provide insight into the characteristics that make FCP effective for preventing crashes. Based on performance in the IIHS FCP test, vehicles with superior-rated FCP systems had earlier TTC thresholds and braked with greater deceleration that increased with speed compared with basic/advanced-rated systems. The linear mixed-effects models that were developed can guide assumptions about AEB response characteristics for superior-rated FCP systems in future simulation studies.

A Bayesian Regression Analysis of the Effects of Alert Presence and Scenario Criticality on Automated Vehicle Takeover Performance

OBJECTIVE

This study investigates the impact of silent and alerted failures on driver performance across two levels of scenario criticality during automated vehicle transitions of control.

BACKGROUND

Recent analyses of automated vehicle crashes show that many crashes occur after a transition of control or a silent automation failure. A substantial amount of research has been dedicated to investigating the impact of various factors on drivers' responses, but silent failures and their interactions with scenario criticality are understudied.

METHOD

A driving simulator study was conducted comparing scenario criticality, alert presence, and two driving scenarios. Bayesian regression models and Fisher's exact tests were used to investigate the impact of alert and scenario criticality on takeover performance.

RESULTS

The results show that silent failures increase takeover times and the intensity of posttakeover maximum accelerations and decrease the posttakeover minimum time-to-collision. While the predicted average impact of silent failures on takeover time was practically low, the effects on minimum time-to-collision and maximum accelerations were safety-significant. The analysis of posttakeover control interaction effects shows that the effect of alert presence differs by the scenario criticality.

CONCLUSION

Although the impact of the absence of an alert on takeover performance was less than that of scenario criticality, silent failures seem to play a substantial role-by leading to an unsafe maneuver-in critical automated vehicle takeovers.

APPLICATION

Understanding the implications of silent failure on driver's takeover performance can benefit the assessment of automated vehicles' safety and provide guidance for fail-safe system designs.

Can non-crash naturalistic driving data be an alternative to crash data for use in virtual assessment of the safety performance of automated emergency braking systems?

INTRODUCTION

Developers of in-vehicle safety systems need to have data allowing them to identify traffic safety issues and to estimate the benefit of the systems in the region where it is to be used, before they are deployed on-road. Developers typically want in-depth crash data. However, such data are often not available. There is a need to identify and validate complementary data sources that can complement in-depth crash data, such as Naturalistic Driving Data (NDD). However, few crashes are found in such data. This paper investigates how rear-end crashes that are artificially generated from two different sources of non-crash NDD (highD and SHRP2) compare to rear-end in-depth crash data (GIDAS).

METHOD

Crash characteristics and the performance of two conceptual automated emergency braking (AEB) systems were obtained through virtual simulations - simulating the time-series crash data from each data source.

RESULTS

Results show substantial differences in the estimated impact speeds between the artificially generated crashes based on both sources of NDD, and the in-depth crash data; both with and without AEB systems. Scenario types also differed substantially, where the NDD have many fewer scenarios where the following-vehicle is not following the lead vehicle, but instead catches-up at high speed. However, crashes based on NDD near-crashes show similar pre-crash criticality (time-to-collision) to in-depth crash data.

CONCLUSIONS

If crashes based on near-crashes are to be used in the design and assessment of preventive safety systems, it has to be done with great care, and crashes created purely from small amounts of everyday driving NDD are not of much use in such assessment.

PRACTICAL APPLICATIONS

Researchers and developers of in-vehicle safety systems can use the results from this study: (a) when deciding which data to use for virtual safety assessment of such systems, and (b) to understand the limitations of NDD.

On the importance of driver models for the development and assessment of active safety: A new collision warning system to make overtaking cyclists safer

The total number of road crashes in Europe is decreasing, but the number of crashes involving cyclists is not decreasing at the same rate. When cars and bicycles share the same lane, cars typically need to overtake them, creating dangerous conflicts-especially on rural roads, where cars travel much faster than cyclists. In order to protect cyclists, advanced driver assistance systems (ADAS) are being developed and introduced to the market. One of them is a forward collision warning (FCW) system that helps prevent rear-end crashes by identifying and alerting drivers of threats ahead. The objective of this study is to assess the relative safety benefit of a behaviour-based (BB) FCW system that protects cyclists in a car-to-cyclist overtaking scenario. Virtual safety assessments were performed on crashes derived from naturalistic driving data. A series of driver response models was used to simulate different driver reactions to the warning. Crash frequency in conjunction with an injury risk model was used to estimate the risk of cyclist injury and fatality. The virtual safety assessment estimated that, compared to no FCW, the BB FCW could reduce cyclists' fatalities by 53-96% and serious injuries by 43-94%, depending on the driver response model. The shorter the driver's reaction time and the larger the driver's deceleration, the greater the benefits of the FCW. The BB FCW also proved to be more effective than a reference FCW based on the Euro NCAP standard test protocol. The findings of this study demonstrate the BB FCW's great potential to avoid crashes and reduce injuries in car-to-cyclist overtaking scenarios, even when the driver response model did not exceed a comfortable rate of deceleration. The results suggest that a driver behaviour model integrated into ADAS collision threat algorithms can provide substantial safety benefits.

An improved automated braking system for rear-end collisions: A study based on a driving simulator experiment

INTRODUCTION

To assist drivers in avoiding rear-end collisions, many early warning systems have been developed up to date. Autonomous braking technology is also used as the last defense to ensure driver's safety.

METHOD

By taking the accuracy and timeliness of automatic system control into account, this paper proposes a rear-end Real-Time Autonomous Emergency Braking (RTAEB) system. The system inserts brake intervention based on drivers' real-time conflict identification and collision avoidance performance. A driving simulator-based experiment under different traffic conditions and deceleration scenarios were conducted to test the different thresholds to trigger intervention and the intervention outcomes. The system effectiveness is verified by four evaluation indexes, including collision avoidance rate, accuracy rate, sensitivity rate, and precision rate.

RESULTS

The results showed that the system could help avoid all collision events successfully and enlarge the final headway distance, and a TTC threshold of 1.5 s and a maximum deceleration threshold of -7.5 m/s² could achieve the best collision avoidance effect. The paper demonstrates the situations that are more inclined to trigger the RTAEB (i.e., a sudden brake of the leading vehicle and a small car-following distance). Moreover, the study shows that driver characteristics (i.e., gender and profession) have no significant association with system trigger. Practical Applications: The study suggests that development of collision avoidance systems design should pay attention to both the real-time traffic situation and drivers' collision avoidance capability under the present situation.

Computational modeling of driver pre-crash brake response, with and without off-road glances: Parameterization using real-world crashes and near-crashes

When faced with an imminent collision threat, human vehicle drivers respond with braking in a manner which is stereotypical, yet modulated in complex ways by many factors, including the specific traffic situation and past driver eye movements. A computational model capturing these phenomena would have high applied value, for example in virtual vehicle safety testing methods, but existing models are either simplistic or not sufficiently validated. This paper extends an existing quantitative driver model for initiation and modulation of pre-crash brake response, to handle off-road glance behavior. The resulting models are fitted to time-series data from real-world naturalistic rear-end crashes and near-crashes. A stringent parameterization and model selection procedure is presented, based on particle swarm optimization and maximum likelihood estimation. A major contribution of this paper is the resulting first-ever fit of a computational model of human braking to real near-crash and crash behavior data. The model selection results also permit novel conclusions regarding behavior and accident causation: Firstly, the results indicate that drivers have partial visual looming perception during off-road glances; that is, evidence for braking is collected, albeit at a slower pace, while the driver is looking away from the forward roadway. Secondly, the results suggest that an important causation factor in crashes without off-road glances may be a reduced responsiveness to visual looming, possibly associated with cognitive driver state (e.g., drowsiness or erroneous driver expectations). It is also demonstrated that a model parameterized on less-critical data, such as near-crashes, may also accurately reproduce driver behavior in highly critical situations, such as crashes.

A model for naturalistic glance behavior around Tesla Autopilot disengagements

OBJECTIVE

We present a model for visual behavior that can simulate the glance pattern observed around driver-initiated, non-critical disengagements of Tesla's Autopilot (AP) in naturalistic highway driving.

BACKGROUND

Drivers may become inattentive when using partially-automated driving systems. The safety effects associated with inattention are unknown until we have a quantitative reference on how visual behavior changes with automation.

METHODS

The model is based on glance data from 290 human initiated AP disengagement epochs. Glance duration and transition were modelled with Bayesian Generalized Linear Mixed models.

RESULTS

The model replicates the observed glance pattern across drivers. The model's components show that off-road glances were longer with AP active than without and that their frequency characteristics changed. Driving-related off-road glances were less frequent with AP active than in manual driving, while non-driving related glances to the down/center-stack areas were the most frequent and the longest (22% of the glances exceeded 2 s). Little difference was found in on-road glance duration.

CONCLUSION

Visual behavior patterns change before and after AP disengagement. Before disengagement, drivers looked less on road and focused more on non-driving related areas compared to after the transition to manual driving. The higher proportion of off-road glances before disengagement to manual driving were not compensated by longer glances ahead.

APPLICATION

The model can be used as a reference for safety assessment or to formulate design targets for driver management systems.

Multiobjective optimization on cooperative control of autonomous emergency steering and occupant restraint system for enhancing occupant safety

Occupant safety remains one of the most challenging and significant design considerations in the automotive and transportation industry. Nevertheless, independently developed active or passive safety systems may lead to unsatisfactory protective performance under the critical driving scenarios. This study aimed to conduct multiobjective optimization of the cooperative controls between autonomous emergency steering (AES) and occupant restraint system (ORS) to explore the potential occupant injury reduction capability as well as mechanisms subjected to a frontal collision. First, a multiple simulation approach comprising PreScan/Simulink, LS-DYNA, Madymo was used to correlate the control parameters of the safety systems and occupant injuries quantitatively. Then the control parameters of AES and ORS were selected as the design variables after sensitivity analysis, and injury responses of the sampling points were extracted by the multiple simulation approach. Surrogate models and multiobjective optimization algorithm were used to determine the optimum design in cooperative controls of AES and ORS maneuvers, from which in-depth effect mechanisms that contributed to the improvement of occupant protection were identified. Compared to the baseline design, the optimum control parameters of AES-ORS integration substantially decreased the occupant injuries of the head, chest and neck, and consequently led to a reduction of 33.02% in the overall injury risk. This study is anticipated to demonstrate a new design approach for the control system, thereby enhancing occupant safety.

Analyzing drivers' hazard recognition: Precursors to single-vehicle collisions

Extensive driver behavior and performance information provided by real-world video surveillance and sensor data in the SHRP2 Naturalistic Driving Study has enabled the examination of new layers and pathways leading to crash outcomes. We note that the prominence of hazards and the importance of recognizing them vary systematically across single vs. multi-vehicle crashes, and address a fundamental question about safety: why do around three-quarters of drivers involved in single-vehicle crashes not recognize, perceive, or react to the precipitating event (PE)? Using a path-analytic framework through marginal effects, this study investigates factors correlated to recognition of the PE in single-vehicle events, and how these correlations may act as crash precursors. Logit models, accounting for heterogeneity among events and drivers by estimating both fixed and random parameters, quantified correlations among key variables, given a crash or near-crash event (N = 543). The type of PE, roadway environment factors, and driving maneuvers heavily influenced recognition chances. Drivers had a harder time recognizing less conspicuous hazards (e.g. departing the travel way, decreased recognition chances by 48.29%), but seemed better at recognizing prominent hazards (e.g. vehicle losing control, increased recognition chances by 46.71%). In addition, drivers are less likely to recognize PEs when executing less involved driving maneuvers in more relaxed environments, such as daylight (decreased recognition chances by 16.00%), but are more adept in environments that already demand more attention. Recognition reduced the chances of a crash by 12.23%, so we found similar correlations with crash outcome. Future intelligent transportation systems may focus on increasing driver recognition of potential hazards by bringing attention to less conspicuous hazards and less involved driving environments and actions.

Truck drivers' behavior in encounters with vulnerable road users at intersections: Results from a test-track experiment

Crashes involving cyclists and pedestrians in Europe cause the deaths of about 7600 persons every year. Both cyclists and pedestrians are especially exposed in crashes with motorized vehicles and collisions with trucks can lead to severe injury outcomes. The two most frequent crash scenarios between trucks and these vulnerable road users (VRU) are: a) when the truck wants to turn right at an intersection, with a cyclist riding parallel and planning to cross the intersection and b) when a pedestrian crosses in front of the truck in perpendicular direction to the movement of the truck. Advanced Driver Assistance Systems (ADAS)-that are expected to prevent or mitigate these crashes-benefit from detailed information about the behavior of truck drivers. This study is a first exploration of this research area, with the aim to assess how drivers negotiate the encounters with VRUs in the two scenarios described above. Thirteen participants drove an instrumented truck on a test-track. After some baseline recordings, the drivers experienced two laps where they encountered a cyclist target and a pedestrian target crossing their path. The results show that the truck drivers adapted their kinematic and visual behavior in the laps where the VRU targets were crossing the intersection, compared to the baseline laps. The speed profiles of the drivers diverged approximately 30 m from the intersection and glances were directed more often towards front right and right, during the scenario with the cyclist in comparison to baseline laps. For the scenario with the pedestrian crossing, the drivers changed their speed about 14 m from the intersection and glances were directed more often towards the front center, compared to baseline laps. As a result, both the speed and distance from the intersection at the end of the maneuver were significantly different between VRU and baseline laps. Overall, the findings provide valuable information for the design of ADAS that warn the drivers about the presence of a cyclist travelling in parallel direction or that intervene to avoid a collision with a cyclist or pedestrian.

Evaluating automated emergency braking performance in simulated car-to-two-wheeler crashes in China: A comparison between C-NCAP tests and in-depth crash data

Automated Emergency Braking (AEB) is an effective way to prevent crashes from happening or mitigate their severity. Because riders of two-wheelers (TWs) are among the most vulnerable road users, New Car Assessment Programs, like the China New Car Assessment Program (C-NCAP), have recently introduced test scenarios for the assessment of AEB for cars encountering TWs (TW-AEB). The main aim of this study was to determine how well two different C-NCAP test scenario datasets reflect real-world crash scenarios for the purpose of assessing TW-AEB performance. We used virtual counterfactual simulations to determine whether the hypothetical TW-AEB's performance, when applied to the two C-NCAP datasets, was similar to its performance when applied to a set of reconstructed car-to-TW crashes representing real-world crashes. The test datasets were the current C-NCAP scenario set and a proposed C-NCAP scenario set; the real-world crash dataset comprised 113 reconstructed crashes from the Shanghai United Road Traffic Safety Scientific Research Center database (SHUFO). The performances were compared with respect to crash avoidance rate and the characteristics of the remaining crashes. A substantially higher proportion of crashes was avoided in the current C-NCAP scenario set than in the other two (with the sensor field of view (FoV) set to 70° and the activation time to 1.1 s TTC). In fact, with these parameter settings, no crashes remained in the current C-NCAP scenarios, while only 37% and 46% of the crashes in the proposed C-NCAP scenario set and SHUFO crash set were avoided, respectively. Our findings show that TW-AEB systems which are optimized for the current C-NCAP test scenarios are likely to provide benefits in real-world crashes. However, including additional test scenarios which reflect real-world crash situations more accurately would likely lead to a higher correlation between C-NCAP scores and real-world TW-AEB performance. In particular, we recommend the introduction of longitudinal same-direction scenarios with the car or TW turning and perpendicular scenarios with high TW traveling speed, in future C-NCAP releases. Inclusion of these scenarios in C-NCAP might reward improvements of future TW-AEBs toward systems that can save more lives. Furthermore, our study shows that there is likely to be a substantial number of crashes with an impact speed higher than 40 km/h still remaining even after the widespread application of TW-AEB. Therefore, passive safety for TW riders on Chinese roads will be still needed.

Steering or braking avoidance response in SHRP2 rear-end crashes and near-crashes: A decision tree approach

OBJECTIVE

The paper presents a systematic analysis of drivers' crash avoidance response during crashes and near-crashes and developed a machine learning-based predictive model that can determine driver maneuver using pre-incident driver behavior and driving context.

METHODS

We analyzed 286 naturalistic rear-end crashes and near-crashes from the SHRP2 naturalistic driving study. All the events were manually reduced using face video (face and forward) and kinematic responses. In this paper, we developed new reduction variables that enhanced the understanding of drivers' gaze behavior and roadway attention behavior during these events. These features reflected how the event criticality, measured using time to collision, related to drivers' pre-incident behavior (secondary behavior, gaze behavior), and drivers' perception of the event (physical reaction and maneuver). The imperative understanding of such relations was validated using a random forest- (RF) based classifier, which efficiently predicted if a driver was going to brake or change the lane as an avoidance maneuver.

RESULTS

The RF presented in this paper effectively explored the nonlinear patterns in the data and was highly accurate (∼96 %) in its prediction. A further analysis of the RF model showed that six features played a pivotal role in the decision logic. These included the drivers' last glance duration before the event, last glance eccentricity, duration of 'eyes on road' immediately before the event, the time instance and criticality when the driver perceives the threat as well as acknowledge the threat, and possibility of an escape path in the adjacent lane. Using partial dependency plots, we also showed how different thresholds of these feature variables determined the drivers' maneuver intention.

CONCLUSIONS

In this paper we analyzed driving context, drivers' behavior, event criticality, and drivers' response in a unified structure to predict their avoidance response. To the best of our knowledge, this is the first such effort where large-scale naturalistic data (crashes and near crashes) was analyzed for prediction of drivers' maneuver and determined key behavioral and contextual factors that contribute to this avoidance maneuver.

A method for predicting crash configurations using counterfactual simulations and real-world data

Traffic safety technologies revolve around two principle ideas; crash avoidance and injury mitigation for inevitable crashes. The development of relevant vehicle injury mitigating technologies should consider the interaction of those two technologies, ensuring that the inevitable crashes can be adequately managed by the occupant and vulnerable road user (VRU) protection systems. A step towards that is the accurate description of the expected crashes remaining when crash-avoiding technologies are available in vehicles. With the overall objective of facilitating the assessment of future traffic safety, this study develops a method for predicting crash configurations when introducing crash-avoiding countermeasures. The predicted crash configurations are one important factor for prioritizing the evaluation and development of future occupant and VRU protection systems. By using real-world traffic accident data to form the baseline and performing counterfactual model-in-the-loop (MIL) pre-crash simulations, the change in traffic situations (vehicle crashes) provided by vehicles with crash-avoiding technologies can be predicted. The method is built on a novel crash configuration definition, which supports further analysis of the in-crash phase. By clustering and grouping the remaining crashes, a limited number of crash configurations can be identified, still representing and covering the real-world variation. The developed method was applied using Swedish national- and in-depth accident data related to urban intersections and highway driving, and a conceptual Autonomous Emergency Braking system (AEB) computational model. Based on national crash data analysis, the conflict situations Same-Direction rear-end frontal (SD-ref) representing 53 % of highway vehicle-to-vehicle (v2v) crashes, and Straight Crossing Path (SCP) with 21 % of urban v2v intersection crashes were selected for this study. Pre-crash baselines, for SD-ref (n = 1010) and SCP (n = 4814), were prepared based on in-depth accident data and variations of these. Pre-crash simulations identified the crashes not avoided by the conceptual AEB, and the clustering of these revealed 5 and 52 representative crash configurations for the highway SD-ref and urban intersection SCP conflict situations, respectively, to be used in future crashworthiness studies. The results demonstrated a feasible way of identifying, in a predictive way, relevant crash configurations for in-crash testing of injury prevention capabilities.

How Do Drivers Respond to Silent Automation Failures? Driving Simulator Study and Comparison of Computational Driver Braking Models

OBJECTIVE

This paper aims to describe and test novel computational driver models, predicting drivers' brake reaction times (BRTs) to different levels of lead vehicle braking, during driving with cruise control (CC) and during silent failures of adaptive cruise control (ACC).

BACKGROUND

Validated computational models predicting BRTs to silent failures of automation are lacking but are important for assessing the safety benefits of automated driving.

METHOD

Two alternative models of driver response to silent ACC failures are proposed: a looming prediction model, assuming that drivers embody a generative model of ACC, and a lower gain model, assuming that drivers' arousal decreases due to monitoring of the automated system. Predictions of BRTs issued by the models were tested using a driving simulator study.

RESULTS

The driving simulator study confirmed the predictions of the models: (a) BRTs were significantly shorter with an increase in kinematic criticality, both during driving with CC and during driving with ACC; (b) BRTs were significantly delayed when driving with ACC compared with driving with CC. However, the predicted BRTs were longer than the ones observed, entailing a fitting of the models to the data from the study.

CONCLUSION

Both the looming prediction model and the lower gain model predict well the BRTs for the ACC driving condition. However, the looming prediction model has the advantage of being able to predict average BRTs using the exact same parameters as the model fitted to the CC driving data.

APPLICATION

Knowledge resulting from this research can be helpful for assessing the safety benefits of automated driving.

Modelling cyclists' comfort zones from obstacle avoidance manoeuvres

This paper introduces a framework for modelling the cyclist's comfort zone. Unlike the driver's comfort zone, little is known about the cyclist's. The framework draws on existing literature in cognitive science about driver behaviour to explain experimental results from cycling field trials, and the modelling of these results. We modelled braking and steering manoeuvres from field data of cyclists' obstacle avoidance within their comfort zone. Results show that when cyclists avoided obstacles by braking, they kept a constant deceleration; as speed increased, they started to brake earlier, farther from the obstacle, maintaining an almost constant time to collision. When cyclists avoided obstacles by steering, they maintained a constant distance from the object, independent of speed. Overall, the higher the speed, the more the steering manoeuvres were temporally delayed compared to braking manoeuvres. We discuss these results and other similarities between cyclist and driver behaviour during obstacle avoidance. Implications for the design of acceptable active safety and infrastructure design are also addressed.

How do drivers overtake pedestrians? Evidence from field test and naturalistic driving data

For pedestrians, the risk of dying in a traffic accident is highest on rural roads, which are often characterized by a lack of sidewalks and high traffic speed. In fact, hitting the pedestrian during an overtaking attempt is a common crash scenario. To develop active safety systems that avoid such crashes, it is necessary to understand and model driver behavior during the overtaking maneuvers, so that system interventions are acceptable because they happen outside drivers' comfort zone. Previous modeling of driver behavior in interactions with pedestrians primarily focused on road crossing scenarios. The aim of this study was, instead, to address pedestrian-overtaking maneuvers on rural roads. We focused our analysis on how drivers adjust their behavior with respect to three safety metrics (in order of importance): 1) minimum lateral clearance when passing the pedestrian, 2) overtaking speed at that moment, and 3) the time-to-collision at the moment of steering away to start the overtaking maneuver. The influence of three factors on the safety metrics was investigated: 1) walking direction (same as the overtaking vehicle or opposite), 2) walking position (on the edge of the vehicle lane or 0.5 m away from the edge on the paved shoulder), and 3) oncoming traffic (absent or present). Seventy-seven overtaking maneuvers in France from the naturalistic driving study UDRIVE and 297 maneuvers in Sweden from field tests were analyzed. Bayesian regression was used to model how minimum lateral clearance and overtaking speed depended on the three factors. Results showed that drivers maintained smaller minimum lateral clearance and lower overtaking speed when the pedestrian was walking in the opposite direction, on the lane edge, or when oncoming traffic was present. Minimum lateral clearance and time-to-collision were only weakly correlated with overtaking speed. The regression models predicted distributions similar to those actually observed in the data. The time-to-collision at the moment of steering away was comparable in value to the time-to-collision used by Euro NCAP for testing active safety systems in car-to-pedestrian longitudinal scenarios since 2018. This study is the first to analyze driver behavior when overtaking pedestrians, based on field test and naturalistic driving data. Results suggest that pedestrian safety is particularly endangered in situations when the pedestrian is walking opposite to traffic, close to the lane, and when oncoming traffic is present. The Bayesian regression models from this study can be used in active safety systems to model drivers' comfort in overtaking maneuvers.

Safety benefit assessment of autonomous emergency braking and steering systems for the protection of cyclists and pedestrians based on a combination of computer simulation and real-world test results

Cyclists and pedestrians account for a significant share of fatalities and serious injuries in the road transport system. In order to protect them, advanced driver assistance systems are being developed and introduced to the market, including autonomous emergency braking and steering systems (AEBSS) that autonomously perform braking or an evasive manoeuvre by steering in case of a pending collision, in order to avoid the collision or mitigate its severity. This study proposes a new prospective framework for quantifying safety benefit of AEBSS for the protection of cyclists and pedestrians in terms of saved lives and reduction in the number of people suffering serious injuries. The core of the framework is a novel application of Bayesian inference in such a way that prior information from counterfactual simulation is updated with new observations from real-world testing of a prototype AEBSS. As an illustration of the method, the framework is applied for safety benefit assessment of the AEBSS developed in the European Union (EU) project PROSPECT. In this application of the framework, counterfactual simulation results based on the German In-Depth Accident Study Pre-Crash Matrix (GIDAS-PCM) data were combined with results from real-world tests on proving grounds. The proposed framework gives a systematic way for the combination of results from different sources and can be considered for understanding the real-world benefit of new AEBSS. Additionally, the Bayesian modelling approach used in this paper has a great potential to be used in a wide range of other research studies.

How many crashes can connected vehicle and automated vehicle technologies prevent: A meta-analysis

The connected and automated vehicle (CAV) technologies have made great progresses. It has been commonly accepted that CV or AV technologies would reduce human errors in driving and benefit traffic safety. However, the answer of how many crashes can be prevented because of CV or AV technologies has not reached a consistent conclusion. In order to quantitatively answer this question, this study used meta-analysis to evaluate the safety effectiveness of nine common and important CV or AV technologies, and tested the safety effectiveness of these technologies for six countries. First, 73 studies about the safety impact of CV or AV technologies were filtered out from 826 CAV-related papers or reports. Second, the safety impacts of these technologies with regard to assistant types and triggering times have been compared. It shows AV technologies can play a more significant role than CV technologies, and the technologies with closer triggering time to collision time have greater safety effectiveness. Third, in the meta-analysis, the random effect model was used to evaluate the safety effectiveness, and the funnel plots and trim-and-fill method were used to evaluate and adjust publication bias, so as to objectively evaluate the safety effectiveness of each technology. Then, according to the crash data of six countries, the comprehensive safety effectiveness and compilation of safety effectiveness of the above technologies were calculated. The results show that if all of technologies were implemented in the six countries, the average number of crashes could be reduced by 3.40 million, among which the India would reduce the most (54.24%). Additionally, different countries should develop different development strategies, e.g., USA should prioritize the development of the lane change warning and intersection warning, the UK should prioritize applications related to intersection warning and rear-end warning. Overall, this study provides comprehensive and quantitative understating of the safety effectiveness of CA or AV technologies and would contribute to government, vehicle companies, and agencies in deciding the development priority of CA or AV technologies.

The role of off-path glances: A quantitative analysis of rear-end conflicts involving Chinese professional truck drivers as the striking partners

INTRODUCTION

Rear-end crashes are one of the most frequent crash types in China, leading to significant economic and societal losses. The development of active safety systems - such as Automatic Emergency Braking System (AEBS) - could avoid or mitigate the consequences of these crashes in Chinese traffic situations. However, a clear understanding of the crash causation mechanisms is necessary for the design of these systems.

METHOD

Manually coded variables were extracted from a naturalistic driving study conducted with commercial vehicles in Shanghai. Quantitative analyses of rear-end crashes and near crashes (CNC) were conducted to assess the prevalence, duration, and location of drivers' off-path glances, the influence of lead vehicle brake lights on drivers' last off-path glance, and driver brake onset, and the influence of off-path glances and kinematic criticality on drivers' response to conflicts.

RESULTS

The results indicate that the Chinese truck drivers in our study rarely engage in distracting activities involving a phone or other handheld objects while driving. Instead, they direct their off-path glances mainly toward the mirrors, and the duration of off-path glances leading to critical situations are shorter compared to earlier analyses performed in Western countries. The drivers also often keep small margins.

CONCLUSIONS

Overall, the combination of short time headway with off-path glances directed toward the mirror originates visual mismatches which, associated to a rapid change in the kinematic situation, cause the occurrence of rear-end CNC. When drivers look back toward the road after an off-path glance, a fast response seems to be triggered by lower values of looming compared to previous studies, possibly because of the short time headways. Practical Application: The results have practical implications for the development of driver models, for the design of active safety systems and automated driving, and for the design of campaigns promoting safe driving.

Control Oriented Prediction of Driver Brake Intention and Intensity Using a Composite Machine Learning Approach

Driver perception, decision, and control behaviors are easily affected by traffic conditions and driving style, showing the tendency of randomness and personalization. Brake intention and intensity are integrated and control-oriented parameters that are crucial to the development of an intelligent braking system. In this paper, a composite machine learning approach was proposed to predict driver brake intention and intensity with a proper prediction horizon. Various driving data were collected from Controller Area Network (CAN) bus under a real driving condition, which mainly contained urban and rural road types. ReliefF and RReliefF (they don’t have abbreviations) algorithms were employed as feature subset selection methods and applied in a prepossessing step before the training. The rank importance of selected predictors exhibited different trends or even negative trends when predicting brake intention and intensity. A soft clustering algorithm, Fuzzy C-means, was adopted to label the brake intention into categories, namely slight, medium, intensive, and emergency braking. Data sets with misplaced labels were used for training of an ensemble machine learning method, random forest. It was validated that brake intention could be accurately predicted 0.5 s ahead. An open-loop nonlinear autoregressive with external input (NARX) network was capable of learning the long-term dependencies in comparison to the static neural network and was suggested for online recognition and prediction of brake intensity 1 s in advance. As system redundancy and fault tolerance, a close-loop NARX network could be adopted for brake intensity prediction in the case of possible sensor failure and loss of CAN message.

Toward Computational Simulations of Behavior During Automated Driving Takeovers: A Review of the Empirical and Modeling Literatures

OBJECTIVE

This article provides a review of empirical studies of automated vehicle takeovers and driver modeling to identify influential factors and their impacts on takeover performance and suggest driver models that can capture them.

BACKGROUND

Significant safety issues remain in automated-to-manual transitions of vehicle control. Developing models and computer simulations of automated vehicle control transitions may help designers mitigate these issues, but only if accurate models are used. Selecting accurate models requires estimating the impact of factors that influence takeovers.

METHOD

Articles describing automated vehicle takeovers or driver modeling research were identified through a systematic approach. Inclusion criteria were used to identify relevant studies and models of braking, steering, and the complete takeover process for further review.

RESULTS

The reviewed studies on automated vehicle takeovers identified several factors that significantly influence takeover time and post-takeover control. Drivers were found to respond similarly between manual emergencies and automated takeovers, albeit with a delay. The findings suggest that existing braking and steering models for manual driving may be applicable to modeling automated vehicle takeovers.

CONCLUSION

Time budget, repeated exposure to takeovers, silent failures, and handheld secondary tasks significantly influence takeover time. These factors in addition to takeover request modality, driving environment, non-handheld secondary tasks, level of automation, trust, fatigue, and alcohol significantly impact post-takeover control. Models that capture these effects through evidence accumulation were identified as promising directions for future work.

APPLICATION

Stakeholders interested in driver behavior during automated vehicle takeovers may use this article to identify starting points for their work.

Forward collision warning based on a driver model to increase drivers' acceptance

Objective: Systems that can warn the driver of a possible collision with a vulnerable road user (VRU) have significant safety benefits. However, incorrect warning times can have adverse effects on the driver. If the warning is too late, drivers might not be able to react; if the warning is too early, drivers can become annoyed and might turn off the system. Currently, there are no methods to determine the right timing for a warning to achieve high effectiveness and acceptance by the driver. This study aims to validate a driver model as the basis for selecting appropriate warning times. The timing of the forward collision warnings (FCWs) selected for the current study was based on the comfort boundary (CB) model developed during a previous project, which describes the moment a driver would brake. Drivers' acceptance toward these warnings was analyzed. The present study was conducted as part of the European research project PROSPECT ("Proactive Safety for Pedestrians and Cyclists"). Methods: Two warnings were selected: One inside the CB and one outside the CB. The scenario tested was a cyclist crossing scenario with time to arrival (TTA) of 4 s (it takes the cyclist 4 s to reach the intersection). The timing of the warning inside the CB was at a time to collision (TTC) of 2.6 s (asymptotic value of the model at TTA = 4 s) and the warning outside the CB was at TTC = 1.7 s (below the lower 95% value at TTA = 4 s). Thirty-one participants took part in the test track study (between-subjects design where warning time was the independent variable). Participants were informed that they could brake any moment after the warning was issued. After the experiment, participants completed an acceptance survey. Results: Participants reacted faster to the warning outside the CB compared to the warning inside the CB. This confirms that the CB model represents the criticality felt by the driver. Participants also rated the warning inside the CB as more disturbing, and they had a higher acceptance of the system with the warning outside the CB. The above results confirm the possibility of developing wellsaccepted warnings based on driver models. Conclusions: Similar to other studies' results, drivers prefer warning times that compare with their driving behavior. It is important to consider that the study tested only one scenario. In addition, in this study, participants were aware of the appearance of the cyclist and the warning. A further investigation should be conducted to determine the acceptance of distracted drivers.

Definition of run-off-road crash clusters-For safety benefit estimation and driver assistance development

Single-vehicle run-off-road crashes are a major traffic safety concern, as they are associated with a high proportion of fatal outcomes. In addressing run-off-road crashes, the development and evaluation of advanced driver assistance systems requires test scenarios that are representative of the variability found in real-world crashes. We apply hierarchical agglomerative cluster analysis to define similarities in a set of crash data variables, these clusters can then be used as the basis in test scenario development. Out of 13 clusters, nine test scenarios are derived, corresponding to crashes characterised by: drivers drifting off the road in daytime and night-time, high speed departures, high-angle departures on narrow roads, highways, snowy roads, loss-of-control on wet roadways, sharp curves, and high speeds on roads with severe road surface conditions. In addition, each cluster was analysed with respect to crash variables related to the crash cause and reason for the unintended lane departure. The study shows that cluster analysis of representative data provides a statistically based method to identify relevant properties for run-off-road test scenarios. This was done to support development of vehicle-based run-off-road countermeasures and driver behaviour models used in virtual testing. Future studies should use driver behaviour from naturalistic driving data to further define how test-scenarios and behavioural causation mechanisms should be included.

Real life safety benefits of increasing brake deceleration in car-to-pedestrian accidents: Simulation of Vacuum Emergency Braking

The objective of this study is to predict the real-life benefits, namely the number of injuries avoided rather than the reduction in impact speed, offered by a Vacuum Emergency Brake (VEB) added to a pedestrian automated emergency braking (AEB) system. We achieve this through the virtual simulation of simplified mathematical models of a system which incorporates expected future advances in technology, such as a wide sensor field of view, and reductions in the time needed for detection, classification, and brake pressure build up. The German In-Depth Accident Study database and the related Pre Crash Matrix, both released in the beginning of 2016, were used for this study and resulted in a final sample of 526 collisions between passenger car fronts and pedestrians. Weight factors were calculated for both simulation model and injury risk curves to make the data representative of Germany as a whole. The accident data was used with a hypothetical AEB system in a simulation model, and injury risk was calculated from the new impact speed using injury risk curves to generate new situations using real accidents. Adding a VEB to a car with pedestrian AEB decreased pedestrian casualties by an additional 8-22%, depending on system setting and injury level, over the AEB-only system. The overall decrease in fatalities was 80-87%, an improvement of 8%. Collision avoidance was improved by 14-28%. VEB with a maximum deceleration in the middle of the modelled performance range has an effectiveness similar to that of an "early activation" system, where the AEB is triggered as early as 2 s before collision. VEB may therefore offer a substantial increase in performance without increasing false positive rates, which earlier AEB activation does. Most collisions and injuries can be avoided when AEB is supplemented by the high performance VEB; remaining cases are characterised by high pedestrian walking speed and late visibility due to view obstructions. VEB is effective in all analysed accident scenarios.