Modeling driver behavior in critical traffic scenarios for the safety assessment of automated driving.

OBJECTIVE: Before market introduction, the safety of highly automated driving systems needs to be assessed prospectively. BMW has developed a holistic approach for the assessment of the traffic safety impact by these systems in which stochastic traffic simulations play a significant role. A driver behavior model which represents realistic driver behavior ranging from performance in non-critical everyday driving toward performance in critical situations is key for this approach. To ensure trustworthy results, validation of the driver model is needed. The paper aims at demonstrating that the presented driver model acts realistically in different critical real-world traffic scenarios. METHODS: BMW has been developing the Stochastic Cognitive Model (SCM) which models cognitive processes in traffic situations. These processes range from information acquisition by gaze behavior, mental representation of the environment, recognition of situations from the visual information and reaction to the situation. The driver model combines these cognitive processes with stochastic driver parameters to obtain a variation in driver behavior in simulations. Especially visual attention modeling is key to realistic traffic interactions in simulations as this is the input for the sequential cognitive processes, i.e., the recognition of situations and the reaction to the situation. Modeling of driver's gaze behavior with SCM is thus shown in this paper. RESULTS: SCM is applied in three critical real-world traffic scenarios in which gaze behavior, brake reaction times and time-to-collisions are evaluated and compared to the real-world data. Due to the stochastic approach not only a single SCM agent but a collective of virtual SCM test drivers is assessed. Results show that SCM is capable to simulate the influence of visual inattention on collision risk. CONCLUSION: Realistic driver behavior in simulations can be achieved by using SCM. Especially in the presented critical scenarios SCM is able to represent real-world driving behavior which is determined particularly by its gaze behavior and subsequent reaction. Driving performance varies over different SCM agents which mean that different driving behavior can be simulated with SCM as well. However, the investigation in this paper included only three real-world cases. Therefore, further critical, and additionally non-critical scenarios need to be investigated in the future.

Modelling discomfort: How do drivers feel when cyclists cross their path?

Many cyclist fatalities occur on roads when crossing a vehicle path. Active safety systems address these interactions. However, the driver behaviour models that these systems use may not be optimal in terms of driver acceptance. Incorporating explicit estimates of driver discomfort might improve acceptance. This study quantified the degree of discomfort experienced by drivers when cyclists crossed their travel path. Participants were instructed to drive through an intersection in a fixed-base simulator or on a test track, following the same experimental protocol. During the experiments, three variables were controlled: 1) the car speed (30, 50 km/h), 2) the bicycle speed (10, 20 km/h), and 3) the bicycle-car encroachment sequence (bicycle clears the intersection first, potential 50 %-overlap crash, and car clears the intersection first). For each trial, a covariate, the car's time-to-arrival at the intersection when the bicycle appears (TTAvis), was calculated. After each trial, the participants were asked to report their experienced discomfort on a 7-point Likert scale ranging from no discomfort (1) to maximum discomfort (7). The effect of the three controlled variables and the effect of TTAvis on drivers' discomfort were estimated using cumulative link mixed models (CLMM). Across both experimental environments, the controlled variables were shown to significantly influence discomfort. TTAvis was shown to have a significant effect on discomfort as well; the closer to zero TTAvis was (i.e., the more critical the situation), the more likely the driver reported great discomfort. The prediction accuracies of the CLMM with all three controlled variables and the CLMM with TTAvis were similar, with an average accuracy between 40 and 50 % for the exact discomfort level and between 80 and 85 % allowing deviations by one step. Our model quantifies driver discomfort. Such model may be included in the decision-making algorithms of active safety systems to improve driver acceptance. In fact, by tuning system activation times depending on the expected level of discomfort that a driver would experience in such situation, a system is not likely to annoy a driver.

Multi-agent traffic simulations to estimate the impact of automated technologies on safety.

Objective: The objective of this article was to develop a multi-agent traffic simulation methodology to estimate the potential road safety improvements of automated vehicle technologies. Methods: We developed a computer program that merges road infrastructure data with a large number of vehicles, drivers, and pedestrians. Human errors are induced by modeling inattention, aimless driving, insufficient safety confirmation, misjudgment, and inadequate operation. The program was applied to simulate traffic in a prescribed area in Tsukuba city. First, a 100% manual driving scenario was set to simulate traffic for a total preset vehicle travel distance. The crashes from this simulation were compared with real-world crash data from the prescribed area from 2012 to 2017. Thereafter, 4 additional scenarios of increasing levels of automation penetration (including combinations of automated emergency braking [AEB], lane departure warning [LDW], and SAE Level 4 functions) were implemented to estimate their impact on safety. Results: Under manual driving, the system simulated a total of 859 crashes including single-car lane departure, car-to-car, and car-to-pedestrian crashes. These crashes tended to occur in locations similar to real-world crashes. The number of crashes predicted decreased to 156 cases with increasing level of automation. All of the technologies considered contributed to the decrease in crashes. Crash reductions attributable to AEB and LDW in the simulations were comparable to those reported in recent field studies. For the highest levels of automation, no assessment data were available and hence the results should be carefully treated. Further, in modeling automated functions, potentially negative aspects such as sensing failure or human overreliance were not incorporated. Conclusions: We developed a multi-agent traffic simulation methodology to estimate the effect of different automated vehicle technologies on safety. The crash locations resulting from simulations of manual driving within a limited area in Japan were preliminary assessed by comparison with real-world crash data collected in the same area. Increasing penetration levels of AEB and LDW led to a large reduction in both the frequency and severity of rear-end crashes, followed by car-to-car head-on crashes and single-vehicle lane departure crashes. Preliminary estimations of the potential safety improvements that may be achieved with highly automated driving technologies were also obtained.

Modelling how drivers respond to a bicyclist crossing their path at an intersection: How do test track and driving simulator compare?

Bicyclist fatalities are a great concern in the European Union. Most of them are due to crashes between motorized vehicles and bicyclists at unsignalised intersections. Different countermeasures are currently being developed and implemented in order to save lives. One type of countermeasure, active safety systems, requires a deep understanding of driver behaviour to be effective without being annoying. The current study provides new knowledge about driver behaviour which can inform assessment programmes for active safety systems such as Euro NCAP. This study investigated how drivers responded to bicyclists crossing their path at an intersection. The influences of car speed and cyclist speed on the driver response process were assessed for three different crossing configurations. The same experimental protocol was tested in a fixed-base driving simulator and on a test track. A virtual model of the test track was used in the driving simulator to keep the protocol as consistent as possible across testing environments. Results show that neither car speed nor bicycle speed directly influenced the response process. The crossing configuration did not directly influence the braking response process either, but it did influence the strategy chosen by the drivers to approach the intersection. The point in time when the bicycle became visible (which depended on the car speed, the bicycle speed, and the crossing configuration) and the crossing configuration alone had the largest effects on the driver response process. Dissimilarities between test-track and driving-simulator studies were found; however, there were also interesting similarities, especially in relation to the driver braking behaviour. Drivers followed the same strategy to initiate braking, independent of the test environment. On the other hand, the test environment affected participants' strategies for releasing the gas pedal and regulating deceleration. Finally, a mathematical model, based on both experiments, is proposed to characterize driver braking behaviour in response to bicyclists crossing at intersections. This model has direct implications on what variables an in-vehicle safety system should consider and how tests in evaluation programs should be designed.

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

As the development and deployment of in-vehicle intelligent safety systems (ISS) for crash avoidance and mitigation have rapidly increased in the last decades, the need to evaluate their prospective safety benefits before introduction has never been higher. Counterfactual simulations using relevant mathematical models (for vehicle dynamics, sensors, the environment, ISS algorithms, and models of driver behavior) have been identified as having high potential. However, although most of these models are relatively mature, models of driver behavior in the critical seconds before a crash are still relatively immature. There are also large conceptual differences between different driver models. The objective of this paper is, firstly, to demonstrate the importance of the choice of driver model when counterfactual simulations are used to evaluate two ISS: Forward collision warning (FCW), and autonomous emergency braking (AEB). Secondly, the paper demonstrates how counterfactual simulations can be used to perform sensitivity analyses on parameter settings, both for driver behavior and ISS algorithms. Finally, the paper evaluates the effect of the choice of glance distribution in the driver behavior model on the safety benefit estimation. The paper uses pre-crash kinematics and driver behavior from 34 rear-end crashes from the SHRP2 naturalistic driving study for the demonstrations. The results for FCW show a large difference in the percent of avoided crashes between conceptually different models of driver behavior, while differences were small for conceptually similar models. As expected, the choice of model of driver behavior did not affect AEB benefit much. Based on our results, researchers and others who aim to evaluate ISS with the driver in the loop through counterfactual simulations should be sure to make deliberate and well-grounded choices of driver models: the choice of model matters.