Thursday 09:00 – 10:30
“How do Vehicle Size, Speed, TTC, Age and Sex Affect Cyclists’ Gap Acceptance When Interacting With (Automated) Vehicles?”
Sabine Springer-Teumer, Daniel Trommler, Josef F. Krems
Abstract— Given its high complexity and the associated space-sharing conflicts, mixed urban traffic depicts a key challenge for automated driving. For safe and efficient traffic, it is therefore necessary that interactions between automated vehicles and vulnerable road users (VRUs; e.g., cyclists, pedestrians) are well-coordinated and fine-tuned to one another. Hereby, vehicle deceleration is a useful informal communication signal to give priority to VRUs. In this context, gap acceptance can be used to obtain information regarding VRUs’ expected deceleration rate and time of deceleration onset for safe crossings. The aim of this study was to explore the gap acceptance of cyclists and to examine the influence of vehicle size, vehicle speed, and different levels of time-to-arrival (TTA) of oncoming vehicles as well as age and gender on cyclists’ gap acceptance. For this purpose, a 2x2x2x2x4 mixed-design video-based online study was conducted comprising the experimental within-factors size (car and truck) and speed (30 km/h and 45 km/h) of the oncoming vehicle as well as the TTA between cyclists at the (theoretical) point of collision (0s, 1s, 3s, 5s). The quasi-experimental between-factors were age (younger: 18-35 years, older: >35 years) and sex (female, male). A total of N = 35 persons (n♀ = 12; Mage = 38 years, SDage = 15) participated in the study that were presented a set of short videos from a cyclist’s perspective riding towards an intersection while a vehicle was approaching from the left. Results revealed that participants (1) decided to cross more likely when the approaching vehicle was a truck compared to a car, (2) chose to cross later with higher speeds of the approaching vehicle compared to lower speeds, and (3) decided to cross earlier with lower (theoretical) TTAs compared to higher TTAs. In tendency, older participants decided later to cross or not to cross compared to younger participants. The results help to align the behaviour of AVs with that of cyclists to establish safe and effective road traffic.
Keywords: automated vehicles, cyclist-vehicle interaction, gap acceptance, crossing decision, age, sex, vulnerable road users.
“Drivers’ Interpretation and Reaction to Vehicle Motion in Narrow Passages: Recommendations for Implicit Communication Strategies of Automated Vehicles”
Linda Miller and Martin Baumann
Abstract— Automated vehicles promise to increase road safety, but they must integrate into an existing sociotechnical system. In this mixed traffic, automated vehicles must coordinate their driving maneuvers with human drivers surrounding them and resolve conflicts that arise. A narrowed road is an example of a situation requiring mutual coordination between humans and technology. Vehicles approach a road passage from opposite sides that is too narrow for both to drive simultaneously, so one of the two agents has to adjust its current driving behavior (e.g., slow down, perform evasive maneuver). In this context, vehicle motion is not regarded as a mere byproduct of driving but as an utility that can be used by automated vehicles to deliberately communicate their driving intention (i.e., implicit communication). As part of a three-year project funded by the German Federal Ministry of Education and Research on connected and automated driving in urban environments – CADJapanGermany: Human Factors – among others, narrow passage situations were investigated in iterative studies using different methodological approaches. In particular, the focus was on how human drivers interpret, react to, and evaluate the vehicle motion of oncoming traffic in narrow passages. To this end, online surveys, video-based online experiments, interviews, and a multi-agent driving simulator study were conducted. Overall, the results of the different studies show that yielding behaviors (e.g., decelerating, stopping) are rated as more cooperative than insisting behaviors (e.g., accelerating, maintaining speed). Findings also implied that the time-to-arrival of the oncoming vehicle has to be more than twice that of the driver’s own vehicle for drivers to choose to drive first rather than to drive second in a narrow passage. A comparison of implicit communication strategies, including speed changes, lateral changes, and a combination of both, showed that drivers mainly used the lateral trajectory to interpret the intention of the oncoming vehicle. Driving patterns including lateral movements were also faster interpreted, rated as more distinct and cooperative than patterns without lateral movements, and compensated for the processing disadvantage of early versus late communication timing. In this regard, movements to the center of the road are considered an efficient way for automated vehicles to communicate the intention to drive first and movements to the edge of the road to communicate the intention to drive second in narrow passages. In the driving simulator, drivers successfully adapted their driving behavior to the communicated intentions of automated vehicles in mixed traffic, just as to manually controlled vehicles. However, drivers generally drove faster when interacting with automated vehicles compared to manually-controlled vehicles. This might be due to the different expectations drivers had towards these two categories of vehicles: while other drivers were expected to drive in an error- and safety-prone manner, automated vehicles were expected to drive in an optimized and safety-promoting manner. Indeed, automated vehicles were expected to represent a nearly perfect driver by optimizing various safety-, interaction-, and driving style-related aspects of driving. However, on a subjective level, yielding behavior was generally rated as more trustworthy and cooperative than insisting behavior regardless of the vehicle category.
In summary, the project results demonstrate that vehicle motion is a good means of communicating a vehicle’s intention to surrounding human road users. The actual vehicle motion was the strongest determinant of drivers’ reactions across all studies, ahead of factors such as vehicle category (automated vs. manually-controlled oncoming vehicle) or preexisting expectations (insist or yield). For an expectation-based design of implicit communication in mixed traffic, this means that automated vehicles are recommended to communicate their driving intention early and as easily detectable and salient as possible, for instance through lateral movements. In this way, automated vehicles might successfully realize their potential to increase traffic safety and efficiency in a new traffic environment of social coexistence between humans and technology.
“Combination of Driver States Monitoring and Take-Over Performance Prediction: Implementation Concept Using Real-Time Monitoring and Cloud Computing”
Toshihisa Sato
Abstract— This paper describes the concept of a system in which a driver monitoring system measures the driver conditions in real time and predicts the appropriateness of manual driving by the driver after the transition based on various threshold values stored in a database in the cloud. The evaluation factors using the driver monitoring system include an assessment of the driver states during the automated driving systems, a prediction of the take-over performances after the transition to manual driving, and decision-making of the threshold values based on behavior data in the cloud databases. We collected the driver’s cognitively loaded and visually loaded conditions data in the automated driving using driving simulator experiments and developed the prediction method of the driving behavior after the driver received a “Request to Intervene” and began to drive manually. The task-capability interface model is applied to the judgment methodology of whether the current driver’s states exceed the thresholds or not. Driving behavior data at that location stored in the cloud could contribute to determining if the system must intervene in the driver state before the transition because the take-over performance after the transition will be lower than the threshold.
“Transition Design for Automated Driving – About the Improvement of Human Reliability and Reduction of Task Switching Costs”
Klaus Bengler, Burak Karakaya, Elisabeth Shi
Abstract— Automated driving (AD) seems to be a foreseeable and increasing technology in future mobility systems. While different levels of automation still raise the question of human involvement and responsibility in different driving scenarios the introduction of AD seems to be a matter of time. Especially realizations of partial and conditional automation require the human driver to take control over the driving task either voluntarily or following a takeover request issued by the automation (SAE International, 2021). Many studies and guidelines focus on reaction time and take over time. These considerations have to be enhanced by the quality of the human driving maneuver and stabilization behavior. Especially in critical scenarios it might be questioned which driving strategy is reducing the risk, if automated stand-still, human intervention or assisted human driving can be chosen. A further question is how to decide on the premise that human intervention should always be allowed to overrule an automation decision. Driving simulation data show that dedicated assistance is necessary in collision avoidance after automation to enhance human reliability after transition from automated driving to manual driving. A clear benefit of automation is the possibility to reduce the workload for the human driver. This situation may result in underload and reduced performance in transition situations. Experiments on the test track show that driver readiness can be moderated and improved by non-driving related tasks with certain characteristics. There is clear evidence that the theoretical framework of task switching theories (e.g. Rubinstein, Meyer & Evans, 2001) is much more adequate to explain these effects than Wickens’ (2002) multiple resource theory (Shi & Bengler, 2022).
“Influences on drivers’ attention levels while applying level 2 automated driving systems”
Bo Yang and Kimihiko Nakano
Abstract— The usage of level 2 automated driving is becoming a common traffic scene in our daily life. To ensure driving safety while using the level 2 automated driving systems in various traffic conditions, it is an essential issue to clarify the influences of level 2 automated driving on driver behaviours, especially their attention levels. A driving simulator experiment was performed in this study, and the differences in driver behaviours under level 2 automated driving and manual driving were analyzed with eye-gaze behaviours and questionnaires. It was observed that the gaze time to the front area and the speedometer significantly decreased within level 2 automated driving, which indicated that drivers’ attention levels to the front might be reduced while using level 2 automated driving, compared to that of manual driving.
“Development and Validation of Simulation-Type Learning Materials for Smooth Takeovers in Automated Driving Level 3”
Yoshiko Goda, Maki Arame, Junko Handa, Masashi Toda, Makoto Itoh, Satoshi Kitazaki
Abstract— To drive a partially automated Level 3 vehicle safely, it must be able to take over smoothly when a request of intervention (RoI) is received. Previous research maintains that takeovers are smooth with the availability of a basic knowledge of automated vehicles. However, it has been reported that even if the handover is smooth immediately after the first experiment, then, the effects of learning disappear, and a takeover is not smooth when the experiment is conducted after a time period such as one month. For a smooth takeover, the regulation and management of drivers’ attention and tasks is also important. Few studies on safety education at automated Level 3 consider drivers’ management of sub-tasks and the regulation of their attention during automated driving. In this research, to help drivers recognize the importance of managing and regulating their resources and attention, even during automated driving for smooth takeovers, two kinds of simulation-type online learning materials were designed and developed for takeover practices in different scenarios, that is, with and without a simple shooting game as a surrogate reference task (SuRT). Immediate and delayed feedback was provided about motorists’ driving with a sequence before and after the RoI. True experimental research was conducted with three conditions: two simulation learning materials as experimental groups and one non-simulation video as a control group. Sixty participants were randomly assigned to one of three conditions. To control the knowledge learning, the same video learning material used in the control group was shown to the other two groups. The experiment groups used the simulation materials after viewing the video. The test on automated driving with twenty questions and the driving simulation test with four scenes, including the RoI, were provided before and after learning. The post-questionnaire was conducted at the end of the experiment concerning attitudes toward safe driving and opinions on the experiment and treatment. The results show that both simulation-type learning materials may have positive effects on the retention of knowledge about automated driving and the reservation of the preferred performance, that is, a smooth takeover. Based on the questionnaire, the experiment group with SuRT may realize the importance of regulating their resources during automated driving for a smooth takeover.
“