Safety assessment for autonomous vehicles: A reference driver model for highway merging scenarios.

Driver models are crucial for the safety assessment of autonomous vehicles (AVs) because of their role as reference models. Specifically, an AV is expected to achieve at least the same level of safety performance as a careful and competent driver model. To make this comparison possible, quantitative modeling of careful and competent driver models is essential. Thus, the UNECE Regulation No. 157 proposes two driver models as benchmarks for AVs, enabling safety assessment of AV longitudinal behaviors. However, these two driver models are unable to be applied in non-car-following scenarios, limiting their applications in scenarios such as highway merging. To this end, we propose a careful and competent driver model for highway merging (CCDM2) scenarios using interpretable reinforcement learning-based decision-making and safety constraint control. We compare our model's safe driving capabilities with human drivers in challenging merging scenarios and demonstrate the "careful" and "competent" characteristics of our model while ensuring its interpretability. The results indicate the model's capability to handle merging scenarios with even better safety performance than human drivers. This model is of great value for AV safety assessment in merging scenarios and contributes to future reference driver models to be included in AV safety regulations.

Resolving uncertainty on the fly: modeling adaptive driving behavior as active inference.

Understanding adaptive human driving behavior, in particular how drivers manage uncertainty, is of key importance for developing simulated human driver models that can be used in the evaluation and development of autonomous vehicles. However, existing traffic psychology models of adaptive driving behavior either lack computational rigor or only address specific scenarios and/or behavioral phenomena. While models developed in the fields of machine learning and robotics can effectively learn adaptive driving behavior from data, due to their black box nature, they offer little or no explanation of the mechanisms underlying the adaptive behavior. Thus, generalizable, interpretable, computational models of adaptive human driving behavior are still rare. This paper proposes such a model based on active inference, a behavioral modeling framework originating in computational neuroscience. The model offers a principled solution to how humans trade progress against caution through policy selection based on the single mandate to minimize expected free energy. This casts goal-seeking and information-seeking (uncertainty-resolving) behavior under a single objective function, allowing the model to seamlessly resolve uncertainty as a means to obtain its goals. We apply the model in two apparently disparate driving scenarios that require managing uncertainty, (1) driving past an occluding object and (2) visual time-sharing between driving and a secondary task, and show how human-like adaptive driving behavior emerges from the single principle of expected free energy minimization.

Scenario Generation for Autonomous Vehicles with Deep-Learning-Based Heterogeneous Driver Models: Implementation and Verification.

Virtual testing requires hazardous scenarios to effectively test autonomous vehicles (AVs). Existing studies have obtained rarer events by sampling methods in a fixed scenario space. In reality, heterogeneous drivers behave differently when facing the same situation. To generate more realistic and efficient scenarios, we propose a two-stage heterogeneous driver model to change the number of dangerous scenarios in the scenario space. We trained the driver model using the HighD dataset, and generated scenarios through simulation. Simulations were conducted in 20 experimental groups with heterogeneous driver models and 5 control groups with the original driver model. The results show that, by adjusting the number and position of aggressive drivers, the percentage of dangerous scenarios was significantly higher compared to that of models not accounting for driver heterogeneity. To further verify the effectiveness of our method, we evaluated two driving strategies: car-following and cut-in scenarios. The results verify the effectiveness of our approach. Cumulatively, the results indicate that our approach could accelerate the testing of AVs.

Efficiency performance and safety evaluation of the responsibility-sensitive safety in freeway car-following scenarios using automated longitudinal controls.

The Responsibility-Sensitive Safety (RSS) model was proposed by Mobileye as a mathematical model that defines the real-time safety distance that the automated vehicle (AV) needs to maintain from surrounding vehicles. However, RSS strategy tends to be overly conservative. This research made modifications to the RSS safe distance to reduce its conservativeness without affecting safety, and evaluated the modified (RSS_X) and original RSS (RSS_O) in freeway car-following scenarios extracted from the Shanghai Naturalistic Driving Study (SH-NDS). The modifications were replacing the maximum acceleration with the positive previous time-step acceleration and adding standstill gap. In this study, 6,146 car-following scenarios were extracted and divided into two groups, normal scenarios (5,923) and safety-critical events (SCEs, near crashes) (223), to evaluate the efficiency and safety performance of RSS. The RSS_O and RSS_X were then embedded into the intelligent driver model (IDM) and model predictive control (MPC), but because the RSS modifications caused a few acceleration stability problems, the IDM and MPC were also modified to accommodate the RSS_X. The efficiency performance results showed that the modified models (IDM + RSS_X, MPC + RSS_X) performed better than the originals (IDM + RSS_O, MPC + RSS_O) in that they had higher average speeds, more comfortable acceleration pattern, and smaller longitudinal clearance between vehicles, which leads to less conservativeness. To evaluate safety, human drivers were compared with the original and modified models. RSS reduced the severity of at least 80 % of the human driver SCEs. MPC + RSS_X increased the mean minimum time to collision (TTC) for the majority of SCEs from 1.65 s to 4.08 s, and IDM + RSS_X increased the mean minimum TTC for the majority of SCEs from 1.53 s to 3.44 s.

Driver models for the definition of safety requirements of automated vehicles in international regulations. Application to motorway driving conditions.

UN Regulation 157, the first global regulation regarding the type-approval of Automated Driving Systems (ADS), has been adopted in 2021. In it, safety performance requirements are being defined for vehicles of automation Level 3, according to the SAE J3016, with a limited Operational Design Domain (ODD). In particular, for three types of events that are related to motorway driving, two models are provided to distinguish between preventable traffic scenarios, for which the ADS is expected to avoid an accident, and unpreventable traffic scenarios, for which accidents cannot be avoided and the ADS can only mitigate their severity. The models recreate the short-term behavior of a driver who reacts to an emergency. Two possible actions are predicted: either no reaction or full braking when danger is identified. In the present paper the two models are analyzed and compared with two additional models: an industry proposed model, the Responsibility Sensitive Safety framework (RSS), and the Fuzzy Safety Model (FSM) proposed by the authors. As in the case of the two regulation models, also the RSS, although more sophisticated, assumes that the possible reaction by the driver is binary. This approach neglects the ability of a human driver to drive defensively and anticipate possible risks. Defensive drivers, indeed, may use comfortable decelerations in anticipation, to avoid finding themselves in an emergency situation. The FSM uses fuzzy logic to mimic this behavior. Results show that anticipation plays a very important role to reduce the number of unpreventable traffic scenarios. In addition, by validating the classification capabilities of the four models with real traffic data, the FSM proved to be the most suitable of the investigated models. On the basis of these results, the FSM has been included in the proposal for amending UN Regulation 157, thus allowing to set higher safety standards for the first automated vehicles that will be introduced into the market.

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.

Detection and response to critical lead vehicle deceleration events with peripheral vision: Glance response times are independent of visual eccentricity.

Studies show high correlations between drivers' off-road glance duration or pattern and the frequency of crashes. Understanding drivers' use of peripheral vision to detect and react to threats is essential to modelling driver behavior and, eventually, preventing crashes caused by visual distraction. A between-group experiment with 83 participants was conducted in a high-fidelity driving simulator. Each driver in the experiment was exposed to an unexpected, critical, lead vehicle deceleration, when performing a self-paced, visual-manual, tracking task at different horizontal visual eccentricity angles (12°, 40° and 60°). The effect of visual eccentricity on threat detection, glance and brake response times was analyzed. Contrary to expectations, the driver glance response time was found to be independent of the eccentricity angle of the secondary task. However, the brake response time increased with increasing task eccentricity, when measured from the driver's gaze redirection to the forward roadway. High secondary task eccentricity was also associated with a low threat detection rate and drivers were predisposed to perform frequent on-road check glances while executing the task. These observations indicate that drivers use peripheral vision to collect evidence for braking during off-road glances. The insights will be used in extensions of existing driver models for virtual testing of critical longitudinal situations, to improve the representativeness of the simulation results.

Driver's Preview Modeling Based on Visual Characteristics through Actual Vehicle Tests.

This paper proposes a method for obtaining driver's fixation points and establishing a preview model based on actual vehicle tests. Firstly, eight drivers were recruited to carry out the actual vehicle test on the actual straight and curved roads. The curvature radii of test curved roads were selected to be 200, 800, and 1500 m. Subjects were required to drive at a speed of 50, 70 and 90 km/h, respectively. During the driving process, eye movement data of drivers were collected using a head-mounted eye tracker, and road front scene images and vehicle statuses were collected simultaneously. An image-world coordinate mapping model of the visual information of drivers was constructed by performing an image distortion correction and matching the images from the driving recorder. Then, fixation point data for drivers were accordingly obtained using the Identification-Deviation Threshold (I-DT) algorithm. In addition, the Jarque-Bera test was used to verify the normal distribution characteristics of these data and to fit the distribution parameters of the normal function. Furthermore, the preview points were extracted accordingly and projected into the world coordinate. At last, the preview data obtained under these conditions are fit to build general preview time probability density maps for different driving speeds and road curvatures. This study extracts the preview characteristics of drivers through actual vehicle tests, which provides a visual behavior reference for the humanized vehicle control of an intelligent vehicle.

Measurement data obtained by an instrumented steering wheel for driver model development.

The data article presents the data acquired by an Instrumented Steering Wheel, able to measure the three force components and the three moment components applied by each of the two driver hands on an Instrumented Steering Wheel (ISW). Additionally, the ISW senses the grip forces at each hand. In order to simulate emergency manoeuvres in a safe environment, a test track with a kick plate is used. Nine different drivers pass over the kick plate six times each. The drivers need to make an action on the steering wheel to counteract the lateral disturbance and recover the straight desired path. The vehicle has been instrumented with an ISW and an inertial measuring unit. Data acquired by the two sensors have been synchronized and analysed. The force components due to mass properties of the ISW have been compensated in a proper way, to highlight the loads exerted only by the driver hands. In the present data article, the data acquired during the described kick-plate test are reported for one driver during a single test. Discussion and conclusion have been presented in [1]. Data are provided in Matlab environment. Videos are provided to show how the manoeuvre occurs. The vehicle that was used for tests was modified with respect to the corresponding production vehicle. Data refer to the specific vehicle used in the tests that does not match with any vehicle produced by Toyota.

Detection of Driving Capability Degradation for Human-Machine Cooperative Driving.

Due to the limitation of current technologies and product costs, humans are still in the driving loop, especially for public traffic. One key problem of cooperative driving is determining the time when assistance is required by a driver. To overcome the disadvantage of the driver state-based detection algorithm, a new index called the correction ability of the driver is proposed, which is further combined with the driving risk to evaluate the driving capability. Based on this measurement, a degraded domain (DD) is further set up to detect the degradation of the driving capability. The log normal distribution is used to model the boundary of DD according to the bench test data, and an online algorithm is designed to update its parameter interactively to identify individual driving styles. The bench validation results show that the identification algorithm of the DD boundary converges finely and can reflect the individual driving characteristics. The proposed degradation detection algorithm can be used to determine the switching time from manual to automatic driving, and this DD-based cooperative driving system can drive the vehicle in a safe condition.

Optimal jam-absorption driving strategy for mitigating rear-end collision risks with oscillations on freeway straight segments.

Traffic oscillations (or stop-and-go waves) in freeway traffic jam cause large variation of vehicle speed and remarkably reduce travel safety. Previous jam-absorption driving strategies focused on the operational side and did not consider the safety effects caused by the controlled vehicle on freeways. In this paper, we proposed an optimal jam-absorption driving strategy to mitigate traffic oscillations and rear-end collision risks on freeway straight segments. Firstly, the proposed strategy determined the starting and ending point of an oscillation at the temporal and spatial dimensions based on the Wavelet Transform (WT) and the steady equilibrium condition of car-following driving. Then different controlled vehicles were evaluated by the given absorbing speeds. Various measurements were considered to evaluate the safety performance of the strategies. The optimal solution was obtained which guided the controlled vehicle to move slowly at the optimal jam-absorbing speed, and created a gap to eliminate the downstream oscillation timely but avoid causing secondary wave in the upstream traffic. The Intelligent Driver Model (IDM) was modified to build the simulation platform in a connected environment. The results showed that our proposed strategy effectively reduced the severity of traffic oscillations, or even fully eliminate the oscillations. The optimal strategy reduced the surrogate safety measures by 93.53 %-94.78 %, and decreased the total travel time by 1.27 %. We also compared our strategy with previous strategies and the results suggested that ours had better performances.

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.

Evaluating the safety impact of connected and autonomous vehicles on motorways.

Recent technological advancements bring the Connected and Autonomous Vehicles (CAVs) era closer to reality. CAVs have the potential to vastly improve road safety by taking the human driver out of the driving task. However, the evaluation of their safety impacts has been a major challenge due to the lack of real-world CAV exposure data. Studies that attempt to simulate CAVs by using either a single or integrating multiple simulation platforms have limitations, and in most cases, consider a small element of a network (e.g. a junction) and do not perform safety evaluations due to inherent complexity. This paper addresses this problem by developing a decision-making CAV control algorithm in the simulation software VISSIM, using its External Driver Model Application Programming Interface. More specifically, the developed CAV control algorithm allows a CAV, for the first time, to have longitudinal control, search adjacent vehicles, identify nearby CAVs and make lateral decisions based on a ruleset associated with motorway traffic operations. A motorway corridor within M1 in England is designed in VISSIM and employed to implement the CAV control algorithm. Five simulation models are created, one for each weekday. The baseline models (i.e. CAV market penetration: 0%) are calibrated and validated using real-world minute-level inductive loop detector data and also data collected from a radar-equipped vehicle. The safety evaluation of the proposed algorithm is conducted using the Surrogate Safety Assessment Model (SSAM). The results show that CAVs bring about compelling benefit to road safety as traffic conflicts significantly reduce even at relatively low market penetration rates. Specifically, estimated traffic conflicts were reduced by 12-47%, 50-80%, 82-92% and 90-94% for 25%, 50%, 75% and 100% CAV penetration rates respectively. Finally, the results indicate that the presence of CAVs ensured efficient traffic flow.

Neuroergonomics of car driving: A critical meta-analysis of neuroimaging data on the human brain behind the wheel.

Car driving, an everyday life activity, has been under the scope of investigation for long. Neurosciences and psychology have contributed to better understand the human processes engaged while driving, to such an extent that a meta-analysis of all available fMRI data is now possible to extract the most relevant information. Using the Activation Likelihood Estimation method, we therefore conducted such a meta-analysis on 9 studies, representing 27 neuroimaging contrasts and 131 participants. We identified a network composed of brain areas underlying the cognitive abilities required for driving: sensorimotor coordination, sensory and attentional processing, high-level cognitive control and allocation of attentional resources. We complemented this meta-analysis with a neuroergonomics approach combining driving control knowledge, distinguishing the strategical, tactical and operational levels, with neuroscientific knowledge and models on cognitive control operated by the prefrontal cortex. The results exposed the distinct neural circuits engaged behind the wheel depending on the task performed. Based on the combination of neuroscientific and ergonomic knowledge, a hybrid car driving framework is also proposed.

Impacts of transportation sector emissions on future U.S. air quality in a changing climate. Part II: Air quality projections and the interplay between emissions and climate change.

In Part II of this work we present the results of the downscaled offline Weather Research and Forecasting/Community Multiscale Air Quality (WRF/CMAQ) model, included in the "Technology Driver Model" (TDM) approach to future U.S. air quality projections (2046-2050) compared to a current-year period (2001-2005), and the interplay between future emission and climate changes. By 2046-2050, there are widespread decreases in future concentrations of carbon monoxide (CO), nitrogen oxides (NOx = NO + NO2), volatile organic compounds (VOCs), ammonia (NH3), sulfur dioxide (SO2), and particulate matter with an aerodynamic diameter ≤ 2.5 µm (PM2.5) due mainly to decreasing on-road vehicle (ORV) emissions near urban centers as well as decreases in other transportation modes that include non-road engines (NRE). However, there are widespread increases in daily maximum 8-hr ozone (O3) across the U.S., which are due to enhanced greenhouse gases (GHG) including methane (CH4) and carbon dioxide (CO2) under the Intergovernmental Panel on Climate Change (IPCC) A1B scenario, and isolated areas of larger reduction in transportation emissions of NOx compared to that of VOCs over regions with VOC-limited O3 chemistry. Other notable future changes are reduced haze and improved visibility, increased primary organic to elemental carbon ratio, decreases in PM2.5 and its species, decreases and increases in dry deposition of SO2 and O3, respectively, and decreases in total nitrogen (TN) deposition. There is a tendency for transportation emission and CH4 changes to dominate the increases in O3, while climate change may either enhance or mitigate these increases in the west or east U.S., respectively. Climate change also decreases PM2.5 in the future. Other variable changes exhibit stronger susceptibility to either emission (e.g., CO, NOx, and TN deposition) or climate changes (e.g., VOC, NH3, SO2, and total sulfate deposition), which also have a strong dependence on season and specific U.S. regions.

Impacts of transportation sector emissions on future U.S. air quality in a changing climate. Part I: Projected emissions, simulation design, and model evaluation.

Emissions from the transportation sector are rapidly changing worldwide; however, the interplay of such emission changes in the face of climate change are not as well understood. This two-part study examines the impact of projected emissions from the U.S. transportation sector (Part I) on ambient air quality in the face of climate change (Part II). In Part I of this study, we describe the methodology and results of a novel Technology Driver Model (see graphical abstract) that includes 1) transportation emission projections (including on-road vehicles, non-road engines, aircraft, rail, and ship) derived from a dynamic technology model that accounts for various technology and policy options under an IPCC emission scenario, and 2) the configuration/evaluation of a dynamically downscaled Weather Research and Forecasting/Community Multiscale Air Quality modeling system. By 2046-2050, the annual domain-average transportation emissions of carbon monoxide (CO), nitrogen oxides (NOx), volatile organic compounds (VOCs), ammonia (NH3), and sulfur dioxide (SO2) are projected to decrease over the continental U.S. The decreases in gaseous emissions are mainly due to reduced emissions from on-road vehicles and non-road engines, which exhibit spatial and seasonal variations across the U.S. Although particulate matter (PM) emissions widely decrease, some areas in the U.S. experience relatively large increases due to increases in ship emissions. The on-road vehicle emissions dominate the emission changes for CO, NOx, VOC, and NH3, while emissions from both the on-road and non-road modes have strong contributions to PM and SO2 emission changes. The evaluation of the baseline 2005 WRF simulation indicates that annual biases are close to or within the acceptable criteria for meteorological performance in the literature, and there is an overall good agreement in the 2005 CMAQ simulations of chemical variables against both surface and satellite observations.

Longitudinal safety evaluation of connected vehicles' platooning on expressways.

Connected vehicles (CV) technology has recently drawn an increasing attention from governments, vehicle manufacturers, and researchers. One of the biggest issues facing CVs popularization associates it with the market penetration rate (MPR). The full market penetration of CVs might not be accomplished recently. Therefore, traffic flow will likely be composed of a mixture of conventional vehicles and CVs. In this context, the study of CV MPR is worthwhile in the CV transition period. The overarching goal of this study was to evaluate longitudinal safety of CV platoons by comparing the implementation of managed-lane CV platoons and all lanes CV platoons (with same MPR) over non-CV scenario. This study applied the CV concept on a congested expressway (SR408) in Florida to improve traffic safety. The Intelligent Driver Model (IDM) along with the platooning concept were used to regulate the driving behavior of CV platoons with an assumption that the CVs would follow this behavior in real-world. A high-level control algorithm of CVs in a managed-lane was proposed in order to form platoons with three joining strategies: rear join, front join, and cut-in joint. Five surrogate safety measures, standard deviation of speed, time exposed time-to-collision (TET), time integrated time-to-collision (TIT), time exposed rear-end crash risk index (TERCRI), and sideswipe crash risk (SSCR) were utilized as indicators for safety evaluation. The results showed that both CV approaches (i.e., managed-lane CV platoons, and all lanes CV platoons) significantly improved the longitudinal safety in the studied expressway compared to the non-CV scenario. In terms of surrogate safety measures, the managed-lane CV platoons significantly outperformed all lanes CV platoons with the same MPR. Different time-to-collision (TTC) thresholds were also tested and showed similar results on traffic safety. Results of this study provide useful insight for the management of CV MPR as managed-lane CV platoons.

Some Aspects of Vehicle Active Safety.

Three selected aspects of vehicle active safety are presented in this article: (a) modeling the driver and the driver-vehicle environment system, (b) the dynamic aspects of vehicle rollover, and (c) an analysis of the process of passing. Sample solutions and results show the need for further research in the field of vehicle safety in order to lower the probability of drivers, passengers, and other road users being involved in road accidents.