Automated Driving Systems Research Articles

New insights into factors affecting the severity of autonomous vehicle crashes from two sources of AV incident records

Superior safety is the main banner value of promoting autonomous vehicle (AV) technology, but it is difficult to responsibly claim it. The potential for AVs to reduce crash and injury risks would be overshadowed by technological limitations, regardless of their ability to mitigate or eliminate human error. This study aims to identify the key factors affecting crash severity by analyzing real-world AV crash data from the U.S. between 2015 and 2022. We integrated two open data sources from the California DMV and NHTSA. Mixed multinomial logit models incorporating the interaction effects were estimated using the crash severity level, addressing the observed and unobserved heterogeneities. Our results show that Advanced Driver Assistance System (ADAS) engagement is associated with a higher likelihood of slight injury crashes, whereas Automated Driving System (ADS) engagements show the opposite effect. In addition, we found that various environmental and road factors, including lighting conditions, weather, road type, and road surface conditions, significantly affect the severity of AV crashes. For instance, daylight conditions contribute to a lower likelihood of slight-injury crashes. On the other hand, driving under unfavorable weather conditions (cloudy, foggy, rainy, or snowy), on the highway, and on non-dry road surfaces are associated with an increase in the likelihood of severe injury crashes. Furthermore, several significant interaction effects were revealed. First, the mitigating effect of ADS engagement on the likelihood of slight injury crashes is reduced by the rear-end collision type. Second, the likelihood of slight injury crashes increases when AV interacts with heavy trucks on highways. Third, the likelihood of severe injuries increases when AVs collide with Vulnerable Road Users (VRUs) on urban streets. Overall, this research is expected to provide policymakers and AV manufacturers with valuable insights into AV safety, stressing that addressing the identified factors will lead to improved AV design and control algorithms.

Effects of Automation Exposure and Non-Driving Related Tasks on Manual Driving Performance after Takeover of Manual Control

Prolonged exposure to an automated driving system (ADS) can lead drivers to acclimate to the ADS in maladaptive ways. For example, exposure to a conditional ADS that maintained a time headway of 0.3 s resulted in drivers adopting smaller time headways postautomation, which is attributed to behavioral adaptation. The current study investigated whether drivers adapt to an ADS’ performance over time and if this adaptation differs based on the duration of automation exposure (15 or 25 min) and the task performed during the automation (NDRT or driving-related task). A greater exposure to the automation did not lead drivers to adopt smaller time headways. Surprisingly, drivers who provided a higher evaluation of the automation adopted smaller time headways post-automation than drivers who provided a lower evaluation. Future research should explore the relationship between trust in ADS and driving behavior to guide the development of safer ADSs.

To Err is Automation: Can Trust be Repaired by the Automated Driving System After its Failure?

To Err is Automation: Can Trust be Repaired by the Automated Driving System After its Failure?

Thermal Cameras and Their Safety Implications for Pedestrian Protection: A Mixed Empirical and Simulation-Based Characterization

Multiple sensing technologies exist for advanced driver support and driving automation systems. Car manufacturers combine different solutions to increase the robustness and operational capabilities of driving automation technology perception systems. One sensing method that did not achieve high market penetration with respect to other solutions is the thermal camera. Thermal cameras can operate in low-light and high-contrast conditions as they do not need external sources of illumination. Instead, they are sensitive to the infrared radiation emitted by any surface above 0 K. Thus, thermal cameras can be a suitable candidate to compensate for the weaknesses of visible cameras. To establish the potential benefit of thermal cameras for pedestrian protection, this paper reports on an experiment using state-of-the-art thermal imaging sensors. The authors’ scientific effort includes a characterization of the infrared sensor capability to identify pedestrians when coupled with a detection algorithm and provides a simulation-based investigation of the potential safety benefits of thermal camera adoption. The analysis additionally includes a heuristic characterization of the detection capabilities of a passenger car equipped with visible cameras to provide a baseline comparison. Ultimately, the thermal sensors return noteworthy detection capability that could lead to a potentially significant improvement in road safety on rural roads provided that the sensor has sufficient resolution.

Failing Consistently and Expectedly: Effects of Error Consistency and Drivers’ Mental Models on Trust in Automated Vehicles

This ongoing study investigates how driver expectations and automation error consistency influence trust in Driving Automation Systems (DASs). Drawing on the Expectation Confirmation Theory (ECT), we hypothesize that trust increases when system performance aligns with or exceeds driver expectations on the DASs. Inconsistencies in error patterns may diminish trust. We conducted a 2 × 3 × 3 mixed-design experiment on STISIM Drive with 150 participants, manipulating expectations (high vs. low) and error patterns (no error, consistent, or inconsistent errors). Trust and response times to takeover requests (TORs) were measured three times, one at each drive. We predict higher trust and longer TOR times in low-expectation conditions. As for error patterns, we predict the highest trust to arise from scenarios with no errors and longer TOR times when errors are inconsistent. Our findings are expected to provide insights into enhancing DAS design by aligning system performance with user expectations and promoting error consistency to improve trust and user satisfaction.

Investigating Drivers’ Awareness of Automated Vehicle Modes Using the State Diagram Method

Driving automation introduces multiple driving modes to maximize the system’s benefits, but drivers must monitor and stay aware of these modes, which can sometimes lead to mode confusion. We modified Degani and Heymann’s state diagram method to assess the mode structure of a hypothetical driving automation system and the likelihood of discrepancies between drivers’ mode awareness and the system’s actual mode. We used the modified method and driving simulation data from participants who weren’t fully informed about all automation modes. The modified method visualized all possible combinations of automation modes and drivers’ mode awareness, highlighting where they diverge and estimating the frequency of the divergence. The diagram identified areas where human-machine interface (HMI) design can help drivers maintain accurate mode awareness. The modified state diagram method provides actionable insights for designing HMIs to reduce mode confusion and can be used to develop computational models of automation mode structures.

Dynamic and quantitative trust modeling and real-time estimation in human-machine co-driving process

The development of automated vehicles (AVs) will remain in the stage of human–machine co-driving for a long time. Trust is considered as an effective foundation of the interaction between the driver and the automated driving system (ADS). Driver’s trust miscalibration, represented by under-trust and over-trust, is considered to be the potential cause of disuse and misuse of ADS, or even serious accidents. The estimation and calibration of trust are crucial to improve the safety of the driving process. This paper mainly consists of the following two aspects. Firstly, a dynamic and quantitative trust estimation model is established. A framework for trust estimation is constructed. Driver’s perceived risk and behavior features were monitored and a Kalman filter was used to dynamically and quantitatively estimate the driver’s trust. We conducted a driver-in-the-loop experiment and generated model parameters through a data-driven approach. The results demonstrated that the model exhibited precision in trust estimation, with the highest accuracy reaching 74.1%. Secondly, a reminder strategy to calibrate the over-trust of the driver is proposed based on the model from the first part. A scenario with four risky events was designed and the ADS would provide voice reminders to the driver when over-trust was detected. The results demonstrated that the reminder strategy proved to be beneficial for safety enhancement and moderate trust maintenance during the driving process. When the driver is over-trusting, the accident rates of the reminder group and the non-reminder group were 60.6% and 13.0%, respectively. Our contribution in this paper can be concluded by four points: (1) A real-time trust estimation model is proposed, which is dynamic and quantitative, considering the evolution pattern of driver’s trust and the perceived risk; (2) Mathematical modeling and machine learning methods are combined; (3) A trust-based reminder strategy that aims to enhance the safety of human–machine co-driving is designed; (4) Driver-in-loop experiment validates the effectiveness in enhancing the safety, maintaining driver’s trust and reducing trust biases in human–machine co-driving.

Comparison of Waymo rider-only crash data to human benchmarks at 7.1 million miles

Objectives This article examines the safety performance of the Waymo Driver, an SAE level 4 automated driving system (ADS) used in a rider-only (RO) ride-hailing application without a human driver, either in the vehicle or remotely. Methods ADS crash data were derived from NHTSA’s Standing General Order (SGO) reporting over 7.14 million RO miles through the end of October 2023 in Phoenix, Arizona, San Francisco, California, and Los Angeles, California, and compared to human benchmarks from the literature. Results When considering all locations together, the any injury reported crashed vehicle rate was 0.6 incidents per million miles (IPMM) for the ADS vs. 2.80 IPMM for the human benchmark, an 80% reduction or a human crash rate that is 5 times higher than the ADS rate. Police-reported crashed vehicle rates for all locations together were 2.1 IPMM for the ADS vs. 4.68 IPMM for the human benchmark, a 55% reduction or a human crash rate that was 2.2 times higher than the ADS rate. Police-reported crashed vehicle rate reductions for the ADS were statistically significant when compared in San Francisco and Phoenix, as well as combined across all locations and the any-injury-reported reductions were statistically significant in San Francisco and in all locations. The any property damage or injury comparison had statistically significant decreases in 3 comparisons but also nonsignificant results in 3 other benchmarks. When excluding ADS crashes with a delta-V less than 1 mph (a measure of sensitivity to lower reporting threshold), about half of the ADS collisions were excluded, resulting in comparisons that showed a large statistically significant reduction in all comparisons except for one comparison from San Francisco. Conclusions The statistically significant reductions in police-reported and any injury reported crash rates indicate a promising positive safety impact of ADS. The direction and significance of comparisons in the any property damage or injury outcome group are inconclusive due to difficulties in estimating a matching human benchmark. More research is needed on defining any property damage or injury benchmarks with clear lower reporting thresholds to reduce the systematic uncertainty in the benchmark rates. Together, these crash rate results contribute to the continuous growth in confidence, together with other methodologies, in a safety case approach.

Evaluation of Automated Driving System Safety Metrics With Logged Vehicle Trajectory Data

Evaluation of Automated Driving System Safety Metrics With Logged Vehicle Trajectory Data

Exploratory analysis of injury severity under different levels of driving automation (SAE Levels 2 and 4) using multi-source data

Vehicles equipped with automated driving capabilities have shown potential to improve safety and operations. Advanced driver assistance systems (ADAS) and automated driving systems (ADS) have been widely developed to support vehicular automation. Although the studies on the injury severity outcomes that involve automated vehicles are ongoing, there is limited research investigating the difference between injury severity outcomes for the ADAS and ADS equipped vehicles. To ensure a comprehensive analysis, a multi-source dataset that includes 1,001 ADAS crashes (SAE Level 2 vehicles) and 548 ADS crashes (SAE Level 4 vehicles) is used. Two random parameters multinomial logit models with heterogeneity in the means of random parameters are considered to gain a better understanding of the variables impacting the crash injury severity outcomes for the ADAS (SAE Level 2) and ADS (SAE Level 4) vehicles. It was found that while 67 percent of crashes involving the ADAS equipped vehicles in the dataset took place on a highway, 94 percent of crashes involving ADS took place in more urban settings. The model estimation results also reveal that the weather indicator, driver type indicator, differences in the system sophistication that are captured by both manufacture year and high/low mileage as well as rear and front contact indicators all play a role in the crash injury severity outcomes. The results offer an exploratory assessment of safety performance of the ADAS and ADS equipped vehicles using the real-world data and can be used by the manufacturers and other stakeholders to dictate the direction of their deployment and usage.

Optimizing the Steering of Driverless Personal Mobility Pods with a Novel Differential Harris Hawks Optimization Algorithm (DHHO) and Encoder Modeling.

This paper aims to improve the steering performance of the Ackermann personal mobility scooter based on a new meta-heuristic optimization algorithm named Differential Harris Hawks Optimization (DHHO) and the modeling of the steering encoder. The steering response in the Ackermann mechanism is crucial for automated driving systems (ADS), especially in localization and path-planning phases. Various methods presented in the literature are used to control the steering, and meta-heuristic optimization algorithms have achieved prominent results. Harris Hawks optimization (HHO) algorithm is a recent algorithm that outperforms state-of-the-art algorithms in various optimization applications. However, it has yet to be applied to the steering control application. The research in this paper was conducted in three stages. First, practical experiments were performed on the steering encoder sensor that measures the steering angle of the Landlex mobility scooter, and supervised learning was applied to model the results obtained for the steering control. Second, the DHHO algorithm is proposed by introducing mutation between hawks in the exploration phase instead of the Hawks perch technique, improving population diversity and reducing premature convergence. The simulation results on CEC2021 benchmark functions showed that the DHHO algorithm outperforms the HHO, PSO, BAS, and CMAES algorithms. The mean error of the DHHO is improved with a confidence level of 99.8047% and 91.6016% in the 10-dimension and 20-dimension problems, respectively, compared with the original HHO. Third, DHHO is implemented for interactive real-time PID tuning to control the steering of the Ackermann scooter. The practical transient response results showed that the settling time is improved by 89.31% compared to the original response with no overshoot and steady-state error, proving the superior performance of the DHHO algorithm compared to the traditional control methods.

Safety of the Intended Functionality Validation for Automated Driving Systems by Using Perception Performance Insufficiencies Injection

Safety of the Intended Functionality Validation for Automated Driving Systems by Using Perception Performance Insufficiencies Injection

Comparative safety performance of autonomous- and human drivers: A real-world case study of the Waymo Driver

After several years of public road testing, the commercial deployment of fully autonomous vehicles—or Automated Driving Systems (ADS)—is poised to scale substantially following significant technological advancements and recent regulatory approvals. However, the fundamental question of whether an ADS is safer than its human counterparts remain largely unsolved due to several challenges in establishing an appropriate real-world safety comparison method. As scaling ensues, the lack of an established method can contribute to misinterpretations or uncertainties regarding ADS safety and impede the continuous and consistent assessment of ADS performance. This study introduces three research developments to define a robust and replicable safety comparison method to address this critical methodological gap. First, we introduce the use of liability insurance claims data to measure the comparative safety between ADS and human drivers. Second, we use Swiss Re insurance claims data to establish the first zip code- and responsibility-calibrated human performance benchmark, composed of over 600,000 private passenger vehicle claims and 125 billion miles of driving exposure. Third, we perform a case study by applying the developed baseline to evaluate the safety impact of the Waymo Driver. We find that when benchmarked against zip code-calibrated human baselines, the Waymo Driver significantly improves safety towards other road users. The comparison method established in this study can be replicated for other regions or ADS deployments to aid the decision-making of ADS safety stakeholders such as regulators, and instill trust in the general public.

Baseline vulnerable road user injury risk in multiple U.S. dense urban driving environments

Objective Understanding and modeling baseline driving safety risk in dense urban areas represents a crucial starting point for automated driving system (ADS) safety impact analysis. The purpose of this study was to leverage naturalistic vulnerable road user (VRU) collision data to quantify collision rates, crash severity, and injury risk distributions in the absence of objective injury outcome data. Methods From over 500 million vehicle miles traveled, a total of 335 collision events involving VRUs were video verified and reconstructed (126 pedestrians, 144 cyclists, and 65 motorcyclists). Data consisted of anonymized video and sensor data (Global Positioning System and accelerometer) from vehicles equipped with Nexar dash cameras. Each event was qualitatively evaluated to assess the collision geometries and other extrinsic collision factors (e.g., time of day, roadway location, presence of occlusions). Previously published injury risk models were utilized to provide estimates of Maximum Abbreviated Injury Scale MAIS2+ and MAIS3+ injury potential. Collision severity distributions and aggregated injury risk estimates were compared. Results Vehicle speeds at the time of impact ranged from 0 to 39 mph, and VRU speeds ranged from 0 to 27 mph. In general, vehicles were traveling at lower speeds at the time of impact when turning in comparison to straight travel. Nearly all events (95%) were associated with MAIS3+ injury risk estimates below 10%. Collisions involving a potential occlusion or the vehicle responding to surprising behavior by the VRU were associated with higher estimates of injury risk than those without occlusion or with the vehicle initiating the conflict. Based on accumulated risk for each event, it can be estimated that approximately 55 persons could be moderately injured (MAIS2+) and approximately 6 persons could be seriously injured (MAIS3+). Conclusion These data indicate that responding to surprising VRU behavior, having visibility be potentially occluded, and vehicle travel behavior were associated with differences in collision speed and injury risk estimation. The specific comparisons made in this study are not intended to be comprehensive but serve as a starting point for considering baseline driving risk associated with VRU collisions in dense urban areas.

Searching for a Cheap Robust Steering Controller

The study of lateral steering control for Automated Driving Systems identifies new control solutions more often than new control problems. This is likely due to the maturity of the field. To prevent repeating efforts toward solving already-solved problems, what is needed is a cohesive way of evaluating all developed controllers under a wide variety of environmental conditions. This work serves as a step in this direction. Four controllers are tested on five maneuvers representing highways and collision avoidance trajectories. Each controller and maneuver combination is repeated on five sets of environmental conditions or Operational Design Domains (ODDs). The design of these ODDs ensures the translation of these experimental results to real-world applications. The commercial software, CarSim 2020, is extended with Simulink models of the environment, sensor dynamics, and state estimation performances to perform highly repeatable and realistic evaluations of each controller. The results of this work demonstrate that most of the combinations of maneuvers and ODDs have existing cheap controllers that achieve satisfactorily safe performance. Therefore, this field’s research efforts should be directed toward finding new control problems in lateral path tracking rather than proposing new controllers for ODDs that are already solved.

The effect of human-machine interface modality, specificity, and timing on driver performance and behavior while using vehicle automation

The effectiveness of the human–machine interface (HMI) in a driving automation system during takeover situations is based, in part, on its design. Past research has indicated that modality, specificity, and timing of the HMI have an impact on driver behavior. The objective of this study was to examine the effectiveness of two HMIs, which vary by modality, specificity, and timing, on drivers’ takeover time, performance, and eye glance behavior. Drivers’ behavior was examined in a driving simulator study with different levels of automation, varying traffic conditions, and while completing a non-driving related task. Results indicated that HMI type had a statistically significant effect on velocity and off-road eye glances such that those who were exposed to an HMI that gave multimodal warnings with greater specificity exhibited better performance. There were no effects of HMI on acceleration, lane position, or other eye glance metrics (e.g., on road glance duration). Future work should disentangle HMI design further to determine exactly which aspects of design yield between safety critical behavior.

Development and validation of automated driving behavior questionnaire (ADBQ)

The Driver Behavior Questionnaire (DBQ), which is rooted in Human Factors, is a common tool used in the context of driving behavior. However, it does not consider driver behavior when automated-driving systems are active. The primary objective of this study is to construct a questionnaire with high levels of reliability to assess the driving behavior of automated-vehicle (AV) users. This paper presents the development and validation process of a 16-item automated driving behavior questionnaire, composed of three factors of the driving process (perception, cognition, and action) and two factors of automated driving systems (user literacy and dependency). Responses from 441 active AV users were collected and analyzed. The application of factor analysis resulted in the identification of a five-factor solution, and the results showed that male drivers exhibited higher levels of literacy, action, and reliance on AVs than female drivers. In terms of trust, those with complete trust reported higher AV dependency, whereas those with low trust reported lower AV dependency and higher cognitive scores. Addressing gender- and trust-based disparities is crucial for traffic safety, especially among male and less-trusting drivers. The developed ADBQ will serve as a supporting tool for system developers and researchers to assess the driving behaviors of AV users.

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.

The Impact of Transparency on Driver Trust and Reliance in Highly Automated Driving: Presenting Appropriate Transparency in Automotive HMI

Automation transparency offers a promising way for users to understand the uncertainty of automated driving systems (ADS) and to calibrate their trust in them. However, not all levels of information may be necessary to achieve transparency. In this study, we conceptualized the transparency of the automotive human–machine interfaces (HMIs) in three levels, using driving scenarios comprised of two degrees of urgency to evaluate drivers’ trust and reliance on a highly automated driving system. The dependent measures included non-driving related task (NDRT) performance and visual attention, before and after viewing the interface, along with the drivers’ takeover performance, subjective trust, and workload. The results of the simulated experiment indicated that participants interacting with an SAT level 1 + 3 (system’s action and projection) and level 1 + 2 + 3 (system’s action, reasoning, and projection) HMI trusted and relied on the ADS more than did those using the baseline SAT level 1 (system’s action) HMI. The low-urgency scenario was associated with higher trust and reliance, and the drivers’ visual attention and NDRT performance improved after viewing the HMI, but not statistically significantly. The findings verified the positive role of the SAT model regarding human trust in the ADS, especially in regards to projection information in time-sensitive situations, and these results have implications for the design of automotive HMIs based on the SAT model to facilitate the human–ADS relationship.

The Design of Operational Design Condition for Automated Driving System

<div>A new revolution has taken place in the automobile industry in recent years, intelligent and connected vehicle (ICV) [<span>1</span>] has achieved a higher market share in recent years and relevant technologies have been quickly developed and widely accepted, so the auto industry needs to make regulations for automated driving system (ADS) on ICVs, mainly to assure the safety of ICV. To meet the requirements above, the definition of operational design domain (ODD) [<span>2</span>, <span>3</span>] was put forward by the Society of Automotive Engineers (SAE) and International Organization for Standardization (ISO) a few years ago. ODD defines necessary external environment conditions for the ADS to operate, but the internal status of the vehicle is also a key part of judging whether ADS can operate safely. Based on that, we propose a novel definition named operational design condition (ODC), which can supersede ODD to play a bigger role in improving regulations and standards, and promoting vehicle safety and technological development. We summarize the research about ODD, analyze the relationship between ODC and ODD, and list the advantages of the proposed definition of ODC. ODC is designed based on the principle of layered description, concerns for safety, perceivable, and minimal collection, which will promote technologies and cooperation based on a deeper and common understanding.</div>