A Matched Case-Control Analysis of Autonomous vs Human-Driven Vehicle Accidents

doi:10.21203/rs.3.rs-3401212/v1

Download PDF

Article

A Matched Case-Control Analysis of Autonomous vs Human-Driven Vehicle Accidents

https://doi.org/10.21203/rs.3.rs-3401212/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 18 Jun, 2024

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

Despite the recent advancements that autonomous vehicles (AV) have shown in their potential to improve safety and operation, considering differences between AV and human-driven vehicles (HDV) in accidents remain unidentified due to the scarcity of real-world AV accident data. We investigate the difference in accident occurrence between AV levels and HDV by utilizing 2,100 AV (ADS and ADAS) and 35,113 HDV accident data. A matched case-control design was conducted to investigate the impact of different variables on the likelihood of accidents involving AV versus HDV. The analysis suggests that AVs tend to be safer than HDVs in many accident situations. Yet, the likelihood of an AV accident during dawn/dusk or while making turns is greater than that of HDVs in the same situations. Moreover, rear end accidents in which an AV hits an HDV are less common than those where an HDV hits an AV. Our research sheds light on the accident risk disparities between AVs and HDVs, informing future developments in AV technology and safety enhancements.

Physical sciences/Engineering/Civil engineering

Scientific community and society/Social sciences/Decision making

The automation of systems has been experiencing rapid development and has brought about a revolution in the transportation industry. The introduction of AV technology has made the vision of a safe transportation system with effortless driving seem attainable. It is anticipated that the automation of systems will significantly reduce the number of accidents, as human errors contribute to 50% to 90% of them¹. While smart transportation has showcased several benefits, these emerging technologies have also exhibited significant drawbacks, particularly regarding safety risks. Accidents of on-road testing have already been documented in limited testing data².

According to findings from the RAND corporation³, a mere advancement in safety features in the initial release of AVs could yield significant life-saving results. The research suggests that if AVs were to be introduced with an average safety level ten percent higher than that of the typical human driver, approximately 600,000 fatalities could be averted in the United States over a span of 35 years. Nevertheless, it is crucial to acknowledge that between 2015 and 2022, there was a yearly rise in both the annual miles traveled by autonomous vehicles (AVMT) and the number of AV accidents on public roads in California, with the exception of a decline in 2020 attributable probably to the COVID-19 pandemic⁴. AV testing on California public roads is permitted by the California Department of Motor Vehicles (CADMV)⁵. Up until June 2023, we have identified and enriched with further parameters 598 ADS accidents in California (AVOID dataset)⁶. These reports provide details about AV accidents and disengagements, which happen when the vehicle switches from autonomous mode due to technological issues or when the test driver or operator assumes manual control for safety purposes. The detailed information included in these reports revealed various factors related to AV accidents. We further collected the corresponding reports of HDV accidents to contrast the differences between AV and matched HDV accidents.

In this study, we thoroughly examined a dataset comprising both AV and HDV accidents. The dataset utilized in this study was compiled from various sources and encompassed information on accident types, road and environmental conditions, pre-accident vehicle movements, and accident outcomes. We use a matched case-control logistic regression model. We further add the NHTSA AV database comprising an additional 495 ADS and 1001 ADAS accidents. These findings can not only illuminate the factors that contribute to accidents involving AVs vs. HDVs, but also offer a new outlook on the risks connected to the rapidly advancing AV technology.

There are many potential benefits of AVs on traffic safety, such as a reduction in human error, reduced fatigue, and distraction. Cascetta et al. (2022) and Friedrich (2016) explored the behavior of SAE automation Level 2 cars and found out that AVs may exhibit some human-like behaviors, being more cautious and following traffic laws more strictly^7,8. Koopman and Wagner (2017) highlighted the importance of understanding how AVs will interact with human drivers, pedestrians, and other road users. However, there are also some possible safety challenges that need to be addressed in AVs⁹. Penmetsa et al. (2021) identified several safety challenges that need to be addressed in the development of AV, such as the need for reliable sensing and perception, robust decision-making algorithms, and fail-safe mechanisms¹⁰. Kalra and Paddock (2016) presented a statistical methodology with 382 accidents per 100 million miles based on binomial and Poisson distributions for estimating the required mileage to establish the reliability of AVs, it shows that AVs would have to be driven hundreds of millions of miles and sometimes hundreds of billions of miles to demonstrate their reliability in terms of fatalities and injuries¹¹.

The comparison of safety performance between AV and HDV is a topic of debate, with conflicting viewpoints¹². Schoettle and Sivak (2015) uncovered that AVs have a higher rate of accidents per million miles traveled compared to HDVs in limited (and generally less demanding) conditions (e.g., avoiding snowy areas) with 11 accident records¹³. Dixit et al. (2016) analyzed the statistical distribution of reports on 69 manual disengagements and accidents, they compared the accident rates of Google self-driving cars to those of human drivers and found that fatal accidents involving Google cars were lower than that of HDV, no fatalities occurred as compared to 1 death for every 108 million miles in California¹⁴. From 2009 to 2015 in Mountain View, California, Google cars demonstrated a significantly lower rate of police-reportable accidents per million vehicle miles traveled (VMT) compared to human drivers (2.19 versus 6.06)¹⁵. Shetty et al. (2021) found that the type of AV technology used was a significant predictor of accident risk, with vehicles that were less autonomous (i.e., requiring more human intervention) having a higher risk of accidents under different scenarios¹⁶. Some studies also have concentrated on analyzing the accident severity involving different types of AV or manual disengagement accidents. Favarò et al. (2017) presented significant insights into the dynamics of accidents involving 26 AV accidents, including the most common type of accidents, damage locations, impact forces, and accident frequencies¹⁷. The study found that rear-end accidents were the most frequent type of accidents from the macro perspective, with AVs being hit from behind by conventional vehicles at a rate twice that of rear-end accidents involving traditional cars by comparing testing vehicles in California with conventional vehicles in the U.S. in 2013. About 60% of these accidents involved relative speeds below 10 mph. Ding et al. (2023) collected a sizable dataset (1280 cases) to compare and analyze various variables related to injury severity between the ADAS and ADS equipped vehicles by random parameter multinomial logit models. Subject vehicle’s contact area, road and environment, and pre-accident movement, significantly impact crash injury severity. ADS-equipped vehicles in work zones have a higher probability of being involved in minor and moderate/severe injury accidents¹⁸.

These studies offered significant insight into the factors that contribute to AV and HDV accidents, while earlier studies may not consider sufficient factors of road environment, accident outcome, pre-accident condition, and accident type for accident analysis due to a lack of matched AV and HDV data for identifying the characteristics of AV accidents and how they differ from regular HDV accidents. The number of the accidents of the relevant studies and their automation level are shown in Fig. 1. Data from 2,100 AV accidents was identified from AVOID database. However, as the manual override and conventional modes of ADAS closely resemble HDV, we first use only 604 ADS cases from CA. Furthermore, to align with the characteristics of HDV accidents and due to difficulties in obtaining or imputing precise road types and corresponding variables for matched case-control logistic regression, some cases were removed. As a result, 548 AV accident accidents were deemed appropriate for the matched case-control design and are discussed in this paper.

3.1 Data Collection

The available data was segregated into two distinct categories for analysis. For AV accidents, a total number of 2,100 accidents have been collected and identified in the AVOID dataset thus far⁶, including pre-accident conditions and accident scene information for both CADMV and NHTSA accident data ¹⁹. The pre-accident condition variables considered in the analysis comprise vehicle manufacturers, AV driving modes, and pre-accident vehicle movement status. Furthermore, the accident scene information encompasses factors such as weather conditions, lighting conditions, accident types, and narrative describing the accident from the reporting agency. This information, along with other relevant factors, is utilized for conducting in-depth analysis of AV accidents. Among this information, the day of week, time of day and road location are typical confounders relevant to the traffic accident risk²⁰. To be specific, the risk of traffic accidents can vary by the day of the week and time of day due to differences in traffic volume and driver behavior (e.g., commuters mostly in peak periods). In addition, the road location can affect the risk of traffic accidents by influencing traffic volume, speed limits, and the presence of other risk factors such as road design factors, intersections, pedestrians, and bicyclists.

The second group of accidents comprised information on accidents involving vehicles operated by human drivers, which was gathered from the Statewide Integrated Traffic Records System (SWITRS)²¹. This format and structure of data can be matched with the AV data, we collected 35,113 cases of HDV accidents according to the year of accidents as the first step. In our analysis of matched case-control samples, we organized the data into N strata according to the occurrence of AV accidents. Each stratum consisted of one case and k corresponding controls. To ensure consistency across strata, we employed a matched case-control logistic regression by adjusting the number of control samples and assessed the resulting estimates for each model. The method begins by utilizing an initial equal proportion of AV accident samples to HDV accident samples (1:1) and progressively increasing the ratio (1:3, 1:5, 1:7, 1:9...) until the coefficients between consecutive models exhibit no significant change. From Fig. 2, it is evident that there are no notable disparities between the models employing sample ratios of 1:5 and 1:6. As a result, we opt for the 1:5 ratio for our analysis. To further support our decision, we evaluate the improvement in log-likelihood across the models, aligning with our hypothesis of selecting an AV accident sample to HDV-accident sample ratio of 1:5. The ultimate dataset comprised 2,740 HDV accidents as controls and 548 AV accidents as cases.

3.2 Data Description

The categories of variables include road and environmental characteristics (such as weather, road condition, road surface, and lighting conditions), pre-accident conditions (including associated factors, pre-accident movement, and accident participants), accident type, and accident outcome (accident severity), as shown in Table 1. In addition, the coordinates of accidents are extracted by Google Map API, then the type of road is identified to conduct matched case-control logistic regression. The full data set includes 2100 (ADS + ADAS) crashes based on AVOID (CADMV and NHTSA’s AV accident databases). In addition, we utilize 35,133 HDV accidents. Table 1 presents the general descriptive statistics of the databases. It provides a summary of the characteristics of the available variables that are classified into four major categories in the final data. It presents all the databases including 2100 AVs and 35,133 HDV accidents, and the ADS data specifically used for the matched case-control analysis. Significant disparities between AV and HDV accidents can be seen in movements such as slowing down, proceeding straight, and moving into opposing lanes, with AVs exhibiting higher rates. It's noteworthy that AVs tend to be involved more frequently in accidents in challenging conditions like traffic events, work zones, or narrowed roadways. Dawn or dusk conditions experience 6.9% of AV accidents, which is higher than the 3.7% rate for HDVs. Analyzing the pre-accident scenario, a distinctive observation is that only 5.9% of AV accidents are attributed to inattention or poor driving behavior, in contrast to a much higher 19.8% for HDVs.

Table 1

Summary statistics for the variables included in the generated datasets.
Variable type	Variable description	AV (ADS + ADAS)		ADS (For matched case control study)			HDV
Variable type	Variable description	Mean	Standard deviation	Mean	Standard deviation			Mean	Standard deviation
Road and environment
Weather	Clear sky indicator (1 if weather is clear, 0 otherwise)	0.7688	0.1404	0.8869		0.1404	0.8343			0.3718
	Cloudy sky indicator (1 if weather is cloudy, 0 otherwise)	0. 0948	0.1042	0.0748		0.1042	0.1105			0.3135
	Rain indicator (1 if weather is rain, 0 otherwise)	0.1294	0.0952	0.0292		0.0952	0.0519			0.2218
	Fog indicator (1 if weather is fog, 0 otherwise)	0. 0070	0.2089	0.0091		0.2089	0.0033			0.0574
Road Condition	Normal road conditions (1 if there are no unusual conditions like construction, maintenance, or obstruction, 0 otherwise)	0.7687	0.3171	0.9799		0.3171	0.9864			0.1160
	Traffic event/work zone (1 if a prior traffic event such as traffic accidents, disabled vehicles, and spilled cargo that adversely affected normal traffic operations separate from the reported incident or work zone, 0 otherwise)	0.2251	0.2633	0.0109		0.2633	0.0126			0.1114
	Reduced Roadway Width (1 if road condition is reduced roadway width, 1 otherwise)	0.006	0.1685	0.0091		0.1685	0.0011			0.0329
Road Surface	Dry surface indicator (1 if surface is dry, 0 otherwise)	0.7926	0.0952	0.881		0.0952	0.8854			0.3185
	Wet surface indicator (1 if surface is wet, 0 otherwise)	0.1151	0.4781	0.0867		0.4781	0.1034			0.1323
	Snow/slush/ice indicator (1 if surface is covered with snow/slush/ice, 0 otherwise)	0.0923	0.0427	0.0323		0.0427	0.0112			0.0321
Lighting	Daylight (1 if daylight, 0 otherwise)	0.7050	0.4023	0.7555		0.4023	0.6546			0.4755
	Dawn/dusk indicator (1 if it is dawn/dusk, 0 otherwise)	0.0694	0.3552	0.0255		0.3552	0.0369			0.1884
	Dark– lighting (1 if it is dark– lit, 0 otherwise)	0.1980	0.1265	0.2153		0.1265	0.2861			0.4520
	Dark– not lighting (1 if it is dark– not lit), 0 otherwise)	0.0276	0.0251	0.0036		0.0251	0.0224			0.1480
Pre-accident conditions
Associated Factor	Inattention (1 if driver is distracted, 0 otherwise)	0.0586	0.0952	0.0420		0.2007	0.1975			0.3981
	Entering/Leaving Ramp (1 if driver is Entering/Leaving Ramp, 0 otherwise)	0.0139	0.1200	0.0146		0.1200	0.0210			0.1433
	Defective Vehicle Equipment (1 if Vehicle Equipment is defective, 0 otherwise)	0.0109	0.1685	0.0292		0.1685	0.0012			0.0350
	Uninvolved Vehicle (1 if Vehicle is uninvolved, 0 otherwise)	0.0117	0.4999	0.0036		0.4999	0.0143			0.1186
	No dangerous driver behavior (1 if there is no dangerous driver behavior, 0 otherwise)	0.9049	0.2748	0.9106		0.2748	0.7660			0.4234
Pre-accident movement	Stopped (1 if pre-accident movement is stopped, 0 otherwise)	0.0435	0.1523	0.0237		0.1523	0.0016			0.0399
	Proceeding straight (1 if pre-accident movement is proceeding straight, 0 otherwise)	0.4835	0.4758	0.3549		0.4758	0.5801			0.4935
	Run-off Road (1 if pre-accident movement is Run-off Road, 0 otherwise)	0.0242	0.1200	0.0182		0.1200	0.0171			0.1297
	Turn (1 if pre-accident movement is turning left or right, 0 otherwise)	0.0711	0.2720	0.178		0.2720	0.1049			0.3065
	Backing (1 if pre-accident movement is backing, 0 otherwise)	0.0055	0.2007	0.0191		0.0952	0.0282			0.1655
	Slowing (1 if pre-accident movement is Slowing, 0 otherwise)	0.2109	0.0604	0.1414		0.0604	0.0005			0.0213
	Changing lanes (1 if pre-accident movement is changing lanes, 0 otherwise)	0.0582	0.2312	0.1277		0.2312	0.0661			0.2485
	Entering traffic lane (1 if pre-accident movement is entering traffic lane, 0 otherwise)	0.075	0.1404	0.0903		0.3910	0.0303			0.1715
	Into opposing lane (1 if pre-accident movement is into opposing lane, 0 otherwise)	0.007	0.1340	0.0146		0.1340	0.1634			0.3697
	Parked (1 if pre-accident movement is parked, 0 otherwise)	0.0086	0.3910	0.0201		0.1404	0.0015			0.0385
	Travelling wrong way (1 if pre-accident movement is travelling wrong way, 0 otherwise)	0.0125	0.3381	0.012		0.3381	0.0062			0.0787
Accident Participation	Pedestrian (1 if accident with pedestrian, 0 otherwise)	0.1023	0.0952	0.1423		0.0952	0.1510			0.580
	Other participants (1 if accident with other participants, 0 otherwise)	0.0203	0.3341	0.0310		0.3341	0.2227			0.4161
	Vehicle (1 if accident with another vehicle, 0 otherwise)	0.8773	0.3789	0. 8266		0.3789	0.6263			0.4838
Accident type
	Head-on (1 if accident type is Head-on, 0 otherwise)	0.1389	0.3534	0.1460		0.3534	0.0490			0.2158
	Sideswipe (1 if accident type is Sideswipe, 0 otherwise)	0.1441	0.4076	0.2099		0.4076	0.1419			0.3489
	Rear-end (1 if accident type is Rear-end, 0 otherwise)	0.5246	0.1921	0.5237		0.1921	0.5057			0.5000
	Broadside (1 if the front of one vehicle slams into the side of another vehicle, 0 otherwise)	0.1365	0.1735	0.0821		0.1735	0.2469			0.4312
	Hit-object (1 if accident type is Hit-object, 0 otherwise)	0.0559	0.3497	0.0383		0.3497	0.0565			0.2309
Accident outcomes
Severity	No-injury (1 if injury severity outcome is no-injury, 0 otherwise)	0.8371	0.4114	0. 6478		0.4114	0.6497			0.4771
	Minor (1 if injury severity outcome is minor, 0 otherwise)	0.1194	0.0604	0. 2026		0.4302	0.2753			0.4467
	Moderate (1 if injury severity outcome is moderate, 0 otherwise)	0.0406	0.4302	0.1478		0.0604	0.0660			0.2484
	Fatal (1 if injury severity outcome is fatal, 0 otherwise)	0.0031	0.1579	0.0018		0.1579	0.0089			0.0940

A matched case-control study is an observational study that involves comparing individuals who have a specific health outcome or disease (the cases) with individuals who do not have the health outcome or disease (the controls)²². The study design involves selecting cases and controls based on their exposure to a particular risk factor or characteristic, and then comparing the frequency of that exposure between the two groups. In the context of this paper, a matched case-control study could be used to investigate the relationship between accident-related risk factors^23,24. Cases would be AVs involved in accidents, and controls would be HDVs in accidents.

A matched control study has been designed to investigate the impact of various factors on the likelihood of accidents in two specific scenarios: AV and HDV.

Conditional logistic regression is a variant of logistic regression that specifically tackles the issue of stratification within matched case-control studies²⁵. In this research, there are N strata denoted by $i=\text{1,2},\dots \dots ,N$. Each stratum has one AV accident and k HDV accident samples denoted by $j=\text{1,2},\dots \dots ,K$.

The probability of observing the data in the $i th$ strata depend on the total number of observations and the number of accidents recorded within that stratum²⁶. The probability of the $j th$ observation being an accident in the ith strata is denoted as ${P}_{ri}$ (${X}_{ji}$), and it is determined by a vector of variables ${X}_{ij}=( {X}_{1ij}, {X}_{2ij}, {\dots \dots ,X}_{pij})$. A logistic regression model with linear parameters is employed to gauge the likelihood of an accident, as described by Abdel-Aty et al. (2004)²⁷:

$${logit\left|{P}_{ri}\right(X}_{ij})=( {a}_{i},{b}_{1}{X}_{1ji}, \dots \dots {b}_{k}{X}_{pji})$$

The controlled variables used to create strata are reflected in the intercept term. To incorporate the impact of stratification in the analysis, it is possible to construct a conditional log-likelihood. This log-likelihood function comprises multiple terms, each representing the conditional probability of an accident occurring within a specific stratum²⁸. The following equation presents the formula for the conditional likelihood function, as stated by Abdel-Aty et al. (2004)²⁷:

$$\pounds \left(\beta \right)={{\varPi }_{i=1}^{N}\left[\right( 1+{\sum }_{j=1}^{k}\text{exp}\left\{{\sum }_{u=1}^{p}{\beta }_{u}\left({x}_{uji}-{x}_{u0i}\right)\right\}\left)\right]}^{-1}$$

The coefficients' estimates in Eq. (1) are identical to the maximum likelihood function values in Eq. (2). These estimates are log-odds ratios that can provide an approximation of the relative risk of an accident and are also referred to as hazard ratios (i.e., the ratio of odds for accident occurrence versus non-occurrence). The hazard ratio is calculated by raising the exponential value to the coefficient's power. For a dummy variable, the odds ratio is a statistic defined as the ratio of the odds of the case. The odds ratio can be written as

$$OR\left({x}_{k}\right)=\frac{\text{Pr}\left({y}_{i0}=1,{x}_{k}=1,Z\right)/[1-\text{Pr}\left({y}_{i0}=1,{x}_{k}=1,Z\right)]}{\text{Pr}\left({y}_{i0}=1,{x}_{k}=0,Z\right)/[1-\text{Pr}\left({y}_{i0}=1,{x}_{k}=0,Z\right)]}=\text{e}\text{x}\text{p}\left({\beta }_{k}\right)$$

where, $Z$ represents the vector of explanatory variables excluding ${x}_{k}$. ${\beta }_{k}$ is the estimated coefficient for ${x}_{k}$.

5.1 General Trends in the Full Accident Data

We first present general trends and comparisons between AV and HDV accidents of the full dataset comprising 2100 AV (1099 ADS and 1001 ADAS related accidents) and 35,133 HDV only accidents. We also contrast the ADS vs the ADAS.

Figures 3 to 6 display the distribution of factors affecting AV (2,100), SAE level 4 ADS (1099), level 2 ADAS (1001) and HDV accidents (35,133), respectively. Few interesting findings can be noted by examining these Figures. Evaluating environmental factors, the majority of accidents involving both AVs and HDVs occur under clear skies. Specifically, HDVs have a slightly higher occurrence rate at 83%, compared to 77% for AVs. However, AVs are more frequently involved in accidents under rainy or foggy conditions. In terms of accident type, rear-end accidents constitute a majority for both AVs and HDVs, accounting for 52% and 51% of accidents, respectively. Notably, vehicles make up 88% of participants in AV accidents, with pedestrians accounting for 10%. In contrast, for HDVs, pedestrians constitute 15% and vehicles 63% of accident participants, as depicted in Figs. 3 and 6. When examining accident outcomes, AVs tend to result in no injuries or minor injuries more frequently than HDVs. Meanwhile, HDVs exhibit a higher propensity for outcomes resulting in moderate to fatal injuries, highlighting the potential safety superiority of autonomous vehicles.

ADAS and ADS display differences across various conditions. Regarding weather and road conditions, ADAS has 15.48% fewer accidents in clear skies but a 13.71% increase in rain compared to ADS, with comparable rates in fog. On road conditions, ADAS accidents experience a 28.94% rise in traffic events in work zones and surpass ADS by 17.28% on wet roads. Analyzing pre-crash conditions, ADAS accidents indicate a 43.02% increase for proceeding straight vehicles, while reporting 25.08% fewer turning vehicle accidents than ADS. In terms of accident types, ADAS surpasses ADS by 8.51% in broadside accidents and lags by 16.88% in sideswipe accidents. From an outcome perspective, ADAS accidents present an 18.84% uptick in no-injury cases but a 10.73% decrease in moderate injuries against ADS.

We have also analyzed the full data to identify the influence of roadway elements and factors related to time using a random parameter logit model. Only a single random parameter demonstrated a significant effect on the "Day of the week" variable. Upon analyzing the model, we found that the dawn/dusk indicator and turn indicator variables exhibit positive coefficients that are statistically significant at a 95% confidence level. This indicates a higher probability of an AV accident occurrence when these variables are present. Furthermore, we discovered that several variables demonstrate high significance and exhibit negative coefficients, suggesting a reduced likelihood of an accident when these factors are present. These variables include the rain indicator, rear-end indicator, broadside indicator (a broadside indicator is a car accident that occurs when the front of one vehicle slams into the side of another vehicle), moderate severity, proceeding straight indicator, run-off road indicator, backing indicator, and entering traffic lane indicator.

Furthermore, we determine the accident type associated with AV and HDV based on the accident angle. The graphical representation of the AV rear-end accident distribution is illustrated in Fig. 7. The analysis ed that 79% of rear-end accidents involve HDV hitting AV, while 21% of rear-end accidents involve AV hitting HDV. Notably, in cases where HDVs hit AVs, 65% of AVs are operating in the autonomous mode. Conversely, when AVs are responsible for hitting HDVs, 72% of AVs are operating in the conventional mode. This observation suggests that conventional mode occurred more frequently than autonomous mode where AV hit the HDV. It may be concluded that the autonomous mode result in fewer rear-end accidents for AVs compared with the conventional mode. This may be attributed to the advance autonomous mode of AVs. Autonomous mode uses advanced algorithms to detect and avoid obstacles and other vehicles in the path of the vehicle²⁹.

It is also shown in Fig. 7 that 69% of the AVs in the autonomous mode are stopping when they are hit by an HDV. When in conventional mode, half of the AV vehicles are moving, and the others are stopping. This means that if a pedestrian or object suddenly appears, the AVs can stop or slow down the vehicle to reduce the severity of an accident. When the AV is in conventional mode when AV hit HDV, most of the AVs are moving. While in autonomous mode, half of the AV vehicles are moving, and the others are stopping. We may conclude that compared with the autonomous mode, human drivers may not react as quickly or may not notice the object in time to take appropriate action. In terms of accident severity, 82% of accidents occur as minor when HDV hit AV. This percentage is 67% when the AV hit HDV, which may contradict common intuition. It is important to note that a majority of moderate and major accidents involving an AV hitting an HDV occur when both vehicles are moving in the conventional mode.

5.2 Matched Case Control Model for ADS Accidents in California

To overcome this challenge of variables that confound the relationship between risk factors and traffic accident outcomes, the first principle is to match cases and controls at the same location. In the case of a location that does not have enough controls, similar locations within a radius of 5 miles were used, and the day of the week and time of day were controlled for to make sure that cases and controls were under similar traffic patterns. Aside from intersections and streets, the location of each stratum for AV and HDV accidents is on the same highways and expressways. In addition, the same road type for each stratum is controlled for to ensure similar geometric design.

To examine the impact of exogenous variables on accident risk for different vehicle types, we conducted a matched case-control logistic regression model for AV (ADS) and HDV accidents. Here we use 548 ADS accidents and their 1:5 matched 2740 HDV accidents. The estimation results and 95% confidence intervals of the odds ratio are presented in Table 2, which was generated using the survival package in R programming³⁰. A total of 11 significant variables were identified by combining road and environment, accident type, accident outcomes, and pre-accident conditions during the estimation process.

Table 2

Matched Case-Control Logistic Regression Model. (1 if accident is AV, 0 HDV; Sampling ratio is 1:5)
Variable	Estimated parameter	Odds Ratio	t-statistic	P value
Road and environment
Rain indicator (1 if weather is rain, 0 otherwise)	-1.098	0.334	-8.038	< 0.001	[0.255,0.436]
Dawn/dusk indicator (1 if it is dawn/dusk, 0 otherwise)	1.639	5.15	8.962	< 0.001	[3.599,7.370]
Accident type
Rear-end (1 if accident type is rear-end, 0 otherwise)	-0.891	0.410	-3.815	< 0.001	[0.260,0.649]
Broadside (1 if accident type is broadside, 0 otherwise)	-1.662	0.189	-2.167	0.030	[0.042,0.853]
Accident outcomes
Moderate (1 if injury severity outcome is moderate, 0 otherwise)	-0.496	0.609	-2.773	0.006	[0.429,0.865]
Fatal (1 if injury severity outcome is fatal, 0 otherwise)	-0.525	0.592	-2.09	0.037	[0.362,0.968]
Pre-accident conditions
Proceeding straight (1 if pre-accident movement is proceeding straight, 0 otherwise)	-0.831	0.436	-6.977	< 0.001	[0.345,0.550]
Run-off Road (1 if pre-accident movement is Run-off Road, 0 otherwise)	-1.126	0.325	-2.918	0.004	[0.152,0.691]
Entering traffic lane (1 if pre-accident movement is entering traffic lane, 0 otherwise)	-1.746	0.175	-5.39	< 0.001	[0.093,0.329]
Turn (1 if pre-accident movement is turn, 0 otherwise)	0.317	1.373	2.036	0.042	[1.012,1.864]
Backing (1 if pre-accident movement is backing, 0 otherwise)	-0.630	0.533	-2.42	0.016	[0.350,0.887]
Number of strata	548
McFadden pseudo-R-squared	0.602
Likelihood ratio test	50.58
Wald test	46.95
Score (log rank) test	48.37

Sample size: 548 ADS accidents and HDV strata

5.3 Findings of Road, Environment, and Accident Type

Based on the results of the matched case control logistic regression, compared with HDV, the odds of an AV (ADS) accident occurring in rainy weather are 0.336 times. This indicates a lower likelihood of an AV accident in rainy weather compared to an HDV accident. To be noticed, several accidents involving vehicles equipped with ADS have occurred under favorable weather conditions, as these vehicles are seldom tested in adverse weather. AVs have a much faster reaction time than humans in responding to changing road conditions, including those caused by rain. They can rapidly process sensor data and adjust the vehicle's speed and trajectory within milliseconds, whereas humans may take several seconds to react¹⁴. Additionally, while rain can increase the likelihood of skidding or loss of control of a vehicle, AVs can employ consistent and precise sensing technologies such as cameras³¹, LiDAR³², radar³³, and GPS³⁴ (note not all AVs necessarily have all these sensors) to detect and accurately perceive road conditions, regardless of the weather conditions³⁵. In contrast, human drivers may have difficulties seeing through heavy rain or fog, leading to a delay in detecting potential hazards or reacting appropriately.

Interestingly, the dawn/dusk odds ratio indicates a 5.15 higher probability of AV accident than HDV accident. AVs are generally regarded as safer than human drivers in low-light scenarios such as dawn and dusk. This is because the sensors and cameras used by AVs may not be able to quickly adapt to changes in lighting conditions, which could affect their ability to detect obstacles, pedestrians, and other vehicles³⁶.At dawn and dusk, for instance, the sun's shadows and reflections may confuse sensors, making it hard for them to distinguish between objects and identify potential hazards. Furthermore, the fluctuating light conditions can impact the accuracy of object detection and recognition algorithms used by AVs, which can result in false positives or negatives³⁷.

Accident types related findings for AVs and HDVs is worth noting. Compared to HDV accidents, AVs experience relatively lower risks in rear-end and broadside (side-impact) accidents (0.4104 times and 0.1898 times, respectively). This finding indicates that AVs can detect and react to potential rear-end and side-impact accident situations much faster than humans can. This is because they are equipped with advanced sensors and software that can quickly analyze the surrounding environment and make decisions based on the data received^38,39. This allows AVs to respond to potential accidents before they occur, which can prevent or mitigate the severity of a rear-end accident.

5.4 Findings of Pre-Accident Conditions and Accident Outcomes

In terms of pre-accident conditions, most of the pre-accident movements made by AVs reduce the probability of accidents, except for turning, which increases the likelihood of an accident by 1.95 times compared to HDVs. One possible reason is a lack of situational awareness. AVs rely on sensors and algorithms to perceive their surroundings and make driving decisions⁴⁰. However, these systems may not detect all obstacles and hazards, particularly in complex and dynamic driving scenarios like turning at intersections⁴¹. This can result in errors in trajectory identification and accident avoidance. Additionally, AVs are programmed to follow predefined rules and scenarios, which may not encompass every possible driving situation⁴². The modifications of scenarios can present difficulty for AVs in perceiving and responding to them, thereby raising the risk of an accident⁴³. Conversely, most human drivers have years of experience and can adapt to unexpected circumstances on the road. This experience enables them to make better decisions while turning, such as adjusting their speed and trajectory based on the actions of other vehicles and pedestrians⁴⁴.

AV accidents are less likely to occur than HDV accidents in situations such as proceeding straight, run-off road (a vehicle leaves the designated roadway and travels onto an area that is not intended for regular traffic) and entering traffic lane conditions (a vehicle transitioning from a stationary or parked position to enter a traffic lane and become physically present within the flow of traffic). When considering the proceeding straight condition, it was found that AV accident resulted in a 0.4356 lower probability of an HDV accident. Remarkably, AV accident risk is 0.3245 times as high as that of an HDV accident in run-off road condition, which can be explained by the faster reaction time of AVs⁴⁵. AVs can detect these situations and apply corrective action, such as adjusting the speed or steering angle^46,47, more quickly and accurately than a human driver⁴⁸.

It is interesting to discover that the statistical analysis revealed a significant correlation between the entering traffic lane indicator and AV accidents, the risk of which is 0.1745 times as high as HDV accident. According to the results of the matched case-control logistic regression, the impact of backing is noteworthy, which show that the AV is less likely to be affected than the HDV. According to the analysis, the model using accidents of AVs showed a decreased probability of accidents for moderate and fatal indicators in comparison to HDV.

A comprehensive examination was performed using a dataset of AV and HDV accidents in this study. A total of 2,100 AV accidents and 35,133 HDV accident records were collected, which accurately reflected the accident details. The analysis first dealt the whole available data that includes both ADS and ADAS (SAE levels 4 and 2, respectively) using general descriptive statistics, percentages and a random parameter logit model (not shown in the paper for brevity and since results are almost consistent with the other matched model). The analysis considered four categories of variables, including accident type, road and environment, pre-accident movement, and accident outcomes.

The accident data of AV and HDV were compared using the matched case-control logistic regression. The impact of different variables on the potential of accidents in AV vs HDV was conducted using a matched case-control logistic regression model. Based on the model estimation results, it can be concluded that AVs (ADS) in general are safer than HDVs in most accident scenarios for its object detection and avoidance, precision control, and better decision-making.

However, the probability of an AV accident happening under dawn/dusk or turning conditions is 5.150 and 1.373 times higher, respectively, than the probability of an HDV accident occurring under the same conditions. The possible reasons are a lack of situational awareness in complex driving scenarios and limited driving experience of AVs¹⁸. Improving the safety of AV under dawn/dusk or turning conditions necessitates a holistic approach that involves advanced sensors, robust algorithms, and smart design considerations. Key strategies include enhancing weather and lighting sensors, implementing redundancy measures, and integrating sensor data effectively. By focusing on these aspects, the safety of AVs can be significantly enhanced in challenging scenarios. Although the results were mostly consistent in the two models, difference in backing accidents were found, indicating better performance of ADS in reducing backing accidents compared to ADAS.

Compared with current studies that only focus on AV accidents ^49–51 or analyze AV and HDV with limited samples ^4,7,52, this paper has provided valuable insights into the factors that contribute to AV in comparison to HDV accidents through the analysis of real-world accidents and multi-source data. Furthermore, the scope of the research has been broadened to encompass both AV and HDV, instead of solely concentrating on different levels of AV accidents without considering a comparison with HDV accidents. Moreover, this study successfully addresses the issue of unbalanced data between AV and HDV accidents by employing a matched case-control study design. Metrics such as location and Average Annual Daily Traffic (AADT) or speed limit are accounted for in the matched case-control design. One of the constraints of this study is analyzing the detailed levels of AV and the specific activated ADAS or AV system in an accident. Understanding and modeling different classifications of AV versus HDV accidents can be challenging and may require more data and advanced modeling techniques.

AUTHORS CONTRIBUTION

M.A. and S.D. conceived the study. S.D. and M.A. wrote the manuscript. S.D. estimated the models and conducted the analysis. M.A. supervised the analysis and edited the manuscript.

COMPETING INTERESTS

The authors declare no competing interests.

DATA AVAILABILITY

The data are open source and available at https://github.com/ozheng1993/AVOID-Autonomous-Vehicle-Operation-Incident-Dataset

ACKNOWLEDGEMENTS

The authors wish to thank Dr. Ou Zheng for his role in creating the AVOID data used in this study⁶

Milakis, D., Van Arem, B. & Van Wee, B. Policy and society related implications of automated driving: A review of literature and directions for future research. Journal of Intelligent Transportation Systems 21, 324–348 (2017).
Lee, D. & Hess, D. J. Regulations for on-road testing of connected and automated vehicles: Assessing the potential for global safety harmonization. Transportation Research Part A: Policy and Practice 136, 85–98 (2020).
Zhang, L. Cruise’s Safety Record Over 1 Million Driverless Miles, <https://getcruise.com/news/blog/2023/cruises-safety-record-over-one-million-driverless-miles> (2023).
Liu, Q., Wang, X., Wu, X., Glaser, Y. & He, L. Crash comparison of autonomous and conventional vehicles using pre-crash scenario typology. Accident Analysis & Prevention 159, 106281 (2021).
DMV, C. AUTONOMOUS VEHICLE COLLISION REPORTS, <https://www.dmv.ca.gov/portal/vehicle-industry-services/autonomous-vehicles/autonomous-vehicle-collision-reports/> (2023).
Zheng, O. et al. AVOID: Autonomous Vehicle Operation Incident Dataset Across the Globe. arXiv preprint arXiv:2303.12889 (2023).
Cascetta, E., Carteni, A. & Di Francesco, L. Do autonomous vehicles drive like humans? A Turing approach and an application to SAE automation Level 2 cars. Transportation research part C: emerging technologies 134, 103499 (2022).
Friedrich, B. The effect of autonomous vehicles on traffic. Autonomous driving: Technical, legal and social aspects, 317–334 (2016).
Koopman, P. & Wagner, M. Autonomous vehicle safety: An interdisciplinary challenge. IEEE Intelligent Transportation Systems Magazine 9, 90–96 (2017).
Penmetsa, P., Sheinidashtegol, P., Musaev, A., Adanu, E. K. & Hudnall, M. Effects of the autonomous vehicle crashes on public perception of the technology. IATSS research 45, 485–492 (2021).
Kalra, N. & Paddock, S. M. Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability? Transportation Research Part A: Policy and Practice 94, 182–193 (2016).
Leilabadi, S. H. & Schmidt, S. in 2019 IEEE Intelligent Transportation Systems Conference (ITSC). 889–893 (IEEE).
Schoettle, B. & Sivak, M. A preliminary analysis of real-world crashes involving self-driving vehicles. University of Michigan Transportation Research Institute (2015).
Dixit, V. V., Chand, S. & Nair, D. J. Autonomous vehicles: disengagements, accidents and reaction times. PLoS one 11, e0168054 (2016).
Teoh, E. R. & Kidd, D. G. Rage against the machine? Google's self-driving cars versus human drivers. Journal of safety research 63, 57–60 (2017).
Shetty, A. et al. Safety challenges for autonomous vehicles in the absence of connectivity. Transportation research part C: emerging technologies 128, 103133 (2021).
Favarò, F. M., Nader, N., Eurich, S. O., Tripp, M. & Varadaraju, N. Examining accident reports involving autonomous vehicles in California. PLoS one 12, e0184952 (2017).
Ding, S. et al. Exploratory Analysis of the Crash Severity between Vehicular Automation (SAE L2-5) with Multi-Source Data. arXiv preprint arXiv:2303.17788 (2023).
NHTSA. Standing General Order on Crash Reporting, <https://www.nhtsa.gov/laws-regulations/standing-general-order-crash-reporting#data> (2023).
Gross, F. & Jovanis, P. P. Estimation of the safety effectiveness of lane and shoulder width: Case-control approach. Journal of transportation engineering 133, 362–369 (2007).
SWITRS. Statewide Integrated Traffic Records System, <https://iswitrs.chp.ca.gov/Reports/jsp/index.jsp> (2023).
Ingram, D., Sanders, K., Kolybaba, M. & Lopez, D. Case-control study of phyto-oestrogens and breast cancer. The Lancet 350, 990–994 (1997).
Abdel-Aty, M. A., Hassan, H. M., Ahmed, M. & Al-Ghamdi, A. S. Real-time prediction of visibility related crashes. Transportation research part C: emerging technologies 24, 288–298 (2012).
Ahmed, M. M., Abdel-Aty, M. & Yu, R. Bayesian updating approach for real-time safety evaluation with automatic vehicle identification data. Transportation research record 2280, 60–67 (2012).
Rahman, M. M. & Lamsal, B. P. Ultrasound-assisted extraction and modification of plant‐based proteins: Impact on physicochemical, functional, and nutritional properties. Comprehensive Reviews in Food Science and Food Safety 20, 1457–1480 (2021).
Rahman, R., Bhowmik, T., Eluru, N. & Hasan, S. Assessing the crash risks of evacuation: A matched case-control approach applied over data collected during Hurricane Irma. Accident Analysis & Prevention 159, 106260 (2021).
Abdel-Aty, M. & Abdalla, M. F. Linking roadway geometrics and real-time traffic characteristics to model daytime freeway crashes: generalized estimating equations for correlated data. Transportation Research Record 1897, 106–115 (2004).
Peck, R. C., Gebers, M. A., Voas, R. B. & Romano, E. The relationship between blood alcohol concentration (BAC), age, and crash risk. Journal of safety research 39, 311–319 (2008).
Ahangar, M. N., Ahmed, Q. Z., Khan, F. A. & Hafeez, M. A survey of autonomous vehicles: Enabling communication technologies and challenges. Sensors 21, 706 (2021).
Therneau, T. A package for survival analysis in S. R package version 2 (2015).
Sadli, R., Afkir, M., Hadid, A., Rivenq, A. & Taleb-Ahmed, A. Map-matching-based localization using camera and low-cost GPS for lane-level accuracy. Sensors 22, 2434 (2022).
Chen, S., Liu, B., Feng, C., Vallespi-Gonzalez, C. & Wellington, C. 3d point cloud processing and learning for autonomous driving: Impacting map creation, localization, and perception. IEEE Signal Processing Magazine 38, 68–86 (2020).
Tang, X., Zhang, Z. & Qin, Y. On-road object detection and tracking based on radar and vision fusion: A review. IEEE Intelligent Transportation Systems Magazine 14, 103–128 (2021).
Cui, J., Liew, L. S., Sabaliauskaite, G. & Zhou, F. A review on safety failures, security attacks, and available countermeasures for autonomous vehicles. Ad Hoc Networks 90, 101823 (2019).
Yeong, D. J., Velasco-Hernandez, G., Barry, J. & Walsh, J. Sensor and sensor fusion technology in autonomous vehicles: A review. Sensors 21, 2140 (2021).
Parekh, D. et al. A review on autonomous vehicles: Progress, methods and challenges. Electronics 11, 2162 (2022).
Khatab, E., Onsy, A., Varley, M. & Abouelfarag, A. Vulnerable objects detection for autonomous driving: A review. Integration 78, 36–48 (2021).
Van Brummelen, J., O’Brien, M., Gruyer, D. & Najjaran, H. Autonomous vehicle perception: The technology of today and tomorrow. Transportation research part C: emerging technologies 89, 384–406 (2018).
Duarte, F. & Ratti, C. The impact of autonomous vehicles on cities: A review. Journal of Urban Technology 25, 3–18 (2018).
Cao, Y. et al. in Proceedings of the 2019 ACM SIGSAC conference on computer and communications security. 2267–2281.
Levin, M. W. & Boyles, S. D. Intersection auctions and reservation-based control in dynamic traffic assignment. Transportation Research Record 2497, 35–44 (2015).
Fagnant, D. J. & Kockelman, K. Preparing a nation for autonomous vehicles: opportunities, barriers and policy recommendations. Transportation Research Part A: Policy and Practice 77, 167–181 (2015).
Cunneen, M., Mullins, M. & Murphy, F. Autonomous vehicles and embedded artificial intelligence: The challenges of framing machine driving decisions. Applied Artificial Intelligence 33, 706–731 (2019).
Kyriakidis, M. et al. A human factors perspective on automated driving. Theoretical issues in ergonomics science 20, 223–249 (2019).
Van Griethuysen, C., Gillissen, F. & Koelmans, A. Measuring acid volatile sulphide in floodplain lake sediments: effect of reaction time, sample size and aeration. Chemosphere 47, 395–400 (2002).
Din, Z. M. U., Razzaq, W., Arif, U., Ahmad, W. & Muhammad, W. in 2019 4th International Conference on Emerging Trends in Engineering, Sciences and Technology (ICEEST). 1–4 (IEEE).
Lee, S., Jeong, E., Oh, M. & Oh, C. Driving aggressiveness management policy to enhance the performance of mixed traffic conditions in automated driving environments. Transportation research part A: policy and practice 121, 136–146 (2019).
Cui, J., Sabaliauskaite, G., Liew, L. S., Zhou, F. & Zhang, B. Collaborative analysis framework of safety and security for autonomous vehicles. IEEE Access 7, 148672–148683 (2019).
Kutela, B., Das, S. & Dadashova, B. Mining patterns of autonomous vehicle crashes involving vulnerable road users to understand the associated factors. Accident Analysis & Prevention 165, 106473 (2022).
Xu, C., Ding, Z., Wang, C. & Li, Z. Statistical analysis of the patterns and characteristics of connected and autonomous vehicle involved crashes. Journal of safety research 71, 41–47 (2019).
Zhu, S. & Meng, Q. What can we learn from autonomous vehicle collision data on crash severity? A cost-sensitive CART approach. Accident Analysis & Prevention 174, 106769 (2022).
Petrović, Đ., Mijailović, R. & Pešić, D. Traffic accidents with autonomous vehicles: type of collisions, manoeuvres and errors of conventional vehicles’ drivers. Transportation research procedia 45, 161–168 (2020).

There is NO Competing Interest.

Download PDF

Journal Publication

published 18 Jun, 2024

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

A Matched Case-Control Analysis of Autonomous vs Human-Driven Vehicle Accidents

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. BACKGROUND AND LITERATURE REVIEW

3. DATA PREPARATION

3.1 Data Collection

3.2 Data Description

4. MATCHED CASE-CONTROL LOGISTIC REGRESSION METHODOLOGY

5. RESULTS

5.1 General Trends in the Full Accident Data

5.2 Matched Case Control Model for ADS Accidents in California

5.3 Findings of Road, Environment, and Accident Type

5.4 Findings of Pre-Accident Conditions and Accident Outcomes

6. CONCLUSIONS

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1