Development of freeway-based test scenarios for applying new car assessment program to automated vehicles

Abstract

As automated driving technology continues to develop, studies are being conducted to develop various scenarios for assessing the functional safety, failure safety, and mobility of automated vehicles (AVs). As the commercialization of AVs progresses, it is necessary to develop a set of test scenarios for new car assessment programs (NCAPs), so as to provide information on the safety and reliability of AVs to consumers. To provide valuable information regarding newly emerged AVs to consumers who are willing to purchase them, it is necessary to derive specific and well-defined test scenarios based on the safety-in-use. Accordingly, to apply NCAPs to AVs, this study established test scenarios targeting freeways where AVs were expected to be commercialized. To this end, based on freeway traffic accident data and opinions of traffic safety and AV experts, we derived possible dangerous situations when an AV is maintaining a lane on a freeway. Functional scenarios were defined based on the derived dangerous situations. The priority of the defined functional scenarios was set using the analytic hierarchy process (AHP). Accordingly, this study presents a logical and concrete scenario construction methodology for deriving the ranges and values of test parameters for functional scenarios.

Categorizing Data-Driven Methods for Test Scenario Generation to Assess Automated Driving Systems

“…Depending on the evaluation scope, different types of data sources, such as accident and real driving databases, may simultaneously be necessary. In the reviewed studies, five methods relied on two different types of primary data sources: police accident and real driving data [19], [46], [75], [76].…”

“…Approximately 68% of the data used in the reviewed methods were different types of real driving data that were used as inputs for the generation process. The most common sensors for real driving data are various camera videos or images from drones [17]- [20], [45], [48], [50], [65], [74], stationary cameras [19], [20], [42], [53], [62], [63], [71], [76], or dash cams/on-board cameras [26], [36], [43], [46], [51], [62], [63], [75]. Dynamic parameters, such as GPS or velocity, represent a large proportion of the real driving data (NDS and BUS data [16], [20]- [22], [24], [25], [27], [31], [32], [39], [44], [49], [51]- [53], [57]- [59], [66], [67], [72]).…”

“…A rarely used data source is a dataset of pre-crafted scenarios or trajectories [20], [23], [68], [73]. Accident databases, as data sources, share approximately 26% of the identified methods, which, in most cases, originate from a set of police accident reports [19], [24], [28]- [30], [33], [35], [43], [46], [47], [55], [61], [67], [69], [70], [75], [76]. In some cases, only accidents involving vehicles with an ADS were considered and filtered from existing accident databases [34], [37], [38], [64].…”

“…A unique example of the RA approach is the minimization of the performance outcome to identify worstcase scenarios and select them for the testing and evaluation of the ADS [72]. The representativity in the natural population is described by RB, which occurs ten times [19], [22], [28], [35], [38], [46], [54], [62], [63], [76].…”

RPRP-SAP: A Robust and Precise ResNet Predictor for Steering Angle Prediction of Autonomous Vehicles

“…However, considering a wide range of unexpected situations in real-world scenarios, (such as response in emergencies, climate factors, traffic jams and complex routes) thorough testing of AVs is expensive and time-consuming [8], [28]. This drives the necessity of computer-aided programs competent in performing extensive testing of AVs for all possible unsafe scenarios that AVs can encounter during their movements on different tracks [41].…”