Safe Scenario Boundaries Determination by Parameter Variation for an Automated Driving System
(2023)
An expanding area of research interest is the scenario-based testing and development of Automated Driving Systems (ADS). In scenario-based testing, a system is examined in a set of pre-defined scenarios to inspect its behavior. Scenarios are described by a set of parameters, such as velocities and distances. For a safety-related system, identifying the parameter limits for safe operation is essential to reduce harm. Hence, there is a need to determine safe boundaries considering the assumed parameter set in a scenario to support the Verification and Validation (V&V) of an ADS. This paper presents a systematic approach to determining safe boundaries of parameters by scenario-based testing. The contributions of this work are: (i) performing scenariobased parameter variation to detect collisions, (ii) identifying safe boundaries of each single parameter from a specific Operation Design Domain (ODD) and (iii) providing safety-related evidence to identify safe boundaries from defined ODD. The results of this work can assist scenario reduction techniques to derive nothazardous scenarios and hazardous scenarios and support the V&V Processes of ADS.
Scenario-based testing is essential for Highly Automated Driving (HAD) vehicles to determine the safety-related input parameters and their boundaries. The increasing complexity, vehicle functions, and operational design pose new challenges for scenario-based testing, as the number of scenarios is enormous. Therefore, an efficient and systematic process is required in the various stages of scenario-based testing. The contribution of this study is to provide sensitivity information of safety related parameters and support logical scenario reduction. This paper presents an approach that supports to optimize the safety-related parameters boundary towards logical scenario reduction. Additionally, sensitivity analysis is applied by computing Variance- Based Sensitivity Analysis (VBSA) indices and prioritize the input parameters. Two datasets are investigated by VBSA based on the input parameters. One dataset is based on the samples from realworld scenarios and other dataset is derived from the samples considering statistic distributions with a specific parameter range. Moreover, the proposed approach is applied to an exemplary use case and the outcomes are demonstrated.
Scenario analysis is essential for the validation of highly automated driving (HAD) systems. The complexity of overall system safety is increasing in terms of Functional Safety (FuSa) and Safety of Intended Functionality (SOTIF). However, field testing of all possible safety-critical scenarios is hardly possible for automated vehicles. Therefore, scenario simulation is necessary for HAD to support the validation process and has gained acceptance in recent years. However, scenariobased analysis leads to an explosion of scenarios, so a scenario database is required at the beginning of the development phase. Hence, scenario reduction approaches need to be integrated in the conceptual phase to reduce the scenario modeling and testing effort. The contribution of this paper is to present simulation-based testing approaches for determining a reduced set of collision scenarios, taking into account the sensitivity of safety-critical parameters. Furthermore, a set of application scenarios is simulated to demonstrate the scenario reduction approaches by considering the detected collisions under a specific or restricted Operational Design Domain (ODD). In addition, this study supports the provision of a safety argument with evidence by introducing a variance-based sensitivity analysis and a scenario database that can be used as an input data set for an artificial intelligence system.