Year: 2021
Type: dissertation
Abstract: The complexity of the development of modern automobiles is steadily increasing, especially due to the topic of automated driving functions. Unlike former driver assistance systems, which only support the driving tasks, modern systems have to take full control over the car. This imposes high demands on the quality and robustness of the systems and their development processes. However, the information exchange between the development phases is still conducted manually by using text-based data (e.g., the specification). Previous attempts to optimize this process by using model-based approaches failed because of the insufficient willingness of the developers to adapt them, mainly due to the initial learning curve or the conversion expense. Because the testing of modern driving functions is more and more based on simulations, the importance of the corresponding traffic scenario descriptions is rising. Those descriptions are currently created manually, either as plain text or directly as simulation model. Additionally, a reliable estimation of the required test effort for multi-sensor systems already during early development phases is crucial for a sound project planning. Such a project planning also includes a selection of the software development processed, methods and tools that should be applied. Available approaches do not offer a quick graphic overview or the possibility to compare quite different approaches. The taxonomy proposed in this work provides such an option, based on the V-Model and further annotations. Neither the customer expectations regarding automated driving systems, nor the rating of the development methods by the developers remain stable. Hence, in this work, the results from three conducted surveys are given to get a review of the current state of the domain. To optimize the transitions between the development phases and to lower the adaptation effort for the developers, two iterative, text-based concepts for the formulation of requirements and scenario descriptions are proposed, each of them combined with a corresponding domain-specific language in the framework JetBrains Meta Programming System (MPS). Although existing approaches are also using textual formats, they do not support multilingual representations, data at different abstraction levels or the automatic generation of the handover artifacts (e.g., models) required for the development. Even though methods to estimate the test effort for multisensor systems are available, they rely either on the physical sensor models or implementation details. In this work, an approach is given, that can determine this effort to be conducted for a defined confidence level already during the specification phase and solely based on the sensor characteristics. These five concepts contribute to the scenario-accompanied development process of automated driving functions.
Author: Bock, Florian
Institution: Friedrich-Alexander-Universität Erlangen-Nürnberg
Language: German
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7054765509
