System Capability Feedback-Cycles in Automotive Software Development

Abstract: Context: The automotive industry is currently going through rapid change, driven by new technology; for example, electrification, autonomous driving, and connected cars. This new technology is largely based on electronics and software, and vehicles are increasingly becoming software-intensive systems. This affects how vehicles are developed, as automotive companies seek to adopt processes used in development of software-only systems, to gain the benefits of development speed and quick learning cycles possible in software development. Where sequential processes were previously the norm, automotive companies now aim to use agile methods at company-scale. Given the safety-critical nature of vehicles, and the mix software, hardware, and mechanical parts, this is challenging. Objective: This thesis explores how system-level feedback capabilities can be achieved in development of automotive systems. Method: To investigate a real-world setting, empirical methods are a natural choice. As an overarching research strategy, field studies are conducted at automotive companies. Over four studies, qualitative data is collected through semi-structured and structured interviews, focus groups, and workshops. The data is analyzed using adaptable methods, such as thematic coding. These qualitative approaches allow for open-ended questions, which are suitable for exploratory research. Findings: Transitioning towards agility changes the role of architecture, requirements, and in general of system-level artifacts previously finalized during early development phases. Nevertheless, what is covered by architecture and requirements still needs to be handled. They contain accumulated expertise, and fundamental concerns, such as safety, remain. However, automotive companies need to handle an increased importance of software for new feature development. Continuing business-as-usual is not an option. Conclusion: To achieve feedback capabilities on the system-level, there is a need for tools and methods allowing artifacts on higher levels of abstraction, for example architecture descriptions and requirements, to be modified and evolve over the entire course of development.