Emerging Results from FAMER
The completed FAMER publications show a clear and connected progression of results. Together, they demonstrate how requirements engineering can better support the development of safe AI-based automotive perception systems, especially in multi-party industrial settings. Details about these publications are available on the FAMER Publications page.
The first set of results focuses on understanding the problem space. The study S1 on requirements representations in machine learning-based automotive perception systems highlights the need for better ways to represent, communicate, and connect requirements across multiple stakeholders. It shows that perception-system development depends on several interrelated requirement representations distributed across organisations, tools, and lifecycle stages.
A second stream of results S2 establishes data annotation as a requirements engineering concern rather than only a data-processing activity. This work shows that annotation decisions directly affect whether system-level goals, safety expectations, and quality requirements can be fulfilled in machine learning systems.
Building on this, the research S3 on managing data annotation requirements in practice identifies how annotation requirements emerge, evolve, and are coordinated in industrial autonomous-driving contexts. These findings make visible the practical challenges of defining, negotiating, and maintaining annotation requirements in real development settings.
Another important result S4 concerns data annotation quality problems. This work identifies recurring quality issues and shows how problems in annotation practices can propagate into downstream perception-system development, validation, and trustworthiness concerns.
Finally, the study S5 on Data Annotation Requirements Representation and Specification (DARS) moves from problem understanding to solution design. It provides a more structured way to represent and specify annotation requirements, helping connect system goals, annotation decisions, and validation needs in a traceable and reusable manner.
What These Results Mean for Industrial Partners
For the industrial partners, these results provide both understanding and practical value.
Volvo Cars benefits from a stronger shared language and clearer ways to connect high-level system concerns with perception-related requirements. The results support more structured requirements communication across complex development settings.
Zenseact benefits from methods and insights that improve multi-party coordination, requirement alignment, and iterative development of AI-based perception systems. The results are especially relevant for distributed agile environments where many actors contribute to the same system.
Kognic benefits from a clearer foundation for defining, refining, and validating annotation requirements. The results support better guidance for clients, more reliable ground-truth production, and improved annotation-tool and workflow design.
RISE benefits from results that strengthen safety assurance, requirement traceability, and industrial best practices for safe perception systems. These outcomes can also be transferred into research, innovation support, and competence-building activities.
University of Gothenburg benefits by translating these industrially grounded findings into scientific publications, teaching, and future research on requirements engineering for AI-based and safety-critical systems.
Overall FAMER Contribution So Far
Overall, the completed papers (see FAMER Publications page) show that FAMER is producing a coherent body of results that moves from:
- Understanding requirements and collaboration challenges,
- Identifying annotation-related requirements and quality problems,
- and developing structured solutions for representing and specifying these requirements.
This means the results are not isolated contributions, but emerging pieces of a broader framework for improving requirements engineering, annotation quality, traceability, and collaboration in safe AI-based perception-system development.