Precog: Requirements Engineering toward Safe Machine Learning-Based Perception Systems for Autonomous Mobility

Purpose and goal:

The goal of the Precog prestudy was to work toward an alignment of high-level expectations on machine learning-based perception systems in autonomous mobility. Precog conducted an interview study to pave the way for a larger project by 1) identifying potential partners and 2) performing an initial challenge elicitation. We identified several potential partners for future projects within Sweden. Our findings will narrow our focus and prepare an application for a future Vinnova FFI call.

Expected results and effects:

We grouped the findings into challenges within eight themes: 1) Data, 2) Perception, 3) AI/ML aspects, 4) System- and software engineering, 5) Quality, 6) Business ecosystem, 7) Requirements engineering and 8) Annotation. Each of these themes could be a focus of future research studies, with results contributing back to the autonomous driving community. Next, we plan to author a paper describing pre-study results. The prioritized results will narrow our scope in the future application.

Approach and implementation:

The Precog study organized group interviews with key stakeholders in the Swedish automotive industry. The study was guided by a series of research questions refined into semi-structured interview questions. The sampling method was a mix of purposive, convenience, and snowball sampling. We interviewed 19 participants from five different companies and performed thematic analysis. Finally, we hosted a hybrid workshop with 20 participants to validate and prioritize future work.