From the Concept of Operations towards an Operational Design Domain for safe AI in Aviation

Kurzfassung

In the project RESILIENZ at the German Aerospace Center, a systematic approach for combining operational scenarios from Concept of Operations (ConOps) to capture specific operational ranges and limitations, is investigated. Based on an exemplary aviation use case - collision avoidance -, the corresponding ConOps is defined and used as a basis to derive the Operational Design Domain (ODD). Since ODD is one of the pillars for developing safe AI-based systems this enables meeting high-level system requirements and in the future resilience for failure tolerant AI systems. In this paper, a model-based system engineering framework is introduced in which scenarios are generated from the ConOps description in a highly automated way. In total 50 scenarios were generated and evaluated. This dataset of the operational scenario can be used for testing the initial ODD coverage. The framework in this paper ensures a systematic and highly automated approach from describing the ConOps towards deriving and testing an ODD in the aviation domain.