Toward Learning Realizable Scenario-Based, Formal Requirements Specifications

Distributed, software-intensive systems such as fully automated cars have to handle various situations employing message-based coordination. The growing complexity of such systems results in an increasing difficulty to achieve a high quality of the systems' requirements specifications, particularly w.r.t. the realizability of the specifications. Scenario-based requirements engineering addresses the message-based coordination of such systems and enables, if underpinned with formal languages, automatic requirements validation techniques for proving the realizability of a requirements specification. However, formal requirements modeling languages require a deep knowledge of requirements engineers and typically require many manual iterations until they find a realizable specification. In order to support requirements engineers in the stepwise development of scenario-based requirements specifications, we propose to evolve a high-quality specification from a (presumably unrealizable) manually created specification employing an evolutionary algorithm. In this paper, we show our results on automatically evolving new assumptions on the systems' environment behavior that guarantee a realizable requirements specification. Based on this contribution, we outline our research roadmap toward our long-term goal of automatically supporting requirements engineers in finding high-quality requirements specifications.