Towards a Hazard Assessment Workflow for Assurance of Increasingly Autonomous Systems

The advent of increasingly autonomous systems poses a significant challenge for design assurance. As human operators are increasingly replaced by automated decision-making, the operational environment must be taken into account at design time. As a result, there is an increasingly large and complex input space to consider during design, and assurance approaches must be adapted to accommodate this paradigm shift. Existing safety assessment methodologies were typically developed for systems under continuous human supervision, and new approaches are needed to assure autonomous systems. In this article, we propose a hazard assessment methodology for the safety assurance of increasingly autonomous systems. The workflow combines Natural Language Processing (NLP)-enabled elicitation of hazards from historical incident data, functional modeling, and scenario-based resilience simulation to augment traditional expert-driven hazard analysis and improve coverage of the hazard space. Requirements derived from this analysis are subsequently formalized in a structured natural language, enabling automatic generation of logical formulae, which can be analyzed for correctness. The framework is realized through two NASA-developed toolsets: the Hazard Analysis DESigner (HADES) and the Formal Requirements Elicitation Tool (FRET). The methodology is demonstrated on TROUPE, a NASA experimental rover swarm designed to autonomously and cooperatively map unknown terrain.