Analysing the Safety of Decision-Making in Autonomous Systems
This paper from Richard Hawkins, John McDermid, and me is being presented at SafeComp 22 in Munich. Please do come along to see me present our work. We’ll update the paper and the link to the camera-ready version once the conference proceedings have been release. For now, enjoy…
We characterise an autonomous system as one that has the capability to take decisions independently from human control. This independent and autonomous decision making could give rise to new hazards or hazard causes not present in an equivalent human-controlled system, e.g. through lack of human real-world understanding. Despite the increased adoption of autonomous systems there has been a dearth of research in the area of safety analysis and assurance of decision-making for autonomous systems. This paper is intended to be a first step to fill this gap. We compare and contrast the differing causal models of autonomous and non-autonomous systems, and build on existing safety engineering techniques in order to define a process (Decision Safety Analysis) for the analysis of autonomous decision-making. We show, using a real-world example, how this process supports the development of safety requirements to mitigate hazardous scenarios.