Removing the Big Red Button: Uncrewed Ground Vehicles in Complex Military Environments

This paper sets out an approach being taken by the Defence Science and Technology Laboratory (Dstl), the University of Exeter and Supacat to derive preliminary safety arguments for autonomous land vehicles operating in complex military environments. The use of machine learning by an autonomous vehicle to achieve human-defined high-level goals leads to variability and a level of unpredictability in how these goals are achieved, which is at odds with the traditional approach to safety case derivation, particularly in the creation and presentation of evidence to support compelling safety arguments of predictably safe behaviour. This may be one cause of lag in these vehicles’ exploitation and conclusive trialling in a military context, promoting the continued reliance on manually-activated motion termination as part of the safety argument. We evaluate the hazards and associated risks associated with operating an optionally crewed vehicle, using a bow-tie diagram approach to define revised safety architectures for more automated motion termination. It defines some of the wider associated Defence Lines Of Development (DLODs) [1] actions and revisions to autonomy-related standards and regulations, which would facilitate development of such systems and help accelerate both the development and deployment of progressively assured autonomous vehicles in the military land domain.