Considerations in Assuring Safety of Increasingly Autonomous Systems

Recent technological advances have accelerated the development and application of increasingly autonomous (IA) systems in civil and military aviation. IA systems can provide automation of complex mission tasks-ranging across reduced crew operations, air-traffic management, and unmanned, autonomous aircraft-with most applications calling for collaboration and teaming among humans and IA agents. IA systems are expected to provide benefits in terms of safety, reliability, efficiency, affordability, and previously unattainable mission capability. There is also a potential for improving safety by removal of human errors. There are, however, several challenges in the safety assurance of these systems due to the highly adaptive and non-deterministic behavior of these systems, and vulnerabilities due to potential divergence of airplane state awareness between the IA system and humans. These systems must deal with external sensors and actuators, and they must respond in time commensurate with the activities of the system in its environment. One of the main challenges is that safety assurance, currently relying upon authority transfer from an autonomous function to a human to mitigate safety concerns, will need to address their mitigation by automation in a collaborative dynamic context. These challenges have a fundamental, multidimensional impact on the safety assurance methods, system architecture, and V&V capabilities to be employed. The goal of this report is to identify relevant issues to be addressed in these areas, the potential gaps in the current safety assurance techniques, and critical questions that would need to be answered to assure safety of IA systems. We focus on a scenario of reduced crew operation when an IA system is employed which reduces, changes or eliminates a human's role in transition from two-pilot operations.