Design of Flight Guidance and Control Systems Using Explainable AI

Advances in Artificial Intelligence (AI) have enabled the development of Flight Guidance and Control Systems (FG&CS) using Machine Learning (ML) from large archives of operational flight data. As the scope of the missions of drones/Small Unmanned Airborne Systems (sUAS) and Urban Air Mobility (UAM) vehicles increase (e.g. operations beyond-visual-line-of-sight (BVLOS) over people), the need for proof-of-compliance for safety regulations is required.Approaches to meet the regulatory airworthiness requirements for software assurance, such as those embodied in DO-178, require coverage of paths through the code. This "path coverage requirement" can be achieved by a sub-class of ML, known as Explainable ML (X-ML) that enables the derivation of formal models, such as Situation-Goal-Behavior Models (SGBM), from the ML FG&CS.This paper describes the results of a case-study of the development of an X-ML FG&CS that highlights the limits of an X-ML only approach and the need for a Model-based System Engineering (MBSE) analysis and design to complete the design to meet safety performance standards. The case study identified, specifically, the presence of five design error archetypes. These design error archetypes can be eliminated (i.e. addressed) by analysis of the SGBM using Model-Based System Engineering (MBSE) Tools as described in this paper. The implications and limitations of this approach for developing airworthy FG&CS are discussed.