Verifying algorithms for autonomous aircraft by simulation - generalities and example

Abstract

An open question in Air Traffic Management is what procedures can be validated by simulation where the simulation shows that the probability of undesirable events is below the required level at some confidence level. The problem is including enough realism to be convincing while retaining enough efficiency to run the large number of trials needed for high confidence. The paper first examines the probabilistic interpretation of a typical requirement by a regulatory agency and computes the number of trials needed to establish the requirement at an equivalent confidence level. Since any simulation is likely to consider only one type of event and there are several types of events, the paper examines under what conditions this separate consideration is valid. This paper establishes a separation algorithm at the required confidence level where the aircraft operates under feedback control as is subject to perturbations. There is a discussion where it is shown that a scenario three to four orders of magnitude more complex is feasible. The question of what can be validated by simulation remains open, but there is reason to be optimistic.