Evolving robust strategies for autonomous flight: A challenge to optimal control theory

摘要

This article represents the second step in a series of papers posing a new challenge to the rigours of modern control theory, namely by developing an adaptive approach to create autonomous flight control systems. Based upon previous work whichdemonstrated that computational evolutionary adaptation is a generic extension of linear optimal control, here we explore the issues concerning the robustness of the technique in more rich and realistic domains. Robustness, like survivability, is a keydriving initiative behind the acceptance of control systems in failure critical scenarios, like automatic flight control. It therefore seems worthwhile to seek ways of adjusting evolutionary optimisation methods to encourage robustness across allindividuals that emerge from an evolutionary simulation. This can be achieved primarily by modifying the Evolutionary Strategy selection process to optimise across multiple simulation conditions, including noisy environments and control failure scenarios. By doing so, we have again empirically demonstrated a further prediction; evolutionary control designs are superior to their linear theoretic counterparts in terms of their robustness and likewise survivability in uncertain environments.