Verification of autonomous systems: Challenges of the present and areas for exploration

Abstract

The words “failure” and “success” are simple binary terms used to describe the outcomes of maritime autonomous behavior, yet how to draw the line separating the two stark realities demonstrates the subtle nature of this problem. If how to judge success or failure is improperly performed, users of systems that rely upon autonomous behavior may quickly find their initial trust misplaced and in some cases, the trust may never be recovered. A loss of trust or an unwillingness to trust newer technologies can greatly hinder the transition and use of such technologies. Acceptance of current and future maritime autonomy so that its use is more ubiquitous and commonplace requires the establishment of rigorous techniques for which clearly defined verification of autonomy can be performed. Therefore, it is critical that verification of maritime autonomous systems be tenably and consistently implemented. For our work, we identify some of the difficulties with verification efforts with respect to general autonomy and not only maritime autonomy. In an attempt to address the difficulties, we present the potential use of techniques found within other fields to assist with some of these difficulties.