Three Fundamental Decisions in the Evaluation of Autonomous Naval Mine Hunting Operations

This article examines three key decisions in evaluating autonomous mine hunting operations using the Mine Shape Bayes Net framework. These decisions concern the classification of mine-like contacts (MILCOs) in side-looking sonar imagery. The first asks which mine classes to consider. The second asks whether to calibrate assessments of MILCO classification performance with mine garden trials or target simulation techniques. The third asks about the optimal division of responsibility for classification between human operators and Artificial Intelligence (AI). Three options for this division are considered: human supervised, shared responsibility, and fusion (this last giving humans the smallest role). Simulation results show that fusion consistently yields the lowest mean sum squared error in estimating mine counts by shape, particularly under conditions of uncertain underwater navigation. When underwater navigation is perfect, human-centered approaches improve. It is not that AI is inherently superior, but rather that it is more robust to navigation uncertainty underwater, giving it an edge in current operational contexts.