Today, the number of autonomous platforms used in semi-coordinated operations is often larger than 10 and this number is rapidly increasing. It is imperative to advance the interdisciplinary science of autonomy to take full advantage of these new capabilities and so maintain our leading scientific and engineering edge. We are researching novel fundamental formalisms and principled methodologies for optimal sensing using heterogeneous and collaborative swarms of autonomous platforms that are smart. Intelligence here refers to the ability to compute and autonomously adapt an optimal sensing plan based on the predicted environment and acoustic performance and their uncertainties, and on the predicted effects of environmental and acoustic sensing on future operations. Our approach is generic and applicable to any Naval swarms that move and sense large-dimension dynamics fields, but our focus is underwater sensing with a range of platforms including AUVs, gliders, ships, moorings and remote sensing.

When compared to other control problems of large dimensions, the differences with our problem are that: (i) naval platforms are heterogeneous and their data are gappy but multivariate; (ii) marine fields are dynamic on multiple-scales and have very large dimensions, but are predictable to some degree; and (iii), the measurements to be collected will affect these future predictions. Therefore, there are feedbacks between optimal sensing and predicting, in time and space, and across variables. We use guidance from ocean-acoustic modeling, dynamical system theory, uncertainty prediction, decision-making under uncertainty, artificial intelligence, bio-inspired algorithms with emergence of global properties, and distributed computing. In all cases, we are interested in the global swarms and high-level optimization, not the detailed control of a single robot. A global objective function defines the optimal dynamic and collaborative autonomy. In our case, objective functions depend on the predicted environment, on the predicted values and positions of the expected data, and on the feedbacks between data and dynamics.

We are researching autonomous sensing swarms and formations that exploit the multi-scale, multivariate, four-dimensional environmental-acoustic dynamics and predictabilities. We are developing new global swarm patterns and high-level optimization schemes based on control theory and dynamical system theory (e.g. artificial potential functions, nonlinear contraction analysis), artificial intelligence (e.g. hybrid evolutionary optimization, particle swarm optimization and reinforcement learning) and bio-inspired behaviors (e.g. distributed flocking/swarming, ant algorithms). From the methods explored, a small subset of candidate schemes are being selected based on accuracy, robustness, predictability and generality. These selected schemes will be augmented to learn from (i) environmental forecasts and their associated uncertainties and (ii) the projected impact of the sensing on the forecasts. Incremental testing of the algorithms will be accomplished using idealized simulations. System tests use hindcasts from our extensive set of at-sea exercises.