Deepak Subramani was awarded the best poster award for the poster titled A Stochastic Optimization Method for Energy-based Path Planning at the Computations for Design and Optimization / Center for Computational Engineering Symposium 2015. Congratulations Deepak!

Real-time cooperation between autonomous vehicles can enable time-critical missions such as tracking and pursuit of a dynamic target or environmental feature, but relies on wireless communications. Underwater, communication over distances beyond about one hundred meters is almost exclusively accomplished through acoustics, which bring challenges such as propagation delays, low data rates, packet loss, and scheduling constraints due to interference and limited bandwidth. These limitations make underwater pursuit missions preeminent applications of networked control. Motivated by such applications, this talk presents contributions towards multi-vehicle feedback control in the presence of severe communication constraints. The first major area of work considers the formulation and solution of new underwater multi-vehicle tracking and pursuit problems using closed-loop control. Specifically, we will discuss field experiments in range-based target pursuit at high tracking bandwidths, as well as a novel networked control methodology for pursuit of dynamic ocean features, such as fronts, using multiple vehicles and ocean model forecasts. The second area of work presents a unified formalism for multi-vehicle control and estimation with measurement, control, and acknowledgment packets all subject to scheduling, delays and packet loss, which we demonstrate in simulation and field experiments.

We derive and apply a methodology for the initialization of velocity and transport fields in complex multiply-connected regions with multiscale dynamics. The result is initial fields that are consistent with observations, complex geometry and dynamics, and that can simulate the evolution of ocean processes without large spurious initial transients. A class of constrained weighted least squares optimizations is defined to best fit first-guess velocities while satisfying the complex bathymetry, coastline and divergence strong constraints. A weak constraint towards the minimum inter-island transports that are in accord with the first-guess velocities provides important velocity corrections in complex archipelagos. In the optimization weights, the minimum distance and vertical area between pairs of coasts are computed using a Fast Marching Method. Additional information on velocity and transports are included as strong or weak constraints. We apply our methodology around the Hawaiian islands of Kauai/Niihau, in the Taiwan/Kuroshio region and in the Philippines Archipelago. Comparisons with other common initialization strategies, among hindcasts from these initial conditions (ICs), and with independent in situ observations show that our optimization corrects transports, satisfies boundary conditions and redirects currents. Differences between the hindcasts from these different ICs are found to grow for at least 2-3 weeks. When compared to independent in situ observations, simulations from our optimized ICs are shown to have the smallest errors.