| This study focuses on two areas needed to deploy unmanned underwater vehicles (UUVs) in an estuary: Planning a path through the currents, and determining position along the way.;For the path planning goal, since small vehicles have a small energy budget, and current velocities can be high, it is critical to be able to plan paths that use environmental currents to advantage rather than attempting to fight them. Given sufficient time and a correct choice of path, a UUV should be able to ride currents to move in almost any direction, reducing energy requirements and increasing endurance. Because there is limited available onboard CPU, the algorithm should ideally be very lightweight to allow for real-time replanning.;In order to search the space of possible paths for the best one, a number of different methods were considered. Gradient methods were found to be ineffective because the solution space is not quadratic. The method used is to define a single cost function, use a shooting algorithm to find an initial solution that is dominated by the current, and then a heuristic search attempts to bend the path toward the minimum-cost solution.;In addition, in the physical testing in the Hudson River, it is important to get an accurate assessment of the actual location of the UUV. In the process of trying to estimate position while underwater, the acoustic beacon system was tested and found to have serious errors in a noisy environment. Instead, simulated beacon distances were generated, and errors were measured using particle filters, the current best algorithm for estimating variables given non-Gaussian distributions. This was then compared to a new algorithm, Multiple Analytical Distribution Filter (MADF). The two algorithms are found to be largely equivalent, with MADF more computationally efficient in certain contexts. |
