Centralized Position Optimization of Multiple Agents in Spatiotemporally-Varying Environment: A Case Study with Relocatable Energy-Harvesting Autonomous Underwater Vehicles in the Gulf Stream

Abstract

This paper evaluates a strategy for using multiple energy-harvesting autonomous underwater vehicles (AUVs) to extract hydrokinetic energy out of a spatiotemporally-varying Gulf Stream (GS) resource. When anchored, the conceptual AUV can generate energy with on-board turbines, and can relocate itself when needed. Model predictive control (MPC)-based centralized optimization is used to optimize the locations of a team of AUVs along a cross-stream transect. To maintain an accurate estimate of the transect flow profile, a Kalman filter-based estimator is used to blend flow speed measurements from the AUVs with intermittent auxiliary flow speed data from a High Frequency Radar Network (HFRNet). To characterize the uncertainty of the GS flow profile estimation, variance of flow speed is modeled using tools from Gaussian Process (GP). The MPC strategy proposed in the paper ensures an appropriate balance between learning the exact location of GS jet (termed `exploration') and maximizing overall power generation (termed `exploitation'). The importance of estimating the flow pattern in the transect with high resolution is shown through simulation results.