Multi-agent acoustic noise field characterization using optimal control

Abstract

Investigations of acoustic-driven control schemes have stayed away from traditional Optimal Control methods because of the computational load and lack of direct solution methods. The method presented here overcomes these challenges and learns a spatially varying noise field by applying the Pontryagin Maximum Principle (PMP) to solve for optimal trajectories over a changing estimate of the noise field. The resulting control drives the agents to the regions of the estimated field with the most uncertainty. This optimization is conducted with multiple agents interested in the noise field for a single frequency as well as for multiple agents simultaneously optimizing over multiple frequencies of interest. The challenges of finding a solution to the mixed boundary value problems is achieved through an eigenvalue decomposition method. The selected cost function is designed to minimize time and thus bound error introduced by the open loop control derived through the PMP methodology. Simulations and preliminary experimental results are presented to demonstrate the effectiveness of this method. The resulting information from utilizing this method are useful in optimized sensor location problems, which are especially relevant for the large dynamic array system used to do the initial sensing. [Research supported by NAVSEA and Raytheon Company.]