Gradient-based Adaptive Navigation of Multiple Diving Autonomous Surface Vehicles

Abstract

A gradient-based method of extrema seeking is implemented for a set of water column profiling vehicles. Measurements used to guide the navigation are sparse because these vehicles must transit along the surface and can only dive intermittently to collect a vertical profile. The vehicles do not have enough energy to repeatedly take samples at small intervals, but too large of steps may cause them to miss a feature of interest altogether. Multiple vehicles are commanded using cluster navigation concepts, abstracting the individual robots into a cohesive unit. Profiling locations are dependent on evaluation of gradients calculated from previous vehicle measurements. Vehicle spacing defined at the cluster-level serves to affect the gradients calculations. Too small of a vehicle spacing increases sensitivity to noise, and too large of a spacing between profiles may result in the cluster failing to detect a feature of interest. Simulations have been run to explore the relationship between cluster parameters and feature parameters to begin preparation of this technique for field applications using an inexpensive profiling vehicle developed for this research.