Active perception for plume source localisation with underwater gliders

摘要

© 2018 Australasian Robotics and Automation Association. All rights reserved. We consider the problem of localising an unknown underwater plume source in an energy-optimal manner. We first develop a specialised Gaussian process (GP) regression technique for estimating the source location given concentration measurements and an ambient flow field. Then, we use the GP upper confidence bound (GP-UCB) for active perception to choose sampling locations that both improve the estimate of the source and lead the glider to the correct source location. A trim-based FMT∗planner is then used to find the sequence of controls that minimise the energy consumption. We provide a theoretical guarantee on the performance of the algorithm, and demonstrate the algorithm using both artificial and experimental datasets.