A Gaussian Process Regression Approach to Cooperative Sampling by Underwater Gliders

Abstract

Cooperative sampling with multiple underwater gliders is becoming popular for oceanographic observations. Development of efficient cooperative approaches has attracted people’s attention because direct energy acquisition from the ocean by the underwater vehicles is still at an early stage of research and the energy carried on board is quite limited. In this paper, we use a Gaussian Process Regression (GPR) approach to cooperating a group of underwater gliders. Assuming the observations follow a multivariate Gaussian distribution, the underwater gliders are directed toward the most informative direction that is predicted using previous observations by all the gliders in the group. Estimation of a simulated tempereature field is presented using the GPR approach. Comparison with the conventional lawnmower approach shows that the GPR approach is superior in accuracy and efficiency.