Experiments on sampling/patrolling with two Autonomous Underwater Vehicles

• A patrolling/sampling strategy for cooperative AUVs is developed.
• Gaussian Processes and Voronoi tessellations are used as main tools.
• Realistic constraints, decentralization and robustness are main features.
• Experiments on a two underwater vehicles setup.

In this paper, an experimental investigation of the sampling/patrolling tasks by means of cooperative underwater vehicles is presented. Both in the case of sampling and patrolling, the mission consists in periodically visiting proper locations of the environment in order to estimate the field of interest or check for unexpected events. In previous work of the same authors, it has been shown that these tasks can be addressed by a proper combination of Gaussian Processes and Voronoi tessellation tools. The Gaussian Processes allow to put these tasks in a probabilistic framework, while the Voronoi tessellation allows to distribute the overall strategy as well as the handling of asynchronous communication or possible loss or adjunct of vehicles. The approach was already tested in simulation by varying all the parameters of interest as well as on a set-up composed by three marine surface vehicles. In this paper, experiments are extended to the still challenging scenario of Autonomous Underwater Vehicles (AUVs) communicating by acoustic modems and moving in the 3D space. The experiments were run in the La Spezia harbour, in Italy, as demo of the European project Co3AUV s.