Surrogate Modeling for ROV Trajectory Planning in Realistic Marine Currents: A Methodology for Data Assisted Underwater Navigation

Accurately predicting and controlling Remotely Operated Vehicles (ROVs) trajectories in complex marine environments is essential for inspection, maintenance, and environmental monitoring. This paper proposes an integrated framework combining experimental measurements, Computational Fluid Dynamics (CFD), and Deep Learning (DL) techniques, specifically employing a U-net architecture as a surrogate model for velocity field and turbulence tensor prediction in agitated tank environments. Experimental data are collected from the CIRTESU facility at Universitat Jaume I, featuring a 12x8x5 m water tank with adjustable currents generated by a controlled impeller system. High-resolution velocity data acquired via acoustic Doppler velocimetry (ADV) validate CFD simulations, significantly enhancing the training dataset. The developed U-net surrogate model effectively predicts comprehensive velocity fields from sparse velocity measurements at strategically selected sensor locations, identified through Principal Component Analysis (PCA). Integrated into a simulated ROV control environment, the model facilitates real-time trajectory adjustments and optimal path planning, reducing energy consumption and improving mission efficiency. This comprehensive approach provides a robust, scalable solution for precise underwater navigation in complex flow conditions.