Randomized MPC for view planning in AUV seabed inspections

Nowadays, AUVs are state-of-the-art technologies for seabed inspection of an area of interest to generate optical/acoustic mosaics, gather representative bathymetric data, or search for objects of interest. As a matter of fact, typically, inspection surveys are designed by skilled operators using a lawnmower or zig-zag patterns. Nevertheless, the outcomes are not predictable since the performance of exteroceptive sensors depends on environmental conditions, seafloor, and targets composition and shape. Thus, in the last years, sensor-driven planning strategies that actively consider the sensor acquisitions as feedback, avoiding multiple attempts, were proposed. This paper presents a sensor-driven planning solution based on a Randomized Model Predictive Control approach. The developed system simulates multiple AUV possible trajectories and selects the one that better steers the AUV toward non-explored regions. Moreover, a novel method based on the Kernel Density Estimation technique for fast evaluating the AUV trajectories is proposed. Realistic simulations are presented to validate the hereby proposed strategy.