Adaptive Ocean Gradient Tracking Using an Autonomous Underwater Vehicle With a Boundless Model

This work presents a method for exploring a dynamic river plume boundary using an autonomous underwater vehicle with an on-board lightweight <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">boundless</i> model. The <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">boundless</i> approach is achieved by not constraining the path evaluations or the Gaussian random field to a predefined geographical area. In-situ decision-making enables targeted sampling of the ocean–river plume interaction. The data-driven and adaptive approaches provide the capability and opportunity to fully utilize the operational window for the vehicle. The method was developed using a simulated plume and vehicle, and results from simulation studies and successful field trials from the Douro River plume outside Porto, Portugal, are presented. The vehicle adapts its path based on the underway real-time assimilation of measurements, seeking to gain new information while not straying away from the front. Owing to the unpredictable shape and size of the river front, a model-based <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">boundless</i> method for adaptive sampling was constructed, generating potential waypoints as a function of the vehicle's position and the accumulated knowledge of the plume. Not bounding the spatial or geographical extent of the method allows for greater variation in plume shape and size. The river plume's extent is defined here as the area within the sharpest spatial salinity gradient, containing less saline water than the surrounding ocean. In the method, the depth of the sharpest vertical salinity gradient, or plume depth, is estimated using a 2-D <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Gaussian Process</i>, where the plume depth is estimated from a dive and ascent envelope of the robot traversing the ocean in an undulating fashion. Computational efficiency is gained from the resulting low number of inputs to the Gaussian process, compared to the number of salinity measurements, ensuring rapid on-board adaption. The next waypoint is chosen as the first waypoint in a path that maximizes the weighted sum of uncertainty, estimated plume depth, and the absolute value of the difference between the current plume depth and the estimated river plume depth along the path. This encourages traversal of the plume in a fashion that enables the extent of the plume to be resolved in high detail. The data-driven method was field verified in the Douro River, proving the ability to track the river plume to balance exploration and exploitation behavior to maximize the information value of the mission in real time onboard the vehicle.