Underwater robots are used for inspecting offshore structures, surveying reefs, and accompanying divers, but most of these tasks are still flown by hand by a pilot. This manual approach raises questions about efficiency, skill requirements, and the potential for automation.

The paper introduces AQUAACT, a dataset and benchmark designed to support research into closed-loop underwater teleoperation. The dataset captures joystick-level control inputs, and the benchmark provides a standardized way to evaluate autonomous or assisted control methods against human performance.

By offering a common testbed, the work could help reduce dependence on continuous human piloting, though the abstract does not detail the dataset's size or collection methods. The significance lies in enabling more reproducible comparisons for underwater autonomy research.