Framework for fair comparisons of underwater vehicle controllers - showcasing the robustness properties of a model-free sliding mode controller tuned with a random-forest-based Bayesian optimization approach

Tasks with underwater vehicles present several challenges that include complex and hazardous environments, and unmodeled and/or unknown uncertainties. The setup of positioning controllers is therefore a difficult and laborious task and very often leads to suboptimal performance results on the field. This paper shows a framework for methodical evaluation and setup of dynamic positioning controllers for underwater vehicles with simulation-based optimization method using performance metrics. The proposed method can be configured to be mission specific and delivers a controller configuration that also allows a fair numerical comparison between control algorithms on similar scenarios.