Disturbance-Compensated Trajectory Tracking of Underactuated UUVs Using Surrogate-Assisted Tuned SMO-NMPC

Trajectory tracking is a fundamental capability for underactuated unmanned underwater vehicles (UUVs), but it remains challenging due to nonlinear coupled dynamics, unmeasured ocean disturbances, and actuator constraints. Sliding-mode observer (SMO)-compensated nonlinear model predictive control (NMPC) can enhance robustness by estimating and compensating for lumped disturbances in the prediction model; however, its performance is highly sensitive to strongly coupled SMO gains and NMPC cost weights. Joint tuning of these parameters is a simulation-driven black-box problem, because each candidate setting requires expensive nonlinear closed-loop simulations with repeated constrained NMPC solutions. To address this issue, this paper proposes a sample-efficient surrogate-assisted tuning framework for disturbance-compensated SMO-NMPC in UUV trajectory tracking. In the proposed framework, differential evolution (DE) generates candidate parameter vectors, while a Gaussian process surrogate and a distance-aware infill criterion determine which candidates receive expensive closed-loop evaluations. The two components therefore have distinct roles: DE defines the candidate search distribution, whereas the surrogate-assisted infill mechanism allocates the limited true-evaluation budget by considering both predicted fitness and parameter-space novelty. Comparative simulations show that, under the same budget for expensive evaluations, the proposed method achieves faster offline tuning and lower final fitness than GA, PSO, standard DE, and SAEA-Random. The resulting tuned controller improves tracking accuracy while satisfying input constraints, and it remains robust under intensified, abrupt, and stochastic disturbances and transfers to four unseen reference trajectories without retuning.