Robust control of USV-UAV heterogeneous systems under complex sea states: A hierarchical optimization approach

To address spatiotemporal response mismatch and cross layer parameter coupling in USV-UAV cooperative control under complex sea states, this paper proposes a three layer architecture optimized by stage wise sequential hierarchical Bayesian optimization (HBO). The bottom layer combines composite gain sliding mode control (CG-SMC) with an extended state observer (ESO) for disturbance rejection; the middle layer applies predictive line of sight (PLOS) guidance to configure cooperative search topology; and the top layer integrates COLREGs based nonlinear model predictive control (NMPC) with dynamic command fusion for smooth collision avoidance. Best feasible parameters are fixed and inherited sequentially from the bottom to the top layer, coordinating control, guidance, and decision making. Simulations using a VIPER USV and quadrotor UAV compare HBO with hierarchical differential evolution and manual tuning. HBO suppresses torque chattering, limits the surge velocity coefficient of variation to 2.44%, and increases average UAV cruise altitude and effective SAR area by up to 23.85% and 41.50%, respectively. Monte Carlo and cross platform tests further demonstrate robustness, predictability, and generalization under stochastic disturbances and platform variations. The framework supports coupled black box optimization of heterogeneous maritime SAR systems.