Shipboard helicopter landing in the near-deck region requires stable attitude regulation and high-precision deck-relative motion control under substantial model uncertainty and environmental disturbances, conditions under which conventional model-based controllers may lose performance or become overly conservative. This paper proposes a task-oriented, learning-enhanced control algorithm for ship-relative near-deck station keeping and landing by integrating a model-based baseline controller with residual reinforcement learning in a deck-relative closed-loop framework. The algorithmic contribution is the deck-relative baseline–residual control architecture: a split-channel incremental nonlinear dynamic inversion (INDI) outer loop and a reduced-order dynamic inversion (DI) inner loop provide the nominal baseline pathway, while a bounded residual Proximal Policy Optimization (PPO) policy supplies compensation in the same physical outer-loop command channels to suppress unmodeled nonlinearities and time-varying disturbances. Simulation results show that Residual PPO improves hover robustness and landing performance relative to the baseline controller and Pure PPO. With approximately 20–30% residual authority, it achieved 90.0% Desired landing rates in both tested descent-and-landing scenes.
Chang et al. (Sun,) studied this question.