Aerial manipulation has a wide variety of applications, such as inspection, maintenance and human-robot cooperation. The addition of a manipulator allows an aerial platform to not only perceive but also actively manipulate its environment through physical interactions.
The Maestro project, initiated by Saxion SMART, focuses on the development of a modular and robust aerial manipulation system for the inspection and maintenance of offshore wind turbines. This research, conducted as a collaboration between RaM and SMART, contributes to the project by investigating learning-based hybrid force–position control for an underactuated aerial manipulator, enabling stable push-and-slide interactions along flat surfaces under offshore wind disturbances and varying contact friction conditions. Deep Reinforcement Learning (DRL) methods are employed, as they have demonstrated promising results in handling complex control tasks, largely due to their adaptability and robustness to environmental uncertainties, scenarios in which classical model-based approaches often fail.
The proposed approach involves learning a hybrid force–position controller that intrinsically incorporates low-level control by directly commanding the individual rotor thrusts. Additionally, safety features are integrated into the controller design to provide formal safety guarantees in operating regimes where such guarantees would otherwise be absent. The performance of the proposed approach is evaluated through both simulation and experimental testing.