Learning-Based Compound Docking Control for UAV Aerial Recovery: Methodology and Implementation

This article focuses on unmanned aerial vehicle (UAV) docking control, which is one of the most difficult and significant issues during UAV aerial recovery. The UAV probe, which is affected by the barycenter translational and attitude rotational motion, needs to be controlled precisely to track and dock with a dynamic drogue towed by a flexible cable under multiple wind perturbations. Thus, a learning-based docking control method compounded of a barycenter translation antidisturbance controller, which has good antidisturbance ability and can ensure the control stability, and an attitude rotation learning controller, which can utilize the effect of rotational motion on the probe motion to further improve the docking precision and is trained by a deep reinforcement learning algorithm using the proposed experience-based value expansion mechanism to improve the training efficiency, is proposed. Then, a distributed mechatronics docking experimental system scheme with the characteristics of accurate simulation, realistic system, and intuitive display is proposed to comprehensively verify the proposed docking control method. The results show that the proposed method can reduce the tracking error by 20% and improve the docking success rate by 10%.