Vertical Recovery Guidance of Unmanned Aerial Vehicles Based on Improved SAC

Addressing the issue of precise guidance of vertical recovery for a three-degree-of-freedom (3 DOF) unmanned aerial vehicle (U A V) in three-dimensional space, a novel improved SAC (Soft Actor-Critic) intelligent controller is designed to achieve vertical recovery guidance for the UAV. The specific improvements are as follows: considering that the controller will be embedded into the U A V in the future, the architecture of the actor and critic networks is specially designed. At the same time, jerk (the time rate of change of acceleration) is incorporated into the action variables and acceleration is incorporated into the state variables to make the controller networks aware of the variation tendency of acceleration and thus achieve the effect of smooth control of apparent acceleration. Additionally, a reward function is designed to ensure that the U A V hovers near the target point (zero point), thereby preventing any exceptional situations during targeting. Simulation results demonstrate that the proposed intelligent controller enables precise landing and hovering control guidance for the U A V in three-dimensional space, exhibiting certain robustness and adaptability to deviations such as mass change, wind interference and suddenly drop of thrust.