Control System Design of a Pitch-Decoupled VTOL UAV Using Reinforcement Learning

The paper presents the design, development, and simulation of an intelligent control system for the pitch-decoupled vertical takeoff and landing (VTOL) unmanned aerial vehicle (UAV). In contrast to conventional tilt-rotor VTOL UAVs, the mechanical structure of the pitch-decoupled VTOL UAV allows passive transition from vertical to horizontal flight modes, and vice versa, without the use of additional servo actuators. A detailed kinematic scheme and the dynamic equations of motion of the pitch-decoupled VTOL UAV equipped with a two-degree-of-freedom robotic arm are developed using Denavit–Hartenberg parameters and Euler–Lagrange equations. The aerodynamic stability of the UAV is investigated, and the aerodynamic coefficients used in the complete dynamics model are derived. On this basis, a multivariable control system is developed. To address the well-known transition-phase issues of VTOL UAVs, a neural controller is designed using the reinforcement learning actor-critic method. It is shown that the proposed neural controller enables a smooth and accurate transition phase and subsequent horizontal flight for the pitch-decoupled VTOL UAV.