Reinforcement Learning-Based Turning Control of Asymmetric Swept-Wing Drone Soaring in an Updraft

Soaring drones can use updrafts to reduce flight energy consumption like soaring birds. With control surfaces that are similar to those of soaring birds, the soaring drone achieves roll control through asymmetric sweepback of the wing on one side. This will result in asymmetry of the drone. The moment of inertia and the inertial product will change with the sweepback of the wing, causing nonlinearity and coupling in its dynamics, which is difficult to solve through traditional research methods. In addition, unlike general control objectives, the objective of this study was to enable the soaring drone to follow the soaring strategy. The soaring strategy determines the horizontal direction of the drone based on the vertical wind situation without the need for active control of the vertical movement of the drone. In essence, it is a horizontal trajectory tracking task. Therefore, based on the layout and aerodynamic data of the soaring drone, reinforcement learning was adopted in this study to construct a six-degree-of-freedom dynamic model and a control flight training simulation environment for the soaring drone with asymmetric deformation control surfaces. We compared the impact of key factors such as different state spaces and reward functions on the training results. The turning control agent was obtained, and trajectory-tracking simulations were conducted.