Aerodynamic Error Compensation of Quadrotor with All-Moving Wings (QAW) via Deep Reinforcement Learning Geometric Control

Incorporating all-moving wings into a quadrotor significantly enhances flight efficiency, but tracking errors induced by inaccuracies in aerodynamic parameter modeling during high-speed flight remain non-negligible. To address this, we propose a geometric control framework enhanced by deep reinforcement learning to balance energy efficiency and control precision. Specifically, a Twin Delayed Deep Deterministic Policy Gradient (TD3)-based compensator fine-tunes rotor outputs to compensate for the aerodynamic force error caused by inaccurate aerodynamic parameter modeling. This approach effectively reduces tracking errors of the all-moving-wing quadrotor (QAW) while incurring only an approximately 6.4% increase in energy consumption. The proposed controller is validated in a high-fidelity simulation environment developed using Unity3D.