It is challenging to model and control a tail-sitter unmanned aerial vehicle (UAV) because its blended wing body generates complicated nonlinear aerodynamic effects, such as wing lift, fuselage drag, and propeller-wing interactions. We therefore devised a hybrid aerodynamic modeling method and model predictive control (MPC) design for a quadrotor tail-sitter UAV. The hybrid model consists of the Newton-Euler equation, which describes quadrotor dynamics, and a feedforward neural network, which learns residual aerodynamic effects. This hybrid model exhibits high predictive accuracy at a low computational cost and was used to implement hybrid MPC, which optimizes the throttle, pitch angle, and roll angle for position tracking. The controller performance was validated in real-world experiments, which obtained a 57% tracking error reduction compared with conventional nonlinear MPC. External wind disturbance was also introduced and the experimental results confirmed the robustness of the controller to these conditions.

翻译：尾座式无人机的建模与控制具有挑战性，因其混合翼机身会产生复杂的非线性空气动力学效应，例如机翼升力、机身阻力和螺旋桨-机翼相互作用。为此，我们针对四旋翼尾座式无人机，提出了一种混合空气动力学建模方法与模型预测控制设计方案。该混合模型由描述四旋翼动力学的牛顿-欧拉方程和用于学习残余空气动力学效应的前馈神经网络组成。该混合模型以较低的计算成本实现了高预测精度，并用于实现混合模型预测控制，通过优化油门、俯仰角和滚转角实现位置跟踪。在实际实验中验证了控制器的性能，与传统非线性模型预测控制相比，跟踪误差降低了57%。同时引入外部风扰动，实验结果确认了控制器对该条件的鲁棒性。