cc-DRL: A Convex Combined Deep Reinforcement Learning Flight Control Design of a Morphing Quadrotor
针对变形四旋翼飞行器难以建立精确数学模型的问题，提出一种结合深度强化学习和凸组合技术的无模型飞行控制算法，用于轨迹跟踪和姿态稳定，仿真验证了有效性。
In comparison to common quadrotors, the structure deformation of morphing quadrotors endows them with better flight performance but also results in more complex flight dynamics. Generally, it is extremely difficult or impossible for these morphing quadrotors to develop an accurate mathematical model that describes their complex flight dynamics. This fact leads to a particularly challenging situation, as the existing mature model-based flight control theory fails to address the flight control design issue of morphing quadrotors. By resorting to a combination of model-free control techniques [e.g., deep reinforcement learning (DRL)] and convex combination (CC) technique, a convex-combined-DRL (cc-DRL) flight control algorithm is proposed for flight trajectory tracking and attitude stabilization of a class of morphing quadrotors with arm-length deformation. In the proposed cc-DRL flight control algorithm, a proximal policy optimization algorithm is utilized to offline train the corresponding optimal flight control laws for some selected representative arm length modes. Hereby, a cc-DRL flight control scheme is constructed by the CC technique. Finally, simulation results are presented to show the effectiveness and merit of the proposed DRL flight control algorithm.