Anti-Disturbance Compensation for Quadrotor Close Crossing Flight Based on Deep Reinforcement Learning

The aim of this article is the design of a feedforward compensator based on deep reinforcement learning (DRL) for cooperative quadrotors in close crossing flight. Quadrotors are described by state-space models that include shearing airflow disturbance from other quadrotors. This disturbance is compensated in a feedforward way using DRL. Both value based compensator and policy based compensator algorithms are proposed for training purposes. Then, Lyapunov stability criteria are used to prove that the reference trajectory can be tracked boundedly even during the training process of the proposed algorithms, and that a smaller bound of tracking error can be achieved when the compensator converges. An indoor experimental system for online training has been developed for validation purposes. Both simulation and experimental results are provided to demonstrate the effectiveness and advantages of the proposed compensator.