Compensation Control of UAV Based on Deep Deterministic Policy Gradient

In this paper, a compensatory control strategy for quadrotor UAV based on deep deterministic policy gradient is proposed. Which solves the problems that the traditional quadrotor UAV control relies heavily on accurate system models, and cannot adapt to unknown external disturbances, and the training efficiency of reinforcement learning autonomous control algorithm based on the trial-and-error mechanism is low. Among them, the basic controller is used to realize the stability control of UAV, which reduces the search space of reinforcement learning algorithm and speeds up the convergence speed. In addition, the external sparse reward is designed to enhance the training efficiency of learning algorithm. Furthermore, the controller based on DDPG is used to compensate unknown external disturbances, adapt to the UAV's large range of parameter changes, and optimize the control accuracy. The effectiveness of the algorithm is verified by simulation experiments.