A Control Method for Quadrotor Based on DDPG

This paper mainly discusses a method of bringing the copter to one target point rapidly and steadily at the same time. By building a physic-based offline simulation environment for a quadrotor, raw data from sensors are emulated. Meanwhile the states and motion locus can be estimated via a Kalman filter fusing the data from multi sensors. With the inputs of states and action of the copter, a control method based on Deep Deterministic Policy Gradient (DDPG) is introduced. In this paper, the feasibility of this approach is analyzed, and its availability is proved by the simulation experiment. The experimental results show this method enhances the robustness and rapidity in the control process, which will improve the performance of the whole copter system.