Enhancing UAV Stability: A Deep Reinforcement Learning Strategy

Abstract

Recently, as the utilization of unmanned aerial vehicles (UAVs) extends into increasingly varied domains, the control of UAV flight dynamics has become crucial. This paper presents an innovative approach for controlling UAV flight dynamics based on the deep deterministic policy gradient (DDPG) algorithm within the reinforcement learning (RL) paradigm. The proposed method is designed to rapidly adapt and maintain stable flight attitudes, outperforming conventional proportional integral derivative (PID) control methods. Utilizing a quadcopter UAV model equipped with four motors as the basis, this paper presents a strategy for UAV flight dynamics control. This strategy demonstrates flexibility and efficiency in a variety of scenarios and is scalable to accommodate UAVs of larger dimensions.