Intelligent Tracking Method of Rotary-Wing UAV Based on Deep Reinforcement Learning
Introduce the dynamic model of quadcopter drones and the basic principles of traditional visual servoing to address the issues of poor visibility constraints and low control efficiency in visual servoing control of rotor drones.A Markov model for visual servo control of rotary-wing UAV based on deep reinforcement learning is proposed to address the issue of visual servo gain adjustment,and combined with the deep deterministic policy gradient(DDPG)algorithm,the model was trained,and finally two decoupled servo gains,namely linear velocity and angular velocity,were obtained.The results show that compared with traditional image-based visual servoing(IBVS)methods,this method can ensure that the target is always in the field of view,the flight trajectory is smooth,and the time required to reach the target position is short,meeting the visibility constraints of visual servoing control for rotary-wing UAV.At the same time,the control efficiency is 23.5%higher than traditional control methods.In the experiment,this method can achieve precise control in complex nonlinear environments,increasing the application scenarios of visual servo technology.
deep reinforcement learningrotary-wing UAVvisual servofield of view constraint
NETL
NSTL
万方数据
