Control Design And Research Of Flapping Wing Flight Robot Based On Reinforcement Learning

In recent years,with the rapid development of microelectromechanical technology,robot technology,new material technology and newly energy technology,the research on flapping wing flight robots have gradually entered the fast lane,and has received the attentions from many countries.Compared to the traditional mode of flights,flapping wing flights have a greater advantage in terms of energy efficiency,flexibility and camouflage.Therefore,the development of a new type of flapping wing flight robot would be an important direction in future aircraft researches.At the same time,the study of smarter and more efficient control methods is urgently needed.This paper mainly studies the application of reinforcement learning in attitude control and position control of flapping wing flight robot.First of all,this paper will start from the system modeling of flapping wing flight robot,consider a six-DOF imitation Hummingbird model as the research object,and obtain the dynamic equation of the system by establishing a suitable coordinate system and kinematics analysis.Then based on the system dynamics equation of the imitation Hummingbird flapping wing flight robot,a typical framework of reinforcement learning is introduced: The Actor-Critic algorithm is designed to control the strategy,and the stability of the system is proved according to the Lyapunov stability theory.In view of the good characteristics of neural networks in dealing with uncertainty and nonlinearity,neural networks are used to build Actor networks and Critic networks,and reinforcement learning controllers are designed based on Actor networks.The Actor neural network is used to realize the approximation of the continuous strategy space,while the Critic neural network is used to realize the approximation of the value function.In addition,the PD controller is designed to be compared with the reinforcement learning controller,and the feasibility and control effect of the controller are verified by MATLAB simulation platform.In addition,the attitude motion and the position movement of flapping wing flight robot are studied respectively in this paper.For attitude motion,the system model is unknown,so by introducing neural network to deal with the uncertainty of the system,taking into account all known and partially known cases of the system state variables,the full state feedback neural network controller and the output feedback neural network controller are designed respectively to realize the control of attitude motion.For position motion,the system model is known,so the model-based controller is designed to control the position motion.The stability of the system and the feasibility and control effect of the controller are verified by Lyapunov theory and MATLAB simulation platform respectively.