| The dynamics of mechanical arm is highly nonlinear and strong coupling features,also exist in the process of motion modeling error and unmodeled dynamics,joint friction between,time-varying external disturbance,parameter measurement error and a series of influence factors,making it difficult to get accurate dynamics model,thus affecting the mechanical arm control effect.Therefore,it is necessary to compensate or approximate the nonlinear dynamics model to improve the control performance.In addition,many existing control methods do not have excellent learning ability and cannot learn adaptive control rules in the control process,so as to adjust the structure or parameters of the control system in real time to adapt to unpredictable uncertain factors through the information in the control process and the state of the controlled object.In this paper,an adaptive sliding mode robust control method based on deep reinforcement learning is proposed for manipulator with uncertain dynamics model.Various uncertainties of dynamics model are integrated into system uncertainties,and the influence of system uncertainties is eliminated by robust control.At the same time,according to the information in the control process and the state of the manipulator as the observation quantity of the deep reinforcement learning agent,the parameters of the sliding mode robust controller are adjusted in the continuous action space to make the system stable in the optimal or suboptimal working state.To ensure the efficient and stable learning of deep reinforcement learning agent,a reward function combining gaussian function and Euclidean distance was proposed.The main research contents of this paper are as follows:(1)Based on the theoretical knowledge of kinematics and dynamics,forward kinematics and inverse kinematics of 2-DOF manipulator are analyzed and deduced.Based on the Lagrange function theory,a dynamic model of a two-degree-of-freedom manipulator was established and its dynamic characteristics were introduced.(2)Based on the theoretical knowledge of sliding mode control and robust control,a sliding mode robust(SMR)controller was proposed,which used the sliding mode control to control the nominal model,and the robust control was used to eliminate the influence of uncertain factors such as modeling error and time-varying interference.The asymptotic stability of the closed-loop control system is proved by Lyapunov stability theory.The simulation results show that the SMR controller has good trajectory tracking performance and robustness for the manipulator with uncertain dynamics model.(3)Based on the basic theoretical knowledge of Deep reinforcement learning,an adaptive sliding mode robust control method(DDPGSMR)based on Deep Deterministic Policy Gradient algorithm(DDPG)was proposed.DDPG algorithm is used to adjust the parameters of SMR controller in real time according to the control process information and the running state of the manipulator,so that the system can keep the optimal or sub-optimal working state.The simulation results show that the performance of trajectory tracking control can be improved effectively and the adaptability and anti-interference of the control system can be improved by introducing DDPG algorithm to adjust and optimize the parameters of the controller based on SMR controller.In order to analyze the control performance of DDPGSMR more comprehensively,PD,adaptive control based on radial basis function neural network(RBFNN)approximation,SMR and DDPGSMR were compared for trajectory tracking control.The results show that DDPGSMR has better trajectory tracking performance,transient performance and steady-state performance.(4)On the basis of the simulation experiment,the experimental platform of two-DOF manipulator was built to carry out the trajectory tracking control experiment of DDPGSMR.Experimental results show that the proposed control method can effectively control the manipulator and has good adaptability. |
PDF Full Text Request