Adaptive Motion Control Of Manipulator In Uncertain Environment

In the process of rapid development of nuclear industry,the maintenance and emergency disposal of nuclear facilities become more and more important.Under the uncertain environment of complex and unpredictable nuclear facility operation and emergency disposal,the contact task,like valve turning and door opening,put forward higher requirements for the manipulator motion control.It is of great significance to carry out the research on the autonomous task disposal of the manipulator in the uncertain environment to ensure the safety of nuclear power production.Due to the complexity of nuclear facility operation and emergency disposal,there may be some adverse factors such as sensor failure,lack of light,mechanical arm inconvenient to install force sensor,etc.,it is difficult to meet the operation requirements only relying on a single sensor.Therefore,in this paper,based on vision and force sense respectively,two kinds of adaptive control methods of manipulator are proposed and studied.A vision control method of manipulator with uncertain environment parameters is proposed.In order to solve the problem of uncertain environment parameters,like the position and rotation of the contact task's operating object are unknown,the adaptive manipulator control method based on the deep reinforcement learning with the visual feedback is proposed.This method uses the control strategy of tracking the label on the operating object indirectly to track the operating object trajectory.The eye in hand vision system is built to obtain the position and rotation of the label relative to the manipulator end-gripper.And then the deep deterministic policy gradient algorithm(DDPG)is used to track the position and rotation of the operating object in real time.Finally,the simulation experiments are completed to verify that the effectiveness of the proposed visual control algorithm.A force control method of manipulator with uncertain environment parameters is proposed.In order to solve the problem of uncertain environment parameters,like the stiffness and position of the contact task's operating object are unknown,the impedance control method based on position is studied and analyzed.In view of the disadvantages of difficult to obtain the position and stiffness of environment and poor tracking performance of impedance control force in complex environment,the impedance control equation is improved,and combined with DDPG algorithm,an adaptive impedance control algorithm based on deep reinforcement learning is proposed.In addition,based on deep reinforcement learning theory and impedance control theory respectively,two comparative algorithms are proposed: the adaptive force control algorithm based on deep reinforcement learning and the adaptive impedance control algorithm.Finally,the simulation experiments are completed to verify that the effectiveness of adaptive impedance control algorithm based on deep reinforcement learning.Based on the typical contact task of valve turning,a V-rep simulation platform and a real scene experiment platform are built,which are composed of UR5 manipulator,gripper,RGBD camera and six-dimensional force sensor.Based on the adaptive motion control methods of the manipulator with different feedback information,the simulation and experiment of turning various sizes of valve wheel are completed.