Research On Adaptive Compliance Control Method Of Manipulator Based On Dynamic Parameter Identification

Contact human-robot interaction has a wide range of applications in industry,military,medical care,etc.Robots interact with the human body through force sense.In order to ensure the safety of the human body while achieving natural,smooth,compliance movement,it requires that the force-sensing system of robots extended from the end to any position.However,the current ‘skin-like' sensor is not mature,and the structure of high integration joints is complex,which dramatically limit the flexibility of human-robot interaction.Therefore,a method of identifying dynamic information and corresponding estimating whole body contact force based on motor driving current and prior dynamics knowledge of robot is proposed.After that,the impedance control method based on position control is used to realize the compliance control of the whole body without additional force or a ‘skin-like'sensor.Firstly,this paper takes the 3-DOF manipulator as the primary research object,and analyzes its mechanism,drive,and control system.And the standard mathematical model of robot is established by using the method of DH.Thus,the relationship between the end pose of the manipulator in space and the joint variables is obtained.The working space is obtained by an inverse solution,which provides a theoretical basis for motion control.Secondly,by analyzing the training process of neural networks,this paper establishes a neural network based on the prior knowledge of manipulator dynamics,the training goal is the relationship between joint variables and joint torques in the operation process.Therefore,the real-time estimation of external force mapping in joint space is realized by pre-trained off-line.In the process of training,a method of identifying the parameters from the end joint to the initial joint is adopted,the advantage of this method is that it overcame the dependence of identification parameters on initial value caused by saddle point in the traditional method.Experiments show that this method can not only reduce the number of samples needed for training,but also effectively generalize the neural network to the unobserved space,reduced the number of saddle points in the optimization process,and realize the accurate identification of ideal dynamic parameters by neural network.Thirdly,after obtaining the precise dynamic model and the force mapping in the joint space,the impedance control based on position control is used to control the compliance of the manipulator.For the outer impedance loop,an improved impedance model with an S-like growth curve function as the weight is proposed to improve the sensitivity of the contact force of the manipulator.In the inner loop,an adaptive sliding mode controller based on RBF neural network is used to achieve precise track following.Thus,the security and efficiency of the human-robot interaction system are guaranteed.Finally,the control system of the 3-DOF manipulator is built by Lab VIEW.The experiment of human-robot interaction is carried out,and the friendly handshake between humans and the manipulator is achieved.The result shows that the force sensing method based on motor driving current and manipulator dynamics proposed in this paper is effective,and the contact force information of the whole manipulator can be sensor precise.On this foundation,the human-robot interaction is realized by the impedance control of the joints.