Identification Of Dynamic Parameters Of Manipulator Based On Bidirectional Long-Short-Time Neural Network

This paper mainly studies the problem of dynamic parameter identification of the manipulator.Aiming at the problem that it is difficult to accurately model nonlinear characteristics such as joint friction and flexibility of the manipulator,a dynamic parameter identification method of the manipulator based on Bidirectional Long Short-Term Memory(BiLSTM)network is proposed.On the basis of dynamic modeling,the estimation algorithm of the external contact force of the manipulator and the collision detection algorithm of the manipulator based on the BiLSTM dynamic model are designed,and the modeling accuracy of the dynamic model is verified experimentally on the UR5 manipulator.The main work content of this paper includes:(1)A model learning method based on BiLSTM networks for manipulators is proposed.The solution of the dynamic model of the manipulator is regarded as a regression problem,and the expected future position and velocity of the joints of the manipulator are used as the input of backpropagation,the actual position and velocity are used as the input of forward propagation,and the joint torque is used as the output for model training.In addition,in order to improve the accuracy of the model,the network structure is designed individually for each joint of the manipulator.Experimental results show that the proposed method has smaller joint torque prediction errors than Back Propagation(BP),Long ShortTerm Memory(LSTM)and Gate Recurrent Unit(GRU)networks.(2)A method for estimating the external contact force of the manipulator based on the BiLSTM dynamic model and BP network is designed.Aiming at the difficulty of modeling joint current and end contact force caused by joint flexibility and friction of the manipulator,a modeling method of joint current and end contact force based on BP network is proposed,and the joint current predicted by the dynamic model is combined with The actual joint current difference is used as the input of the network,and the contact force at the end of the manipulator is used as the output for network training.(3)A collision detection algorithm of the manipulator based on the BiLSTM dynamic model is designed.Using the dynamic model obtained by BiLSTM network training,the collision detection experiment was carried out on the UR5 manipulator,and different parts of the human body were collided with the end of the manipulator.The experimental results all conform to the ISO/TS15066 standard.Then use the UR5 self-collision detection algorithm to compare with the algorithm designed in this paper.The results show that the collision response time of this method is shorter than the UR5 self-collision detection algorithm,and the collision impact force is smaller.