Research On Modeling And Position Control Of Soft Manipulator Based On Neural Network

Traditional rigid robots are widely used in the industrial field due to their high rigidity,high strength,high precision and high speed.The soft manipulator has sufficient flexibility,adaptability,super redundancy or infinite degrees of freedom,and can even change its shape at will to adapt to the environment and goals.It can complete many tasks that traditional robots cannot complete.Compared with traditional rigid manipulators,it can meet Working in complex environments such as small,chaotic,crowded,etc.,the soft manipulator has more degrees of freedom and the nonlinear characteristics of the soft material itself,which makes the modeling and control of the soft manipulator more complicated.This paper studies the neural network modeling and position control methods of the soft manipulator.The main research contents are as follows:First of all,this article takes the three-cavity soft manipulator as the research object,builds the soft manipulator test experimental platform,uses a three-dimensional digitizer to collect the steady-state position information of the end points corresponding to the different driving pressures of the soft manipulator;then uses BP neural network,LSTM neural network,BP and LSTM combined neural network to establish the model of the soft manipulator,study the correspondence between the steady-state position of the end point of the soft manipulator and the driving pressure,and compare the error value of the predicted air pressure at the target position and the real air pressure,It is found that the prediction effect of the soft manipulator model based on the combination of BP neural network and LSTM neural network is the best;then based on the target position and random target position of the test set,this model is experimentally verified,and the model predicts the corresponding driving pressure value of the target position load the software manipulator again,and calculate the error between the position reached by its end point and the target position is 6.95mm;on this basis,combined with the PI controller,adjust the driving air pressure value predicted by the model,and the position error of the manipulator have been greatly improved.