Research On Control Of Teleoperation Robot System Based On Artificial Neural Network

As a method of extending human perception and behavior,teleoperation can help humans to work in unreachable or dangerous environment.Teleoperation technology has broad application prospects in aerospace,medical,nuclear,security and other fields.The teleoperation system always have master side and salve side,and the master-slave side in a teleoperation system was connected by the communication network.When the operator controls the master end,the control signal is transmitted to the slave through the communication channel,thereby manipulating the slave to complete the specified action.However,in the practical application of teleoperation,there are still many problems to be solved,such as system stability and transparency under random delay and force feedback.Stability and transparency,as the most important performance indicators of teleoperation systems,are always mutually influential and contradictory.Therefore,this paper studies the stability and transparency of teleoperation system based on artificial neural network technology.During the teleoperation process,the operator usually relies on the force feedback and can not accurately sense the exact contact force in the slave side,especially the fine operation requiring high transparency,so the teleoperation contact force prediction method based on GRNN was proposed.The arm electromyography signal,the master end position signal,the slave end velocity and acceleration signals were used as inputs,and the slave end contact force was used as the output to train the GRNN model,which was used to predict the contact force of the slave end.Finally,the experimental results of two sets of contrast experiments show that the method can more accurately predict the contact force of the slave end.And in order to improve the execution efficiency of the operator to complete the repetitive task during the teleoperation,the teleoperation control method with virtual force base on LSTM was proposed.The LSTM neural network have the memory function of the time series,and could predict the slaveposition of the next moment through the slave motion state of the previous moment and the current moment.The predicted slave position can be converted into a virtual force according to the master-slave mapping relationship,and then the virtual force acts on the master end by virtual spring,so that the operator's hand can be moved.Finally,through two sets of grabbing experiments,the experimental results show that the method can improve the efficiency of the action,reduce the task time,and improve the stability of the system.