Prediction and Elimination of Physiological Tremor During Control of Teleoperated Robot Based on Deep Learning

Abstract

Currently, teleoperated robots, with the operator’s input, can fully perceive unknown factors in a complex environment and have strong environmental interaction and perception abilities. However, physiological tremors in the human hand can seriously affect the accuracy of processes that require high-precision control. Therefore, this paper proposes an EEMD-IWOA-LSTM model, which can decompose the physiological tremor of the hand into several intrinsic modal components (IMF) by using the EEMD decomposition strategy and convert the complex nonlinear and non-stationary physiological tremor curve of the human hand into multiple simple sequences. An LSTM neural network is used to build a prediction model for each (IMF) component, and an IWOA is proposed to optimize the model, thereby improving the prediction accuracy of the physiological tremor and eliminating it. At the same time, the prediction results of this model are compared with those of different models, and the results of EEMD-IWOA-LSTM presented in this study show obvious superior performance. In the two examples, the MSE of the prediction model proposed are 0.1148 and 0.00623, respectively. The defibrillation model proposed in this study can effectively eliminate the physiological tremor of the human hand during teleoperation and improve the control accuracy of the robot during teleoperation.