Dynamic response prediction of hydraulic soft robotic arms based on LSTM neural network

Abstract

Hydraulic soft robotic arms possess advantages of high flexibility and good adaptability by simulating biological organs, such as elephant trunks and octopus tentacles, and therefore have broad application prospects in unstructured environments. However, the dynamic models established by physical mechanism have some problems, such as lack of accuracy and poor real-time performance due to complex nonlinear relationships between hydraulic pressures and arm deformations. This article investigated the dynamic analysis and response prediction of a double-section hydraulic soft arm based on long short-term memory neural network. The real-time tip coordinates of the hydraulic soft arm were collected under randomly generated pressure excitations. The long short-term memory neural network was built by taking the hydraulic pressures as inputs and the tip coordinates as outputs. The experimental data were divided into training set, validation set, and test set, where the training set was used to learn the parameters of the long short-term memory network, the validation set was used to optimize the hyperparameters, and the test set was used to evaluate the performance. The results showed that the dynamic model based on the long short-term memory neural network could predict the dynamic response of the hydraulic soft arm under given inputs efficiently and accurately. This data-driven model provides a candidate for the application of dynamic response prediction and motion control of the hydraulic soft arm.