Deformation Prediction and Trajectory Tracking of a Multidirectional Bending Fiber-Reinforced Soft Pneumatic Actuator Using Deep Learning

The nonlinear material properties of soft pneumatic actuators (SPA) make precise prediction of their deformation challenging. Especially, the deformation behavior of actuators with internal multi-chambers under varying pressure combinations has not been sufficiently studied. This study focuses on a multidirectional bending fiber-reinforced soft pneumatic actuator (MDB-FRSPA) and develops a deep learning (DL) neural network model to predict its deformation under different combinations of chamber pressures. First, a deformation dataset for the actuator under different pressure combinations is constructed using the finite element method (FEM). Based on this dataset, a prediction model mapping the four-channel pressure input to the centerline deformation of the actuator is developed using a long short-term memory (LSTM) neural network. The results show that the coefficient of determination (R²) of the model on the test set reaches 0.998. Finally, the integration of a deep learning neural network inverse model with iterative learning optimization enables the end-effector of theMDB-FRSPA to track both circular and square trajectories. In the FEM validation, the average position error is confined to less than 2 mm.