Motion Control Utilizing Surrogate Model for A Soft Actuator Driven by Airbag-typed Cells

Most control methods for soft robots are developed based on kinematic models derived from deformation and force analysis. However, due to the non-linearity and uncertainty of soft robotic structure, it is difficult to establish accurate models, leaving a great gap in the precise control of soft robots. Recent research has shown that machine learning provides a highly effective solution. In this work, we propose a back-propagation neural network based on particle swarm optimization algorithm to establish the surrogate kinematics of a airbag-type soft actuator. Using the motion data of the soft actuator to train the network model, the corresponding relationship between the soft actuator and the end position can be obtained. The results show that the surrogate model has a good prediction effect, and the average relative error of the model is 6.4%, enabling control the motion of the soft actuator accurately enough.