Review of Modeling and Control Methods of Soft Robots Based on Machine Learning

Abstract

The soft robot is a kind of robot with high degree of freedom and continuous deformation. It is mostly made of soft materials, which has the characteristics of high flexibility and high task adaptability. Due to the superelasticity of the material and the continuity of the structure of the soft robots, the traditional modeling and control methods suitable for rigid robot are no longer applicable. Methods of machine learning are being gradually applied. Deep learning can avoid the complex physical modeling process of soft robots, thus realizing data-driven learning. Reinforcement learning can be used to solve the problem of lack of accurate model in robot control. This paper reviewed the application of machine learning in modeling and control of soft robots including kinematics modeling and dynamics modeling of soft robot based on deep learning, and motion control of soft robot based on reinforcement learning. By analyzing and sorting out the bottleneck problems and feasible solutions of data-driven modeling and control schemes of soft robot, the development trend of soft robot technology is discussed.