High-Dimensional Nonlinear Dynamics Based on LSTM Modeling for Accurate Manipulation of Soft Manipulators

The continuous soft manipulator is gradually regarded as a feasible alternative to the traditional rigid link robot because of its good flexibility. However, due to its obvious nonlinear and time-varying characteristics, actual modeling and control are still difficult. This paper focuses on the motion control method based on long short-term memory (LSTM), combs the basic theory of soft manipulator and LSTM. The existing work can be roughly divided into three categories: modeling and control of a single manipulator, multi-manipulator collaboration, and applications in complex environments. The LSTM method has certain advantages in dealing with time-varying and hysteresis characteristics, it still faces problems such as complex models, data dependence, and limited real-time performance. Combining physical models with data-driven methods and designing lighter control strategies may be a more practical direction. This paper provide some reference for the design of adaptive and intelligent control methods of soft robots.