Recurrent neural networks for modeling soft manipulator in simulated and real environments

Abstract

Neural networks have emerged as a potent tool for addressing kinematic challenges in the development of soft robotics. This study aims to explore inverse kinematics and forward model for soft robots. Two datasets were created: one comprising simulated soft robots and another featuring a real manipulator of varying sizes, to train recursive neural networks. After training and validation, the optimal hyperparameters were identified, and the performance of the models was evaluated. The results demonstrate that the models were able to achieve significantly high accuracy, underscoring the effectiveness of neural networks in complex robotics applications.