Description
Advances in the rapidly growing field of soft robotics show that robotic systems and devices made from soft materials surpass rigid-links robots in terms of adaptability and flexibility. As such, soft robots are believed to bridge the gap between humans and autonomous machines. Despite an increasing sophistication in the development of soft robots, research on closed loop control of soft robots still lags behind. This can be partly attributed to the high nonlinearities, which complicate accurate modeling. Artificial neural networks (ANN) can be a very powerful tool for capturing even those non-linearities that are very often neglected. In this article, we extend our previous research on finite element (FE) based training of artificial neural networks for modular soft robots. We present a method by which sufficiently large amounts of training data can be generated in order to learn the kinematic model of a soft pneumatic actuator which can move in three-dimensional space. The method is generic and can be employed for learning of kinematic models for simulation and control.
