Shape Estimation of Travel Path Using Reservoir Computing for Water-Driven Soft Robot

Abstract

A soft robot has a significant advantage regarding its locomotion that it can change its shape passively based on its path shape variations. In this study, we adopt reservoir computing (RC), a kind of recurrent neural network, to classify five types of path shapes based on the inner pressure in the flexible chambers of the developed robot. By using small sets of the time-varying pressure waveform, we found that RC with an appropriate number of neurons could accurately classify the path shapes.