A Recurrent Neural Network Model for Lamprey-Like Robot Movement

Abstract

One of the most important problems in controlling robot locomotion, is how to generate an online trajectory for robots with a large number of degrees of freedom. In this paper, we present a biomimetic approach which is based on Central Pattern Generator (CPG) inspired from the lamprey to solve this problem. CPG is a neural network found in animal and has the ability of producing rhythmic pattern without receiving rhythmic input. We used a recurrent neural network(RNN) to model the CPG and proposed the genetic algorithm to train the RNN to follow a predefined oscillatory pattern. It is furthermore demonstrated that recurrent neural networks do indeed exhibit oscillatory behavior and may in this way be used to imitate the function of the CPG responsible for locomotion in animate creatures.