Neural network based control of legged hopping systems

Abstract

Controlling the model of a movement system based on the dynamics of biological hopping and running is investigated. This movement system consists merely of a massless spring attached to a point mass. It describes a fast three-dimensional legged locomotion on even grounds. Rapidly moving legged autonomous systems require different hardware layouts and control approaches in contrast to slow moving ones. The spring mass system is a model that describes this principle movement as well as the principle control task. Multilayer-perceptrons are used to implement neurocontrollers suitable for such a movement system. They are proved to be suitable for exact control of the movement with a relatively small number of neurons. This is also shown by an experiment where the environment of the spring-mass system has been changed from even to uneven ground. The neurocontroller performs well without being trained for it.