Designing rhythmic motions using neural oscillators

Abstract

Neural oscillators offer simple and robust solutions to problems such as locomotion and dynamic manipulation. However, the parameters of these systems are notoriously difficult to tune. This paper presents an analysis technique which alleviates the difficulty of tuning. The method is based on describing function analysis, and can predict the steady state motion of the system, analyze stability, and be used to determine robustness to system changes. The method is illustrated using a number of design examples including an implementation of juggling on a real robot.