Particle Swarm Optimization of Matsuoka's oscillator parameters in human-like control of rhythmic movements

In the field of neuroscience, the Matsuoka's nonlinear neural oscillator is commonly used to model Central Pattern Generator (CPG) in humans/animals. How the parameters of such structure should be selected is not always clear. It was generally done in past studies thanks to a trial-and-error method that needs to be reiterated each time the task changes. Recent studies using a Describing Function Analysis (DFA) of this CPG model provide interesting analytical tuning methods. Nevertheless, as they are based on a linear approximation, they might have a limited efficiency in the particular case of timing-sensitive task, such as the ball-bouncing task considered in this study. A Particle Swarm Optimization (PSO) is thus proposed to select the parameters of a novel neural oscillator-based human-like control architecture able to face disturbances and to adapt to new reference set-points during the ball-bouncing task. The general method presented in the present paper can also be used for other Matsuoka's oscillator tunings and other tasks.