Input trajectory transfer in heterogeneous dynamic systems for muscular stimulation

Abstract

Transferring input or output trajectories between dynamic systems is a key element in multi-agent learning control or to reuse previously learned trajectories if the dynamics have changed. However, if the dynamics are unknown or non-linear, this transfer is not trivially found. We propose a method to estimate this transfer by looking at the difference in dynamics for structurally similar systems. The method is introduced and applied to the stimulation dynamics of muscles. Despite the muscles having complex, non-linear dynamics we show the method can successfully estimate the difference in dynamics and transfer an input trajectory from one muscle to the other.