Entrainment-enhanced Neural Oscillator for Imitation Learning

摘要

To achieve biologically inspired robot control architectures based on neural oscillator networks, goal-directed imitation is addressed with respect to the problem of motion generation. It would be desirable to easily acquire appropriate motion patterns for skill learning between dissimilar bodies to attain the goal of the demonstrated motion. This requires neural oscillator networks to adapt to the non-periodic nature of arbitrary input patterns exploiting their entrainment properties. However, even in the most widely-used Matsuoka oscillator, when an unknown quasi-periodic or non-periodic signal is applied, its output signal is not always closely entrained. Therefore, current neural oscillator models may not be applied to the proposed goal-directed imitation for skill learning. To solve this problem, a supplementary term is newly included in the equation of Matsuoka oscillator. We verify general properties of the proposed model of the neural oscillator and illustrate in particular its enhanced entrainment by numerical simulation. We also show the possibility of controlling dynamic responses of oscillator-coupled mechanical systems. Technical implications of the results are discussed