发布日期:Oct 01, 2013 2:57:6 PM
Abstract:
In this paper, we proposed a machine learning method which realizes human-robot interaction without pattern recognition. In most previous researches, action patterns were recognized as discrete states, e.g., “kick”, “punch” or “jump”. Discrete states are a kind of low-dimensional representation of movements. In this work, we try to realize human-robot interaction without performing pattern recognition. In particular, we use Gaussian process latent variable model to acquire a nonlinear low-dimensional state representation. As an example, we developed a robot which plays paper-stone-scissors with a human player.