Statistical manipulation learning of unknown objects by a multi-fingered robot hand

Abstract

This paper proposes a learning method for multi-fingered manipulation of unknown objects. The method is a combination of higher-order local autocorrelation (HLAC), principal components analysis (PGA), and mean-shft clustering. Our results show that the different geometric restrictions of manipulation maximize the variance in the space of Feature vectors identified by HLAC analysis. As a result, the data corresponding to each manipulatory act are clustered in a high-dimensional space in accordance with the restrictions via PCA. Mean shift clustering method classify the clusters which correspond the restrictions. The efficacy of the proposed method is shown by means OF handling experiments of given diameter caps subjected to rotational restriction.