A Posture Planning Method in Clustered Synergy Subspace for HIT/DLR Hand II

In this paper, a dexterous hand posture planning method in clustered synergy subspace is proposed, and the HIT/DLR Hand II is used as verification platform. Firstly, posture dataset of the hand is obtained by recording the hand’s motion while human teleoperating it. Secondly, the synergy subspace of the hand is created by applying Gaussian process latent variable model on posture data. Thirdly, the data in synergy subspace is further clustered by K-means, because similarities between different postures can be predicted from their inter-distance in synergy subspace. Finally, posture of the dexterous hand is generated from the synergy-level data in specific cluster instead of the whole subspace. To evaluate the method proposed in this paper, comparation of posture reconstruction error between this method and directly posture planning in synergy space is shown. The results show that method proposed in this paper is more accurate and anthropomorphic, and the control paraments of the hand have been considerable reduced.