Analysis of human hand posture reconstruction under constraint and non-constraint wrist position

As a compactness unit, the human hand shows high versatility and sophisticated grasp functionality. How to design a robot hand replicating the human grasp posture is a challenging task. Mechanical implementation of postural synergies provides new hope for resolving this problem. Generally, these posture synergies are extracted from a large data set consisting of a variety of hand grasp postures that can be reconstructed through synergies with an acceptable error. In the daily life, people can successfully grasp an object within a grasp tolerance (acceptable scope of relative position between human hand and objects). However, the relative position between human hand and objects is almost ignored in previous studies on the reconstruction of human grasp postures. In this paper, we tend to analyze the difference of the reconstruction of hand postures in two different scenarios: constraint and non-constraint wrist positions. The principal component analysis (PCA) is applied to the posture data sets acquired under two different data-acquisition paradigms, with steady and varying wrist-object position, respectively, for reconstructing the hand postures. The reconstruction differences between these two data-collection paradigms are analyzed. The information transmission rates given by different number of PCs are qualified. The distributions of the first four PCs elements under the two paradigms are also presented, respectively. Our results show that the specific hand postures in changing relative position within grasp tolerance can be faithfully reconstructed only when the relative position between the human hand and the object are fully considered.