The Study of Dexterous Hand Manipulation: A Synergy-Based Complexity Index

Abstract

In this work we tackle the question of how to analyze and objectively quantify the complexity of a manipulation task. The study investigates the kinematic behavior of the hand joints in three different manipulation tasks of growing complexity: reaching-to-grasp, tool use and piano playing. The collected data were processed to extract the kinematic synergies of the hand by means of singular value decomposition. A novel, unbiased metric to determine hand manipulation complexity was based on the cumulative variance accounted for. This Variance-Accounted-For Complexity Index (VAF-CI) reliably distinguished between different manipulation tasks. Moreover, an unsupervised learning method (k-means clustering) was able to use the index to accurately identify the 3 distinct manipulation tasks. These results may be leveraged to improve the control of biomimetic dexterous robots during manipulation tasks.