Reach-to-Grasp Planning for a Synergy-Controlled Robotic Hand Based on Grasp Quality Prediction

Abstract

Grasp analysis of multi-fingered robotic hands has been actively investigated via the use of postural synergies in the past a few years. During grasp planning, the variables for the hand's low-dimensional representation are often optimized together with the hand's position and orientation. The planning optimization terminates when a stable grasp is reached, usually after repetitive trials and resulting in relatively low planning efficiency. This paper proposes a gradient-based algorithm to plan reach-to-grasp task for a robotic hand. The key feature is to predict the grasp quality and adjust the hand's postural synergies, position and orientation in the hand's approaching phase, in order to arrive at a stable grasp with minimal trials. The measure, used for the grasp quality assessment, is modified from the Q distance. The Q distance is a differentiable measure that has been shown highly efficient in quantifying grasp quality. Derivations and formulations of this algorithm are elaborated. The planning for multi-fingered pinches of various objects is successfully realized. Inclusion of additional contact points on the palm would enable the proposed algorithm applicable for grasp planning, eventually leading to a unified framework for efficient reach-to-grasp planning in the near future.