A Compliance-centric View of Grasping

We advocate the central importance of compliance for grasp performance and demonstrate that grasp algorithms can achieve robust performance by explicitly considering and exploiting mechanical compliance of the grasping hand. Specifically, we consider the problem of robust grasping in the absence of a priori object models, focusing on object capture and grasp stability under variations of object shape for a given robotic hand. We present a simple characterization of the relationship between hand compliance, object shape, and grasp success. Based on this hypothesis, we devise a compliance-centric grasping algorithm. Real-world experiments show that this algorithm outperforms compliance-agnostic grasping, eliminates the need for explicit contact state planning, and simplifies the perceptual requirements when no a priori information about the environment is available.