Estimating tactile data for adaptive grasping of novel objects

We present an adaptive grasping method that finds stable grasps on novel objects. The main contributions of this paper is in the computation of the probability of success of grasps in the vicinity of an already applied grasp. Our method performs adaptions by simulating tactile data for grasps in the vicinity of the current grasp. The simulated data is used to evaluate hypothetical configurations and thereby guide the robot in the right direction. We demonstrate the applicability of our method by constructing a system that can plan, apply and adapt grasps on novel objects. Experiments are conducted on objects from the YCB object set, [1], and our method increases the robot's success rate from 71.4% to 88.1%. Our experiments show that the application of our grasp adaption method improves grasp stability significantly.