Estimating part tolerance bounds based on adaptive Cloud-based grasp planning with slip

We explore setting bounds on part tolerances based on an adaptive Cloud-based algorithm to estimate lower bounds on achieving force closure during grasping. We consider the most common robot gripper: a pair of thin parallel jaws, and a conservative class of push-grasps allowing slip that can enhance part alignment for parts that can be modeled as extruded polygons. The grasp analysis algorithm takes as input a set of candidate grasps and perturbations of a nominal part shape. We define a grasp quality metric based on a lower bound on the probability of achieving force closure. We present two extensions to our previous highly-parallelizable algorithm that adaptively reduce the number of grasp evaluations and improve the lower bound by including slip. We develop a procedure for finding the effect of increasing tolerance in vertices on grasp quality, which allows part tolerances to be bounded to ensure minimum grasp quality levels. We find that including slip improves grasp quality estimates by 16%, and our adaptive extension reduces grasp evaluations by 91.5% while maintaining 92.6% of grasp quality.