Push-Grasp Quality Evaluation for Polygonal Parts under Pose Uncertainty using Quasi-static Simulation

We present a framework for analyzing the quality of push-grasps of extruded polygonal parts under pose uncertainty using a parallel jaw gripper. Performing a push-grasp involves using the first jaw to push the part into stable alignment, after which the second jaw closes onto the part. The motion of parts pushed on a work surface is difficult to predict if the precise interactions between the part and the work surface are not known. However, the relative motion of the part can be easily calculated. We use a quasi-static simulation to predict this relative motion. Given a nominal part shape and anticipated uncertainty in the part pose, we use a Monte Carlo sampling approach to evaluate grasp quality under pose uncertainty. We sample poses from the uncertainty distribution, and execute simulations to evaluate if the grasp is successful, ie, if force-closure is achieved. We then calculate the overall grasp quality as a weighted average across samples, where the weight for each sample is the probability of that sample occurring. Since each sample can be analyzed in parallel, this approach is well-suited for Cloudbased execution. We also present a sensitivity analysis of the grasp quality for a given grasp on a non-convex polygonal part under varying position and orientation uncertainty. Our experiments suggest that while position uncertainty has a direct effect on quality, orientation uncertainty has complex effects which depend on part shape and symmetry. This supports our hypothesis that a simulation-based grasp quality metric is important for comparing different grasps under varying levels of pose uncertainty.