Estimating Part Pose Statistics With Application to Industrial Parts Feeding and Shape Design: New Metrics, Algorithms, Simulation Experiments and Datasets

摘要

Part feeders take an unsorted bulk of identical parts and output them in a uniform orientation. Common feeders sort out those items that settle initially on one specific face, and further reorient them as desired. Quick estimators of the probability of settling on a given face facilitate the design of parts for efficient feeding and of the feeding lines themselves. Nevertheless, the evaluation and the development of such estimators have been hindered by the lack of data. Here, I create and analyze a large, simulated dataset; evaluate estimators available in the literature by comparing their predictions to simulation results with the help of a custom-made metric; and propose new estimation algorithms. The new estimators offer viable alternative to the direct dynamic simulation of parts due to their low average errors.