Grasp Synthesis of Dishware Using Mean Shape Fitting for a Table Bussing Robot

Abstract

As the robotics industry has developed, robots are frequently encountered in various environments, including coffee barista robots, chicken frying robots, cleaning robots and so on. Recently, robots serving food from the kitchen to the customers’ tables are often seen in restaurants. However, the development of table bussing robots, which are designed to clear dishes from the customers’ tables and are expected to effectively reduce manpower, is still in the beginning stage. In order to provide stable service, table bussing robots need the ability to robustly detect and grasp a variety of dishes. Therefore, in this paper, we propose a method for grasp synthesis of dishware using mean shape fitting for a table bussing robot. First we employ off-the-shelf instance segmentation network to detect dishes. Then pose information of the detected dishes is estimated by applying a category-level object pose estimation network with mean shape fitting. Once the object’s pose information is obtained, grasp synthesis is performed by analyzing the deformed mean shape with the PCA(principal component analysis) algorithm.