Manipulate a Pile of Objects Using Pose Analysis

Abstract

We address the problem of manipulating a pile of objects on the table using pose analysis approach. The stacked objects are separated with RANSAC and the gradient segment algorithm. The CVFH feature of the point cloud of separated objects is extracted, and SVM is used for the pose estimation. Then the largest success ratio of grasping is computed by analyzing the success ratio of grasping objects from different directions. According to the current pose of objects and the largest success ratio of grasping objects, the robot hand-joints is driven to access grasping points on objects. We tested our approach using a RGB-D sensor (Xtion Pro Live) and a Nao humanoid. Results proved that our approach is with high efficiency for robustly manipulating a pile of objects on the table top.