Point Pair Feature based 6D pose estimation for robotic grasping

Abstract

6D pose estimation is a crucial research topic for flexible and autonomous systems. With the development of 3D sensors, methods using range data or point clouds show great potentials. This paper proposes an effective approach to estimate the target object's 6D pose based on point pair features. Several improvements including cluster-based downsampling, neighbor search using K-D tree, multi-frame point clouds fusion and pose verification were made to optimize the performance of the approach. Based on the object's pose, we propose a strategy to grasp the object. We tested our approach in real environment and get 97.5 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> success rate of pose estimation and 95.8% success rate of grasping objects.