A concept to solve Bin Picking problem in the robotics based on CAD Model and Measurement data

Abstract

Bin picking is one of the core tasks in robotics. In terms of automating industrial processes, bin-picking shows to be a fundamental step. Although highly researched, it is still a challenging task as it combines 3D object recognition and pose estimation, handling and path planning. This paper focuses on recognition and pose estimation. Here a concept for picking Lego blocks laying randomly in a bin using CAD models of objects to be handles and 3D point cloud is presented. The approach relies on aligning the point cloud captured by a Cognex camera with the CAD models, then uses a best-fit method to determine the gripping position and orientation of the object. The concept has been validated in an experimental setup and compared to a state-of-the-art tool. Results demonstrate the efficiency and robustness of the proposed approach.