Design of Digital Planner and 3D Vision System for Robot Bin Picking

Abstract

Robot bin picking plays an important role in modern manufacturing process. In order to make these manufacturing systems more efficient and productive, it is essential to make a valid grasping plan from gripper design, industrial part recognition, pose estimation, to grasping evaluation. This paper proposes such a planning framework that enables the robot to learn to grasp an industrial part and improve the performance in two phases. First, prior knowledge of 3D model is utilized for gripper selection, database generation and grasping point evaluation in a design phase. Next, attempts for single part grasping are made in a test phase, and grasping failures will trigger the redesign in the previous phase. The grasping plan is then used for grasping randomly distributed parts. A risk assessment is made per part for selection of best candidate of parts, taking both grasping efficiency and potential collision into account. At last, pose adjustment is applied on the robot to improve grasping capability. Through simulation and field test, we demonstrate that the two-phase planning framework is a practical solution for robot bin picking applications.