Zvyšování přesnosti pozice a orientace objektů při jejich umisťování manipulátorem

Abstract

The presented dissertation deals with increasing the position accuracy of objects during their placement or assembly by a robot. Industrial robots are commonly used in assembly lines, where bin-picking, which is the removal of disordered objects from a pallet or box, is becoming increasingly used. This pick-and-place application is solved by a vision system and a robot that places the grasped objects on a conveyor that moves the objects on the technology line for further processing. This work presents the possibility to eliminate the conveyor which can be achieved by performing the assembly process directly after grasping the object via bin-picking system. The thesis focuses on refining the object pose estimation in a gripper of the robot (in 3D space) by Iterative Closest Point algorithm. To refine the pose estimation, the correct input data to the ICP algorithm are crucial, which is achieved by scanning the relevant features, the geometric primitives of the object.

The introductory part of this dissertation is devoted to commercially available bin-picking systems and an overview of the current state of the art with possibilities for refining the assembly or manipulation process. Then, the objectives of the thesis are stated based on a survey of the current state of the art and in the context of projects carried out by the Department of Robotics. The actual part of the thesis is divided into subchapters according to the different objectives of the thesis. This article describes a methodology for positioning sensors relative to the scanned object (and various materials) to increase the reliability of data collection. Based on the obtained characteristics, a simulation model is developed for virtual scanning and simulation purposes. Subsequently, a sensor placement methodology is developed to refine the pose estimation – finding the optimal pose of the sensors relative to the scanned object with respect to the collected data for input to the ICP algorithm. The simulation model and pose estimation are verified on a real system.