Visual guided robotic picking system for the grocery industry

Abstract

People spend a lot of time and energy doing grocery shopping. Stores have become bigger and more centralised leading to more people having to use cars for their grocery shopping. The market for online grocery shopping has been increasing rapidly, and many new companies are emerging in a market traditionally ruled by giants. Ordering groceries online can be convenient for the customer, but in the end, someone has to do the picking in a warehouse. This thesis presents a vision guided robotic picking system designed to pick groceries from vertical shelves. A suggested solution of using the feature detector and descriptor algorithm SIFT to locate and estimate the objects pose is presented. By using a known image of each product, four corner points can be located and used to estimate the homography. The system is implemented on a vertically mounted 3-axis gantry robot mounted in front of shelves. The scope of this project extends to controlling the motors on the robot as well as an industrial vacuum system that is used together with suction cups to pick items. A solution for controlling the robot using the open software library SOEM as well as Robotic Operating System (ROS) is presented. Results show that the pose estimation algorithm can provide a positional accuracy within 1cm on items with a flat surface. This is good enough for the robot to place a suction cup. A short video demonstrating the functionality of the system can be seen using the following link: https://goo.gl/vQDF44