Cobot + Vision to remove plastic straps from pallets

Abstract

This thesis aims at assisting Volvo Trucks to automate the de-strapping of the incoming pallets on a conveyor line using a vision-based robotic solution. Volvo provides a UR10e cobot, a Photoneo 3D-scanner and Aurora Design Assistant (ADA) to carry out this thesis. In return, Volvo requires the answer to the question whether this robot, camera, and software configuration is satisfactory to get the job done. The solution, however, needs to be divided into two main parts - building the system architecture along with integration of the given equipments and development of methods for de-strapping of the pallets. This thesis proposes the system architecture along with integration of the given equipments and the three methods that have been developed for de-strapping of the pallets. Each method is a combination of the force mode functionality of the robot and the image processing functionality via ADA after image acquisition via the camera. A python script is used to act as the middle man between the robot’s controller and ADA, facilitating data exchange and logging of data. The methods will be judged on the ability of the camera to identify and localize the strap, the ability of the UR controller to form a valid robot path to grab the straps and the cycle time. The accuracy and dependability of the robot, camera and software is judged by doing 30 cycles of grabbing the straps at various pallet offset and/or rotated from/at a fixed point. This will give quantitative data. Thus, answering Volvo’s question of "Can a UR10e cobot, a Photoneo 3D-scanner and ADA be used to de-strap their pallets?". The calibration error of the estimated homogeneous transformation matrix was found to be 3.5 mm. The fastest method was Method 3 with an average cycle time of 39.24 s. The camera, when deployed on the production line, was able to correctly determine the presence of straps with a success rate of 76.79% and the success rate of the colour identification of the straps was 75.00%. The camera was found to be able to solve the strap identification and colour identification but having a 3D camera is deemed excessive, and the image quality could be the limitation of the robustness of strap identification and colour identification.