A Collaborative Robot Cell for Random Bin-picking based on Deep Learning Policies and a Multi-gripper Switching Strategy

Abstract

This paper presents the details of a collaborative robot cell assembled with off-the-shelf components designed for random bin-picking and robotic assembly applications. The proposed work investigates the benefits of combining an advanced RGB-D vision system and deep learning policies with a collaborative robot for the assembly of a mobile phone. An optimised version of YOLO is used to detect the arbitrarily placed components of the mobile phone on the working space. In order to overcome the challenges of grasping the various components of the mobile phone, a multi-gripper switching strategy is implemented using suction and multiple fingertips. Finally, the preliminary experiments performed with the proposed robot cell demonstrate that the increased learning capabilities of the robot achieve high performance in identifying the respective components of the mobile phone, grasping them accurately and performing the final assembly successfully.