Cognitive Grasping System: A Grasping Solution for Industrial Robotic Manipulation using Convolutional Neural Network

Abstract

In the modern era, object grasping has thousands of use cases across industries and loads of manual effort is devoted to repetitive tasks. Automating this task is important and the use of robots with embedded artificial intelligence is the key for improving grasping operations. Over the years many researches have been working on object grasping to make this operation as flexible as possible. Starting from the latest results of the use of Convolutional Neural Network, the proposed work aims at optimizing the results of the grasping tasks to make it reliable for an industrial use. Limitations are analyzed and new parameters are defined in order to make the manipulation task repeatable in terms of robot grasp position. In fact, in an automated production line, this is an important problem to consider because in many situations the object has to be positioned always in the same position with the same orientation.