Robotic Grasping Based on Deep Learning: A Survey

The progress achieved in the robotics field in the last decade made the deployment of robots possible in real - world. One of the essential skills required to do these tasks is object grasping and placing. Grasping is the process of lifting an object without dropping it by the end effector of a robot manipulator. To accomplish the grasping task, a robot has to automatically identify the grip region of a specific item via tactile or visual information. Deep learning, a branch of machine learning developed by simulating human brain activities, can be beneficial in achieving robot grasping tasks. In addition, the development in hardware computational abilities and data availability helped Deep Learning succeed in many fields such as computer vision, image segmentation, and object detection. Derived by this success, many researchers developed Deep Learning-based methods for robot object grasping and achieved promising results. This survey focuses on data-driven (or sometimes called learning-based) approaches based on Deep Learning and used for vision-based robotic grasping. The purpose of this review paper is to provide an overview of object grasping and placing based on Deep Learning, it's used in this field with exploring the most recent works.