The Comprehensive Review of Vision-based Grasp Estimation and Challenges

Abstract

Robotic grasping has emerged as a fundamental skill and a vital task for a robotic manipulator in various sectors over recent decades. Although a preprogramming method is now a general application, the challenges to handling complicated and unstructured scenarios remain. Machine vision, therefore, has become a focus of interest from many researchers as a primary perception to provide flexible manipulation in unknown and uncertain environments rather than control working space. This research presents a comprehensive review of vision-based grasp detection for a parallel gripper, analyzing potential techniques, existing challenges, and future directions. It delves into fundamental concepts of grasp detection and estimation, including traditional and learning-based methods. Additionally, the study explores essential benchmark datasets and metrics. This paper not only offers opportunities to develop grasp detection methodologies but also applications in the real world, such as fruit picking in agriculture, pick-and-pack items in supermarkets and logistics, and pick-and-sort objects in manufacturing. This will enable substantial changes and impacts of the robotic manipulator in the modern world.