A Survey on Learning-Based Robotic Grasping

Abstract

Purpose of Review This review provides a comprehensive overview of machine learning approaches for vision-based robotic grasping and manipulation. Current trends and developments as well as various criteria for categorization of approaches are provided. Recent Findings Model-free approaches are attractive due to their generalization capabilities to novel objects, but are mostly limited to top-down grasps and do not allow a precise object placement which can limit their applicability. In contrast, model-based methods allow a precise placement and aim for an automatic configuration without any human intervention to enable a fast and easy deployment. Summary Both approaches to robotic grasping and manipulation with and without object-specific knowledge are discussed. Due to the large amount of data required to train AI-based approaches, simulations are an attractive choice for robot learning. This article also gives an overview of techniques and achievements in transfers from simulations to the real world.

Robotics Dexterous Grasping: The Methods Based on Point Cloud and Deep Learning

“…With the exception of visual perception approaches, Luo et al ( 2017 ) and Wang C. et al ( 2020 ) exhibit there are many other perception methods can help improve robot performance. Caldera et al ( 2018 ), Kroemer et al ( 2019 ), Li and Qiao ( 2019 ), and Kleeberger et al ( 2020 ) focus on the overview of robot manipulation methods based on deep learning. Mohammed et al ( 2020 ) and Zhao W. et al ( 2020 ) introduce the techniques in robot learning on the basis of reinforcement learning.…”

Cyber-Physical Systems for Micro-/Nano-assembly Operations: a Survey

“…Later researches introduced vision recognition of micro/ nano-components and learning algorithms applied to recognition and tracking of micro-objects [35] presents a complete review on learning-based approaches to perform general gripping operations at the macro-scale. To enhance dexterity and optimize the grasping trajectories, authors in [36] [37•] exploited physical models of grasping forces and pull-off forces during micro-manipulation operations using search algorithms for a real-time application which computes the gripping trajectory in less than 0.1 s. van Vuuren JJ et al [38] propose a learning-based methodology for identifying novel objects and evaluating different candidate grasping strategies for an optimal grasping and handling which might be applied to the manufacturing of consumer electronic products.…”

Memory Efficient Grasping Point Detection of Nontrivial Objects

“…they extract the information from sample data in order to detect the grasp pose without having to be analytically programmed [7]. The following paragraphs are focused on a brief categorization of data-driven methods, since they have, in most cases, already been shown to outperform the analytic approaches to the grasping point detection in terms of computational complexity and accuracy [8], [9]. For analytic approaches, the surveys of various methods can be found in [10], [11].…”

“…Moreover, some approaches deal only with grasping of a single separated object, while others are able to handle grasping in a dense clutter [18]. A comprehensive review of various categorizations of data-driven approaches can be found in the recent work of Kleeberger et al [9].…”