Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Survey of Learning-based Approaches for Robotic In-Hand Manipulation

Abraham Itzhak Weinberg, Alon Shirizly, Osher Azulay, Avishai Sintov

TL;DR

The developments of learning approaches for in-hand manipulations are tracked and a glossary of terms is designed as an introduction for novices in the field with a glossary of terms as a guide of novel advances for advanced practitioners.

Abstract

Human dexterity is an invaluable capability for precise manipulation of objects in complex tasks. The capability of robots to similarly grasp and perform in-hand manipulation of objects is critical for their use in the ever changing human environment, and for their ability to replace manpower. In recent decades, significant effort has been put in order to enable in-hand manipulation capabilities to robotic systems. Initial robotic manipulators followed carefully programmed paths, while later attempts provided a solution based on analytical modeling of motion and contact. However, these have failed to provide practical solutions due to inability to cope with complex environments and uncertainties. Therefore, the effort has shifted to learning-based approaches where data is collected from the real world or through a simulation, during repeated attempts to complete various tasks. The vast majority of learning approaches focused on learning data-based models that describe the system to some extent or Reinforcement Learning (RL). RL, in particular, has seen growing interest due to the remarkable ability to generate solutions to problems with minimal human guidance. In this survey paper, we track the developments of learning approaches for in-hand manipulations and, explore the challenges and opportunities. This survey is designed both as an introduction for novices in the field with a glossary of terms as well as a guide of novel advances for advanced practitioners.

Survey of Learning-based Approaches for Robotic In-Hand Manipulation

TL;DR

The developments of learning approaches for in-hand manipulations are tracked and a glossary of terms is designed as an introduction for novices in the field with a glossary of terms as a guide of novel advances for advanced practitioners.

Abstract

Human dexterity is an invaluable capability for precise manipulation of objects in complex tasks. The capability of robots to similarly grasp and perform in-hand manipulation of objects is critical for their use in the ever changing human environment, and for their ability to replace manpower. In recent decades, significant effort has been put in order to enable in-hand manipulation capabilities to robotic systems. Initial robotic manipulators followed carefully programmed paths, while later attempts provided a solution based on analytical modeling of motion and contact. However, these have failed to provide practical solutions due to inability to cope with complex environments and uncertainties. Therefore, the effort has shifted to learning-based approaches where data is collected from the real world or through a simulation, during repeated attempts to complete various tasks. The vast majority of learning approaches focused on learning data-based models that describe the system to some extent or Reinforcement Learning (RL). RL, in particular, has seen growing interest due to the remarkable ability to generate solutions to problems with minimal human guidance. In this survey paper, we track the developments of learning approaches for in-hand manipulations and, explore the challenges and opportunities. This survey is designed both as an introduction for novices in the field with a glossary of terms as well as a guide of novel advances for advanced practitioners.
Paper Structure (31 sections, 5 figures, 2 tables)

This paper contains 31 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Overview on in-hand manipulations
  3. Dexterous and non-dexterous manipulation
  4. Types of in-hand manipulations
  5. In-hand manipulations that maintain continuous contact
  6. In-hand manipulations that do not maintain continuous contact
  7. In-hand manipulation with Non-dexterous hands
  8. Parallel grippers
  9. Soft hands
  10. In-hand manipulation with Dexterous hands
  11. Perception
  12. Vision
  13. Haptics
  14. Simulation of In-hand Manipulation
  15. Datasets of in-hand manipulation motions
  16. ...and 16 more sections

Figures (5)

  • Figure 1: Statistics on paper publications which addressed or mentioned robotic in-hand manipulations over the past five years in three learning sub-fields: Model-driven learning, Reinforcement Learning (RL) and Imitation Learning (IL), along with papers that do not use any learning method. The search is based on Google Scholar and may include publications with merely a single mention of the topic and non-peer-review publications.
  • Figure 2: Taxonomy of robotic in-hand manipulation.
  • Figure 3: Various dexterous and non-dexterous hands. (a) Non-dexterous parallel jaw gripper model 2F-85 by Robotiq. The gripper has only a single DOF for opening and closing on an object. (b) The four-finger dexterous and anthropomorphic Allegro hand with 16 DOF. (c) A four finger non-dexterous soft hand operated by pneumatic bending actuators Abondance2020. (d) Underactuated compliant hand model-O from the Yale OpenHand project Ma2017YaleOP. Images (a),(b) and (d) were taken by the authors.
  • Figure 4: Illustration of (left) basic RL, (top right) actor-critic architecture, and (bottom right) a multi-network architecture.
  • Figure 5: Flowchart of policy training with Imitation Learning (IL). The policy is first learned based on expert demonstrations and then iteratively refined using a chosen RL algorithm.