A Survey of Multifingered Robotic Manipulation: Biological Results, Structural Evolvements, and Learning Methods

Abstract

Multifingered robotic hands (usually referred to as dexterous hands) are designed to achieve human-level or human-like manipulations for robots or as prostheses for the disabled. The research dates back 30 years ago, yet, there remain great challenges to effectively design and control them due to their high dimensionality of configuration, frequently switched interaction modes, and various task generalization requirements. This article aims to give a brief overview of multifingered robotic manipulation from three aspects: a) the biological results, b) the structural evolvements, and c) the learning methods, and discuss potential future directions. First, we investigate the structure and principle of hand-centered visual sensing, tactile sensing, and motor control and related behavioral results. Then, we review several typical multifingered dexterous hands from task scenarios, actuation mechanisms, and in-hand sensors points. Third, we report the recent progress of various learning-based multifingered manipulation methods, including but not limited to reinforcement learning, imitation learning, and other sub-class methods. The article concludes with open issues and our thoughts on future directions.

Journey from human hands to robot hands: biological inspiration of anthropomorphic robotic manipulators

“…In recent years, advances in materials science, sensor technology, and artificial intelligence have continued to drive the development of more advanced and sophisticated robotic hands [16][17][18][19][20][21]. Today, there are a wide variety of robotic hands and prosthetic devices that are capable of mimicking the complex movements and dexterity of the human hand, and these devices are being used in a variety of applications, from manufacturing and logistics to healthcare and rehabilitation [22].…”

“…Dynamic modelling allows researchers to go further, simulating the hand's behavior under different loading conditions and aiding in the development of control strategies [102]. These modeling processes involve techniques of not only dynamics and kinematics algorithms based on geometric and numerical optimization, but also data-driven approaches based on experimental measurements [22,[103][104][105].…”

Affordance Segmentation Using Tiny Networks for Sensing Systems in Wearable Robotic Devices

“…The architecture uses a segmentation head connected to 4 feature maps corresponding to as many layers of the backbone. We opted for these two general configurations with respect to single instances of recent solutions from robotics literature because their target setup is substantially different making the comparison unfair [44]. The last baseline was SegFormer [45].…”

DenseTact: Optical Tactile Sensor for Dense Shape Reconstruction

“…Since dexterity similar to that of human fingers is often required for precise manipulation tasks, robotic grippers of various operating principles have been explored. Soft-robotic hand-like grippers have been proposed to pick up objects of various sizes, largely depending on the size and dimension of the gripper design [1], [2]. Some gripper designs lean on van der Waals forces for specifically picking up micro-scale objects [3], or maximizing them in a gecko-inspired fashion [4].…”