Underactuated anthropomorphic hands: Actuation strategies for a better functionality
Introduction
Anthropomorphic robotic hands that have been developed so far for industrial and prosthetic purposes can be roughly categorized as fully actuated hands [1], [2], [3], [4] and highly underactuated [5], [6], [7], [8].
The former group has many actuators, placed on the forearm or in the hand itself, by the sacrifice of the weight and the size of the hand and the actuation power. These hands can grasp and manipulate a wide range of objects by applying a complex control strategy. Yet, due to their electromechanical and control architecture complexity, they are not the best option for many applications that demand simplicity, lightness and low cost.
The underactuated hands usually contain 1–6 actuators (compared to the 34 muscles that control human hand). They offer the following advantages [9] over the former group:
- •
Simpler electromechanical structure.
- •
Lower weight, size and price.
- •
Simpler control architecture.
The human hand has an average weight of 400 g [10]. However, prosthetic terminal devices of similar weight have been described as being too heavy by users [11]. With the current technologies it is not possible to construct a fully actuated hand lighter than 400 g. For this reason fully actuated prosthetic hands are not being commercialized for prosthetic applications. Here we present some of the underactuated hands, which are either a prototype in a research center or commercialized product, and categorize them based on their number of actuators. For a simpler description of these prototypes, we depicted the name of the bones and joints of the human hand. In the following paragraphs DOF indicates “Degrees-Of-Freedom” and DOA stands for “Degree-Of-Actuation” or in other words Number-Of-independent-Actuators.
Vincent hand is a 6-DOF (Degrees-of-Freedom), 6-DOA (Number-of-Actuators) commercial prosthetic hand. It has 1 actuator for each finger except from the thumb that is double actuated, one for the thumb abduction/adduction movement and another one for the thumb flexion (total 6 actuators) [13]. The metacarpophalangeal joint flexion is moved directly by gears. The metacarpophalangeal joint of the thumb, for flexion and abduction/adduction, is moved using two separate actuators. The hand prosthesis has 10 movable joints, which can be actively moved in the direction of flexion and extension [14]. The joints of the human hand can be seen in Fig. 1.
iLIMB is a 6-DOF, 5-DOA commercial prosthetic hand. It includes 1 actuator per each finger (total 5 actuators) which directly actuates the respective metacarpophalangeal joint flexion. The thumb abduction/adduction movement is not motorized and is performed manually. The abduction/adduction movement is actually limited to two possible positions at approximately 90 around an axis parallel to the wrist axis. At these positions a locking mechanism locks the thumb position. To alternate this position an external force is applied by the other hand, so that the thumb rotates around the mentioned axis until it locks in the second position [14], [15].
BeBionic is a 6-DOF, 5-DOA commercial prosthetic hand. It includes 1 actuator per each finger (total 5 actuators) but in contrast to iLIMB hand the motors are placed in the mid-hand (metacarpus). The thumb abduction/adduction is manual. [14], [16].
Meka H2 is a 12-DOF, 5-DOA compliant four finger hand (the little finger is eliminated). There is 1 actuator per each finger and another one for the thumb abduction/adduction movement (total 5 actuators) [17], [18].
Smarthand is a 16-DOF, 4-DOA underactuated anthropomorphic hand. It includes 4 actuators all located inside the palm structure. The first actuator drives the thumb flexion/extension, the second one drives the index flexion/extension, the third one actuates the middle, ring and little finger flexion/extension and the fourth one actuated the thumb abduction/adduction [19]. The flexion/extension metacarpophalangeal joint is directly connected onto an extension of the brushed DC motor shaft; a certain degree of non-back-drivability is achieved by means of a high reduction, this actually, allows slight adaptation of the thumb axis while it is closed against the other fingers in a precision grasp [20], [21].
Kazuki Mitsui et al. developed a 3-DOA underactuated anthropomorphic hand. It includes 3 actuators and one solenoid. One actuator for the thumb flexion, one for the index flexion and another one for the other three fingers. Yet through an innovative design a solenoid is embedded into the thumb axis, which enables the ab/ad of the thumb when necessary [22].
MANUS hand is a 3-DOF, 3-DOA research anthropomorphic prosthetic hand. It includes 3 actuators and one Geneva wheel. One actuator for the thumb, one for the other four fingers and one for the wrist. A different design purpose was tested using a Geneva-wheel based mechanism which enables the thumb to move into 2 planes with only one motor actuating it, the thumb has a coupled flexion and abduction/adduction [6].
Michelangelo is a 2-DOF, 2-DOA commercial anthropomorphic prosthetic hand. It includes 2 actuators, one for closing all the fingers and another one for changing the angle of the thumb. A small motor changes the path that the thumb will take when the main motor actuates to close the hand either in a palmer or lateral grasp [23], [24].
KNU hand is a research prosthetic hand with 16-DOF and 2-DOA. It includes 2 actuators, one for the thumb flexion, another actuator for the other 4 fingers and one Geneva wheel. The approach used for the ab/ad of the thumb is the same as in the MANUS hand, but as the interphalangeal joint is coupled with the carpometacarpal joint a complex driven path is required to transmit the motor’s power. Thus to achieve this feature an external Geneva and crank-slider mechanism is used [25].
Single actuator hands are especially interesting because they are usually light-weight and simple to control. In such hands usually researchers try to use innovative mechanisms in order to increase the hand’s adaptivity to a wide range of objects.
Sensorhand is a 1-DOF, 1-DOA commercial prosthetic hand. It includes only 1 actuator that actuates the flexion of the thumb, index and middle finger. The other two fingers have only an aesthetic purpose [26].
TUAT hand is a 21-DOF, 1-DOA research prosthetic hand. The thumb’s rotation axis is placed at from the wrist axis. The thumb only operates to fix the object position as a fulcrum [27].
Researchers have also analyzed kinematic and dynamical behavior of the finger flexion in order to design optimal adaptive mechanisms that can grasp different objects with a single actuator [28], [29], [30]. In some cases in addition to the adaptive flexion mechanism, researchers integrated a friction-based proximal joint on the thumb, that can be manually actuated by the other hand.
In the case of the Pisa-IIT hand [31], boundary surfaces are used in order to block some of the fingers from bending and to form the desired postures on the hand for grasping objects with different geometries. Baril et al. incorporated a mechanical blocking system that allows the amputee to manually switch between 3 possible configurations. In other words the blocking mechanism blocks some of the fingers from closing [32].
In some designs, some innovative features allowed to use a single actuator for actuation of more than one axis, mostly by combining the thumb flexion and abduction, thus allowing for reduction of the total number of actuators. This can be seen in the 3DOF prosthetic hand presented in [6], [22], where a solenoid enables or disables the abduction/adduction movement of the hand, and thus one actuator with the help of a small solenoid operates both flexion and abduction of the thumb. In some cases the thumb flexion and ab/ad movements are coupled in a single movement [33], [34], [35], [31].
Some performance metrics have been already developed to evaluate different designs and performance aspects of the anthropomorphic hands. Bettler et al. reviewed the mechanical design and performance specifications of anthropomorphic prosthetic hands until 2013. Their main objective was to report the performance of several commercially available myoelectric prosthetic hands and studied the finger design and kinematics, mechanical joint coupling, and actuation methods of these commercial prosthetic hands. However hands are not compared in detail in terms of their grasping functionality [36]. Most of the metrics that have been previously developed in the field of grasping are designed to measure the quality of each grasp. M.A. Roa et al. made an extensive review of grasp quality measures and their performance [37]. When a hand is planned to grasp an object, there are usually several grasps that can fulfill the task. The problem which is addressed in these benchmarks is determination and planning for the best contact points, the hand configuration, and also determination and control of the contact forces in order to offer the best tolerance to the disturbance, the best dexterity and the best equilibrium respectively (see for instance [38], [39]).
However, these studies focus on the quality of each single grasp with the objective of grasp planning, which is mostly a problem in the domain of industrial robotic hands. Here we are looking at the overall performance of a bionic hand for prosthetic applications for successful grasping of several objects. Based on a benchmark that is designed for that purpose, we would like to study the effect of different actuation strategies of the hands on their functionality on grasping. Such study should consider all possible grasps that humans perform in daily basis.
In summary, this study aims to investigate number of actuators as well as the actuation strategy against a set of performance metrics in order to suggest the best actuation strategy for a better functionality and better performance of anthropomorphic hands. Such study can be useful for designers in the early stages of designing a prosthetic terminal, for deciding the number of actuators and how such actuators are allocated to the DOFs of the hand. In other words, this study shows which joints of the hands should be coupled together and driven by a single actuator, in order to get the best performance with minimum number of actuators. This is important for developing hands, which have a small number of actuators (i.e. less than 5 actuators), and thus benefit from a simple electromechanical structure.
Section snippets
Problem statement
As can be seen from the introduction, the underactuated hands developed up to now may include 1–6 actuators, with several different configurations. While in some hands, each finger is actuated separately, in others several fingers are connected to the same actuator. Even with the same number of actuators, the actuation strategy varies among the anthropomorphic hands developed so far.
In this regard, we divide the underactuated hands into 5 groups based on their number of actuators, and within
Categorization of the grasps
Several analyses have been performed in order to define the grasps used by human hand. From those we found the analysis in [40] the most comprehensive study, containing 33 different grasps.
For a detailed and a valid analysis, we first defined several criterias. Based on these criterias the grasps will be categorized, which facilitates the analysis for the “Grasp Diversity” in Section 6 and consequently for the “Grasp Functionality” in Section 7.
Description
Sixteen configurations were created for this analysis. A number/symbol is assigned to each finger and movement of the hand, where numbers were given to all different fingers except for the thumb that represented by the capital letter ‘T’. Index, middle, ring and little fingers are given by numbers 2, 3, 4 and 5 respectively (Fig. 4). Here, actuators are allocated to one finger or set of fingers or to abduction/adduction movement of the hand because the metacacarpal (MCP) joint allows two kinds
Case study: validation of the categorization
Here we perform a case study, in order to compare the grasps that can be performed by an anthropomorphic hand with the grasps that can be performed based on this analysis. In this case we used ISR-Softhand [41] for the analysis. ISR-Softhand has the same actuation strategy as configuration 3.0M.
ISR SoftHand can perform 20 of the 33 grasps, with an exact or very similar imitation of the images shown in the grasp list, as can be seen in Table 8. These grasps are shown in Table 8 by sign ‘’. 6
Grasp diversity
The idea behind this part of the analysis is to find out the number of achievable grasps out of the 33 possible grasps presented in [40] for each configuration. We reviewed the status of each finger and its relation to other fingers for each grasp in order to find out if that grasp is achievable with the defined configurations. For instance considering the medium wrap grasp (Table 9), we can observe that the thumb requires to apply lateral force to the cylinder. Thus a configuration which does
Grasp functionality
In the previous analysis we focused on grasp diversity without considering their usage frequency. Many of the 33 grasps introduced in [43] are not frequently used by humans. In order to analyze each configuration in terms of functionality we need to analyze them against the highly used grasps. To do so, we need to know what are the most used grasps by humans. Some analyses were made in order to find out the most used grasps by humans [44], [45], [46]. But a more recent study [47] showed to
Hand gestures
Besides grasping or more skillfully, manipulating objects, sometimes we use our hands to express ourselves while we speak or act, for example, when we wave to someone across the street, point or hand shake.
From the first analysis, we analyzed which of the gestures are achievable with each configuration. We used the configurations and their achievable grasps from Table 9 and related them to the following gestures (results are shown in Table 11):
- •
Hand shaking;
- •
Waving;
- •
Pointing/Clicking;
- •
Like (thumb
Results and discussion
Both analyses lead us to some very interesting conclusions, that will be discussed here. It is clear that increasing the number of actuators improves the hand’s functionality in terms of number of achievable grasps. But based on the results of our analysis, we discuss how those actuators should be allocated to the hand’s degrees of freedom. For this reason we will mostly compare and discuss the configurations with the same number of actuators, with a special focus on the hands with 3 actuators.
Conclusion
The actuation strategy of robotic hands plays an important role in its performance. While it is clear that usually more actuators lead to a better performance, it was not clear how to allocate these actuators to the hands DOFs. We performed two separate analyses. In the first analysis, the Grasp Diversity, we only analyzed the number of achieved grasps for each actuation strategy. But since many of the grasps in the list of the 33 possible grasps are rarely used in routine tasks, we performed a
Acknowledgment
This research work was partially supported by the Portuguese Foundation of Science and Technology, SFRH/BPD/70557/2010 & PEst-C/EEI/UI0048/2013.
Mahmoud Tavakoli received his Ph.D. in Automation and Instrumentation from University of Coimbra, Portugal in 2010. Currently he is an auxiliary Investigator in ISR-UC. His research interests include field and service robots, Medical robots, Grasping, inspection robots and climbing robots. He is the author of more than 40 publications as patents, books, book chapters, and articles in international peer reviewed journals and conferences.
References (47)
- et al.
Multiple finger, passive adaptive grasp prosthetic hand
Mech. Mach. Theory
(2001) - et al.
Robot Hands and the Mechanics of Manipulation
(1985) - et al.
Review of gifu hand and its application
Mech. Based Des. Struct. Mach.
(2011) - et al.
Robonaut 2-the first humanoid robot in space
- et al.
Antagonistically driven finger design for the anthropomorphic DLR hand arm system
- et al.
Underactuated Robotic Hands, Vol. 40
(2008) - et al.
The manus-hand dextrous robotics upper limb prosthesis: mechanical and manipulation aspects
Auton. Robots
(2004) - et al.
Optimized design of the underactuated robotic hand
- et al.
A compliant, underactuated hand for robust manipulation
Int. J. Robot. Res.
(2014) - et al.
The use of underactuation in prosthetic grasping
Mech. Sci.
(2011)
Performance characteristics of anthropomorphic prosthetic hands
Anthropomorphic hand optimization based on a latent space analysis
Objectives, criteria and methods for the design of the smarthand transradial prosthesis
Robotica
The smarthand transradial prosthesis
J. Neuroeng. Rehabil.
Mechatronic design of a transradial cybernetic hand
An under-actuated robotic hand for multiple grasps
Cited by (35)
Feasibility of proportional EMG control for a hand exoskeleton: A Fitts’ Law approach
2018, IFAC-PapersOnLine
Mahmoud Tavakoli received his Ph.D. in Automation and Instrumentation from University of Coimbra, Portugal in 2010. Currently he is an auxiliary Investigator in ISR-UC. His research interests include field and service robots, Medical robots, Grasping, inspection robots and climbing robots. He is the author of more than 40 publications as patents, books, book chapters, and articles in international peer reviewed journals and conferences.
Baptiste Enes is a M.Sc. student of Biomedical Engineering in the field of Biomedical Instrumentation and Biomaterials, from the Faculty of Science and Technology, University of Coimbra, Coimbra, Portugal. He is currently performing an investigation toward realization of his M.Sc. Thesis at the Institute of Systems and Robotics (ISR), University of Coimbra, Portugal, which focuses on the optimization of anthropomorphic under actuated hands.
Joana Rita Santos is an M.Sc. student of Biomedical Engineering in the field of Biomedical Instrumentation and Biomaterials, from the Faculty of Science and Technology, University of Coimbra, Coimbra, Portugal.
She is currently performing an investigation toward realization of her M.Sc. Thesis at the Institute of Systems and Robotics (ISR), University of Coimbra, Portugal, which focuses the design of adaptive synergies for the ISR-Softhand.
Lino Marques has a Ph.D. degree in Electrical Engineering from the University of Coimbra, Portugal. He is currently a Senior Lecturer at the Department of Electrical and Computer Engineering, University of Coimbra, and he heads the Embedded Systems Laboratory of the Institute for Systems and Robotics from that university (ISR-UC). His main research interests include embedded systems, mechatronics, robotics for risky environment.
Aníal T. de Almeida is Full Professor of the Electrical Engineering Department of University of Coimbra and Director of the Institute for Systems and Robotics (ISR)-Coimbra. He has participated in several research projects and is author of more than 200 research articles. He was the General Chair of several conference including ROS-2012 in Portugal.