Design of A Novel Compact Dexterous Hand for Teleoperation
Proceedings of 2001 IEEE International Symposium on Computational Intelligence in Robotics and Automation July 29 - August 1, 2001, Banff, Alberta, Canada
Design of A Novel Compact Dexterous Hand for Teleoperation Guowu Jia, Guang Chen, Ming Xie School of Mechanical and Production Engineering Nanyang Technological University Singapore 639798 Abstract This solution, however, has some major problems: • The usage of multi-motors results in heavy and bulky actuation units, which makes the hand less adaptive to be mounted to different robot arms or bodies when necessary. • The more lengthy tendons are used, the more damping and friction are introduced, which will cause both force and position errors. The friction between the cable and the outer coating can never be avoided. Since most times dexterous hands in teleoperation are used to handle delicate operation, a small amount of force and position errors might make task completion extremely difficult (for example in telemedicine). • Researchers also point out that the control system of the dexterous hand with multiple motors will be very complex because of the large number of motors and the actuation system coupling [11].
This paper focuses on the mechanism design of a novel compact dexterous hand of 3 fingers, 9 joints and 7 DOF (degrees of freedom). The technology of single-motor actuation is used and a kind of compact bilateral clutch is designed to reduce the size of actuation and control part of the hand. Different from most dexterous hand designs, no lengthy tendons are needed in this design since the actuation and control parts are all integrated closely to the fingers. Furthermore, passive joints are added to reduce the number of control DOF and make the hand behave like a human hand. The notion of perfect dexterous hands is discussed from the viewpoint of degrees of freedom. The hand’s kinematics and its control architecture in our experiment system of teleoperation are also briefly presented.
1. Introduction
Using direct driven technology, the compact Belgrade hand has only 4 DOF. It emphasizes more on local autonomy during grasping than maximizing dexterity. Using TorqueSwitch to control two joints with only one motor, the compact BarrettHand overcomes the shortcomings of conventional grippers and achieves dexterous grasping. However it is more like an intelligent gripper than a dexterous hand because only 4 axes can move at the same time and it can’t implement most dexterous manipulations.
Currently most of the industrial robot manipulators consist of a robot arm and a simple end-effector or gripper. Apparently the simple end-effectors or grippers lack flexibility for various tasks, especially for fine teleoperation. Only with dexterous hands can robot manipulate objects of different shapes, sizes and stiffness as human being. Although much research work have been done on grasp and manipulation theory [1], [2], kinematics [3], [4] and dynamics [5], [6] of multifingered dexterous hands, the progress in the research of hand mechanism remains slow. Most current dexterous hands [7], [8] have bulky actuation parts and can’t be mounted directly to the ends of robot arms, the rests [9], [10] are portable but not flexible enough to handle different operation requirements.
Therefore, to design a compact hand with high dexterity has always been a big challenge since the emergence of robotics. Aiming to integrate the actuation part with the fingers as a whole and cut off lengthy tendon overheads so as to improve position and force accuracy and portability, we designed a novel compact dexterous hand with high dexterity. In this paper, the notion of perfect dexterous hands is discussed from the viewpoint of degrees of freedom before the mechanism of our proposed compact dexterous hand is described in detail, at last the kinematics and the control scheme are highlighted briefly.
In the traditional hand designs, multi-motors are usually adopted, and lengthy tendons are used for motion transmission. For example, the Utah/MIT hand has 32 pneumatic actuators placed in a remote actuator pack and uses ploymeric tendons and pulleys to transmit the power. So as in the Standford/JPL dexterous hand, 12 electric motors are used as actuators in a remote site and long cables are used to transmit the motion.
0-7803-7203-4/01/$10.00 2001 IEEE
5
2. Perfect Dexterous Hand
along X and Z axes and to rotate objects around Y axis as well. Z
There is no specific definition for dexterous hands presented in literatures. Many researches associate the dexterity problem with particular tasks. Paul Michelman and Peter Allen [12] consider that the dexterous manipulation is driven by task function. Datseris and Palm [13] maintain that dexterous hands should be able to manipulate a class of objects in three rotations and two translations. Nevertheless, in general manipulation problem, it is reasonable to relate the dexterity to the capability of fingertips’ manipulation in arbitrary rotations and translations of a rigid object as in Fuentes and Nelson’s research [14]. The notion of perfect dexterous hand from which the design principles of our hand mechanism derive is presented as the following.
X
Figure 1. Two finger manipulation If another finger with two pitch joints joins in and their relative positions are properly configured as in Figure 2, then the object rotations around X and Z axes can be achieved. At the same time, a stable grasp is guaranteed by three point contact (ideal point contact is assumed between the fingertip and the object). If each finger has at least one yaw joint, then the rest translation of objects along Y axis can be achieved by simultaneous actuation of two pitch joints and one yaw joint, which is demonstrated in Fuentes and Nelson’s experiment [14].
A perfect dexterous hand mimics human hands with high dexterity and it should have basic functions as human hand. The two basic functions of human hands are grasp and manipulation. For the first purpose, it can hold objects in arbitrary orientation tightly with stability. Iberall classifies the grasp behavior of human hand into three categories in terms of “opposition”:1, Pad grasp with force between the pads of the finger and thumb; 2, Palm grasp with force between the fingers and the palm; 3, Side grasp with force between the thumb and the side of the index finger. In Napier’s scheme, grasps are divided into power grasp and precision grasp. Usually palm grasp and side grasp is power grasp while pad grasp belongs to precision grasp [15].
Z X
Figure 2: Three finger manipulation Therefore, one conclusion that can be made is that a perfect dexterous hand must have at least 3 fingers with totally 9 DOF and each finger must have two pitch joints and one yaw joint.
For the second purpose, it can exert arbitrary motion on the objects, which means it can translate objects freely along X, Y and Z axes respectively and rotate objects arbitrarily around X, Y, Z axes within certain space.
Good examples of perfect dexterous hands are the two famous hands, the Utah/MIT hand with 16 DOF and the Stanford/JPL hand with 9 DOF. While the Belgrade/USC hand and the BarrettHand are not perfect dexterous hands. The Stanford/JPL exactly satisfies the minimum requirements for a perfect dexterous hand.
We assume that a perfect dexterous hand must meet all the above requirements. In perfect dexterous hand design, precision grasp is more concerned than power grasp in order to obtain high dexterity and sensitivity. Therefore fingertip manipulation of a rigid object rather than palm manipulation or digit manipulation of fingers is given the highest priority in design process. Since the number of DOF determines the dexterity and complexity of the hand design, then there rises a problem: how many DOF should a perfect dexterous hand have at least?
3. Compact Dexterous Hand Design In our Laboratory, a dexterous hand model has been previously developed with single-motor actuation technology [16]. However it still uses tendons to transfer the remote motion to each finger and has a bulky size as most dexterous hands. Based on this model, our research focuses on compact design by integrating the actuation and control part with fingers as a whole. The benefits of this integration are obvious: a) the lengthy tendon overheads are avoided; b) The whole hand can be installed at robot arm ends directly, no additional support parts are needed. Therefore the hand has high
First, naturally all joints are assumed as rotation joints as those of human hands. If only two fingers are considered as illustrated in Figure 1, apparently two pitch joints of each finger can make objects to do rectilinear motion
0-7803-7203-4/01/$10.00 2001 IEEE
6
proximal digit, one seldom, if not totally, use this joint for object translation along Y axis. Paul Michelman and Peter Allen point out that the human capacity to translate objects by finger motions is quite limited, especially for translations parallel to palm [12]. The main functions of the yaw movement are to adjust the distance between the index and the middle finger and to help to rotate object around Z axis more conveniently. Hence for compact purpose, the yaw joints of both index and middle fingers are given up but that of the thumb is kept. Then with 3 fingers, 7 DOF and arranged in this way as shown in Figure 2., the hand is able to manipulate a rigid object in three rotations and two translations.
adaptability to various robot applications. In this design, the numbers of fingers and DOF are firstly optimized according to the above discussed notion of a perfect dexterous hand and two observations. A new kind of bilateral clutch is designed to reduce the size and weight of the control part, at the same time the compact integration makes the hand more portable. 3.1 Optimization of the number of DOF Generally the more flexibility of the hand we require, the more DOF are needed and the more complex the hand is. A human hand has over 25 DOF and each finger has 4 DOF. Fortunately, most tasks themselves limit the number of DOF. Therefore limited DOF are required to manipulate objects, and others only function to facilitate the completion of tasks. In short, for compact design purpose, the highest priority should be given to optimize the number of DOF while maintaining most static and dynamic characteristics of human hands and manipulations.
3.2 Finger joint actuation Since the index and middle fingers have only two pitch DOF, they are driven by worm gears directly at the finger ends. The worms connect with the output shafts of the clutches protruding out of the actuation box attached behind the fingers. There are two advantages in using worms and worm gears. One is that it can achieve a large motion transmission ratio while occupying the smallest space; the other is that it has self-block attribute (not backdrivable). This attribute provides the essential function to manipulate objects using clutches. When one finger (or joint) is required not to move, the clutches for that finger must be powered off. At this time, due to the worm gears, the finger will not lose contact with the object but maintain it by receiving passive contact force exerted by other fingers. Thin stranded steel ropes are used to transfer motion between digits, from finger ends to middle joints and tip joints. The finger motion transmission is illustrated in the following Figure.
The thumb, middle, and index fingers compose the basic configuration of our dexterous hand, which meets the requirement of a perfect dexterous hand. Three fingers of human hand have totally 12 DOF, but two observations prompt us to reduce the number of DOF from 12 to 7. The first observation is that apparently there is certain relationship between the rotation of the tip digit and that of the middle digit. In natural movement, the middle joint and the tip joint always rotate together. One can not move just one of them but keep the other one stationary without external force or internal force constrain. As shown in the Figure 3., when the middle joint rotates, the tip joint will follow its movement naturally and vice versus.
Base joint
Tip joint
Figure 4. Finger joint actuation
a. Unnatural way b. Natural way Figure 3. Human finger behavior
To simplify mechanical design, the thumb finger has only one yaw joint at the finger base and no passive joint at the fingertip. The yaw joint is directly driven by the worm, in the meantime the middle joint and tip joint are independently driven by steel ropes which route over the base yaw joint to avoid jam with the yaw joint.
According to this observation, the tip joint is made passive to the middle joint in our hand design. In this way, 3 control DOF are omitted while one finger still has 3 active pitch joints and is more anthropomorphic. The ratio between middle joint and tip joint as 2:1, which is similar to that of a human finger.
3.3 Compact bilateral clutch
The second observation is that although each human finger has a yaw joint for lateral movement at the end of
0-7803-7203-4/01/$10.00 2001 IEEE
Middle joint
7
The 7 clutches are the key components of our hand and they occupy most volume of the actuation part. Each clutch controls one joint’s movement. The specially designed compact bilateral clutch mainly consists of an input shaft, an output shaft, an active armature, two coils and two rotors which connect a pulley and a spur gear respectively. The active armature always rotates with the input shaft. Once one coil is powered on, the active armature will be attracted to corresponding rotor and make it rotate too. With the help of pulleys and spur gears, bilateral motion transmission is achieved. The whole clutch size is constrained in a space of about 35x25x70 mm. The clutch structure is illustrated in Figure 5. input shaft pulley
active armature
actuation box index and middle finger thumb finger
Figure 6. Compact dexterous hand model
gear
4. Hand Kinematics Since index and middle finger have identical kinematics, only index and thumb finger’s kinematics are presented here. Define {Ti } and {Bi } are frames attached to the
rotor
coil
finger tip and finger base respectively as shown in Figure 7. The absolute control angles for index and thumb finger are α i , β i and γ m , α m , β m , the corresponding
output shaft
relative joint angles are θ i1 , θ i 2 , θ i 3 and θ m1 , θ m 2 , θ m 3 .
Figure 5. Structure of bilateral clutch
Finger lengths are denoted as lnk , n={i,m} and k={1,2,3}.
3.4 Hand model One motor and 7 clutches serve as the main actuation part. All the clutches and motor are integrated into a cylinder box with a size of 104 mm in diameter and 92 mm in length. The actuation box includes three parts: one DC servo motor, motion distribution part and 7 clutches. A high torque dc servo motor with digital encoder is used. The motor output rotation is distributed to 7 input shafts of the clutches with a star configuration evenly. The output shafts of clutches directly drive the worm gears at each finger base, which greatly reduces the negative effect of long tendons and improves the accuracy. The thumb finger is located exactly in the middle and opposite to the index and middle fingers, so that the thumb fingertip can choose to approach either the index or middle fingertip. Three fingers have the same length of 115mm and the distance between the index and middle finger is 40mm. The base, middle and tip digits of each finger are the same as human scale: 1:0.666:0.444 [17].
Figure 7. Kinematics model of the hand To solve the forward kinematics of the hand, two special relationships, which are determined by the unique mechanism of the hand, should be noticed firstly. The first one is that the tip joint is passive to the middle joint for the index and middle fingers. The second one is that the middle joint and the base joint are both independently controlled. Rotation of one of them will not change the orientation of the digit controlled by the other joint,
Shown in Figure 6 is the proposed compact dexterous hand.
0-7803-7203-4/01/$10.00 2001 IEEE
8
high level includes a Server PC, a image grabber Card and a camera. The Server PC is responsible for image process, hand motion planning and communication with the remote Client PC through LAN. The low level mainly includes a microprocessor 68HC11 to control joint motion of hand and interface with the high level.
which is different from the nature of human hand. These two relationships can be expressed in matrices in (1) and (2).
1 0 − 1 1 0 1 2 θ m1 1 0 θ = 0 1 m2 θ m 3 0 − 1 θ i 1 θ = i2 θ i 3
0 αi 0 β i 0 0
(1)
0 γ m 0 α m 1 β m
(2)
Sensory Card
Camera
ABTmm
0 sin32βi li3sin32βi +li2 sinβi +li1 sinαi 1 0 0 0 cos32βi li3 cos32βi +li2 cosβi +li1 cosαi 0 0 1
cos β m − sin γ sin β m m = cos γ m sin β m 0
0 cos γ m
sin β m − cos γ m cos β m
sin γ m 0
cos γ m cos β m 0
− l m3 sin β m − lm 2 sin α m − lm 3 sin γ m cos β m − lm 2 cos α m sin γ m − l m1 sin γ m lm3 cos γ m cos β m + lm 2 cos α m cos γ m + lm1 cos γ m 1
Dexterous hand
Client PC
Haptic Hand
Operator
Figure 8. Architecture of experiment system In this experiment, the operator uses a haptic hand (a data glove) to send motion command to remote dexterous hand through campus LAN, while the dexterous hand follows the movement and feedback the visual information in real time.
(3)
6. Conclusion and Future Work A novel compact dexterous hand for teleoperation is developed in this paper. The main objective is achieved by adopting one motor actuation and optimized and integrated mechanism. The overall design features make our compact dexterous hand unique in the following ways: 1) Single-motor actuation; 2) Passive joints for the index and middle fingers; 3)multiple compact bilateral clutches; 4) the first compact dexterous hand with 7 DOF.
(4)
5. Control Architecture
Previous experiment preliminarily proves the validity of one motor actuation technology and the hand design. Extensive experiment is undergoing to evaluate the manipulation capability of this hand. Future work also include development of advanced manipulation algorithm, mainly for high-speed manipulation, and addition of force control for this compact dexterous hand.
One DC servo motor with a digital encoder is used as the only power source. The motor works with a constant speed specified through program and the output motion is transmitted to each clutch evenly. The rotation direction of clutch output depends on the clutch’s status. The angular position θ and velocity ω are determined by:
7. References
ω = λω 0 (5) θ = λω 0 t where λ is the engaging time ratio of clutches, t is the operation time and ω 0 is the constant motor speed.
[1] Matthew Mason and Kenneth Salisbury, Robot Hands and the Mechanics of Manipulation, MIT Press, 1985. [2] M. R. Cutkosky, Robotic grasping and fine manipulation, Kluwer Academic Publisher, 1985.
The developed dexterous compact hand is under testing in our teleoperation experiment system. The system control architecture is divided into two main parts, the high level and the low level. As shown in Figure 8, the
0-7803-7203-4/01/$10.00 2001 IEEE
LAN
68HC11
Based on the above relationships, the forward transformation matrices of the index and thumb fingers can be easily deduced in (3) and (4). cos3 βi 2 0 ATBii = −sin3 βi 2 0
Server PC
9
Models” Proceedings of the IEEE International Symposium on Intelligent Control, September, 1996.
[3] D.J. Montana, “The kinematics of multi-fingered manipulation,” IEEE Transactions on Robotics and Automation 11(4) 491-503, 1995.
[15] S.T. Venkataraman and T. Iberall, Dexterous Robot Hands. New York: Springer-Verlag, 1990.
[4] J.K. Salisbury, B. Roth, “Kinematics and force analysis of articulated mechanical hands,” Journal of Mechanisms, Transmissions and Automation in Design 105 35-41, 1983.
[16] Chen W. J., Xie M., “On the Design of a Novel Dexterous Hand,” The 9 th International Conference on Advanced Robotics, Tokyo, Japan, p61-65, October 25-27,1999.
[5] Suguru Arimoto, Pham Thuc Anh Nguyen etc. “Dynamics and control of a set of dual fingers with soft tips,” Robotica vol.18, pp. 71-80. 2000.
[17] Charles D. Engler, Mikell P. Groover, “Design of an anthropomorphic electro-mechanical hand with exoskeletal control to emulate human hand dexterity,” SME technical paper; MS89-274, Society of Manufacturing Engineers, 1989.
[6] Joseph Chan, Yunhui Liu, “Dynamic simulation of multi-fingered robot hands based on a unified model,” Robotics and Autonomous Systems 32 p185-201, 2000 [7] S. C. Jacobsen, B. K. Iverson, D. F. knutti, R. T. Johnson, K. B. Biggers, “Design of the Utah/MIT dexterous hand,” in proceedings of IEEE International Conference on Robotics and Automation, San Francisco, CA, 1986, pp. 15201532. [8] J. K. Salisbury and J. J. Craig, “Articulated hands: Force control and kinematic issues,” International Journal of Robotics Research, vol. 1, pp. 4-17, 1982. [9] R. Tomovic et al., “A strategy for grasp synthesis with multifingered robot hand,” in proceedings of IEEE International Conference On Robotics and Automation, Raleigh, NC 1987, pp.83-89. [10] William T. Townsend, “Description of a Dexterous Robotic Grasper,” Jounal of the Robotics Society of Japan, Vol. 18, No.6 2000. [11] Gongliang Guo, William A. Gruver, and Xikang Qian, “A new design for a dextrous robotic hand mechanism,” IEEE Control systems, August, 1992. [12] P. Michelman and P. Allen, “Compliant manipulation with a dextrous robot hand,” IEEE International Conference on Robotics & Automation, volume 3, pages 711-716, 1993. [13] P. Datseris and W. Palm, “Principles on the development of mechanical hands which can manipulate objects by means of active control,” Transactions of ASME Journal of Mechanisms, Transmissions and Automation in Design, 107: 148156, June 1985. [14] O. Fuentes and R. Nelson, “Experiments on Dextrous Manipulation without Prior Object
0-7803-7203-4/01/$10.00 2001 IEEE
10
Design of A Novel Compact Dexterous Hand for Teleoperation Guowu Jia, Guang Chen, Ming Xie School of Mechanical and Production Engineering Nanyang Technological University Singapore 639798 Abstract This solution, however, has some major problems: • The usage of multi-motors results in heavy and bulky actuation units, which makes the hand less adaptive to be mounted to different robot arms or bodies when necessary. • The more lengthy tendons are used, the more damping and friction are introduced, which will cause both force and position errors. The friction between the cable and the outer coating can never be avoided. Since most times dexterous hands in teleoperation are used to handle delicate operation, a small amount of force and position errors might make task completion extremely difficult (for example in telemedicine). • Researchers also point out that the control system of the dexterous hand with multiple motors will be very complex because of the large number of motors and the actuation system coupling [11].
This paper focuses on the mechanism design of a novel compact dexterous hand of 3 fingers, 9 joints and 7 DOF (degrees of freedom). The technology of single-motor actuation is used and a kind of compact bilateral clutch is designed to reduce the size of actuation and control part of the hand. Different from most dexterous hand designs, no lengthy tendons are needed in this design since the actuation and control parts are all integrated closely to the fingers. Furthermore, passive joints are added to reduce the number of control DOF and make the hand behave like a human hand. The notion of perfect dexterous hands is discussed from the viewpoint of degrees of freedom. The hand’s kinematics and its control architecture in our experiment system of teleoperation are also briefly presented.
1. Introduction
Using direct driven technology, the compact Belgrade hand has only 4 DOF. It emphasizes more on local autonomy during grasping than maximizing dexterity. Using TorqueSwitch to control two joints with only one motor, the compact BarrettHand overcomes the shortcomings of conventional grippers and achieves dexterous grasping. However it is more like an intelligent gripper than a dexterous hand because only 4 axes can move at the same time and it can’t implement most dexterous manipulations.
Currently most of the industrial robot manipulators consist of a robot arm and a simple end-effector or gripper. Apparently the simple end-effectors or grippers lack flexibility for various tasks, especially for fine teleoperation. Only with dexterous hands can robot manipulate objects of different shapes, sizes and stiffness as human being. Although much research work have been done on grasp and manipulation theory [1], [2], kinematics [3], [4] and dynamics [5], [6] of multifingered dexterous hands, the progress in the research of hand mechanism remains slow. Most current dexterous hands [7], [8] have bulky actuation parts and can’t be mounted directly to the ends of robot arms, the rests [9], [10] are portable but not flexible enough to handle different operation requirements.
Therefore, to design a compact hand with high dexterity has always been a big challenge since the emergence of robotics. Aiming to integrate the actuation part with the fingers as a whole and cut off lengthy tendon overheads so as to improve position and force accuracy and portability, we designed a novel compact dexterous hand with high dexterity. In this paper, the notion of perfect dexterous hands is discussed from the viewpoint of degrees of freedom before the mechanism of our proposed compact dexterous hand is described in detail, at last the kinematics and the control scheme are highlighted briefly.
In the traditional hand designs, multi-motors are usually adopted, and lengthy tendons are used for motion transmission. For example, the Utah/MIT hand has 32 pneumatic actuators placed in a remote actuator pack and uses ploymeric tendons and pulleys to transmit the power. So as in the Standford/JPL dexterous hand, 12 electric motors are used as actuators in a remote site and long cables are used to transmit the motion.
0-7803-7203-4/01/$10.00 2001 IEEE
5
2. Perfect Dexterous Hand
along X and Z axes and to rotate objects around Y axis as well. Z
There is no specific definition for dexterous hands presented in literatures. Many researches associate the dexterity problem with particular tasks. Paul Michelman and Peter Allen [12] consider that the dexterous manipulation is driven by task function. Datseris and Palm [13] maintain that dexterous hands should be able to manipulate a class of objects in three rotations and two translations. Nevertheless, in general manipulation problem, it is reasonable to relate the dexterity to the capability of fingertips’ manipulation in arbitrary rotations and translations of a rigid object as in Fuentes and Nelson’s research [14]. The notion of perfect dexterous hand from which the design principles of our hand mechanism derive is presented as the following.
X
Figure 1. Two finger manipulation If another finger with two pitch joints joins in and their relative positions are properly configured as in Figure 2, then the object rotations around X and Z axes can be achieved. At the same time, a stable grasp is guaranteed by three point contact (ideal point contact is assumed between the fingertip and the object). If each finger has at least one yaw joint, then the rest translation of objects along Y axis can be achieved by simultaneous actuation of two pitch joints and one yaw joint, which is demonstrated in Fuentes and Nelson’s experiment [14].
A perfect dexterous hand mimics human hands with high dexterity and it should have basic functions as human hand. The two basic functions of human hands are grasp and manipulation. For the first purpose, it can hold objects in arbitrary orientation tightly with stability. Iberall classifies the grasp behavior of human hand into three categories in terms of “opposition”:1, Pad grasp with force between the pads of the finger and thumb; 2, Palm grasp with force between the fingers and the palm; 3, Side grasp with force between the thumb and the side of the index finger. In Napier’s scheme, grasps are divided into power grasp and precision grasp. Usually palm grasp and side grasp is power grasp while pad grasp belongs to precision grasp [15].
Z X
Figure 2: Three finger manipulation Therefore, one conclusion that can be made is that a perfect dexterous hand must have at least 3 fingers with totally 9 DOF and each finger must have two pitch joints and one yaw joint.
For the second purpose, it can exert arbitrary motion on the objects, which means it can translate objects freely along X, Y and Z axes respectively and rotate objects arbitrarily around X, Y, Z axes within certain space.
Good examples of perfect dexterous hands are the two famous hands, the Utah/MIT hand with 16 DOF and the Stanford/JPL hand with 9 DOF. While the Belgrade/USC hand and the BarrettHand are not perfect dexterous hands. The Stanford/JPL exactly satisfies the minimum requirements for a perfect dexterous hand.
We assume that a perfect dexterous hand must meet all the above requirements. In perfect dexterous hand design, precision grasp is more concerned than power grasp in order to obtain high dexterity and sensitivity. Therefore fingertip manipulation of a rigid object rather than palm manipulation or digit manipulation of fingers is given the highest priority in design process. Since the number of DOF determines the dexterity and complexity of the hand design, then there rises a problem: how many DOF should a perfect dexterous hand have at least?
3. Compact Dexterous Hand Design In our Laboratory, a dexterous hand model has been previously developed with single-motor actuation technology [16]. However it still uses tendons to transfer the remote motion to each finger and has a bulky size as most dexterous hands. Based on this model, our research focuses on compact design by integrating the actuation and control part with fingers as a whole. The benefits of this integration are obvious: a) the lengthy tendon overheads are avoided; b) The whole hand can be installed at robot arm ends directly, no additional support parts are needed. Therefore the hand has high
First, naturally all joints are assumed as rotation joints as those of human hands. If only two fingers are considered as illustrated in Figure 1, apparently two pitch joints of each finger can make objects to do rectilinear motion
0-7803-7203-4/01/$10.00 2001 IEEE
6
proximal digit, one seldom, if not totally, use this joint for object translation along Y axis. Paul Michelman and Peter Allen point out that the human capacity to translate objects by finger motions is quite limited, especially for translations parallel to palm [12]. The main functions of the yaw movement are to adjust the distance between the index and the middle finger and to help to rotate object around Z axis more conveniently. Hence for compact purpose, the yaw joints of both index and middle fingers are given up but that of the thumb is kept. Then with 3 fingers, 7 DOF and arranged in this way as shown in Figure 2., the hand is able to manipulate a rigid object in three rotations and two translations.
adaptability to various robot applications. In this design, the numbers of fingers and DOF are firstly optimized according to the above discussed notion of a perfect dexterous hand and two observations. A new kind of bilateral clutch is designed to reduce the size and weight of the control part, at the same time the compact integration makes the hand more portable. 3.1 Optimization of the number of DOF Generally the more flexibility of the hand we require, the more DOF are needed and the more complex the hand is. A human hand has over 25 DOF and each finger has 4 DOF. Fortunately, most tasks themselves limit the number of DOF. Therefore limited DOF are required to manipulate objects, and others only function to facilitate the completion of tasks. In short, for compact design purpose, the highest priority should be given to optimize the number of DOF while maintaining most static and dynamic characteristics of human hands and manipulations.
3.2 Finger joint actuation Since the index and middle fingers have only two pitch DOF, they are driven by worm gears directly at the finger ends. The worms connect with the output shafts of the clutches protruding out of the actuation box attached behind the fingers. There are two advantages in using worms and worm gears. One is that it can achieve a large motion transmission ratio while occupying the smallest space; the other is that it has self-block attribute (not backdrivable). This attribute provides the essential function to manipulate objects using clutches. When one finger (or joint) is required not to move, the clutches for that finger must be powered off. At this time, due to the worm gears, the finger will not lose contact with the object but maintain it by receiving passive contact force exerted by other fingers. Thin stranded steel ropes are used to transfer motion between digits, from finger ends to middle joints and tip joints. The finger motion transmission is illustrated in the following Figure.
The thumb, middle, and index fingers compose the basic configuration of our dexterous hand, which meets the requirement of a perfect dexterous hand. Three fingers of human hand have totally 12 DOF, but two observations prompt us to reduce the number of DOF from 12 to 7. The first observation is that apparently there is certain relationship between the rotation of the tip digit and that of the middle digit. In natural movement, the middle joint and the tip joint always rotate together. One can not move just one of them but keep the other one stationary without external force or internal force constrain. As shown in the Figure 3., when the middle joint rotates, the tip joint will follow its movement naturally and vice versus.
Base joint
Tip joint
Figure 4. Finger joint actuation
a. Unnatural way b. Natural way Figure 3. Human finger behavior
To simplify mechanical design, the thumb finger has only one yaw joint at the finger base and no passive joint at the fingertip. The yaw joint is directly driven by the worm, in the meantime the middle joint and tip joint are independently driven by steel ropes which route over the base yaw joint to avoid jam with the yaw joint.
According to this observation, the tip joint is made passive to the middle joint in our hand design. In this way, 3 control DOF are omitted while one finger still has 3 active pitch joints and is more anthropomorphic. The ratio between middle joint and tip joint as 2:1, which is similar to that of a human finger.
3.3 Compact bilateral clutch
The second observation is that although each human finger has a yaw joint for lateral movement at the end of
0-7803-7203-4/01/$10.00 2001 IEEE
Middle joint
7
The 7 clutches are the key components of our hand and they occupy most volume of the actuation part. Each clutch controls one joint’s movement. The specially designed compact bilateral clutch mainly consists of an input shaft, an output shaft, an active armature, two coils and two rotors which connect a pulley and a spur gear respectively. The active armature always rotates with the input shaft. Once one coil is powered on, the active armature will be attracted to corresponding rotor and make it rotate too. With the help of pulleys and spur gears, bilateral motion transmission is achieved. The whole clutch size is constrained in a space of about 35x25x70 mm. The clutch structure is illustrated in Figure 5. input shaft pulley
active armature
actuation box index and middle finger thumb finger
Figure 6. Compact dexterous hand model
gear
4. Hand Kinematics Since index and middle finger have identical kinematics, only index and thumb finger’s kinematics are presented here. Define {Ti } and {Bi } are frames attached to the
rotor
coil
finger tip and finger base respectively as shown in Figure 7. The absolute control angles for index and thumb finger are α i , β i and γ m , α m , β m , the corresponding
output shaft
relative joint angles are θ i1 , θ i 2 , θ i 3 and θ m1 , θ m 2 , θ m 3 .
Figure 5. Structure of bilateral clutch
Finger lengths are denoted as lnk , n={i,m} and k={1,2,3}.
3.4 Hand model One motor and 7 clutches serve as the main actuation part. All the clutches and motor are integrated into a cylinder box with a size of 104 mm in diameter and 92 mm in length. The actuation box includes three parts: one DC servo motor, motion distribution part and 7 clutches. A high torque dc servo motor with digital encoder is used. The motor output rotation is distributed to 7 input shafts of the clutches with a star configuration evenly. The output shafts of clutches directly drive the worm gears at each finger base, which greatly reduces the negative effect of long tendons and improves the accuracy. The thumb finger is located exactly in the middle and opposite to the index and middle fingers, so that the thumb fingertip can choose to approach either the index or middle fingertip. Three fingers have the same length of 115mm and the distance between the index and middle finger is 40mm. The base, middle and tip digits of each finger are the same as human scale: 1:0.666:0.444 [17].
Figure 7. Kinematics model of the hand To solve the forward kinematics of the hand, two special relationships, which are determined by the unique mechanism of the hand, should be noticed firstly. The first one is that the tip joint is passive to the middle joint for the index and middle fingers. The second one is that the middle joint and the base joint are both independently controlled. Rotation of one of them will not change the orientation of the digit controlled by the other joint,
Shown in Figure 6 is the proposed compact dexterous hand.
0-7803-7203-4/01/$10.00 2001 IEEE
8
high level includes a Server PC, a image grabber Card and a camera. The Server PC is responsible for image process, hand motion planning and communication with the remote Client PC through LAN. The low level mainly includes a microprocessor 68HC11 to control joint motion of hand and interface with the high level.
which is different from the nature of human hand. These two relationships can be expressed in matrices in (1) and (2).
1 0 − 1 1 0 1 2 θ m1 1 0 θ = 0 1 m2 θ m 3 0 − 1 θ i 1 θ = i2 θ i 3
0 αi 0 β i 0 0
(1)
0 γ m 0 α m 1 β m
(2)
Sensory Card
Camera
ABTmm
0 sin32βi li3sin32βi +li2 sinβi +li1 sinαi 1 0 0 0 cos32βi li3 cos32βi +li2 cosβi +li1 cosαi 0 0 1
cos β m − sin γ sin β m m = cos γ m sin β m 0
0 cos γ m
sin β m − cos γ m cos β m
sin γ m 0
cos γ m cos β m 0
− l m3 sin β m − lm 2 sin α m − lm 3 sin γ m cos β m − lm 2 cos α m sin γ m − l m1 sin γ m lm3 cos γ m cos β m + lm 2 cos α m cos γ m + lm1 cos γ m 1
Dexterous hand
Client PC
Haptic Hand
Operator
Figure 8. Architecture of experiment system In this experiment, the operator uses a haptic hand (a data glove) to send motion command to remote dexterous hand through campus LAN, while the dexterous hand follows the movement and feedback the visual information in real time.
(3)
6. Conclusion and Future Work A novel compact dexterous hand for teleoperation is developed in this paper. The main objective is achieved by adopting one motor actuation and optimized and integrated mechanism. The overall design features make our compact dexterous hand unique in the following ways: 1) Single-motor actuation; 2) Passive joints for the index and middle fingers; 3)multiple compact bilateral clutches; 4) the first compact dexterous hand with 7 DOF.
(4)
5. Control Architecture
Previous experiment preliminarily proves the validity of one motor actuation technology and the hand design. Extensive experiment is undergoing to evaluate the manipulation capability of this hand. Future work also include development of advanced manipulation algorithm, mainly for high-speed manipulation, and addition of force control for this compact dexterous hand.
One DC servo motor with a digital encoder is used as the only power source. The motor works with a constant speed specified through program and the output motion is transmitted to each clutch evenly. The rotation direction of clutch output depends on the clutch’s status. The angular position θ and velocity ω are determined by:
7. References
ω = λω 0 (5) θ = λω 0 t where λ is the engaging time ratio of clutches, t is the operation time and ω 0 is the constant motor speed.
[1] Matthew Mason and Kenneth Salisbury, Robot Hands and the Mechanics of Manipulation, MIT Press, 1985. [2] M. R. Cutkosky, Robotic grasping and fine manipulation, Kluwer Academic Publisher, 1985.
The developed dexterous compact hand is under testing in our teleoperation experiment system. The system control architecture is divided into two main parts, the high level and the low level. As shown in Figure 8, the
0-7803-7203-4/01/$10.00 2001 IEEE
LAN
68HC11
Based on the above relationships, the forward transformation matrices of the index and thumb fingers can be easily deduced in (3) and (4). cos3 βi 2 0 ATBii = −sin3 βi 2 0
Server PC
9
Models” Proceedings of the IEEE International Symposium on Intelligent Control, September, 1996.
[3] D.J. Montana, “The kinematics of multi-fingered manipulation,” IEEE Transactions on Robotics and Automation 11(4) 491-503, 1995.
[15] S.T. Venkataraman and T. Iberall, Dexterous Robot Hands. New York: Springer-Verlag, 1990.
[4] J.K. Salisbury, B. Roth, “Kinematics and force analysis of articulated mechanical hands,” Journal of Mechanisms, Transmissions and Automation in Design 105 35-41, 1983.
[16] Chen W. J., Xie M., “On the Design of a Novel Dexterous Hand,” The 9 th International Conference on Advanced Robotics, Tokyo, Japan, p61-65, October 25-27,1999.
[5] Suguru Arimoto, Pham Thuc Anh Nguyen etc. “Dynamics and control of a set of dual fingers with soft tips,” Robotica vol.18, pp. 71-80. 2000.
[17] Charles D. Engler, Mikell P. Groover, “Design of an anthropomorphic electro-mechanical hand with exoskeletal control to emulate human hand dexterity,” SME technical paper; MS89-274, Society of Manufacturing Engineers, 1989.
[6] Joseph Chan, Yunhui Liu, “Dynamic simulation of multi-fingered robot hands based on a unified model,” Robotics and Autonomous Systems 32 p185-201, 2000 [7] S. C. Jacobsen, B. K. Iverson, D. F. knutti, R. T. Johnson, K. B. Biggers, “Design of the Utah/MIT dexterous hand,” in proceedings of IEEE International Conference on Robotics and Automation, San Francisco, CA, 1986, pp. 15201532. [8] J. K. Salisbury and J. J. Craig, “Articulated hands: Force control and kinematic issues,” International Journal of Robotics Research, vol. 1, pp. 4-17, 1982. [9] R. Tomovic et al., “A strategy for grasp synthesis with multifingered robot hand,” in proceedings of IEEE International Conference On Robotics and Automation, Raleigh, NC 1987, pp.83-89. [10] William T. Townsend, “Description of a Dexterous Robotic Grasper,” Jounal of the Robotics Society of Japan, Vol. 18, No.6 2000. [11] Gongliang Guo, William A. Gruver, and Xikang Qian, “A new design for a dextrous robotic hand mechanism,” IEEE Control systems, August, 1992. [12] P. Michelman and P. Allen, “Compliant manipulation with a dextrous robot hand,” IEEE International Conference on Robotics & Automation, volume 3, pages 711-716, 1993. [13] P. Datseris and W. Palm, “Principles on the development of mechanical hands which can manipulate objects by means of active control,” Transactions of ASME Journal of Mechanisms, Transmissions and Automation in Design, 107: 148156, June 1985. [14] O. Fuentes and R. Nelson, “Experiments on Dextrous Manipulation without Prior Object
0-7803-7203-4/01/$10.00 2001 IEEE
10
