Dexterous Anthropomorphic Robot Hand With Distributed Tactile Sensor: Gifu Hand II

Transcription

1 296 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 7, NO. 3, SEPTEMBER 2002 Dexterous Anthropomorphic Robot Hand With Distributed Tactile Sensor: Gifu Hand II Haruhisa Kawasaki, Tsuneo Komatsu, and Kazunao Uchiyama Abstract This paper presents an anthropomorphic robot hand called the Gifu hand II, which has a thumb and four fingers, all the joints of which are driven by servomotors built into the fingers and the palm. The thumb has four joints with four-degrees-of-freedom (DOF); the other fingers have four joints with 3-DOF; and two axes of the joints near the palm cross orthogonally at one point, as is the case in the human hand. The Gifu hand II can be equipped with six-axes force sensor at each fingertip and a developed distributed tactile sensor with 624 detecting points on its surface. The design concepts and the specifications of the Gifu hand II, the basic characteristics of the tactile sensor, and the pressure distributions at the time of object grasping are described and discussed herein. Our results demonstrate that the Gifu hand II has a high potential to perform dexterous object manipulations like the human hand. Index Terms Force sensor, hand, humanoid, multifinger, robot, tactile sensor. I. INTRODUCTION IT IS HIGHLY expected that forthcoming humanoid robots will execute various complicated tasks via communication with a human user. The humanoid robots will be equipped with anthropomorphic multifingered hands very much like the human hand. We call this a humanoid hand robot. Humanoid hand robots will eventually supplant human labor in the execution of intricate and dangerous tasks in areas such as manufacturing, space, the seabed, and so on. Further, the anthropomorphic hand will be provided as a prosthetic application for handicapped individuals. Many multifingered robot hands (e.g., the Stanford JPL hand by Salisbury et al. [1], the Utah/MIT hand by Jacobsen et al. [2], the JPL four-fingered hand by Jau [3], and the Anthrobot hand by Kyriakopoulos et al. [4]) have been developed. These robot hands are driven by actuators that are located in a place remote from the robot hand frame and connected by tendon cables. The elasticity of the tendon cable causes inaccurate joint angle control, and the long wiring of tendon cables may obstruct the robot motion when the hand is attached to the tip of the robot arm. Moreover, these hands have been problematic commercial products, particularly in terms of maintenance, due to their mechanical complexity. Manuscript received November 10, 2000; revised March 28, Recommended by Technical Editor M. Meng. H. Kawasaki is with the Department of Mechanical and Systems Engineering, Gifu University, Gifu , Japan ( h_kawasa@cc.gifu-u.ac.jp). T. Komatsu was with the Department of Mechanical and Systems Engineering, Gifu University, Gifu , Japan. He is now with Mitsubishi Electric Company, Inazawa , Japan. K. Uchiyama was with the Department of Mechanical and Systems Engineering, Gifu University, Gifu , Japan. He is now with Honda Motor Company Company, Ltd., Hamamatu , Japan. Publisher Item Identifier /TMECH To solve these problems, robot hands in which the actuators are built into the hand (e.g., the Belgrade/USC hand by Venkataraman et al. [5], the Omni hand by Rosheim [6], the NTU hand by Lin et al. [7], and the DLR s hand by Liu et al. [8]) have been developed. However, these hands present a problem in that their movement is unlike that of the human hand because the number of fingers and the number of joints in the fingers are insufficient. Recently, many reports on the use of the tactile sensor [9] [13] have been presented, all of which attempted to realize adequate object manipulation involving contact with the finger and palm. The development of the hand, which combines a 6-axial force sensor attached at the fingertip and a distributed tactile sensor mounted on the hand surface, has been slight. Our group developed the Gifu hand I [14], [15], a five-fingered hand driven by built-in servomotors. We investigated the hand s potential, basing the platform of the study on dexterous grasping and manipulation of objects. Because it had a nonnegligible backlash in the gear transmission, we redesigned the anthropomorphic robot hand based on the finite element analysis to reduce the backlash and enhance the output torque. We call this version the Gifu hand II. In this paper, we present design concepts and mechanical specifications of the Gifu hand II equipped with a distributed tactile sensor. The Gifu hand II has a thumb and four fingers; the thumb has four joints with four-degrees-of-freedom (DOF) and the finger has four joints with 3-DOF; and the two joint axes of the thumb and the finger near the palm are orthogonal. Moreover, a developed distributed tactile sensor with 624 sensing points can be attached to the hand s surface and it is equipped with a six-axes force sensor at each fingertip. The design concepts and the specifications of the Gifu hand II, the basic characteristics of the tactile sensor, and the pressure distributions at object grasping are described and discussed herein. Our results show that the Gifu hand II has a high potential to perform dexterous object manipulations like the human hand. II. ROBOT HAND DESIGN An overview of the developed anthropomorphic right and left version of the Gifu hand II is shown in Fig. 1, in which the right hand is equipped with force sensors and tactile sensors. The right and left hands are designed symmetrically and have a thumb and four fingers. The design mechanism of the thumb and finger are shown in Fig. 2 and, respectively. The structure of the left hand is shown in Fig. 3. Servomotors and joints are numbered from the palm to the fingertip. The thumb has four joints with 4-DOF and the fingers have four joints with 3-DOF. Movement of the first joint of the thumb and of the fingers allows adduction and abduction, and that of the second /02$ IEEE

2 KAWASAKI et al.: DEXTEROUS ANTHROPOMORPHIC ROBOT HAND: GIFU HAND II 297 Fig. 1. Fig. 2. Developed Gifu hand II. Mechanism of the thumb and the finger. Thumb. Finger. joint to fourth joint allows anteflexion and retroflexion. The main difference between the thumb and the fingers is that the fourth joint of the fingers is actuated by the third servomotor through a planar four-bar linkage mechanism. Thus, the Gifu hand II has 20 joints with 16-DOF. Each servomotor (Maxon dc motor, made by Interelectric AG) has a magnetic encoder with 16 pulses per revolution. Table I summarizes the characteristics of the Gifu hand II. A. Design Concept We conducted this study using the Gifu hand II as a platform for research on dexterous grasping and manipulation of the multifingered hand. The hand is designed to be compact, lightweight, and anthropomorphic in terms of geometry and size, such that it performs grasping and manipulations like the human hand. The design concept is as follows: 1) Size: It is desirable for the robot hand to resemble the human hand in size and geometry for purposes of skillful manipulation. The robot hand was designed to be similar to a relatively large human hand, and has a thumb and four fingers. 2) Number of Joints and Number of DOF: In a human hand, both the thumb and fingers have four joints [6, p. 192]. It is difficult for humans to control the outermost two joints of the fingers independently, because the fourth joint engages with the third joint. However, humans can control the joint angles of the thumb almost independently. The thumb is more dexterous and powerful than the fingers. The independent joint needs an independent actuator in the robot hand. This makes it hard to design a light weight hand. The finger can be modeled as a link mechanism with four joints and 3-DOF, and the thumb can be modeled as a link mechanism with four joints and 4-DOF. Controlling the finger made with coupled joints may be more difficult than controlling the finger made with all independent joints, in terms of grasping and manipulation. However, the finger made with coupled joints, which has more links than the finger made with independent joints, can grasp and manipulate more objects of various shapes than the finger made with fewer links. This is due to the fact that the area for grasping in the finger made with coupled joints is larger than that of the grasping area of the finger made with fewer links. Therefore, coupling will augment the dexterity of the hand. The number of joints and number of DOF of the robot hand were designed to mimic those of the human hand. The thumb is actuated by four servomotors and the fingers actuated by three servomotors. The fourth joint of the fingers are driven by the third servomotor through a planar four-bar linkage mechanism. The first joint and the second joint of human finger cross almost orthogonally at one point. Hence, the hand was designed such that the first joint and the second joint of each finger cross orthogonally at one point by means of an asymmetrical differential gear. Moreover, the asymmetrical differential gear enables the second joint axis to be placed near the surface of the palm, which make an effect to resemble a finger motion of the human. 3) Opposability of the Thumb: The thumb of the human hand can move in opposition to the fingers. Dexterity of the human hand in object manipulation is caused by this opposability. The robot hand was designed such that it has an opposable thumb. 4) Built-In Servomotor: For easy attachment to the robot arm, the robot hand was designed such that all joints are driven by built-in dc servomotors with a rotary encoder. To produce a high stiff hand, the transmission system was created by using high stiff gears such as a satellite gear and a face gear instead of low stiff gears such as a harmonic drive gear, and without using tendon cable. 5) Unit Design: Easy maintenance and easy manufacture of the robot hand are very important, so each joint was designed as a module and each finger was designed as a unit. Due to the unit design of the finger, hands having from two to five fingers are easily made. 6) Force Sensor: Each finger of the robot hand was designed to be equipped with a six-axes force sensor (nano sensor made by BL. AUTOTEC Company) for compliant pinching. 7) Distributed Tactile Sensor: There are many sense organs in the human hand. These permit the human hand to manipulate an object dexterously. It is expected that more tactile sensors

3 298 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 7, NO. 3, SEPTEMBER 2002 Fig. 3. Structure of the left hand. TABLE I SPECIFICATIONS OF THE GIFU HAND II enable more dexterous manipulations. The robot hand was designed to be mounted with a developed distributed tactile sensor with 624 detecting points. 8) Wiring: Wiring is important in robotic mechanisms. All of the wires in the motor, force sensor, and tactile sensor should not prevent object manipulation by the robot hand. We designed all of the wires to be located along the back of each finger and palm. The hard wire used in the commercialized force sensor was changed to a soft wire, so as not to cause an external force to arise due to motion in the hard wire. B. Improvement of Hand Mechanism The mechanism of the Gifu hand II is a dramatic improvement over the Gifu hand I [14], in the following ways. 1) Reduction of Backlash: The Gifu hand I employed an asymmetrical differential reduction gear consisting of bevel gears at the first and second joints of the thumb and fingers. This asymmetrical differential gear enabled the second joint to Fig. 4. Frequency characteristics of the first joint of the thumb. be placed near the surface of the palm and the axes of the first joint and the second joint to be orthogonal. However, a backlash occurred because the axial force of the bevel gear moved the shaft of the bevel gear. Moreover, abrasion of the shank made by the aluminum material increased the backlash. To reduce the backlash in the Gifu hand II, face gear, in which the axial force is not generated, was adopted instead of bevel gear, and the frame material was change to the titanium alloy, which shows strength and excellent performance against abrasion. By these modifications, the backlash of the first joint decreased from 8 to 1.0. The backlashes of the other joints also decreased by nearly the same amount. 2) Higher Output Torque and Higher Response: For higher output torque, a higher-power motor and high-reduction ratio were adopted at the first and second joints. As a result, output torques of the first joint and the second joint are 3.46 Nm. The output force at the fingertip of the thumb is 4.9 N, which is about

4 KAWASAKI et al.: DEXTEROUS ANTHROPOMORPHIC ROBOT HAND: GIFU HAND II 299 Fig. 5. Finite element analysis of No. 1 gear-head housing. Equivalent stress distribution. Deformation. TABLE II EXAMPLES OF SAFETY FACTOR. NO. 1: GEAR-HEAD HOUSING OF FIRST AND SECOND JOINTS. NO. 2: FORCE-SENSOR BRACKET. NO. 3: GEAR-HEAD HOUSING OF THIRD JOINT eight times that of the Gifu hand I. The frequency gain characteristics of the first joint of the thumb in velocity feedback control is shown in Fig. 4. Its measurement was made by setting the first joint axis vertical in order to ignore the effect of gravity. The velocity signal of the joint was obtained by the digital differential with regard to the position signal. The velocity feedback gain was set such that a proportional and differential (PD) control with a desired angle of 15 generates no vibration; the results of this are shown later. The bandwidth is 8.6 Hz. The other joints were also measured similarly, and the results are shown in Table I. The minimum bandwidth of the robot hand is 7.5 Hz, which exceeds the responsibility of the human finger, the bandwidth of which is, at most, 5.5 Hz. This means that the robot hand can move more quickly than the human hand. 3) Higher Stiffness: By stress analysis using a finite element analysis system, the regulation of the safety factor of the mechanism was attempted. On the condition that the largest force acting on the fingertip is 12 N, the force affecting each part impacted and the safety factor of each part were calculated. After reviewing the part or parts that had too low or too high a safety factor, the thickness and form of the relevant parts were revised. Table II shows examples of the safety factor analysis. We assumed that a standard safety factor was 10. In the Gifu hand I, the safety factors of some parts were less 10 or over 50. As an Fig. 6. Distributed tactile sensor for the Gifu hand II. example, results of the finite element analysis of the No. 1 gearhead housing in Gifu hand I are shown in Fig. 5. Fig. 5 and are an equivalent stress distribution diagram and deformation figure, respectively. Redesign reduced the maximum stress of the gear-head housing from 120 to 26 Mpa. The maximum deformation was also reduced from to mm. The Gifu hand II was greatly improved in terms of the safety factor and deemed to have good balance. III. DISTRIBUTED TACTILE SENSOR The distributed tactile sensor for the Gifu hand II was developed with the cooperation of Nitta Corporation. The shape of the tactile sensor is shown in Fig. 6. Tactile sensors are distributed on the surface of the fingers and palm. The thumb and the fingers of Gifu hand II each consist of four links and four

5 300 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 7, NO. 3, SEPTEMBER 2002 TABLE III CHARACTERISTICS OF DISTRIBUTED TACTILE SENSOR Fig. 7. Hysteresis characteristic curve of tactile sensor. joints, in which the base link is mounted in the palm, and other three links are equipped with the tactile sensors. The base link and the other links of the fingers of Gifu hand II correspond to the metacarpal bone and three phalanges of a human hand. In a human hand, the metacarpal bone is located in the palm, and is barely mobile. However, in Gifu hand II, the base link and the other three links of the thumb correspond to the carpal bone, metacarpal bone, and two phalanges of a human thumb, with the difference being that the metacarpal bone is mounted in the palm and can move independently, and the carpal bone is immobile. Hence, three parts of the finger and thumb are equipped with the tactile sensors. The tactile sensor has a grid pattern electrode and uses conductive ink in which the electric resistance changes in proportion to the pressure on the top and bottom of a thin film. Characteristics of the tactile sensor are shown in Table III. Numbers detecting points on the palm, the thumb, and the finger are 312, 72, and 60, respectively, and the total number of measurement points is 624. Electrode width is 2 mm, column pitch is 4 mm, and row pitch is 6 mm. Maximum load is about 74 kn/m, resolution of measurement is 8 bits, and sampling cycle is 10 ms/frame. A response characteristic of the tactile sensor is more than 1 khz. Detected data on the sensor sheet is transported to a PC through a special interface board. This tactile sensor has nonlinear characteristics such as hysteresis, creep, and dispersion in measurement sensitivity. The hysteresis characteristic curve and creep characteristic curve at three arbitrarily selected detecting points are shown in Fig. 7 and Fig. 8, respectively. Vertical axes in these figures are output of the tactile sensor through an analog-to-digital (A/D) converter. These figures also show the dispersion in measurement sensitivity. Signal processing taking such nonlinear characteristics into account is required. Fig. 9 shows an overview of the 6-DOF robot arm (VS6354B, made by DENSO Company) and the Gifu hand II that is equipped with the six-axes force sensor at each fingertip and the developed tactile sensor. The Gifu hand II can be easily attached to another industrial robot by changing a base plate of the hand. Fig. 10 shows computer graphics of output patterns of the tactile sensor, measured while Gifu hand II is holding a soft spherical object 95 mm in diameter, and a hard cylindrical object 66 mm in diameter. The height of the pole listed in Fig. 10 refers to the output level. Some outputs may Fig. 8. Fig. 9. sensor. Creep characteristic curve of tactile sensor. Gifu hand II with five six-axes force sensors and the distributed tactile include noise because of an inadequacy of attachment between the tactile sensor and the hand frame. The extended output patterns of the tactile sensor are shown in Fig. 11. The vertical line in Fig. 11 is the output of the tactile sensor of which the unit is millinewtons. The points on the horizontal plane are the measurement points of the tactile sensor, which are opened to show the measurement data of the force. It is clear that the

6 KAWASAKI et al.: DEXTEROUS ANTHROPOMORPHIC ROBOT HAND: GIFU HAND II 301 Fig. 10. Computer graphics of at object grasping. Spherical soft object. Cylindrical hard object. Fig. 11. Output patterns of tactile sensor by the Gifu hand II. Spherical object. Cylindrical object. Fig. 12. Output patterns of tactile sensor by the human hand. Spherical object. Cylindrical object. output pattern depends on the shape of the object. When the human hand grasps the same objects using the globe attached to the tactile sensor, the output patterns of the tactile sensor are those shown in Fig. 12. In this case, a part of the tactile sensor was cut to fit the size of the human hand. The output pattern by the human hand differs from that by the robot hand in part because the human hand is covered with soft derma. However, these results show that the Gifu hand with the distributed tactile sensor has a high potential for dexterous grasping and manipulation. IV. EXPERIMENTS We constructed a PC-based robot hand control system as shown in Fig. 13. Four 4ch counter boards, two 8ch digital-to-analog (D/A) boards, four 16ch A/D boards, and a timer board are connected to the PCI bus. Signals of D/A are inputted to servomotors through a 16ch linear amplifier. The operating system of the PC is Windows98. A tactile sensor is connected to another PC for measurement and display, and the maximum sampling cycle of detecting the 624 points is 100 Hz. The force

7 302 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 7, NO. 3, SEPTEMBER 2002 Fig. 13. Hand control system. Fig. 15. Experiment of contact task. Fig. 16. force. Responses of contact task. Errors of joint position. Contact (c) (d) Fig. 14. Step responses of the thumb. First joint. Second joint. (c) Third joint. (d) Fourth joint. resolutions coupled with the A/D converter are 32, 32, and 98 mn along the,, and axes, and the moment resolutions are 0.20, 0.20, and 0.26 m N m at about the,, and axes, respectively. The Coulomb frictions of the first joint to the fourth joint of the thumb are 3.01, 0.77, 1.30, and 0.46 m N m, respectively. As shown in Table I, these Coulomb frictions are at most 1.5% of the output torque of the thumb. Step response of the joints of the thumb is shown in Fig. 14. Each joint angle moved from 0 rad to the desired joint angle at 0.26 rad in at least 0.15 s. The rising times of the first joint and the second joint are larger than those of the third joint and fourth joint due to saturation of servomotor input, which is caused by high gear reduction ratio. The results show that the robot finger moved more quickly than the human finger, the response time of which is 5.5 Hz, at most. We examined the characteristics of force control during a contact task by using the thumb as shown in Fig. 15. Each joint position error and contact force are presented in Fig. 16. The initial position of the tip of the thumb was 105 mm from a constraint surface and a desired joint position was given by a 5 polynomial in time, which interpolates between the initial joint position and the terminal joint position within 0.5 s. A desired contact force in the anti-normal direction of the constraint surface was given by a step with 0.49 N after contact. In the noncontact state, proportional, integral, and differential (PID) control was adopted as a joint position control, and in the contact state a position and force hybrid control [16] was adopted. The sampling cycle was 2 ms. This figure shows that the contact force converges to the desired value with a rapid response.

8 KAWASAKI et al.: DEXTEROUS ANTHROPOMORPHIC ROBOT HAND: GIFU HAND II 303 V. CONCLUSION Anthropomorphic robot hands have been actively developed during past two decades. However, development of a five-fingered hand with built-in actuators, and equipped with a force sensor at the fingertips and a tactile sensor on the fingers and palm has only demonstrated in the NTU hand [7], in which the bandwidth of the first joint of the thumb is 0.15 Hz, and the number of detecting points of the tactile sensor is 18. The Gifu hand II has a much higher number of sensors and higher response than any previously developed robot hand, and can move more quickly than the human hand. We consider that the Gifu hand II is useful as a research tool for dexterous robot manipulation using force sense and tactile sense. We have presented herein the design concept and the mechanism features of the Gifu hand II, which is designed to be used as a standard anthropomorphic robot hand. The Gifu hand II is actuated by built-in servomotors, distributed tactile sensors can be attached to its surface and it is equipped with a six-axes force sensor at each fingertip. The mobile space and geometrical size of the fingers are similar to those of the fingers on the human hand. We intend to use the Gifu hand II for future study of dexterous manipulation by the robot arm. ACKNOWLEDGMENT The authors would like to thank all members of the Gifu Robotic Hand Study Group for their support. [11] M. Shimojo, S. Sato, Y. Seki, and A. Takahashi, A system for simulating measuring grasping posture and pressure distribution, in Proc. IEEE Int. Conf. Robotics and Automation, 1995, pp [12] D. Johnston, P. Zhang, J. Hollerbach, and S. Jacobsen, A full tactile sensing suite for dextrous robot hands and use in contact force control, in Proc. IEEE Int. Conf. Robotics and Automation, 1996, pp [13] J. Jockusch, J. Walter, and H. Ritter, A tactile sensor system for a three-fingered robot manipulator, in Proc. IEEE Int. Conf. Robotics and Automation, 1997, pp [14] H. Kawasaki and T. Komatsu, Development of an anthropomorphic robot hand driven by built-in servo-motors, in Proc. 3rd Int. Conf. ICAM, vol. 1, 1998, pp [15], Mechanism design of anthropomorphic robot hand: Gifu hand I, J. Robot. Mechatron., vol. 11, no. 4, pp , [16] M. H. Raibert and J. J. Craig, Hybrid position/force control of manipulators, Trans. ASME, vol. 102, pp , Haruhisa Kawasaki was born in Gifu, Japan, in He received the M.S. and Ph.D. degrees from Nagoya University, Nagoya, Japan, in 1974 and 1986, respectively. From 1974 to 1990, he was a Research Engineer at NTT s Laboratories, Tokyo, Japan. From 1990 to 1994, he was a Professor at Kanazawa Institute of Technology, Kanazawa, Japan. Since 1994, he has been a Professor of Engineering at Gifu University. From July 1998 to January 1999, he was a Guest Professor at the University of Surrey, Surrey, U.K. He is on the Editorial Board of Journal of Robotics and Mechatronics. His research interests include robot control, humanoid robot hand, robot teaching in virtual reality environment, and computer algebra of robotics. REFERENCES [1] J. K. Salisbury and J. J. Craig, Articulated hands: Force control and kinematic issues, Int. J. Robot. Res., vol. 1, no. 1, pp. 4 17, [2] S. C. Jacobsen et al., The Utah/MIT dexterous hand: Work in progress, Int. J. Robot. Res., vol. 3, no. 4, pp , [3] B. M. Jau, Dexterous telemanipulation with four fingered hand system, in Proc. IEEE Robotics and Automation, 1995, pp [4] K. J. Kyriakopoulos, A. Zink, and H. E. Stephanou, Kinematic analysis and position/force control of the anthrobot dextrous hand, IEEE Trans. Syst., Man Cybern. B, vol. 27, pp , Feb [5] G. A. Bekey, R. Tomovic, and I. Zeljkovic, Control architecture for the Bergrade/USC hand, in Dexterous Robot Hand, S. T. Venkataraman and T. Iberall, Eds. New York: Springer Verlag, 1990, pp [6] M. Rosheim, Robot Evolution. New York: Wiley, 1994, pp [7] L. R. Lin and H. P. Huang, Integrating fuzzy control of the dexterous National Taiwan University (NTU) hand, IEEE/ASME Trans. Mechatron., vol. 1, pp , Sept [8] H. Liu, J. Butterfass, S. Knoch, P. Meusel, and G. Hirzinger, Multisensory articulated hand, IEEE Control Syst. Mag., vol. 19, pp , Apr [9] R. S. Fearing, Tactile sensing mechanisms, Int. J. Robot. Res., vol. 9, no. 3, pp. 3 23, [10] R. D. Howe, Tactile sensing and control of robotic manipulation, Adv. Robot., vol. 8, no. 3, pp , Tsuneo Komatsu received the B.S. and the M.S. degrees from Gifu University, Gifu, Japan, in 1997 and 1999, respectively. Since 1999, he has been working at Mitsubishi Electric Company, Inazawa, Japan. Kazunao Uchiyama received the B.S. and the M.S. degrees from Gifu University, Gifu, Japan, in 1998 and 2000, respectively. Since 2000, he has been working at Honda Motor Company, Ltd., Hamamatu, Japan.