Anda di halaman 1dari 10

Mechatronics 11 (2001) 1053±1062

Technical Note
A new approach for the design of robot joint
transmission
P.J. Turner *, P. Nigrowsky, G. Vines
Department of Electronic and Computer Engineering, Brunel University, Uxbridge UB8 3PH, UK
Received 4 April 1999; received in revised form 17 April 2000; accepted 5 May 2000

Abstract

Until recently the only alternative to traditional transmission techniques for robot ma-
nipulators was direct-drive (DD), and associated transmission mechanisms. These method-
ologies were developed to overcome the inherent problems of friction, backlash and
compliance of gearboxes. The DD concept is however not ideal as the absence of a reduction
ratio, means that dynamic coupling and joint interaction become important at the motor
shaft. An alternative approach that keeps the directness of DD, while providing dynamic
coupling and joint interaction attenuation, is the non-linear direct transmission. The new
concept is applied to the prototype SPRINTA robot and its performance demonstrates the
potential of the non-linear direct transmission philosophy. Ó 2001 Elsevier Science Ltd. All
rights reserved.

Keywords: Non-linear direct transmission; Direct-drive; Robotics

1. Introduction

Traditional robotic arms use gearboxes as transmission elements between the


actuators and the joints, which means that the load seen by the motor is reduced and
the motor can be relocated from the arm. The drawback of this method is the
presence of friction and either backlash or compliance in the gearbox. As robotic
applications in the manufacturing industries demand more and more accurate
and faster manipulations, the limitations imposed by the use of gearboxes become
unacceptable, see [1,2], for a comprehensive study on robot transmissions. The same

*
Corresponding author. Fax: +44-0-1895-258-728.
E-mail addresses: dr.p.j.turner@brunel.ac.uk (P.J. Turner), pierre.nigrowsky@fr.bosch.com
(P. Nigrowsky).

0957-4158/01/$ - see front matter Ó 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 5 7 - 4 1 5 8 ( 0 0 ) 0 0 0 5 3 - 2
1054 P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062

argument holds for harmonic drives [3], even though, they diverge from traditional
gearboxes. The development of low speed and high torque motors [4,5], means that
there is no longer a need for a gearbox that magni®es the torque and reduces the
speed between the motor and the joint. A design approach, called direct-drive (DD),
which is based on the use of these special motors has been proposed some 15 years
ago [6,7]. In this approach, the shafts of the motors are directly coupled to the joints
of the manipulator. Friction, backlash and compliance associated with the gearboxes
are then eliminated. The CMU DD Arm model I [8] and the MIT I serial DDarm [9]
are robots designed on the DD concept. In its strict form, DD presents serious
limitations due to the motor shaft being directly connected to the joint so that it must
be mounted on the link, thereby reducing the payload. To avoid this extra loading on
the link, a transmission element is introduced that keeps as much directness as
possible while allowing motor relocation. Robotic arms using this technique diverge
from pure DD as a transmission element is allowed to be introduced between the
shaft of the motor and the joint to be actuated. This transmission element provides
the maximum relocation freedom whilst introducing the minimum of detrimental
e€ects associated with conventional transmissions. No distinction is made however
between this technique and DD, and it is commonly referred to as DD. Vines [17]
proposes to distinguish the two and refers to those that use some form of trans-
mission as direct transmission (DT).
Robot manipulators using DTs still need torque motors, as the directness of the
transmission is kept. A number of robot manipulators using this technique have been
reported; the MIT DD model II, III and IV [6,10], the Minnesota DDarm [12],
the Adept-One [11] and the DD SCARA [13]. The three MIT models as well as the
Minnesota DDarm use linkage mechanisms as transmission elements, to relocate the
motors from the links and to keep as much directness as possible. The Adept-One
and the DD SCARA involve a system of pulleys and steel bands as DT elements.
The DT philosophy has been proposed to achieve a greater level of actuator re-
location freedom than is possible under the DD concept, but without introducing the
majority of the negative aspects of the conventional transmissions. The enhanced
relocation allows reduced actuator mobility, suggesting higher performance than for
DD.

2. The non-linear direct transmission

The DD and DT have some serious consequences on the control method. Since
the torque reducting e€ect of the gearbox is absent, dynamic complexity, such as
coupling and non-linearities are directly re¯ected at the motor shafts. Furthermore,
higher resolution feedback sensors are needed because of the lack of a reduction
ratio. The SISO control technique which is commonly used for traditional manip-
ulators is not appropriate anymore to deal with the high level of dynamic complexity
re¯ected at the motor shafts [10]. More advanced control methods such as model-
based [9,15] or computed torque control [14] are needed. Another method to deal
with dynamic complexity, is to re-design the manipulator in such a way that the
P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062 1055

dynamic complexity is kept to a minimum. This has been done by redistributing the
masses and reformulating the structure of the manipulator. Takase et al. [16] reduced
the gravitational torque by designing an arm with a counterweight but this increases
arm inertia, and thus reduces the acceleration potential. The Minnesota DD robot
[12] uses a four-bar linkage with a speci®c design structure so that the functional
moving parts are balanced. This design arrangement eliminates gravitation torque
thus requiring smaller actuators and ampli®ers and a simpler controller. This bal-
ancing however is only valid for robots having the same speci®c geometrical struc-
ture.
Until now, the complexity introduced by DD and DT has been tackled using
more complex controller algorithms or from a mechanical re-design of the manip-
ulator. Vines [17] proposes a new design, that combines the advantages of the DT
philosophy with the reduction ratio property of gearboxes. This design is based on
the use of a speci®c type of transmission, referred to as a non-linear direct trans-
mission. The non-linear direct transmission keeps the directness of DD whilst pro-
viding a varying reduction ratio. Such transmissions provide all the advantages of
the DT, i.e., no friction, backlash and compliance, whilst providing motor reloca-
tion. The varying reduction ratio can also allow some compensation for the variation
in joint requirements. Depending on the task, the torque and the speed, as well as the
accuracy requirements, vary within the workspace. The dynamic complexity which
dominates in DD and DT, can also be attenuated by providing a reduction ratio. For
example when the manipulator is in a position where the gravitational torque is
signi®cant, the non-linear transmission can be designed such that the load on the
motor is reduced.
A number of mechanical devices satisfy the criteria of a non-linear transmission
topology, for example a non-parallel linkage [18], the Peaucellier linkages [18, p.
181], bands with elliptical pulleys [19]. Vines [17] proposes the gimbal drive (Fig. 1),
which is an extension of the sinusoidal reciprocator [20, p. 83] as a non-linear
transmission element for robot manipulators. Fig. 1(a) represents the model of the
gimbal drive transmission, while Fig. 1(b) is the prototype transmission which has
been built to investigate the transmission properties.

2.1. The gimbal drive

The gimbal drive consists of a truncated cylinder, a rocking shaft (outer ring) and
an inner ring. The input of the system is given by rotating the truncated cylinder and
the output is obtained by the swing of the outer ring. The gimbal drive transmission
has the following matching function, Eq. (1) and see Fig. 2
1
hout ˆ hof out ‡ tan … tan hc  cos hin †; …1†

where hout is the output angle, hof out the o€set angle at the output, hc the gradient of
the truncated cylinder, and hin is the input angle. The matching function of the
gimbal drive has the following features: non-linearity, smoothness, bounded output
travel, cyclic nature and quadrant symmetry.
1056 P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062

Fig. 1. The gimbal drive.

Fig. 2. Transfer characteristics of the gimbal drives.


P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062 1057

The reduction ratio (Krr ) for the gimbal drive can be found from the derivative of
the matching function, so di€erentiating Eq. (1) gives:

dhout …cos hin  tan hc †2 ‡ 1


Krr …hin † ˆ ˆ : …2†
dhin sin hin  tan hc

Fig. 3 shows the reduction ratio against both the input angle hin and the output angle
hout . It appears that Krr ! 1 as hc ! nP, where n is an integer. Thus the gradient hc
appears as a design parameter that de®nes the worst Krr . For values of hc less than
45°, the reduction ratio is always greater than 1. The variation of Krr with angle
position (either input or output) clearly underlines the potential for gravity com-
pensation. When applying the gimbal drive, the transmission must be o€set so that
maximum reduction ratio occurs when the arm is in positions where the gravity
loading is signi®cant.
With such a non-linear relationship between the input and the output of the
transmission, the joint distribution of the resolution of a motor position sensor is
distorted. Regions of high resolution are created at the expense of poorer resolution
in other areas. Fig. 4 indicates the resolution distribution for the gimbal drive with a
gradient hc ˆ 45°. This variable resolution matches the needs of robotic applications
like pick and place, i.e., enhanced resolution is desirable at sites of interaction,
poorer resolution is acceptable between such sites.
The gimbal drive is a novel device which completely satis®es the requirement of a
non-linear direct transmission. The non-linear relation between the input and output
angle provides some reduction ratio in some de®ned regions, thus reducing the joint
coupling as seen by the motor. Also gravity compensation can be introduced if such
reduction ratios occur in positions where the gravity loading is signi®cant. The direct

Fig. 3. The varying reduction ratio of the gimbal drive: (a) reduction ratio vs input angle; (b)reduction
ratio vs output angle.
1058 P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062

Fig. 4. Resolution distribution for hc ˆ 45°.

nature of the transmission means that there is no backlash and compliance, and very
low friction.

2.2. The SPRINTA prototype

The non-linear direct transmission philosophy has been evaluated via the gimbal
drive on the SPRINTA (Serial Parallel Robot Incorporating Non-linear Transmis-
sioned Actuation) arm, Fig. 5. The SPRINTA prototype is not being developed for a
particular application, hence a three DOF anthropomorphic like topology was se-
lected.
The SPRINTA robot uses two gimbal drive elements, one for the actuation of the
elbow (last link) and the other for the shoulder (middle link). The waist (base) is
directly couple to the shaft of the motor, forming a pure DD element. Most of the
robot structure is made of aluminium to reduce the link mass and inertia. The rest is
made of standard mechanical elements, as a deliberate design choice to demonstrate
the potential of the non-linear direct transmission.
Three NSK Megatorqueâ motors are used to action the joints, two RS0608FN003
and one RS0408FN003 [21]. Each motor has its own driver which is con®gured so that
it responds to the command data with a torque proportional to the maximum torque
available. The command signals are provided by three interface cards which are
connected to the QT960 controller board [22]. The motor positions are sensed by
incremental encoders, producing 614 400 pulses per revolution. Interface cards are
used to feedback the position information to the QT960. The controller board is based
on the INTEL 80960KB operating at 20 MHz, which is equipped with 128K of RAM
and 128K of FLASH EPROM. Since the QT960 has neither ®le storage nor operating
system, it is connected via a RS-232 serial port to a PC which acts as a terminal. File
P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062 1059

Fig. 5. The SPRINTA.

handling and software development are performed on the PC. The software is written
entirely in ANSI`C' and the core routines is written in the form of a `C' library.
The control algorithm which is presently used on the SPRINTA is based on in-
dependent joint PID control, with no dynamic modelling included. The algorithm is
implemented on the QT960 controller board with a sample rate of 1 ms.
The SPRINTA prototype obtained a static repeatability as low as 10 lm. This
measurement is obtained by viewing a target mounted upon the endpoint from a
stationary microscope. The test is repeated for various arm con®gurations, and it is
observed that the repeatability is dependent upon the arm's position. These re-
peatability tests do not represent a rigorous measurement of the arm's performance
but indicate some of the potential of the design approach.
1060 P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062

Fig. 6. End point motion for the goalpost test: (a) plot of reconstructed demand path; (b) actual endpoint
motion.

The dynamic performance is assessed by means of the industry standard goalpost


test. The test indicates the maximum speed of motion a robot arm can perform along
a path similar to a `pick and place' action, see Fig. 6(a). The path is 300 mm across
and 25 mm height. The arm is capable of executing the goalpost test (forward and
return movement) in 0.7 s in a permanent cyclic manner. The assessment of the
dynamic tracking is performed at the maximum speed (0.7 s) at which the goalpost
test can be executed. This speed of execution is simply limited by the maximum speed
setting of 1.25 rev/s for the motors. This setting can be updated to 3 rev/s which
means that the robot can physically respond to faster trajectory demands.
Fig. 6(b) shows the photographic record of the actual endpoint path. The actual
path is close to the demand one. The dynamic tracking error on the endpoint
goalpost path presents a maximum deviation less than 40 mm horizontally and less
than 10 mm vertically. The deviation in the path is due to the simplicity of the
control algorithm used (independent joint PID), with advanced control methods
however these performances can be greatly improved [23].

3. Conclusion

A new design philosophy for robot transmissions is presented, and illustrated with
a speci®c example, namely the gimbal drive. The new approach for robot trans-
mission overcomes the problems associated with conventional transmissions (i.e.,
backlash, friction, compliance) and with DD (i.e., motor directly coupled to joint).
Further, the non-linear direct transmission philosophy provides new features to the
P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062 1061

overall robot manipulator. As illustrated with the gimbal drive, a variable input±
output ratio is obtained which allows a reduction in the dynamic coupling at the
motor and compensation for gravity loading in particular arm positions.
The gimbal drive as an example of a non-linear direct transmission element for
robot manipulators allows the relocation of the motors from the link to the base of
the manipulator, and provides gravity compensation as well as dynamic decoupling
for operating positions. With position sensors mounted at the motors, regions of
higher resolution occur due to the varying input±output ratio of the transmission.
This varying resolution ratio combined with an adequate design can be very useful.
It can give increased resolving power in locations where high accuracy is required
without having to use more precise sensors. The implementation of the gimbal drive
on a robot manipulator results in a compact and elegant design. The performance
for the ®rst prototype is excellent, the SPRINTA can execute the goalpost test in 0.7
s and has a repeatability less than 0.01 mm. Such performances are remarkable and
can be compared to those of the Adept-One from Adept Technology, Inc. This robot
performs the goalpost test in 1.0 s [24], while its static repeatability is 0.025 mm in
the horizontal plane and 0.038 mm in the vertical plane [11].

Acknowledgements

The authors would like to thank the EPSRC and the Department of Electronic
and Computer Engineering, Brunel University for supporting this research.

References

[1] Schempf H. Comparative design, modeling, and control analysis of robotic transmissions. Ph.D.
Thesis, Woods Hole Oceanographic Institute, MIT, 1990.
[2] Schempf H, Yoerger DR. Study of dominant performance characteristics in robot transmissions.
J Mech Des 1993;115(3):472±82.
[3] Kircanski NM, Goldenberg AA. Experimental study of nonlinear sti€ness, hysteresis, and friction
e€ects in robot joints with harmonic drives and torque sensors. Int J Robot Res 1997;16(2):214±39.
[4] Welburn R. Ultra high torque motor system for direct drive robotics. In: Robot, vol. 8. Detroit; 1984.
p. 19±71.
[5] Pal SK. Direct drive high energy permanent magnet brush and brushless DC motors for robotic
applications. In: IEE Colloquium in Robot Actuators, Digest No. 1991/146. London: Savoy Place;
October 1991.
[6] Asada H, Youcef-Toumi K. Direct-drive robots: theory and practice. Cambridge, MA: MIT Press;
1987.
[7] Davis S, Chen D. High performance brushless DC motors for direct drive robot arm. J Power-
conversion Intelligent Motion 1985;11(8):34±8.
[8] Asada H, Kanade T. Design of a direct-drive mechanical arm. Technical Report CMU-RI-TR-81-1,
Robotics Institute, Carnegie Mellon University, 1981.
[9] An CH, Atkeson CG, Hollerbach JM. Model-based control of a robot manipulator. Cambridge, MA:
MIT Press; 1988.
[10] Asada H, Youcef-Toumi K. Analysis and design of a direct-drive arm with a ®ve-bar-link parallel
drive mechanism. Trans ASME, J Dyn Syst Meas Cont 1984;106:225±30.
1062 P.J. Turner et al. / Mechatronics 11 (2001) 1053±1062

[11] Adept Technology Inc., 1212 Bordeaux Drive, Sunnyvale, CA 94089, Adept-One Manipulator
System: Product Description and Speci®cations, August 1985.
[12] Kazerooni H, Kim S. On the design and construction of direct-drive robots. Trans ASME, J Engrg
Industry 1990;112:197±201.
[13] Lee DG, Kim KS, Kwak YK. Manufacturing of a SCARA type direct-drive robot with graphite ®bre
epoxy composite material. Robotica 1991;9:219±29.
[14] An CH, Atkeson CG, Griths JD, Hollerbach JM. Experimental evaluation of feedforward and
computed torque control. IEEE Trans Robot Auto 1989;5(3).
[15] Berghuis H. Model-based robot control: from theory to practice. Ph.D. Thesis, University of Twente,
1993.
[16] Takase K, Hasegawa T, Suehiro T. Design and control of a direct drive manipulato. In: Proceedings
of the International Symposium on Design and Synthesis. Japan: Tokyo; July 1984. p. 333±8.
[17] Vines G. Linear and non-linear direct transmissions: a new approach to the design of robot joint
transmissions. Ph.D. Thesis, Electrical Department, Brunel University, 1996.
[18] Hartenberg RS, Denavit J. Kinematic synthesis of linkages. New York: McGraw-Hill, Library of
Congress Catalog Card Number 64-23251; 1964.
[19] Peters RM. Speed variation with elliptical wheels. In: Mechanisms 1973. p. 1±6, The Institute of
Mechanical Engineers, ISBN 0-85298-316-6; 1974.
[20] Chironis NP. Mechanisms and mechanical devices sourcebook. New York: McGraw-Hill, ISBN 0-07-
010918-4; 1991.
[21] Nippon Seiko KK. Europe: Middlesex House, 29-45, High Street, Edgware, Middlesex. HA8 7DH,
Megatorque Motor System User's Manual, 4th ed., May 1990, Document number: C20003-004.
[22] Intel Corporation, Pipers Way, Swindon, Wilts, SN3 1RJ, 80960KB Datasheet, December 1989,
Order number: 270565-004.
[23] Nigrowsky P, Turner PJ. Hybrid controller algorithms for the purpose of fast tracking of a non-
linearly actuated robotic arm. In: Proceedings of the Third International ICSC Sympoium on IIA and
SOCO. Italy: Genova; June 1999.
[24] Hartley J, Hollingum J. Direct drive robots in the limelight. Industrial Robot 1986:43±5.

Dr. P.J. Turner graduated from City University, London with a B.Sc in Control, Instrumentation and
Systems Engineering in 1983. In the following years, he completed an M.Sc. in Control Systems and a
Ph.D. in Robotics at Imperial College. Dr. Turner worked for 3 years as a lecturer at the University of
Manchester before taking up his present position lecturing Control Engineering at Brunel University.

Dr. P. Nigrowsky completed his ®rst Degree in Electrical Engineering at Glamorgan University. He then
went on to do a M.Sc. in Electronic Control Engineering at Salford University, and later a Ph.D. in
Control and Robotics, at Brunel University in 1999. Dr. Nigrowsky works now for Robert Bosch Brake
Systems.

Dr. G. Vines graduated from Brunel University in Electrical Engineering and Electronics in 1991. He was
awarded a Ph.D. in Robotics from the same university in 1998. Dr. Vines is presently working for Oxford
Intelligent Machines.

Anda mungkin juga menyukai