Professional Documents
Culture Documents
net/publication/267593193
Article in Proceedings of the Institution of Mechanical Engineers Part I Journal of Systems and Control Engineering · June 2010
DOI: 10.1243/09596518JSCE1015
CITATION READS
1 143
3 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Farid Tajaddodianfar on 04 May 2016.
The manuscript was received on 7 February 2010 and was accepted after revision for publication on 11 June 2010.
DOI: 10.1243/09596518JSCE1015
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
2 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015
Robust stable control of haptic devices based on transparency maximization 3
Although the order of the hand dynamics may vary effect. But also as explained before, it is necessary to
during the operation but this second-order approxima- take into account the human hand dynamics
tion is fit to the actual hand dynamics in haptic (represented by its admittance Yh) in order to
applications, then it is suggested by the literature [7, 17]. evaluate the stability of the interaction. This means
The feedback connection between robot impe- that in an interactive system the transfer function
dance Zr and uncertain hand admittance Yh can be which is needed for stability analysis, ZrYh, differs
represented by an upper linear fractional transform from the one which is required for performance
of a generalized control plant and a 363 perturba- analysis, Zr. On the contrary, in a servo system only
tion matrix D. Robust stability of the perturbed the loop transfer function is analysed for both
closed-loop system can be determined by the stability and performance. This is the first intrinsic
application of small gain theorem with the struc- difference between servo and interactive control
tured singular value [20]. Perturbation matrix D is systems design.
diagonal and real valued. Although the structured In an interactive system the robotic device is
singular value m has no closed-form solution, its modelled as a two-port linear network. One port is
upper and lower bounds can be calculated by assigned to the interaction of the robot with the
computational methods, and it can provide the controller and the remote side. The other port, in
system with the necessary and sufficient condition which the robot impedance is evaluated, is assigned
(m , 1) for interaction stability in the presence of all to the interaction with the human hand. Thus, the
hands H described by equations (1) and (2). controller transfer function appears neither linearly
Complementary stability refers to the robust nor multiplicatively in the robot impedance Zr
stability of the coupled system of the robot and the needed for performance or the loop transfer function
user hand. Passivity is replaced by complementary ZrYh needed for stability. This reveals the second
stability and, according to the necessity and suffi- intrinsic difference between servo and interaction
ciency of this stability criterion, the device transpar- control design.
ency is expected to be enhanced. Because of the two differences just mentioned,
none of the techniques used in servo control design
can be implemented in interactive system control
2.2 Stability and performance evaluation
design. Performance and stability of an interactive
Two fundamental differences exist between control system should be evaluated separately using differ-
design for servo systems and interactive systems ent transfer functions.
[19]. Performance in an interactive system is
evaluated by the measure of transparency and, as
described before, relates to the impedance of the 3 SYNTHESIS OF THE CONTROLLER
robot Zr regardless of human hand dynamics. In
other words, for evaluating the transparency of a The device model is generally supposed to be as
haptic device, the dynamics of the user hand has no shown in Fig. 1. In Fig. 1, G1 and G2 represent the
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
4 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
transfer functions from user hand displacement and Figure 2(A) displays the arrangement of compensa-
force set point to the force output of the device, tors and the feedback controller, according to which
respectively. Figure 1 also presents the basic feed- the device output impedance is given by
back structure which makes the robot capable of
virtual impedance emulation. fh G1 zLðC1 zC2 Zv Þ
Zr ~ ~ ð5Þ
According to Fig. 1 the impedance transmitted to xh 1zL
the user is given by
where L~G2 Gc is the open-loop transfer function. C1
fh and C2 are chosen as
Zr ~ ~G1 zG2 Zv ð3Þ
xh
{G3
C1 ~ ð6Þ
It is clear in this equation that the user hand L
dynamics has no effect on the robot impedance
and
and, hence, it has no effect on the system transpar-
ency. In the ideal transparent case, robot impen- 1zG4 Gx
dence Zr should be reduced to the virtual impedance C2 ~ ð7Þ
L
Zv; but, according to equation (3), the device
dynamics negatively affects the device transparency. where G3 and G4 are the indentified models of G1
Also, user hand dynamics should be considered in and G2, respectively. Gx equals Gc and its role is
the stability analysis of the coupled system. There- discussed in section 3.3. Application of this type of
fore, according to Fig. 1, robust stability of the compensator causes the robot impedance to reach
transfer function from x0 to fh is studied based on the target impedance and leads to the case of ideal
the concept of complementary stability. transparency. However, C1 and C2 cannot be selected
Compensation of the effects of device dynamics in as equations (6) and (7) for at least two reasons.
a stable manner is the main objective. To achieve First, L21 which consists of inverses of G2 and Gc
this aim, a certain compensatory structure is added would be an improper transfer function and does
to the virtual environment. If the assumed compen- not have a state space form. Second, the transfer
satory structure can provide the system with accep- function Gc is unknown; in fact, Gc is the target of the
table transparency, then the control problem is design procedure. In the following, certain ap-
reduced to the synthesis of a stabilizing controller proaches are given for overcoming these two
since the transparency objective is already satisfied. difficulties.
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015
Robust stable control of haptic devices based on transparency maximization 5
Fig. 2 Compensatory structure and stabilizing controller (A); final control diagram and
weighting functions (B)
equal to p, and if the order of the numerator of G2{1 frequency range at which the virtual impedance can
be n times greater than the order of its denominator, be emulated.
then the transfer function Gp is defined as
G2{1
Gp ~ ð8Þ 3.3 Iterative controller design
ð1zs=pÞn
The controller transfer function Gc, which is used in
Using equation (8), the device output impedance the compensatory structure, is unknown. In fact, the
according to Fig. 2(B) is calculated as objective of the design procedure is a primary
assumption and, consequently, the controller trans-
fh fer function appears in two various places in the
Zr ~
xh control loop. This reveals another problem since it is
G1 {ð1zs=pÞ{n G3 zð1zs=pÞ{n ð1zG4 Gx ÞZv e{sT not possible to force the automatic controller
~ synthesis techniques to give two similar outputs.
1zG2 Gc
ð9Þ In order to solve this problem, an iterative method
is used. To do this, the controller transfer function
Regarding equation (9), it is clear that the device appearing in the compensatory structure is replaced
capability in the emulation of the virtual impedance by an assumed transfer function Gx. In this case, the
is dependent on the frequency of the added poles. controller appears once in the control loop and, as in
Therefore p, as a design parameter, determines the the following, the problem is solved to derive a
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
6 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
controller satisfying stability and performance ob- diagram. This is shown in Fig. 2(B), where Wn
jectives. The derived controller would differ from the represents the performance weighting function for
primary assumption of Gx. However, Gc is estimated the noise reduction objective and Wu is the perfor-
by a lower-order transfer function and Gx is replaced mance weighting function accounting for the control
by Gc. The design procedure is repeated using the input limitations.
new transfer function of Gx; and this is done over
and over again until the algorithm converges to a
function Gc which fits Gx in a wide frequency range. 3.6 Time delay effects
Therefore, the problem of appearance of the con- If the total amount of time delay is assumed to be T,
troller in more than one place in the control loop is then its effects are taken into account by adding the
removed. time delay block of e2sT to the control diagram. A
first-order Pade estimation is used in the calcula-
tions
3.4 Stability analysis
According to the concept of complementary stabi- 1{ðT =2Þs
e{sT ~ ð11Þ
lity, user hand effects are taken into account and the 1zðT =2Þs
stability of the closed-loop transfer function of the
robot impedance and user hand is studied. The This enables the time delay effects to be accounted
closed-loop transfer function from the displacement for in the calculations.
set point x0 to the force output fh is given by
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015
Robust stable control of haptic devices based on transparency maximization 7
fh {Ks ts
G1 ~ ~ ð12Þ
xh R2 Ks zR2 ts
and
fh Ks
G2 ~ ~ ð13Þ
fd tszKs
Fig. 4 Series elastic actuator model (A); corresponding block diagram (B)
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
8 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
G3, which is the identified model of the device, is three types of Gx are selected, and the resulted
given by outputs of the algorithm are studied in the following.
As the first assumption, assume Gx to be equal to a
G3 ~G1 zWm D ð15Þ low-pass filter with low d.c. gain, namely Gx11.
Considering uncertainties and weighting functions
where kDk‘ v1 and Wm represents the uncertainty as described before, the algorithm is performed; after
weighting function which is described by a high-pass three iterations, the algorithm converges to a con-
filter troller Gc1 which fits to the preceding assumption
Gx31 in a wide frequency range. Figure 5 displays
200sz400
Wm ~ ð16Þ intermediary controller dynamics in the iterative
sz20 000 procedure. It should be noted that, in each step, the
output of the algorithm is estimated by a second- or
Performance weighting functions, Wn and Wu, are
third-order transfer function. The resulting control-
given by equations (17) and (18)
ler transfer function, Gc1, is given by
10sz1:4
Wn ~ ð17Þ 0:06104s2 z1951sz1:787|106
sz1400 Gc1 ~ ð19Þ
s2 z1629sz4134
and
As the second assumption, Gx is assumed to be
2sz6:1 equal to a low-pass filter with a relatively high d.c.
Wu ~ ð18Þ
sz61 gain, namely Gx12. Convergence occurs after three
iterations, as shown in Fig. 6, and the resulting
Time delay parameter T is assumed to be equal to controller Gc2 is given by
0.05 s, which represents a relatively slow sampling
rate. 0:2286s3 z6654s2 z5:833|104 sz1:328|107
Gc2 ~
Definition of an appropriate primary assumption s3 z1582s2 z7739sz2:58|104
of Gx is the main part of the design procedure. In ð20Þ
fact, the dynamics of the resulting controller directly
depends on this primary assumption. However,
Fig. 5 Iteration beginning with Gx11 converges to Gc1 Fig. 6 Iteration beginning with Gx12 converges to Gc2
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015
Robust stable control of haptic devices based on transparency maximization 9
Finally, as the third assumption, it is assumed that input x0 is assumed to increase up to the value of
Gx equals a high-pass filter, namely Gx13. After two x1 5 20 mm in Dt seconds, retain this value for a few
iterations, the algorithm converges to Gc3, as given seconds, and decrease back to zero in Dt seconds.
by Low-acceleration excitations which represent low-
frequency behaviour are simulated by relatively high
0:2079s2 z1951sz1:787|106 values for Dt, and high-acceleration excitations are
Gc3 ~ ð21Þ
s2 z2646sz5:165|104 simulated by lower values of Dt. It is very useful to
plot the force transmitted to the user hand versus its
In order to evaluate the device transparency, the displacement. In the ideal case of transparent
output impedances of the device in the presence of emulation of a spring, loading and unloading curves
the three controllers are given in Fig. 7 in which the coincide and turn out to be a straight line with slope
objective is to emulate the nominal virtual spring. It of Kv. System dynamics affects responses and causes
is clear from Fig. 7 that the designed controllers can them to separate. Figure 8 displays low acceleration
improve device transparency up to the frequency of behaviour of the device during the process in the
100 rad/s, which is approximately ten times greater presence of the proposed compensatory structure
than the maximum transparency of the passive and the Gc1 controller, and compares it with the
control case. In addition, studying the structured regular passive control case. Effects of the device
singular value for the coupled system of the device dynamics are more challenging in high acceleration
impedance and user hand, m, confirms robust behaviour. Figure 9 displays the high acceleration
stability of the interaction since mmax does not exceed behaviour of the device. In the regular passive
0.64, 0.88, and 0.90 in the presence of Gc1, Gc2, and control case, loading and unloading curves differ
Gc3, respectively. considerably; however, application of the proposed
control structure causes the two curves to approach
each other, and this leads to enhancement of the
5 SIMULATION RESULTS device transparency.
Effects of modelling uncertainty in the low and
Time-domain analyses are performed to study high acceleration behaviours are shown in Figs 10
effects of the designed controllers on the transpar- and 11, respectively. In Figs 10 and 11, performance
ency of the device. It is assumed that a virtual spring of the proposed control structure for the nominal
with stiffness of 300 N/m is emulated. Reference system is compared with the resulting performance
Fig. 7 Output impedance of the device in the passive control case and in the presence of each of
the three controllers
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
10 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
Fig. 8 Virtual spring simulated with passive control and Gc1 controller; the device experiences
low acceleration
of the system containing 30 per cent and 40 per cent but in the high acceleration behaviour this type of
modelling uncertainty in G3. It is clear that in the uncertainty negatively affects the device transpar-
low acceleration behaviour, modelling uncertainty ency. As described before, this is due to the effects of
does not seriously affect the system transparency; the high-frequency added poles. Since the struc-
Fig. 9 Virtual spring simulated with passive control and Gc1 controller; the device experiences
high acceleration
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015
Robust stable control of haptic devices based on transparency maximization 11
Fig. 10 Virtual spring simulated with the nominal system and various modelling uncertainties
(unc.) in the G3 controller experiencing low acceleration excitation
tured singular value of the coupled system does not Figures 8 to 11 are plotted in the presence of Gc1;
exceed unity, the system does not face instability; the other two controllers result in relatively similar
but higher uncertainties negatively affect the system responses.
transparency, especially in the high acceleration To study effects of measurement noise, a 1 per
behaviours. cent power white noise generator followed by a low-
Fig. 11 Virtual spring simulated with the nominal system and various modelling uncertainties
(unc.) in the G3 experiencing high acceleration excitation
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
12 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
Fig. 12 Sensed force in the presence of measurement noise for the passive control case (a), Gc1
(b), Gc2 (c) and Gc3 (d) controllers
pass filter is implemented. This causes the system to command sent to the device is slightly bigger than
experience noises with frequencies up to 60 rad/s or the desired output force, and this is consistent with
10 Hz approximately. This is consistent with the the main objective in this study.
20 Hz sampling frequency since the highest fre-
quency of the signals in the control system should
not exceed half of the sampling frequency. Figure 12 6 CONCLUSIONS AND FUTURE WORK
displays the force transmitted to the user during the
process in the presence of the three controllers, from A method of controller synthesis for haptic systems
which the effects of the three controllers, especially is given which not only provides the system with
Gc3, are clear. robust stability but also maximizes transparency of
The control command which is sent to the device the device. To do this, a compensatory structure is
in the presence of measurement noise and the Gc2 fitted to the virtual environment which guarantees
controller is plotted in Fig. 13. Comparing this plot device transparency by compensating for its dy-
with those given in Fig. 12 reveals that the control namics. Since the controller appears in two distinct
places in the control diagram, it is replaced by an
assumed transfer function Gx, and then, regarding
various uncertainties and performance weighting
functions, the D-K iteration algorithm is performed
to derive a robust controller. The derived controller,
Gc, replaces the primary transfer function Gx. This
iteration on Gx continues until Gc and Gx fit in a wide
frequency range.
The algorithm is applied to a series elastic actuator
model. The maximum achievable transparency
Fig. 13 Control command in the presence of mea- directly depends on the added pole frequency; an
surement noise for the Gc2 controller added pole with frequency of 300 rad/s enables the
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015
Robust stable control of haptic devices based on transparency maximization 13
device to emulate the target impedance up to the the Routh–Hurwitz criterion. IEEE Trans. Contr.
frequency of 100 rad/s, which is almost ten times Syst. Technol., 2004, 12(4), 583–588.
greater than that of the regular passive control case. 11 Adams, R. J. and Hannaford, B. Stable haptic
interaction with virtual environments. IEEE Trans.
Assuming three different types of transfer functions Robot. Autom., 1999, 15(3), 465–474.
for Gx, three various controllers are derived. Time- 12 Mahvash, M. and Okamura, A. Friction compen-
domain simulations confirm that these controllers sation for enhancing transparency of a teleoperator
improve the device transparency while they are with compliant transmission. IEEE Trans. Robot.,
robust to various uncertainties, and also they can 2007, 23(6), 1240–1246.
reduce measurement noise. 13 Vlachos, K. and Papadopoulos, E. Transparency
maximization methodology for haptic devices.
The proposed method is applied to a single-DOF
device; however, extension to multi-DOF devices
IEEE Trans., 2006, 11(3), 249–255 . >
14 Kuchenbecker, K. J., Fiene, J., and Niemeyer, G.
would be necessary. It should be noted that the main Improving contact realism through event-based
problem in the development of multi-DOF transpar- haptic feedback. IEEE Trans. Visual. Comput.
ent haptic interfaces comes from the driving system Graph., 2006, 12(2), 219–229.
which is necessary to have minimized mechanical 15 Black, B. and Book, W. Dynamic compensating
impedance. Single-DOF methods that lead to lower controller for passive haptic manipulators in tele-
impedance of the drive system enable development operation. In Proceedings of the IEEE International
Conference on Robotics and Automation, Kobe,
of transparent multi-DOF haptic interfaces. Japan, 2009. ?
16 Radi, M., Artigas, J., Preusche, C., and Roth, H.
F Authors 2010 Transparency measurement of telepresence sys-
tems. Lect. Notes Comput. Sci., 2008, 766–775 . @
REFERENCES 17 Cheung, Y., Chung, J. H., and Hong, D. Adaptive
force reflecting teleoperation with local force
1 Hayward, V. and Maclean, K. E. Do it yourself compensators. Robotica, 2007, 25(4), 433–444.
haptics: Part I. IEEE Robot. Autom. Mag., 2007, 14(4), 18 Mobasser, F. and Hashtrudi-Zaad, K. Transparent
88–104. rate mode bilateral teleoperation control. Int. J.
Robot. Res., 2008, 27(1), 57–72.
2 Hashtrudi-Zaad, K. and Salcudean, S. E. Analysis
19 Buerger, S. P. and Hogan, N. Complementary
of control architectures for teleoperation systems
stability and loop shaping for improved human–
with impedance/admittance master and slave
robot interaction. IEEE Trans. Robot., 2007, 23(2),
manipulators. Int. J. Robot. Res., 2001, 20(6),
232–244.
419–445.
20 Doyle, J. Analysis of feedback systems with
3 Hannaford, B. and Ryu, J. H. Time domain
structured uncertainties. IEE Proc. Contr. Theor.
passivity control of haptic interfaces. In Proceed-
Applic., 1982, 129(6), 242–250.
ings of the IEEE International Conference on
; Robotics and Automation, 2001.
4 Ryu, J. H., Hannaford, B., and Hirzinger, G. Time APPENDIX
domain passivity control with reference energy
following. IEEE Trans. Contr. Syst. Technol., 2005, Notation
13(5), 737–742.
5 Stramigioli, S., Secchi, C., Van der Schaft, A. J., bh damping of the user hand model
and Fantuzzi, C. Sampled data systems passivity fd desired force
and discrete port-Hamiltonian systems. IEEE fh sensed force
Trans. Robot., 2005, 21(4), 574–587.
f̂h weighted sensed force
6 Khalil, H. K. Nonlinear systems, edition 3, 2002
< (Prentice Hall) . kh stiffness of the user hand model
7 Speich, J. E. and Goldfarb, M. Modeling human Ks stiffness of the elastic element of the
hand as it interacts with a telemanipulation series elastic actuator
= system. Mechatronics, 2005, 15, 1127–1142 . Kv virtual stiffness coefficient
8 Miller, B. E., Colgate, J. E., and Freeman, R. A. On Kvn nominal virtual stiffness coefficient
the role of dissipation in haptic systems. IEEE mh mass of the user hand model
Trans. Robot. Autom., 2004, 20(4), 768–771. n measurement noise
9 Diolaiti, N., Niemeyer, G., Barbagli, F., and Salis-
x0 displacement set point
bury, J. K. Stability of haptic rendering: discretiza-
tion, quantization, time delay, and Coulomb xh net device displacement
effects. IEEE Trans. Robot., 2006, 22(2), 256–268. R handle arm of the series elastic
10 Gil, J. J., Avello, A., Rubio, A., and Florez, J. actuator
Stability analysis of a 1 DOF haptic interface using s Laplace variable
JSCE1015 Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering
14 M T Ahmadian, G Vossoughi, and F Tajaddodianfar
Proc. IMechE Vol. 224 Part I: J. Systems and Control Engineering JSCE1015