Proceedings of the American Control Conference

Anchorage, AK May 8-10,2002

Tutorial on Input Shaping/Time Delay Control

of Maneuvering Flexible Structures
Tarunraj Singh William Singhose
Dept. of Mech. & Aero. Eng., Dept. of Mech. Eng.,
University of Buffalo, Georgia Institute of Technology
Buffalo, NY 14260 Atlanta, Georgia, 30332
[email protected] [email protected]
http://code.eng.buffalo.edu/tdf/ http://www.me.gatech.edu/inputshaping/

Abstract variation in the location of the poles of the sys-

tem caused by nonlinear components in the sys-
Precise position control and rapid rest-to-rest mo- tem or variation of the parameters of the system
tion is the desired objective in a variety of applica- as a function of temperature.
tions. The desire for reducing the maneuver time
requires reducing the inertia of the structure which Between the late 50’s and the publication of the
subsequently results in low frequency dynamics. Input Shaping paper by Singer and Seering [8],
The requirement of precise position control im- there was some work on the shaping of input
plies that the residual vibration of the structure profiles for control of residual vibration [4], [5].
should be zero or near zero. This paper presents Swigert [5] proposed techniques for the determi-
techniques to shape the input to the system so nation of torque profiles which considered the sen-
as to minimize the residual vibration. A second sitivity of the terminal states to variations in the
class of problems which includes design of input model parameters. Publication of the Input Shap-
profiles for systems with rigid-body modes driven ing paper renewed interest in prefiltering refer-
by actuators with finite control authority, is also ence inputs for robust vibration control, which
presented. The common thread which connects all has resulted in dozens of papers with application
the techniques presented in this paper is related to to spacecrafts, robots, cranes, chemical processes,
the design of controllers which are robust to mod- etc. A chronological listing of papers relevant to
eling uncertainties. The proposed techniques are robust vibration control of slewing structures is
illustrated by simulations and experiments. presented in the bibliography.

This paper will consider two classes of problems:

The first, involves real-time shaping or time-delay
1 Introduction filtering of the reference command to stable sys-
tems with the objective of minimizing the resid-
Smith [l],Calvert and Gimple [2] proposed a sim- ual vibration [l][8], [16], [23]. This will be dealt
ple technique to generate non-oscillatory response in detail in Section 2. The second class of prob-
from an lightly-damped system subject to a step lems considered is the design of controllers for sys-
input. This is achieved by exciting two transient tems with rigid body modes with constraints on
oscillations so as to result in constructive cancel- the control input [7], [9], [14], [29], [49]. Th’1s can
lation of the oscillations. Smith termed the tech- be further classified into two categories which will
nique Posicast motivated by what he states: ‘‘ This be expounded in Section 3. Sections 2 and 3 will
is what happens when a fisherman drops his fly describe in detail, technique for reducing the sen-
in the water at the maximum-position and zero sitivity of the control profile to errors in model
velocity instant” [l].Tallman and Smith [3] illus- parameters such as damping ratio, and natural fre-
trated the Posicast technique using an analog com- quency. Section 4 will briefly address the design
puter and noted the sensitivity of the controller to of robust controllers for nonlinear systems [41].

O-7803-7298-0/02/$17.00 0 2002 AACC 1717

-_
Finally, Section 5 will describe some applications
where the proposed techniques have been success-
fully implemented.

2 Real Time Command Shaping

/ / / / A / / / I /

7 - -*---Payload

6 -
Button On

Figure 1: Overhead Gantry Crane

There are all types of possible solutions to the

problem of flexible dynamics including feedback
control, feedforward control, command shaping,
and even redesigning the physical hardware. A
simple example of this challenging area is pre-
sented by an overhead gantry crane like the one Figure 2: Crane Response: a) Unshaped Command
shown schematically in Figure 1 . The payload is b) Shaped Command
hoisted up by a cable. The upper end of the cable
is attached to a trolley which travels along a rail it is helpful to start with the simplest such com-
to position the payload. mand. We know that giving the system an impulse
will cause it to vibrate; however, if we apply a sec-
Cranes are controlled by a human operator who
ond impulse to the system, we can cancel the vi-
moves levers or presses buttons to cause the trol-
bration induced by the first impulse. This concept
ley to move. If the operator simply presses the
is shown in Figure 3.
control button for a finite time period, then the
trolley will move a finite distance and come to rest. At this point, it is useful to derive the amplitudes
The payload on the other hand, will usually oscil- and time locations of the two-impulse command
late about the new trolley position. The payload shown in Figure 3. If we have a reasonable es-
position for a typical trolley movement is shown timate of the system's natural frequency, w , and
in Figure 2a. damping ratio, C, then the residual vibration that
results from a sequence of impulses can be de-
An experienced crane operator can sometimes pro-
scribed by:
duce the desired payload motion with a small
amount of residual vibration by pressing the but-
ton multiple times at the proper instances. The
payload position for such a situation is shown in where,
Figure 2b.

2.1 Simple Zero Vibration Command

As a first step to understanding how to generate
commands that move systems without vibration,

1718
 because they are squared in (1). Therefore, the
0.6 impulses must satisfy:

0.4 0 = Ai + A2eCWt2COS(Wdt2) (5)

g 0.2 o = A2eCWt2sin(wdt2) (6)

.-*
.I

Equation (6) can be satisfied in a nontrivial man-

g o ner, when the sine term equals zero. This occurs
-0.2 when:

-0.4
0 0.5 1 1.5 2 2.5 3
Time where Td is the damped period of vibration. This
result tells us that there is an infinite number of
Figure 3: Two Impulse Response possible values for the location of the second im-
pulse - they occur a t multiples of the half period of
Ai and ti are the amplitudes and time locations vibration. To cancel the vibration in the shortest
of the impulses, n is the number of impulses in amount of time, choose the smallest value for t2:
the impulse sequence, and W d = w d p . Equa-
Td
t2 = -
tion l is actually the percentage residual vibration.
2
It tells us how much vibration a sequence of im-
pulses will cause, relative to the vibration caused For this simple case, the amplitude constraint
by a single, unity-magnitude impulse. By setting given in (3) reduces to:
( 1 ) equal to zero, we can solve for the impulse
amplitudes and time locations that would lead to A1+A2=1 (9)
zero residual vibration. However, we must place
Using the expression for the damped natural fre-
a few more restrictions on the impulses, or the quency and substituting (8) and (9) into (5) gives:
solution will converge to zerevalued or infinitely-
valued impulses. To avoid the trivial solution of all
zero-valued impulses and to obtain a normalized
result, we require the impulses to sum to one:
Rearranging (10) and solving for A1 gives:
C A i = 1. (3)
eXP(&)
Impulses could satisfy (3) by taking on very large A1 = (11)
positive and negative values. One way to obtain 1 + 4*)
a bounded solution is to limit the impulse ampli-
tudes to finite values or to positive values: Defining K = exp(*), the sequence of two
l/iT
Ai > 0, i = 1,2,....,n impulses that leads to zero residual vibration can
(4)
now be summarized as:
The problem we want to solve can now be stated K
explicitly: find a sequence of impulses that makes
0 0.5Td
( 1 ) equal to zero, while also satisfying (3) and (4).
For a two-impulse sequence, the problem has four
unknowns - the two impulse amplitudes ( A I ,A2) 2.2 Using Zero-Vibration Impulse Se-
and the two impulse time locations ( t l , t2). With- quences to Generate Zero-Vibration
out loss of generality, we can set the time location Commands
of the first impulse equal to zero, tl = 0. The prob- Real systems cannot be moved around with im-
lem is now reduced to finding three unknowns ( A I , pulses, so we need to convert the properties of the
Aa, t2). In order for ( 1 ) to equal zero, the expres- impulse sequence given in ( 1 2 ) into a usable com-
sions in ( 2 ) must both equal zero independently mand. This can be done in a very simple way.

1719
 there are errors in these values (and there always
are), then the impulse sequence will not result in
zero vibration. In fact, for the two-impulse se-
quence discussed above, there can be a lot of vi-
0 2 0 A bration for a small modeling error. This lack of ro-
Initial Command Input Shaper bustness was a major stumbling block for the orig-
I inal formulation that was developed in the 1950’s.
This problem can be visualized by plotting a sen-
sitivity curve that shows the amplitude of residual
0 2 A 2+A vibration as a function of the system parameters.
Shaped Command One such sensitivity curve for the zero-vibration
(ZV) shaper is shown in Figure 6 with the normal-
Figure 4: Multi Pulse Shaped Input ized frequency on the horizontal axis and the per-
centage vibration on the vertical axis. Note that
as the actual frequency deviates from the model-
The impulse sequence is convolved with any d e
ing frequency, the amount of vibration increases
sired command signal. The convolution product
rapidly. The robustness can be measured quan-
is then used as the command to the system. If
titatively by measuring the width of the curve at
the impulse sequence causes no vibration, then the
some low level of vibration. This non-dimensional
convolution product will also cause no vibration.
robustness measure is called the shaper’s insensi-
This command generation process, called input tivity. The 5% insensitivity has been labeled in
shaping, is demonstrated in Figure 4 for an ini- Figure 6 .
tial command that is a pulse function and a two-
impulse input shaper. Note that the convolution
product in this case is the two-pulse command sim- -2V Shaper
------Robust (ZVD) Shaper
ilar to that shown in Figure 2b. But in most cases
the impulse sequence will be much shorter than
the command profile. When this occurs, the com-
ponents of the shaped command that arise from
the individual impulses run together as shown in
Figure 5.

Normalized Frequency (Ww,,,)

Figure 6: Sensitivity Curves

Initial Command Input Shaper

In order to increase the robustness of the input

shaping process, the shaper must satisfy addi-
tional constraints. One such constraint sets the
OA D+A derivative of (1)’with respect to frequency, equal
Shaped Command to zero [SI. That is:
d
Figure 5 : Continuous Shaped Input 0 = -V(w, <)
dw
When this additional constraint is satisfied with V
2.3 Robustness to Modeling Errors = 0, the result is a Zero Vibration and Derivative
The amplitudes and time locations of the impulses (ZVD) shaper [8]. By comparing the 5% insen-
depend on the system parameters (w and <). If sitivities shown in Figure 6, it can be concluded

1720
 Residual
35 r
e 30
8E 25
s
s 20

0
0.6 0.8 1 1.2 1.4
Normalized Frequency (y/w,,J

, 1.L
Figure 8: Specified-Insensitivity Shaper
Frequency (Hz)
= 5%. Another technique related to SI shaping
Figure 7: Three Dimensional Curve minimizes the expected level of residual vibration
over a specified frequency range 1521. This tech-
that the ZVD shaper is significantly more robust nique has the advantage of taking into account any
to modeling errors than the ZV shaper. known distribution of the frequencies in the range
being suppressed.
Since the development of the ZVD shaper, several
other robust shapers have been developed. In fact, Any shaped command will have its rise time in-
shapers can now be designed to have any amount creased by the duration of the shaper as is shown
of robustness to modeling errors [38]. The sen- in Figure 9a. Because the duration of the ZVD
sitivity curve for a very robust shaper is shown shaper is twice that of the ZV shaper, the ZVD
in Figure 6 . Robustness is not restricted to er- shaper increases the rise time more than the ZV
rors in the frequency. Figure 7 shows a three- shaper. This increased rise time is the price that
dimensional sensitivity curve for a shaper that is paid for the increased robustness to modeling
was designed to suppress vibration between 0.7 errors. With the SI shapers, increasing robustness
Hz and 1.3 Hz and also over the range of damp- increases rise time in a nonlinear manner. This
ing ratios between 0 and 0.2. The shapers corre- leads to certain operating points that are advan-
sponding to these curves were designed using the tageous [3].
Specified-Insensitivity (SI) approach. The most
straightforward method for generating a shaper For example, the ZVD shaper has a duration of
only 1 period of the natural frequency. This time
with specified insensitivity to modeling errors is
the technique of frequency sampling [19], [38]. penalty is a small price to pay for the excellent ro-
This method requires repeated use of the vibra- bustness to modeling errors. To demonstrate this
tradeoff, Figure 9 shows the response of a spring-
tion amplitude equation, (1). In each case, V(w,c)
mass system to step commands shaped with the
is set less than or equal to a tolerable level of vi-
bration, hol: three shapers shown in Figure 6. Figure 9a shows
the response when the model is perfect and Fig-
vtol > e-'w3tnJ~(w,, +
~ 1 2 ~ ( w , c, > 2 , s = 1,..., m ure 9b shows the case when there is a 30%error in
(14) the frequency estimate. The increase in rise time
where w, represents the m unique frequencies at caused by the shapers is apparent in Figure 9a,
which the vibration is limited. while Figure 9b shows the vast improvement in vi-
bration reduction that the robust shapers provide
For example, if the shaper needs to suppress vi- in the presence of modeling errors.
bration for frequency errors of 20%, then the con-
straint equations limit the vibration to below K0l The techniques mentioned above produce robust-
a t specific frequencies between 0 . 8 and
~ ~ 1 . 2 ~ ~ . ness built into the design of the input shaper.
This procedure is illustrated in Figure 8 for There are other approaches to achieve robust in-

1721
 disturbance rejection and stability, which are its
natural strengths. Given this realization, the ques-
I tion arises as to how to optimize the combined de-
sign of the feedback and command shaping com-
ponent s.
E
One method assumes a PD feedback controller and
then concurrently chooses the PD gains and the in-
put shaper impulses while satisfying performance
n I 2 1 4 5 specifications [56], [63]. The design method takes
Time into account limits on allowable overshoot, resid-
ual vibration, and actuator effort. Furthermore,
the structure of the method allows a wide range
I .5
of performance requirements, such as disturbance
rejection, to be integrated into the design. The re-
sults indicate that PD feedback control enhanced
.-B I with input shaping provides better performance
e
.I

than PD control alone. This effect is demonstrated

s
0.5
in Figure 10 where the settling time is shown as a
function of the damping ratio of the system. With
PD control alone, the minimum settling time oc-
0 curs near <=0.7 - the classic solution. However,
0 1 2 3 4 5
Time when input shaping is used, the settling time can
be greatly reduced and this occurs when the con-
troller is tuned to have a small damping ratio.
Figure 9: System Response: a) Perfect Knowledge,
b) 30% Error

put shaping. Several researchers have used adap-

tive input shaper modification techniques to ob-
tain robustness. Sensor feedback is used to tune
the input shaper such that the residual vibration
is decreased. Some of these approaches include
that of Tzes and Yurkovich [26] and Khorrami, et
I 1
al. [25], who used on-line adaptive schemes to u p 0' 1 I I
0 0.2 0.4 0.6 0.8 1
date the input shaper parameters. Bodson used a Damping Ratio (0
recursive least-squares technique to tune the input
shaper parameters [36]. Magee and Book modified Figure 10: Settling Time vs. Damping Ratio
the input shaper as a function of the system con-
figuration [18].
2.5 Shaper Design in the S-Plane and Z-
2.4 Concurrent Design of Command Shap- Plane
ing and Feedback Control The effect of command shaping is to place zeros at
Given that command shaping can greatly reduce or near the flexible poles of the system. This idea
the vibration of the system, it reduces the burden was well documented by Bhat and Mui [lo]. This
on the feedback controller. Therefore, the feed- realization leads to straightforward design proce-
back control design becomes easier; it does not dures in the z-plane that were first discussed by
have to be concerned with reducing vibration in- Murphy and Watanabe [16]. Singh and Vadali [23]
duced by the reference command. The design of illustrated that a time-delay filter designed to can-
the feedback controller can be primarily based on cel the poles of the system results in the same solu-

1722
tion as the posicast control developed by Smith [ 11. tional approaches presented to deal with the effect
They also illustrated that cascading multiple ver- of flexibility. Most of these deal with single in-
sions of the time-delay filter resulted in the robust put rest-to-rest problems under two classes: near-
shaper that was proposed by Singer and Singer. minimum time control and exact minimum-time
Singh and Vadali also proposed a simple technique control.
to design time-delay filters using the specified time
spacing of the sampling period [32]. The first category is based on smooth approxima-
tions to the time-optimal control for an equiva-
Seth used z-plane analysis to design a digital lent rigid body. This is applicable where the ap-
shaper for reducing vibration in a coordinate mea- plied input can be smoothly varied and are not
suring machine [27]. Tuttle developed a simple restricted to an on-off set. Junkins et al. [9] pa-
stepby-step method to design multiple-mode in- rameterize a single switch bang-bang profile using
put shapers in the discrete time domain by bring- cubic polynomials in time and illustrate that the
ing together previous methods [30]. Additionally, residual vibration of a flexible structure can be sig-
Tuttle directly addressed the issue of time opti- nificantly reduced for a small penalty in maneuver
mality for digital shapers by presenting a method time. Vadali et al. [33] used the arctan to ap-
for finding a positive impulse shaper that had the proximate the signum function and used a param-
shortest time duration. Like Seth, Jones used z- eterized smooth control profiles to determine near-
plane analysis to design a digital shaper for re- time optimal control profiles for three dimensional
ducing vibration in a coordinate measuring ma- attitude control of ASTREX,a flexible spacecraft
chine [54]. Additionally, Jones indicated the re- testbed.
quirements on shaper duration to obtain an input
shaper with only positive impulses. The second category studies the exact time-
optimal control problem. The determination of
Magee applied a digital shaping filter to a system time-optimal control profiles for flexible slewing
with varying parameters by modifying the input structures with limited control authority has been
shaper duration to account for system parameter addressed by Singh et al. [7]. They illustrate that
variations [18]. This work verified the difficulty the time-optimal control profile for un-damped
of changing shaper duration that was predicted systems is antisymmetric about the mid-maneuver
by Murphy and Watanabe. More recently, Park time. Hablani [ll] studied the same problem,
et al. extended the z-plane based design of digi- but with damped modes. Ben-Asher et al. [14],
tal input shapers to more robust shapers [59]. In present an elegant technique to prove the time-
particular, Park devised a discrete time sensitiv- optimality of the control profiles. It is well known
ity expression. This expression was used to design that the time-optimal control profile is highly sen-
very robust multiple hump input shapers directly sitive to errors in system parameters. Liu and Wie
in the z-plane [47]. [15] present a technique to “robustify” the time-
optimal control by including additional switches
to the control profile. Singh and Vadali [29] pro-
3 Saturating Controllers pose a frequency domain approach for the design
of time-optimal controllers for flexible structures.
The motivation behind their work is the fact that
The problem of design of optimal control with lim-
a bang-bang input can be viewed as a summation
its on control authority has been of interest for
of time-delayed step commands. They pose the
decades. When cost functions such as maneuver
problem as the design of a time-delay filter de-
time, or fuel or a weighted combination of fuel and
time are considered, the resulting optimal control signed to cancel all the poles of the system and
satisfy the rigid body boundary conditions. They
profiles are bang-bang or bang-off-bang implying
use the knowledge that locating multiple zeros of
that the controller is turned on to the extreme
values or is turned off. The problem of design the time-delay filter at the estimated location of
of time-optimal control profiles for flexible struc- the poles of the system, results in robustness to
tures has been of increasing interest over the past modeling uncertainties. They illustrate their tech-
two decades. There have been numerous computa- nique by designing time-optimal and robust time-

1723
optimal control profiles for rest-to-rest and spin- The equations of motion can be decoupled by the
up maneuvers for the benchmark floating oscillator similarity transformation
problem. 1 1

This section will discuss the design of robust time

and weighted fuel-time optimal controllers. The resulting in the equations of motion
benchmark floating oscillator problem [21], will be
used to illustrate the design technique. The design
e..= - u1
problem is posed as the design of a time-delay fil- 2
ter which generate the bang-bang or bang-off-bang
control profile when it is driven by a step input. q + 2q = --U,21 (21)
The knowledge that locating multiple zeros of the
time-delay filter at the estimated location of the and the corresponding boundary conditions are
poles of the system will result in robustness to un- e(0) = q(0) = 0 and d(0) = q(0) = 0
certainty in the location of the pole of the system
will be used in the design process. For a function e(t,) = l , q ( t f )= 0 and d(tf) = q ( t p ) = 0. (22)
f(s) = 0, to have a minimum of two roots at s =
The decoupled equations are used to derive the
so requires that
constraint equations for the optimization prob-
lem since this approach can be generalized for any
(15) number of modes.

This fact is utilized to develop constraint equa-

tions to design time-delay filters with multiple ze-
ros at specified locations.

I m1 w k
m2 I Figure 12: Time Delay Filter Structure

In this paper, the design of control profiles subject

to the aforementioned control constraint is posed
Figure 11: Floating Oscillator as the design of a time-delay filter. The output
of this time-delay filter subject to a step input
The equations of motion of the benchmark floating is the optimal switching control profile, as shown
oscillator problem illustrated in Figure 11 are in Figure 12. For instance to generate a single
switch bang-bang control profile which is the time-
I;[ ; 2 ] [ ; ? ] + [ -k - kk ] [ ; : ] = [ ; I u . optimal control profile for a rigid body system, the
transfer function of the time-delay filter is
(16)
The state constraint for all the optimization prob- G ( s ) = 1 - 2exp(-sT) + exp(-2sT), (23)
lems considered in this section are
and the optimal control profile is given by the
yl(0) = y2(0) = 0, and yl(0) = $2(0) = 0 equation
Y l ( t f ) = Y 2 ( t f ) = 1, and ? h ( t p ) = Ll2(tp) = 0 (17)
1
U(.) = -(1- 2exp(-sT) +exp(-2sT)). (24)
and the normalized control is subject to the con- S
straint The generic transfer function of a time-delay filter
-1 5 U 5 1. (18) is represented as
The nominal values of the parameters of the sys- N
tem are
ml = m2 = k = 1 (19)
G(s)= 1Aiexp(-sTi), where TO= 0, (25)
i=O

1724
and where Ai belongs to the set constraints are
N
A i = [ -2 -1 1 21 AiTiexp(-aTi)cos(wT)= 0, (33)
i=O
to guarantee that the output of the timedelay fil-
ter is either bang-bang or bang-off-bang. For rest- and
N
to-rest maneuver of flexible structures, the con-
AiT,exp(-aT,)sin(wT,)= 0 (34)
straint which guarantees zero residual vibration is
i=O
derived by requiring a set of zeros of the time-delay
filter to cancel the under-damped poles of the sys- which are equivalent to the zero derivative con-
tem. For a system with a set of under-damped straint given in Equation 13.
poles at
Since the constraints are nonlinear, there are po-
S = a fj w , (27) tentially numerous parameter sets which satisfy all
the constraint equations are of the constraints. The sufficient conditions for the
optimality of the control profile are dependent on
N the cost function to be optimized for and a gen-
A;exp(-aT,)cos(wT,)= 0, (28) eral approach to verify the optimality is not avail-
i=O
able. For un-damped system, it has been shown
and that the control profile is anti-symmetric about the
N mid-maneuver time. This fact can be exploited to
Aiexp(-aTi)sin(wT,)= 0, (29) reduce the number of parameters to be optimized
i=O for.
which are derived by forcing the real and imag-
inary parts of the transfer function of the time- 3.1 Time-Optimal Control
delay filter to zero at s = a f j w . Note, that this is The time-optimal control profile for the un-
equivalent to the conditions given in Equations 2. damped benchmark problem can be determined
by solving the following parameter optimization
To satisfy the boundary conditions for the rigid problem which is derived by exploiting the anti-
body for the rest-to-rest maneuver, the transfer symmetric properties of the control profile (Fig-
function of the time-delay filter should have two ure 13):
zeros at the origin of the complex plane to cancel
the rigid body poles, resulting in the constraint min J = T; (35)
equation
N
+ +
- ~ c o s ( w ( T-~T I ) ) 1 C O S ( W T=~0) (36)
1
CAi=O (30) L
Z(2T; - (2Tz - +T; - T ; ) = 1, (37)
i=O

and where
N
C A ~= oT ~
i=O
(31)
A0 = 1, A1 = -2, A2 = 2 , A3 = - 2 , A4 = 1. (38)
The constraint to satisfy the total rigid body dis-
To desensitize the controller to the frequency of
placement is the flexible mode, two switches are added to the
control profile and the problem is
N
1 (TN - T,)2
O(t, = T N )= - C A i (32) min J = T: (39)
2 i=o
2 . -2COS(WT31) + 2COS(WT32) +1+ COS(WT3) = 0 (40)

Finally, to desensitize the control profile to uncer-

-2T31SZn(WT31) + 2T32SiR.(WT32) +
T 3 S i n ( W T 3 ) = 0 (41)

2T: - ( 2 T 3 - TI)' + (2T3 -T +

z)' - T i T; - T f = 2. (42)
tainties in the location of the under-damped poles
of the system, constraints are derived which place where the double subscript represent Pij = Pi-Pj.
multiple zeros of the time-delay filter at the esti- Having solved for the switch times, their optimal-
mated location of the poles of the system. These ity have to be verified. This is achieved by solving

1725
 ~ 7
TI-- - - - - - - - 2T21T1

2~ ~
~
erate time-optimal commands for a wide range of
flexible systems [55].

3.2 F'uel/Time Optimal Control

The F'uel/Time optimal control profile for the
benchmark problem can be represented as shown
in Figure 15 where the anti-symmetric property
for undamped systems is exploited.

Figure 13: Antisymmetric Control Profile

for the costates and the resulting switching curve

is used to corroborate the optimality of the solu-
tion. Details of these techniques have been pre-
sented in [7], [15], [14], [29] etc.

Figure 14 illustrates the time-optimal control pro-

file and the robust control profile and the corre- Figure 15: Fuel-Time Optimal Control Profile
sponding energy sensitivity curves. It is clear that
local to the nominal spring stiffness there is a sig- Minimizing the cost function
nificant improvement of the insensitivity of the ro-
bust control profile. (43)

where the cost function J is a weighted ( a > 0)

combination of the maneuver time and fuel con-
sumed results in the parameter optimization prob-
lem:
min J = 2 T 4 + 2a [T43 +Th] (44)
0
I
1 2 3 4 5 6
I COS(WT43) - COS(WT42) - cOs(WT41) +COS(WT4) 0 (45)
Tim 1
2(-Tf - Tz T: + 2T4(T1 + T2 - T3)) = 1. (46)
- lim3-cpllmal Pmlk
- - Robust cOnrol Profile Solving for the optimal control profile as a func-
tion of a , the control profile changes from a three
switch bang-bang profile for a: = 0, to a six switch
bang-off-bang control profile which simplifies to a
----- two switch bang-bang profile for a: > 0.6824, for
005
---
06 07 08 OS 1
k,
11 12 13 14
the benchmark problem, as shown in Figure 16.

Figure 14: Control and Residual Energy Variation Hartmann and Singh [50] present a general devel-
opment of the necessary and sufficient conditions
for optimality of the fuel/time optimal control pro-
The variation of the structure of time-optimal con- files for system of order higher than the benchmark
trol profiles as a function of damping has been problem.
illustrated by Pao [37] and Singh [35]. For a two-
mass system connected by a spring and a damper, 3.3 Minimax Control
the control profile changes from a three switch for The techniques to desensitize the controller to
un-damped systems to a five switch and back to modeling errors which have been presented to this
a three switch control profile, as the damping is point, only require information about the nomi-
increased. Time-optimal control profile for mult- nal values of the model parameters. The result-
mode systems have been derived in [45]. Tuttle ing controllers are robust in the vicinity of the
and Seering, developed a Matlab toolbox to gen- nominal parameters of the system. However, in

1726
 final displacement states of the system. The above
equation will be referred to as the pseudo-energy
function since it is associated with a hypothetical
spring whose potential energy is zero when y =
yf. The pseudo-energy function is evaluated at the
final time, i.e., the end of the maneuver. The last
0 2 4 6
term is added to guarantee that the cost function
is positive definite.

Minimax bang-bang controllers are designed for

the floating oscillator benchmark problem to illus-
trate the proposed technique. The time-optimal
control of the benchmark problem is a 3-switch
0 1 2 3 4 0 2 4 6 8 1 0 anti-symmetric bang-bang control profile. The
thin solid line in Figure 17 illustrates the vari-
Figure 16: Spectrum of Fuel/Time Optimal Control ation of the residual energy of the time-optimal
control to variations in the spring stiffness. A 3-
numerous applications, the range of uncertainty is switch minimax controllers is designed without the
know and often the distribution of the uncertainty constraint that the residual energy be zero at the
is know. There is thus a desire to formulate opti- nominal value of the spring stiffness. The thin
mization problems which includes the information dashed line illustrates that the maximum mag-
about the range of uncertainty, in the design of nitude of the residual energy over the uncertain
controllers. The technique expounded in this sec- range (0.7 < k < 1.3) has been minimized, but at
tion is germane to the design of the input shapers, the cost of non-zero residual energy for the nomi-
as well as for the design of saturating controllers. nal model. Next, the 5-switch robust time-optimal
control profile is designed to force the slope of the
For a mechanical system undergoing rest-to-rest sensitivity curve for the nominal system to go to
maneuvers, the model can be represented as zero. The think solid line illustrate the signifi-
cant reduction of the residual energy over the en-
My + C ( p ) $+ K(p)y = Dr (47) tire uncertain range. However, this is achieved
at a cost of increased maneuver time. Finally, a
where M is a positive definite mass matrix, and K
5-switch minimax controller is designed and the
and C, the stiffness and damping matrices. K is
thick dashed line illustrates the improvement over
positive semi-definite when the model of the sys-
the robust 5-switch time-optimal controller. The
tem includes rigid body modes and is positive defz
resulting sensitivity curve is similar to the curve
inite otherwise. p is a vector of uncertain param-
for extra-insensitivity input shapers [40].
eters whose elements satisfy the constraints:
lb
pi I pi 1.Piub (48) 3.4 Finite Jerk Time-Optimal Control
Recently, Muenchhof and Singh [58] and Lim et
where pfb and pyb represent the lower and upper al. [53] have proposed a optimal control formula-
bounds on the parameters respectively. The goal tion which includes limits on the rate of change of
here is to design a saturating controller with the control (Jerk). The control rate profiles are bang-
objective of minimizing the maximum value of the bang or bang-off-bang as a function of the per-
residual energy mitted.jerk and the maneuver distance, for a rest-
min max F to-rest maneuver. The problem in [58], is posed
“ P as the design of time-delay filter which is parame-
l T 1 terized to generate a bang-bang or bang-off-bang
F = 5$ M $ + 5(Y-Yf - Yf ) + 5(Yr - ! I d 2 profile whose integral is the control input to the
(50) plant. Figure 18 illustrates the variation of the
where x is a vector of parameters which define switches and change of the structure of the con-
the saturating controller and yf corresponds to the trol profile as a function of permitted jerk for the

1727
 used for the design of robust control profiles for
nonlinear systems. Liu and Singh [41] proposed
a technique where the sensitivity state equations
are included in the problem formulation with the
constraint that the sensitivity states be forced to
zero at the final time. For the nonlinear system
2 = f(zU , &
- (51)
where p- is the vector of uncertain parameters, the
sensitivity state equations are

The control profile should in addition to satisfying

Figure 17: Residual Energy Distribution
the boundary conditions of the system states x,
must also force
benchmark floating oscillator problem.
(53)

To illustrate this approach, consider the bench-

mark problem with a nonlinear spring whose
model is
+
miyi + kl(yi - y2) ka(y1- y213 = 21 (54)
m2ji2 - ~ ( Y -I YZ) - k2(yl - ~ 2 =)0. ~ (55)
The sensitivity state equations are

ml-
dYl
dk2
+ kl-dY12
dk2
+ y;2 + 3k2yl2--
2 dYl2
dk2
=0 (58)
Ah J

Figure 18: Switch Time Trajectories

The time-optimal control profile is designed to sat-
It can be seen that the structure of the control pro- isfy the boundary conditions
file changes significantly. The thick vertical lines
indicate the value of jerk where switches collapse
or are spawned. It is interesting to note that the
maneuver time only increases marginally. as the
permitted jerk is decreased from 00 to 2.
(60)
Figure 19 illustrates the reduction in residual vi-
4 Nonlinear Systems bration of the desensitized time-optimal control,
compared to the time-optimal control in the vicin-
The technique presented for the design of con- ity of the nominal values of the spring coefficients.
troller which are robust to modeling uncertainties The maneuver time of the two time-optimal con-
included location of multiple zeros of the time- trol profile is t f = 4.1514, and for the desensitized
delay filter at the estimated location of the poles control profile is t f = 6.2439. It is clear the insen-
of the system. This approach obviously cannot be sitivity to model parameters is achieved at a cost
of increased maneuver time.
1728
 High tech manufacturing is perhaps the area with
the highest number of input shaping applications.
Shaping was an important component of a control
system developed for a wafer stepper [48]. Multi-
ple modes of a silicon-handling robot were elimi-
nated with input shaping [31]. Accuracy of coordi-
nate measuring machines has been improved with
command shaping [54], [27], [39]. The throughput
of a hard-disk-drive-head testing machine was sig-
nificantly improved with shaping [46]. Figure 21
shows the position response of the reading heads
14
during testing both before and after input shap-
ing was implemented. The greatly reduced set-
tling time allowed for much higher throughput.
Command shaping has been combined with vision
Figure 19: Sensitivity Plot for the 319 switch Control sensing and learning control on x-y-z gantry-type
automation machinery [62]. Figure 22 shows the
decrease in tip vibration of the machine as the
5 Applications learning controller adapts the input shaper to the
system vibration during repeated motion.
Given the simple approach and ease of implemen-
tation of basic input shaping techniques, they have 250
been used in a variety of applications. The wide
spread use is also attributable to the robustness
that can be added to many of the techniques. .I
v1
One major area of success has been on cranes and
crane-like structures. Starr wrote an early pa-
per that implemented a ZV-like shaping scheme
on a crane [6]. Groups a t Sandia and Oak Ridge
National labs have been especially active in this 3
-100
area [28], [20], [43]. Their approach has been sim- -’‘(!I 0.bl Oh2 0.b3 0.h 0.b5 Oh6
ilar to the input shaping described in Section 2; Time (sec)
however, the shaping filter utilized has often been Figure 21: Head Response During Testing
an IIR filter instead of a FIR filter. Their tech-
nique has also been utilized to control sloshing Applications involving shaping the commands to
fluids [43]. Another approach, which has been im- on-off actuators or saturating actuators are fewer
plemented on some large gantry cranes, designed than with real-time shaping, but there have still
input shapers to suppress vibration over the ex- been a number of successes. In fact, the crane
pected operating ranges of the cranes [44]. Fig- control scheme whose results are shown in Fig-
ure 20 shows the reduction in residual vibration ure 20 uses an on-off actuator switching algorithm
as a function of the hoist cable length. to accomplish the vibration reduction. Recently,
a technique for shaping the momentum dumping
E 0 Measured of spacecraft was adopted as a baseline design for
the next generation space telescope [60].
40
g 20- References
[l] Smith, 0. 3. M., “Posicast Control of Damped Oscil-
& 05 10 15 20 25 30 latory Systems”, Proc. of the IRE, 1957, pp 1249-1255.
Cable length (ft) [2] Calvert, J . F. and Gimpel, D. J., “Method and A p
paratus for Control of System Output Response to System
Figure 20: Vibration Reduction vs. Cable Length Input”, U.S. Patent #2,801,351, 1957.

1729
 [16] Murphy, B. R. and Watanabe, I., 1992 “Digital S h a g
ing Filters for Reducing Machine Vibration”, ZEEE Trans-
actions on Robotics and Automation, Vol. 8(April), pp. 285-
289.
[17] Singer, N. C. and Seering, W. P., “An Extension of
Command Shaping Methods for Controlling Residual Vi-
bration Using Frequency Sampling”, ZEEE International
Conference on Robotics and Automation, Nice, France,
-8 L 1992.
0 2000 4000 6000 8000
Time (msec) [18] Magee, D. and Book, W., “The Application of
Input Shaping to a System with Varying Parameters”,
Figure 22: Adaptive Command Shaping Japan/USA Symposium on Flexible Automation, 1992.
(191 Singer, N. C. and Seering, W. P., “An Extension of
Command Shaping Methods for Controlling Residual Vi-
[3] Tallman, G. H., Smith, 0. J. M., 1958, “Analog
bration Using Frequency Sampling”, IEEE International
Study of Posicast Control”, IRE Transactions on Automatic
Conference on Robotics and Automation, Nice, France,
Control, Vol. 3, pp 1421.
1992.
[4] Farrenkopf, R. L., 1979, “Optimal Open-Loop Ma-
neuver Profiles for Flexible Spacecraft”, J . of Guidance, and [20] Noakes, M. W. and Jansen, J. F., “Generalized In-
puts for Damped-Vibration Control of Suspended Pay-
Control, Vol. 2, No. 6, pp 491-498
loads”, Robotics and Autonomous Systems, vol. 10, pp. 199-
[5] Swigert, C . J., “Shaped Torques Techniques”, J. of 205, 1992.
Guidance and Control, Vol. 3, 1980, pp 460-467.
(211 Wie, B. and Bernstein, D., “Benchmark Problems
(61 Starr, G. P., “Swing-Free Transport of Suspended for Robust Control Design”, J . of Guidance, Control and
Objects With a-Path-Controlled Robot Manipulator”, J. Dynamics, Vol. 15, No. 5, 1992, pp 1057-1058.
of Dynamic Systems, Measurement and Control, vol. 107,
pp. 97-100, 1985. [22] Singh, T. and Vadali, S. R., 1993, “Input-Shaped
Control of Three-Dimensional Maneuvers of Flexible Space-
[7] Singh, G., Kabamba, P. T., McClamroch, N. H., craft”, J . of Guidance, Control, and Dynamics, Vol. 16(6),
1989, “Planar, Time-Optimal, Rest-to-Rest Slewing Ma-
pp. 1061-1068.
neuvers of Flexible Spacecraft”, J . of Guidance, Control,
and Dynamics, Vol. 12, No. 1, pp 71-81. [23] Singh, T. and Vadali, S. R., 1993, “Robust Time-
Delay Control”, A S M E J. of Dynamic Systems, Measure-
[SI Singer, N. C., and Seering, W. P., “Preshaping Com-
ment, and Control, Vol. 115, pp. 303-306.
mand Inputs to Reduce System vibrations”, A S M E J .
of Dynamic Systems, Measurement and Control, Vol. 112, [24] Singh, T., Heppler, G. R., 1993, “Shaped Input for
1990, pp 76-82. Multimode System” A S M E J . of Dynamic Systems, Mea-
surement and Control, Vol. 115, 1993, pp 341-347.
[9] Junkins, J., L., Rahman, Z., Bang, H., 1990, “Near-
Minimum Time Maneuvers of Flexible Vehicles: A Lia- (251 Khorrami, F., Jain, S., and Tzes, A., “Adaptive Non-
punov Control Law Design Method, Mechanics and Con- linear Control and Input Preshaping for Flexible-Link Ma-
trol of Large Flexible Structures, Published by the American nipulators”, American Control Conf., San Francisco, CA,
Institute of Aeronautics and Astronautics, Inc., Washing- 1993.
ton. [26] Tzes, A., and Yurkovich, S., “An Adaptive Input
[lo] Bhat, S. P. and Miu, D. K., 1990, “Precise Point- Shaping Control Scheme for Vibration Suppression in Slew-
to-Point Positioning Control of Flexible Structures”, J . of ing Flexible Structures”, ZEEE Transactions on Control
Dynamic Sys., Meas., and Control, Vol. 112(4), pp. 667-674 Systems Technology, Vol. 1, pp. 114-121, 1993.
[ll] Hablani, B. H., 1990, “Zero-Residual-Energy, Single- [27] Seth, N., Rattan, K. and Brandstetter, R., “Vibra-
Axis Slew of Flexible Spacecraft with Damping, Using tion Control of a Coordinate Measuring Machine”, IEEE
Thrusters: A Dynamic Approach”. Proc. of the 1990 A I A A Conf. on Control Apps., Dayton, OH, 1993.
Guidance, Navigation, and Control Conference, (281 Feddema, J. T., “Digital Filter Control of Remotely
[12] Singhose, William E., Singer, Neil C., Seering, War- Operated Flexible Robotic Structures”, American Control
ren P.,1990, “Shaping Inputs to Reduce Vibration: A Vec- Con!., San Francisco, CA, 1993.
tor Diagram Approach” Proc. of the 1990 IEEE Znter- [29] Singh, T. and Vadali, S. R., “Robust Time-Optimal
national Conference of Robotics and Automation, Vol. 2, Control: Frequency Domain Approach”, A I A A J . of Guid-
Cincinnati, Ohio, pp. 922-927. ance, Control and Dynamics, Vol. 17, No. 2, 1994, pp 3 4 6
(131 Hyde, J. M., Seering, W. P., 1991, “Multiple Mode 353.
Vibration Suppression in Controlled Flexible Systems” M I T [30] Tuttle, T. D. and Seering, W. P., “A Zero-
Space Engineering Research Center report, S E R C #3-91 placement Technique for Designing Shaped Inputs to S u p
(141 Ben-Asher, J., Burns, J. A. and Cliff, E. M., 1992, press Multiple-mode Vibration”, American Control Conf.,
“Time-Optimal Slewing of Flexible Spacecraft”, J . of Guid- Baltimore, MD, 1994.
ance, Control, and Dynamics, Vol. 15(2), pp. 360-367. [31] Rappole, B. W., Singer, N. C., and Seering, W. P.,
[15] Liu, Q . and Wie, B., 1992, “Robust Time-Optimal “Multiple-Mode Impulse Shaping Sequences for Reducing
Control of Uncertain Flexible Spacecraft”, J . of Guidance, Residual Vibrations”, 23rd Biennial Mechanisms Confer-
Control, and Dynamics, Vol. 15(3), pp. 597-604 ence, Minneapolis, MN, 1994.

1730
1321 Singh, T., Vadali, S. R., 1995, “Robust Time-Delay [49] Pao, L. Y. and Singhose, W. E., “Robust Minimum
Control of Multimode Systems”, International 3. of Control, Time Control of Flexible Structures”, Automatica, 34(2):
Vol. 62, NO. 6, pp 1319-1339. 229-236, Feb. 1998.
(331 Vadali, S. R., Carter, M. T., Singh, T., and Ab- [50] Hartmann, R., and Singh, T., “Fuel/Time Optimal
hyankar, N. S., 1995, “Near-Minimum-Time Maneuvers of Control of Flexible Structures: A Frequency Domain A p
Large Structures: Theory and Experiment”, J. of Guid- proach’, Journal of Vibration and Control, Sept., 1999, Vol.
ance, Control and Dynamics, Vol. 18, No. 6, pp 1380-1385. 5, No. 5, pp 795-817.
[34] Singh, T., 1995, “Fuel/Time Optimal Control of the [51] Singhose, W., Singh, T., Seering W., 1999, “On-Off
Benchmark Problem”, J . of Guidance, Control, and Dy- Control with Specified Fuel Usage”, ASME J. of Dynamic
namics, Vol. 18(6), pp. 1225-31. Systems, Measurement and Control, Vol. 121(2), pp 2 0 6
[35] Singh, T., 1996, “Effect of Damping On the Structure 212.
of Time-Optimal Controllers”, J. of Guidance, Control and [52] Pao, L. and Lau, M. A., “The Expected Residual Vi-
Dynamics, Vo1.19, No. 5, pp 1182-1184. bration of Traditional and Hybrid Input Shaping Designs”,
[36] Bodson, M., “Experimental Comparison of Two In- J . Guid., Contr., €4 Dyn., vol. 22, pp. 162-165, 1999.
put Shaping Methods for the Control of Resonant Systems”, [53] Sungyung Lim, Homer D. Stevens, and Jonathan P.
IFAC World Congress, San Francisco, CA, 1996. How “Input Shaping Design for Multi-Input Flexible Sys-
(371 Pm. L. Y., “Minimum-Time Control Characteristics tems”, ASME J. of Dynamic Systems, Measurement and
of Flexible Structures”, J . Guidance, Control, and-Dynam- Control, Vol. 121(3), pp 443-447.
ics, 19(1): 123-129, Jan.-Feb. 1996.
[54] Jones, S. and Ulsoy, A. G., “An Approach to Con-
[38] Singhose, W. E., Seering, W. P. and Singer, N. C., trol Input Shaping with Application to Coordinate Measur-
“Input Shaping for Vibration Reduction with Specified In- ing Machines”, J. of Dynamics, Measurement, and Control,
sensitivity to Modeling Errors”, Japan- USA Sym. on Flex- Vol. 121, pp. 242-247, 1999.
ible Automation, Boston, MA, 1996.
[55] Tuttle, and Seering, W. P. “Creating Time-Optimal
[39] Singhose, W., Singer, N., and Seering, W., “Improv- Commands with Practical Constraints”, J. Guid., Contr.,
ing Repeatability of Coordinate Measuring Machines with d Dyn., Vol. 22, NO. 2, 1999, pp. 241-250.
Shaped Command Signals”, Precision Engineering, Vol. 18,
[56] Kenison, M. and Singhose, W. ,“Concurrent Design
pp. 138-146, 1996,
of Input Shaping and Feedback Control for Insensitivity to
[40] Singhose, W., Derezinski, S. and Singer, N., “Extra- Parameter Variations”, Sixth Int. Workshop on Advanced
Insensitive Input Shapers for Controlling Flexible Space- Motion Control, Nagoya, Japan, 2000.
craft”, AIAA J. of Guidance, Control, and Dynamics, vol.
[57] Hindle, T., Singh, T., 2001, “Robust Minimum
19, pp. 385-91, 1996.
Power/Jerk control of Maneuvering Structures”, J. of Guid-
(411 Liu, S-W., and Singh, T., 1997, “Robust Time-
ance, Control and Dynamzcs, Vol. 24, No. 4, pp 816-826.
Optimal Control of Nonlinear Structures with Parameter
Uncertainties”, ASME J . of Dynamic Systems, Measure- [58] Muenchhof, M., Singh, T., 2001, “Jerk Limited Time
ment and Control, Vol. 119, No. 4, 1997, pp 743-748. Optimal Control of Structures”, Proc. of the 2001 ASME
International Mechanical Engineering Congress and Expo-
[42] Singhose, W., Banerjee, A. and Seering, W., 1997, sition
“Slewing Flexible Spacecraft with Deflection-Limiting In-
put Shaping”, J. of Guidance, Control, and Dynamics, Vol. [59] Park, U. H., Lee, J. W., Lin, B. D. and Sung, Y.
20(2), pp. 291-298. G . , “Design and Sensitivity Analysis of an Input Shaping
Filter in the z-Plane”, J. of Sound and Vibration, vol. 243,
[43] Feddema, J., Dohrmann, C. , Parker, G., Robinett,
pp. 157-171, 2001.
R., Romero, V. and Schmitt, D., “Control for Slosh-Free
Motion of an Open Container”, ZEEE Control Systems, vol. [60] Banerjee, A., Pedreiro, N., and Singhose, W., “Vibra-
17, pp. 29-36, 1997. tion Reduction for Flexible Spacecraft Following Momen-
tum Dumping with/without Slewing”, AZAA J. of Guid-
[44] Singer, N., Singhose, W., and Kriikku, E., “An In-
put Shaping Controller Enabling Cranes to Move Without ance, Control, and Dynamics, Vol. 24, 2001.
Sway”, ANS 7th Topical Meeting on Robotics and Remote [Sl] Lau, M. A. and Pm, L. Y., “Comparison of Input
Systems, Augusta, GA, 1997. Shaping and Time-Optimal Control of Flexible Structures”,
1451 Singhose, W., P m , L. Y., and Seering, W. P., “Slew- Proc. American Control Conf., Arlington, VA, pp. 1485-
ing Multi-Mode Flexible Spacecraft Using Zero Derivative 1490, June 2001.
Robustness constraints”, J . of Guidance, Control and Dy- [62] Rhim, S. and W. J. Book, “Noise Effect on Time-
namics, Vol. 20, 1997, pp 204-206. domain Adaptive, Command Shaping Methods for Flexible
[46] Singhose, W., Singer, N., and Seering, W., “Time- Manipulator Control”, IEEE Transactions of Control Sys-
Optimal Negative Input Shapers”, J . of Dynamic Systems, tems Technology, Vol. 9, No. 1, 2001, pp. 84 - 92.
Measurement, and Control, vol. 119, pp. 198-205, 1997. [63] Kenison, M. and Singhose, W., “Concurrent Design
(471 W. E. Singhose, L. J . Porter, T. D. Tuttle, and N. of Input Shaping and Proportional Plus Derivative Feed-
C. Singer, “Vibration Reduction Using Multi-Hump Input back Control”, Accepted to the ASME J. of Dynamic Sys-
Shapers”, J . of Dynamic Systems, Measurement, and Con- tems, Measurement, and Control.
trol, vol. 119, pp. 320-326, 1997.
[48] deRoover, D., Sperling, F. B. and Bosgra, 0. H.,
“Point-to-Point Control of a MIMO Servomechanism”,
American Control Conference, Philadelphia, PA, 1998.

1731