349

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL 41, NO 3, MARCH 1996

String Stability of Interconnected Systems

D. Swaroop and J. K. Hedrick

Abstract-In this paper we introduce the notion of string

stability of a countably infinite interconnection of a class of
nonlinear systems. Intuitively, string stability implies uniform
boundedness of all the states of the interconnected system for all
time if the initial states of the interconnectedsystem are uniformly
bounded. It is well known that the input-output gain of all the
subsystemsless than unity guaranteesthat the interconnectedsystem is input-output stable. We derive sufficient (weak coupling)
conditions which guarantee the asymptotic string stability of a
class of interconnectedsystems. Under the same weak coupling
conditions, string-stable interconnected systems remain string
stable in the presence of small structuralhingular Perturbations.
In the presence of parameter mismatch, these weak coupling
conditions ensure that the states of all the subsystems are all
uniformly bounded when a gradient-basedparameter adaptation
law is used and that the states of all the systems go to zero
asymptotically.

This paper is organized as follows: In Section I, we define

string stability and asymptotic string stability, we present
weak coupling conditions that guarantee string stability for
a class of interconnected systems, and we demonstrate that
exponential string stability is preserved under small structural
perturbations. In Section 11, we prove that every exponentially
string-stable interconnected system is string stable in the presence of small singular perturbations. In Section 111, we discuss
direct adaptive control of such interconnected systems, In
Section IV, we provide an example of longitudinal controller
design for vehicle-following systems.

11. STRINGSTABILITY
We use the following notations: Ilf,(.)llm, or simply Ilfil,
denotes SUpt>o
- Ifi(t)I, and Ilf,(O)II,
denotes SUP, Ifi(0)l. For
all P

I. INTRODUCTION

ARLIER research on interconnected systems focused on

vehicle-following applications [17], [14], [8], [23], [ll],
control of distributed systems (e.g., regulation of seismic
cables, vibration control in beams, etc.) [7], [19], signal
processing [4], and power systems [5]. Loosely speaking,
string stability of an interconnected system implies uniform
boundedness of the state of all the systems. For example, in
automated vehicle-following applications, tracking (spacing)
errors should not amplify downstream from vehicle to vehicle
for safety. Similarly, deflection at any point in a beam or a
rod should remain bounded at all times. Spatial discretization
and control of such distributed systems have a relevance to
the problem of string stability for interconnected systems.
Although a precise definition of string stability was not coined,
Kuo and Melzer [171 and Levine and Athans [ 141 were seeking
optimal control solutions to the automated vehicle-following
problem. Chu defined string stability in the context of vehicle
following [ l l ] . In [4], Chang introduces a stronger version of
stability for interconnected systems, namely, y-stability for
infinite interconnection of linear digital processors. Intuitively,
y-stability ensures that the state of all the systems decays to
zero exponentially in time and system index. In this paper, we
generalize the concept of string stability to a class of interconnected systems and seek sufficient conditions to guarantee their
string stability. We also examine their robustness to structural
and singular perturbations.
Manuscript received July 22, 1994; revised July 3, 1995. Recommended by
Associate Editor, A. M. Bloch. This work was supported in part by the PATH
program at the University of California, Berkeley.
The authors are with the Department of Mechanical Engineering, University
of California, Berkeley, CA 94720 USA.
Publisher Item Identifier S 0018-9286(96)02099-5.

< 03,llft(.)llp

Ilfi(0)llP denotes

or

llfzllp

denotes

(SFIf%(t)lPdt)tand

(E;If%(O)lP)t.

Consider the following interconnected system:

= f ( & ,z%--l,.
. ., z%-r+l)

&%

where i E N , x,-~= 0 V i 5 j , x E

(1)

R,f : R x . . . x Rn
T

-+

R and f(O,...,O) = 0.

times

Dejnition 1: The origin z, = 0, i E Af of (1) is string

stable, if given any E > 0, there exists a S > 0 such that

llx%(o)l)co< =+ SUPz 11x%(~)1lCu

< 6.

Definition 2: The origin x, = 0, i E N of (1) is asymptotically (exponentially) string stable if it is string stable and
z z ( t )+ O asymptotically (exponentially) for all i E N .
A more general definition of string stability is the following
one.
Definition 3 ( l p String Stability): The origin x, = 0, i E Af
of (1) is lp string stable if for all E > 0, there exists a S such
that

Definition 1 of string stability can be restated as -,I

string stability of Definition 3. Henceforth, we will deal
with string stability according to Definition 1. The following
theorem proves, under some weak coupling conditions,
that any countably infinite interconnection of exponentially
stable nonlinear systems is string stable. Clearly, a string of
uncoupled exponentially stable systems is exponentially string
stable. Intuitively, any interconnection of exponentially stable
systems is string stable, if the interconnections are sufficiently
weak. The following lemmas will be useful in proving the
theorems in this paper.

0018-9286/96$05.00 0 1996 IEEE

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 41, NO. 3, MARCH 1996

350

Lemma 1: Let r be a constant positive integer. Define

< 1, the rth
zT - C;/3,zr--3,p, > 0. If
degree polynomial PT(z)
has all its roots inside the unit circle.
Proof: Let zg be such that Pr(z0) = 0 and lzol > 1.Then

For the sake of convenience, we denote V ( x , ) by V,. Then

Pr(z)=

av,

= -f(xi,

8x2

0, ' . . , O )

which is a contradiction. This proves the lemma.

Lemma 2: Let &(t)2 0 V t 2 0, i E hi and if

with
> 0 and p, 2 0, j = 1, 2 , . . . and Po > E;" p j . For
all j I0, V, should be read as zero. Then, given any e > 0,
there exists a 6 > 0 such that

Using the inequality that xy 5

results in

w,

the above equation

llv,(O)llcO < 6 =+ sup ((v,((,< 6 .

2

Proof: Let M =

I IV ,I Ioo 5 MI I V ,(0) I I

Po-CT

01

> 1. It suffices to show that

We prove this by induction. From the

inequality, it follows that:

For i = 1, IIV,lloo 5 K(0) and the induction hypothesis

is valid. Assuming that the hypothesis is valid through the
integer i

If

Ci=2E,is

sufficiently small such that

Ci=213<

w-sc;=2> zc;=213
> 0.
I,

h a 1
CXg(CY[+Q'h)'
then
eh
Consequently, string stability follows from Lemmas 1 and 2.
Let d > 1. Define V(d-', t ) =
K(t)d-'. Clearly,
V(d-', t ) is defined whenever the weak coupling conditions
are satisfied and whenever /~x,(0)~~OO
exists
00

v=

K ( t ) d - 2 5 -vd-(r-l)P~(d)

Q1

ah

,=1

Here P,(,z) = zr -

- yx;=21,

Z
TP,zr-J

where p, =

2el-$"c

1,

I,.

Clearly, Pr(d)
> 0 whenever d > 1 > p(Pr(z)),
the spectral
radius of the polynomial Pr(z).
V -+ 0 exponentially and
hence, K ( t ) , x,(t) + 0 exponentially.
The above theorem can easily be generalized to nonautonomous interconnected systems. Consider the following
nonautonomous interconnection:

= ~llK(0)Ilm.

This proves that the induction hypothesis is valid for all i.

Therefore, SUP, IlV,llm i MllV,(O)llm.
Theorem I (Weak Coupling Theoremfor String Stability): If
the following conditions are satisfied:
f is globally Lipschitz in its arguments, i.e.,

where i E N , 2 , - ,
R" x . . . x RnX R
--f

= 0 Vz 5 j , x E R",f :
Rnand f(O,...,O)= 0.

r times

l f ( Y l , . . . ,YT)-f(~l,...,ZT)l
I ~ l I Y l - z l l + . . . + ~ T l y , - ~ , l . (2)

Then for sufficiently small I,, z = 2, . . . ,T , the interconnected

system is globally exponentially string stable.
Proof: Since the origin of 2 = f(x,0 , . . . , 0 ) is exponentially stable, by the converse Lyapunov theorem, there
exists a Lyapunov function V ( z )and four positive constants

Remark: If the following conditions are satisfied:

f is globally Lipschitz in its arguments, i.e.,

The origin of x = f ( z , 0, . . . , 0, t ) is globally exponentially stable, i.e., there exists a Lyapunov function, V(x)
such that
~z11x112I
V ( x ,t ) 5

Qh112ll2

( 5 ) Then for sufficiently small l,, i = 2, . . . ,T , the interconnected

system is globally exponentially string stable.

SWAROOP AND HEDRICK: STRING STABILITY

35 1

Proofi Let ~ ( x ,t ,) = v,. Clearly, rj, I

-~311~,11~
a411x211C;
1311x,-311. By the same arguments used in Theorem

1, the desired conclusion follows.

Another simple class of interconnected systems that arise in
the context of vehicle-following systems is given by
This condition is satisfied when
where i E N , xi-j 2 0 V i 5 j , x E R",
f : R"xR" xR" -+
R" and f ( 0 , 0, 0) = 0. If the following conditions are
satisfied:
f is globally Lipschitz in its arguments, i.e.,

The origin of i = f ( z , 0, 0) is globally exponentially

stable, i.e., there exists a Lyapunov function, V(x) such
that

This concludes the proof that string stability is robust to small

structural perturbations.
Remark (Weak Coupling for 12 string stability): Consider
the following interconnected system in which every subsystem
is connected only to its neighboring subsystems:
xi

= f(x,-1, x;,2,+1)

+ l21xz - Yzl + k123

- Y31.

The origin of j: = f ( 0 , x, 0) is globally exponentially

stable.
Then for sufficiently small 11 +Z3, the interconnected system
is globally exponentially 12 string stable.
Proofi Since the origin of x = f ( 0 , x, 0) is exponentially stable, by the converse Lyapunov theorem, there exists
a Lyapunov function V(x) such that

Q l l l ~ I l 2I
V ( 2 )I Qhl1412

Let V ( x i ) = V,. Then

+ (K-1 + dlK-2 + . . . + d"l-2V1)].

By the above lemma, string stability follows for sufficiently
for any d > 1.
small I1 12 dl. Define V ( t )=
Then, V
- K V where K > 0. Consequently, exponential
stability is guaranteed.
From the definition of string stability, it is clear that the
string stability of an interconnected system guarantees the
stability of every subsystem. Under some stronger coupling
,Z any finite interconnections of
condition, a1 >
one is asymptotically string stable. In the vehicle-following
applications, although the number of vehicles in every platoon (electronically interconnected system of vehicles) will
be finite, it is necessary that the stability of the platoon
be independent of the size of the platoon to prevent the
saturation of the input actuators. Another interesting feature
about the string stabilizy of an interconnection of exponentially
stable systems is that it is preserved under small structural
perturbations. Consider
2%= f ( & ,. . 1 %-,+1)
f f p ( Z , , . . . ,&-,+1).

+ +

Cf"

YE;=,

'

As before, x, = 0 V i 2 0. If the following conditions are

satisfied:
f is globally Lipschitz in its arguments, i.e.,
lf(x1, 5 2 , 23)-f(Yi, Yz, Y3))
5 lilxi - Y i 1

Then for sufficiently small 12 d l , the interconnected system

is globally exponentially string stable.
Proof: From the Lipschitz property

i E

For the sake of convenience we denote V ( x ; )by

Theorem 1, we obtain

K.

As in

d-"(t)
for d sufficiently close to and
Define V(d, t ) =
greater than unity. Note that V(d, 0) is defined if 11x2(0)112is
defined. Differentiating V , we obtain

9( l l + l d )

cyh
- Z ( l 1 d - 1 + Z3d). If I1
then
Pz(1)
>
0 and Pz(d0) < 0 for all
/ 3 < oIB(cyi+Cyh)'
By
the
intermediate
value theorem, there exists
do 2
a d* > 1 such that Pz(d*) = 0 and Pz(d) > 0 for all

Define &(d) =
2a1cyi

(cy1

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 41, NO. 3, MARCH 1996

352

1< d

<

There exist positive constants,

d*. As a result

V ( d , t ) I -Pz(d)V(d, t ) 5 0

1< d

< d*

I V ( d ,0).

This guarantees exponential 12-string stability of the interconnected system. d* is a performance measure associated
with this interconnected system. For a general interconnected
system, only 12-string stability can be guaranteed.
It is desirable that the string-stability property be preserved
in the presence of parasitic actuator dynamics. In the next
section, we present the conditions which guarantee string
stability of the origin of the interconnected system in the
presence of such parasitic actuator dynamics. From here on,
we consider "look-ahead or lower-triangular systems" only,
and therefore the results would apply for 1,-string stability.
The results for lower-triangular systems that follow can easily
be extended to 12-string stability of general interconnected
systems.
111.

STRING STABILITY OF SINGULARLY

PERTURBED INTERCONNECTED SYSTEMS

Before proceeding to study the string stability of the interconnected system, we present a result on the stability of a
singularly perturbed system from [ 131.
Theorem 2 (Robustness of Exponentially Stable Nonlinear
Systems to Singular Perturbations): Consider the autonomous
singularly perturbed system

where x E R",z E R" and assume that the origin is an

isolated equilibrium point and the functions f 1 and 1 are
locally Lipschitz in an open connected set that contains the
origin. Let z = h l ( z )be an isolated root of 0 = g(z, z),such
that hl(0)= 0. Let y = z- h l ( s ) . If the following conditions
are satisfied:
The reduced system is exponentially stable, i.e., there
exists positive constants a1, ah, a1, a3 and a Lyapunov
function V ( X )such that

The boundary layer system is exponentially stable, uniformly for frozen X, i.e., there exists positive constants
pz, p h , a2, ~ 1and
, a Lyapunov function w(z, y) such
that

and y such that

which implies that

V ( d ,t ) 5 V(d,O)e-P"(d)t

P2

. Then the origin of the singularly perturbed

Let t* =
system is exponentially stable for 0 < E < E * .
Proof: See Theorem 2.1 and [13, Corollary 2.21.
Intuitively, the origin of the perturbed interconnected system
will be string stable if the origin of every perturbed subsystem
is stable and the origin of the "reduced" interconnected system
is string stable. This observation leads us to the following
theorem.
Consider the following perturbed interconnected system:
li.2

22,

= f ( G , 2 2 , zz-l,..',zz-T+l)
= g(z,, 2,)

where
f : R" x

R"

i EN

(10)
(11)

R" x ... x R" t R"

( r - 1 ) times

g:RnxRm+Rm.Letf(O,~..,O)=O,g(O,O)=O,andlet
z, = h(z,,... , ~ , - ~ + 1be) an isolated root of 0 = g(x,, 2,).
Let 7 ~ 2= z, - h(x,),h(0) = 0, and f,g, and h be sufficiently
smooth Lipschitz functions.
Theorem 3 (Robustness of Exponentially Stable Interconnected Systems to Singular Perturbations): If the following
conditions are satisfied:
1) Let there exist a Lyapunov function, V ( X , )such
,
that

j=2

These conditions imply the string stability of the interconnected of reduced (unperturbed) systems.
2) There exists a Lyapunov function W ( z , ,y;) such that

I PzllXzll I l Y z l l + ~ 1 1 Y z 1 1 2 + ~ y 3 1 1 ~ ~ - J + l l 1 2
3=2

with yJ > 0. This condition implies the exponential

stability of the singularly perturbed individual systems.
Then the singularly perturbed interconnected system is string
stable.

SWAROOP AND HEDRICK STRING STABILITY

353

Iv. ADAPTIVECONTROL OF INTERCONNECTED SYSTEMS

Consider the following open-loop interconnected system:

").
E,=,
'

Otherwise, let 5 = min \-%-

Define v(z,, y,) =

Et

Ph

i ( V ( x , ) + kW(x,, y,)). Using shorthand notation v, for

v(z,,g,), V , for V(z,), and W, for W(x,, y,), there exists
a

PI

such that

= fo(Ez,

p=
.(P)
2%
.(P+1)
2%

Yz

where the equation is shown at the bottom of the page. Since

A() is a continuous function of E , define

Since

kc;=, < E,'=,

a l j from Assumption 1, it follows

k ) 0.
2 ) By the
that F ( 0 ) > 0 and F ( 4 a 1 7 ~ ~ ~ ~ ; & <
intermediate value theorem, there exists Ed such that 0 < Ed <
and "0 < E < E d , F ( E )> 0. Therefore
4a17k+(p;+P2k)2
4aiU k
r

L;, 5

-W(lI4l2+ llVz1l2) + C("1j+ k7,)

j=2

11~%-3+1112

By an argument similar to that in Theorem 1, there exists

a constant K > 0, such that Ilv;(.)llm 5 KIIv,(0)llm. This
proves that the interconnection of singularly perturbed systems
is string stable VO < E < Ed.
It also follows, by an argument similar to that in Theorem
1, that v; -+ 0 exponentially.
The above theorem justifies the use of control based on the
reduced (unperturbed) system model.

".

,Et-r+l)

+ g(tz)u,

where ,$, E R p + q + l and where p and q are positive integers.

As assumed earlier, & 3 0 for all j 5 0. f o , g are smooth
vector fields, U, E R and z E N . The output of the zth
subsystem is h, = h(&)with h, E R. The objective is to find
a control such that the states of the closed-loop interconnected
system are always bounded and go to the origin, Ez = 0,
i E N , asymptotically.
The following assumptions are used for obtaining the control effort and analyzing the closed-loop behavior of the
interconnected system:
There exists a global diffeomorphism x, = 41(E,), yz =
42(&)with z, E RP+', gz E Rq,and
$)

Et-1,

(2)

z,

(3)
2,

(PS1)

z,
= OTW
f
f(E,,

Ez-1,.

. . , E,-r+l)

+ Q,TW,(E,)u,

= V ( 2 % Y, J

where 2:) is the jth component of the vector z, and

zZ(l) = h,. The above condition implies the adaptive
linearizability of the open-loop system with a strict relative degree equal to p
1. For details on adaptive
linearizability, see Sastry and Isidori [22]. The vector
fields, f o , g are implicitly assumed to be linearly parameterizable in the con;tant papmeters and will, henceforth,
be represented by 8f and Os, respectively. Similarly, the
parameter estimation errors are given by O f , and 8,. y,
represents the state that will be rendered unobservable by
an input-output linearizing control. In other words, the
dynamics of y, represent the internal dynamics of the zth
system.
(The origin 00 yz = ~ ( 0yz)
, is globally exponentially
stable. This assumption states that the zero dynamics
of every system in the open-loop interconnected system
is exponentially stable. This assumption is required to
establish that y, is bounded uniformly in i when xz is
uniformly bounded in i . A more general form of internal
dynamics that arises in such interconnected systems is of
the following form:

Yz

= ~ ( z z yz,
,

~z-1,

yz-1,.

' . > xz-r+l,

yz-r+l).

To analyze the closed-loop interconnected system with

such an internal dynamics, additional weak coupling conditions similar to those in Theorem 1 (on the magnitude

E E E TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 41, NO. 3, MARCH 1996

354

of the Lipschitz constants associated with y arguments of

7 ) have to be imposed to conclude that yi is uniformly
bounded in i if z; are uniformly bounded in i. To keep
the analysis simple, we will, however, not use this general
form of internal dynamics.
Every system avails the information of its state and the
information of the states of "T" systems preceding it. This
assumption is necessary to generate a feedback linearizing
law which guarantees that the states of all the systems are
bounded uniformly in i . The above assumption enables
us to define Si such that S; = 0 describes the desired
closed-loop (string stable) dynamics. For example, one
could define

si =

,p +

+ . . . + 6P z!l)

Glzzp-l)

behavior of the internal dynamics associated with this system.

Any adaptively linearizable nonlinear system with a coupling
(interconnecting) control law yields this form of equations.
To analyze the effect of parameter adaptation, we assume the
following:
1) There exists a Lyapunov function V ( x , ) (for convenience, K),such that

-l1llZi1I2

+~ ~ j l l ~ i - i + l l 1 2
j=2

(1)

&+lzi-,

where s p 61sp-' +. . . 6, is a Hunvitz polynomial with

real roots. On the surface, Si = 0, llz;llm < IIzi--lIlm
if Sp+l is sufficiently small. In matrix form, Si can be
defined compactly as
i i

with 11 >
1,.
2) There exists a Lyapunov function W,(y,) (for convenience, W,), such that

= f d ( ~ i , .. . , ~ i - ~ + i )bxSi

PlllVzll2 IWz IPhllYzllZ

aw,
--4(zz,

where bx = [O, . . . , 0 , '1 and 5 ; = [zjl),

. . . ,z?']*. f d
is a smooth vector field and it satisfies the weak coupling
conditions described in Theorem 1 so that the dynamics
on the surface Si = 0 is string stable. Algebraically, Si
should be understood as

si = zz!p) + $ d ( Z i ,

5i-1,.

. . , zz-r+l).

Here $d is a smooth scalar function.

The control input ui should be chosen to drive Si to the
surface Si = 0. To obtain the control effort, differentiate Si

+ 4Jd(xi,. > xi-r+1)

= Opqti,.. . , t i - r + l ) + e;wg(li)ui

Si= z(P+l)
,

Choose ui such that

Obtaining control effort requires inversion of jgW, which

may be singular. If it is known that 10gWgl > C where C
is a generic positive constant, projection algorithms could be
employed to counter this problem.
As seen earlier, the closed-loop dynamics of any adaptively
linearizable nonlinear systems with a coupling (interconnecting) control law can be cast in the following form:
xi-1,. . . ,xi-r+i)

+ g:W(x;,

si

= -AS,

!A

= 4 ( Z i , yi,

Si)

yz,

+ Q4llYzll l l 4
We assume the exponentially stable behavior of the
zero dynamics. Assumptions 1 and 2 enable the string
stability of the interconnected system in the absence of
parameter mismatch.
3 ) W ( z , ,y, S,, ~ ~ - 1. ., ,.x,-,+1) is bounded for all its
bounded arguments.
Theorem 4 (Effectiveness of Purameter Adaptation for Interconnected Systems): Under the above mentioned conditions,
the following parameter adaptation law:

SUP,

52-1,.

SUP, IISZ(.)lloo>

SUP,

11mm

xz(t),S Z ( t )--t 0 asymptotically for all i.

are

+ ~ ~ r -Using
l ~ ,the. adaptation

Proofi Let V,, = S,"

law

va,= -2xs,"

. . ,5ipr+1)
(12)

where bx = [0 . . . 0 1IT, = 0,- 8; where 0, is the estimate

of the parameter, and 0; is the actual (constant) value of
the parameter. From the first equation, Si = 0 describes the
desired closed-loop dynamics. The second equation describes
the dynamics of Si, and the third equation indicates the

I l ~ Z ( ~ > l l O o ~

bounded;

+ bxSi

si,

sz, Yz) I -Q211Yzl12 + a311vzll IS21

8, = -rw(z,, . . . , z z - - r + l ) ~ z , r > 0
guarantees that for all bounded I15, (0) 11 m , I IS,
(0) 1 I oo,
I I e, (0)I1

BTWf(J2,. . . , ti-r+l) + e p g ( E i ) u i
f '$d(xi,. ' ' , %i-r+l) = -As,.

ii = f d ( x i ,

ay2

Similarly

SWAROOP AND HEDRICK STRING STABILITY

355

3) The dynamics of S; are usually given by

and

S, = -Xsign(S,) + & ~ ( x ,Y,,

, s,, x 2 - 1 1 * . . , ~ t - r + i )
then the following adaptation law should be used:

0; = - r W ( x ; , . . .
Calculating

along the trajectories of z;

. a&
% = ~ [ f d ( x i , * ~ * : x Z - r fbASi1
+l)
r

z ; - ~ + ~sign
)

r >0

(Si),

to conclude that sup, IIxilloo,supil(S;l(,, supillslloo

are bounded and that z;(t),S ; ( t ) -+ 0 asymptotically
for all i .
The proof of the above remark is similar to the proof of
Theorem 4.
V. EXAMPLE:VEHICLE-FOLLOWING
SYSTEMS

Define e; =

&.

For a good overview on vehicle-following systems, the

readers are referred to [2], [3], 1111, [8], and 1231. The longitudinal constant spacing vehicle-following controller designed
by Hedrick [8] and used for parameter adaptation by Swaroop
[28] will be considered here. A simple longitudinal vehicle
dynamic model for the ith vehicle in the platoon is given by

Then

.. = U , - czx; - F,

2,

M,

where 5%is the position of the ith vehicle in the platoon

with respect to an inertial frame, U, represents the propulsivehraking effort, c,x: is the aerodynamic drag force, and F,
is the tire drag acting on the ith vehicle. The control objectives
wherep = 2, 00. Since l1 >
~ ~ Z J ~5 ~MllS,(lp
e , ~ where
~ p
are :
M > 0 is a constant. Since sup, {IIS,lloIlS,ll2}
o,
< max
. ~ , ( tis) defined as x,(t) - x,-l(t) = L,, where L, is
the desired constant intervehicular spacing. E, ( t ) should
{K,
< 00, it follows that sup, {Ile,ll,,
Ilezl12}
go to zero asymptotically (exponentially) for every lead
< K1 for some positive K1. This implies that sup,{JIx,lloO,
vehicle maneuver.
11x,1)2} < a.
By Assumption 2 (that the zero dynamics
Y?
String stability of the platoon should be guaranteed, i.e.,
of every individual system is minimum phase), sup, Ily,(.)lloO
given
E > 0, there exists a 6 > 0 such that 11~,(0)11, <
exists. By Assumption 3, W ( x , , S,, y,, %,-I,. . , x,-,+l) is
6 =+ SUP,llEzllcc < E .
bounded. Therefore, S, E L,. Consequently, by Barbalats
In designing the controller [8], [28], it is assumed that the
lemma, S, -+ 0.
lead vehicle velocity and acceleration and lead vehicle relative
Observe that sup, 116, JIoo is bounded, since
position information can be communicated to every controlled
vehicle. Define

dg}

s, = 4 + 41% + q3(v, - 211) + q4

Since sup; {I(e;(Jco,Ile;ll2} are bounded, by Barbalats lemma,
ei -+ 0. Therefore, Vi, xg -+ 0.
Remarks:
1) In Assumption 1, S
i = 0 yields the desired string
stable dynamics.
2) Designing decentralized adaptive controllers for interconnected systems can be done in two steps:
a) Identify the desired closed-loop (string stable) dynamics. Design a controller to achieve the desired
closed-loop dynamics in the absence of parametric
uncertainty.
b) Use a gradient adaptation law to update the parameters.

x, - x1
(

3
L,

and U , is chosen such that S, + AS, = 0 for some X

U, is given by

The spacing error dynamics are given by

+ 43

--Ll

+ (41 + X ) i i - l + q1X.E;-1].

>

0.

IEEl3 TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 41, NO. 3, MARCH 1996

356

Let zi = ii
system is

+ XE,.

The dynamics of the

25

interconnected

Platoon string stability is guaranteed if the above interconnected system is string stable. The above interconnected
system is in the same form as in (7). String stability is
guaranteed if q l , q3, and q4 are chosen appropriately.
The theorems developed in this paper are sufficient conditions to guarantee string stability for nonlinear systems. For
linear systems, one could use input-output stability and some
results of this paper to conclude about string stability. For
example, Laplace transformation of the above equation yields

If 41, q3, q4 are chosen such that qlq3

> q4

for some positive K1, K2. String stability and, consequently,

exponential string stability follow immediately.

VI. CONCLUSIONS
In this paper we defined string stability, asymptotic, and
exponential string stability, for countably infinite interconnection of nonlinear systems. We derived sufficient conditions to
guarantee string stability for a class of interconnected systems
and demonstrated their robustness to small singular/structural
perturbations. The interconnections considered look ahead
lower-triangular interconnected systems) or are banded (finite look ahead and look back). We presented parameter
adaptation law (gradient type) to regulate the states of all
systems and to ensure uniform boundedness of all the states
and parameter estimation errors. Finally, we have illustrated
the theory developed in this paper with a controller design for
vehicle-following systems.
REFERENCES
[I] E. Barbieri, Stability analysis of a class of interconnected systems,

[8] J. K. Hedrick, D. H. McMahon, V. K. Narendran, and D. Swaroop,

Longitudinal vehicle controller design for IVHS systems, in Proc.
1990 Amer. Contr. Con$, San Diego, CA.
[9] L. R. Hung, R. Su, and J. Myer, Global transformations of nonlinear
systems, IEEE Trans. Automat. Contr., vol. AC-26, no. 1, pp. 24-31,
1983.
[ 101 A. Isidori, Nonlinear Control Systems New York: Springer-Verlag,
1989.
[ l l ] K. C. Chu, Decentralized control of high speed vehicle strings,
Transportation Res., pp. 361-383, June 1974.
[12] -,
Optimal decentralized regulation for a string of coupled systems, IEEE Trans. Automat. Contr., vol. AC-19, no.-6, pp. 243-246,
June 1974.
E131 P. Kokotovic, H. K. Khalil, and J. OReilly, Singular Perturbation
Methods in Control: Analysis and Design. New York: Academic, 1986.
[14] J. Levine and M. Athans, On the optimal error regulation of a string
of moving vehicles, IEEE Trans. Automat. Contr., vol. AC-11, no. 11,
pp. 355-361, NOV. 1966.
[15] L. E. Peppard, String stability of relative motion PID vehicle control
systems, IEEE Trans. Automat. Contr., vol. AC-19, no. 3, pp. 529-531,
Oct. 1974.
[I61 A. N. Michel, Stability analysis of interconnected systems, SIAM J.
Contr., vol. 12, no. 3, pp. 554-579, Aug. 1974.
[17] S. M. Melzer and B. C. Kuo, Optimal regulations of systems described
by a countably infinite number of objects, Automatica, vol. 7, pp.
359-366, 1971.
[18] K. S. Narendra and A. M. Annaswamy, Stable Adaptive Systems.
Englewood Cliffs, NJ: Prentice-Hall, 1989.
[19] U. Ozguner and E. Barbieri, Decentralized control of a class of
distributed parameter systems, in Proc. IEEE Con$ Decision Contr.,
Dec. 1985, pp. 932-935.
1201 W. Rudin, Principles of Mathematical Analysis. New York McGrawHill, 1964.
[21] S. S. Sastry and M. Bodson, Adaptive Control: Stability, Convergence,
and Robusmess. Englcwood Cliffs, NJ: Prentice-Hall, 1989.
[22] S. S. Sastry and A. Isidori, Adaptive control of linearizable systems,
IEEETrans. Automat. Contr., vol. 34, no. 11, pp. 1123-1131, Nov. 1989.
[23] S. Sheikholeslam and C. A. Desoer, Control of interconnected systems:
The platoon problem, IEEE Trans. Automat. Contr., Feb. 1992.
[24] -,
Indirect adaptive control of a class of interconnected nonlinear
dynamical systems, Int. J. Contr., vol. 57, no. 3, pp. 743-765, 1993.
[25] D. D. Siljak, Stability of large-scale systems under structural perturbations, ZEEE Trans. Syst., Man, Cybern., vol. SMC-2, pp. 657-663,
1972.
[26] -,
On stability of large-scale systems under structural perturbations, IEEE Trans. Syst., Man, Cybern., vol. SMC-3, pp. 415417,
1973.
[27] J. J. E. Slotine and W. Li, Applied Nonlinear Control. Englewood
Cliffs, NJ: Rentice-Hall, 1991.
[28] D. Swaroop and J. K. Hedrick, Direct adaptive longitudinal control of
vehicle platoons, Con$ Decision Contr., Orlando, FL, Dec. 1994.
[29] W. E. Thompson, Exponential stability of interconnected systems,
IEEE Trans. Automat. Contr., vol. AC-15, pp. 504-506, 1970.
[30] M. Vidyasagar, Nonlinear Systems Analysis. Englewood Cliffs, NJ:
Prentice-Hall, 1978.
[3 1] __ , Input-output analysis of large-scale interconnected systems,
vol. 29 of Lecture Notes in Control and Information Sciences. New
York Springer-Verlag. 1981.
[32] M. Vidyasagar and A. Vanelli, New relationships between input-output
and Lyapunov stability, IEEE Trans. Automat. Contr., vol. 27, no. 2,
pp. 481433, April 1982.

J. Dynamic Syst., Measurements, Contr., vol. 115, no. 3, pp. 546551,

Sept. 1993.
[2] R. E. Fenton and J. G. Bender, A study of automatic car following,
IEEE Trans. Veh. Technol., vol. VT-18, no. 3, Nov. 1969.
[3] R. J. Caudill and W. L. Garrard, Vehicle follower longitudinal control
for automated transit vehicles, J. Dynamic Syst., Measurements, Contr.,
vol. 99, no. 4, pp. 241-248, Dec. 1977.
[4] S. S. L. Chang, Temporal stability of n-dimensional linear processors
and its applications, IEEE Trans. Circuits Syst., vol. CAS-27, no. 8,
pp. 716-719, Apr. 1980.
[5] E. J. Davison and N. Tripathi, The optimal decentralized control of a
large power system: Load and frequency control, IEEE Trans. Automat.
Contr., vol. AC-23, no. 2, pp. 312-325, 1978.
[6] C. A. Desoer and M. Vidyasagar, Feedback System: Input-Output
Properties. New York Academic, 1975.
[7] M.L. El-Sayed and P. S. Krishnaprasad, Homogenous interconnected
systems: An example, IEEE Trans. Automat. Contr., vol. AC-26, no.
4, pp. 894-901, Aug. 1981.

D. Swaroop received the B.Tech degree from the

Indian Institute of Technology, Madras, in 1989 and
the M.S. and Ph.D. degrees from the University
of California, Berkeley, in 1992 and 1994, respectively, all in mechanical engmeermg.
Since 1995 he has been a Visiting Postdoctoral
Researcher for the University of California PATH
program. His research interests include modeling,
applied nonlinear and adaptive control theory, realtime control of mechanical systems, and vibrations.
Dr. Swaroop is a member of ASME.

SWAROOP AND HEDRICK: STRING STABILITY

J. K. Hedrick was a Professor of mechanical engineering at Massachusetts Institute of Technology

from 1974-1988. He is currently the Director of
the Vehicle Dynamics Laboratory in the Department
of Mechanical Engineering at the University of
California, Berkeley. His research interests include
the development of advanced control theory and
its application to a broad variety of transportation systems including high-speed ground vehicles,
passenger and freight rail vehicles, automobiles,
heavy trucks, and aircraft. He has consulted for
many industries in the transportation area including problems in design,
vibration, isolation, and electronic controller design. He has authored over
60 publications and edited several texts.
Dr. Hedrick has served on many national committees including the Transportation Research Board, the American National Standards Institute, ISO,
and the NCHRP. He is currently a member of the Board of Directors of
the International Association of Vehicle Systems Dynamics and Editor of the
Vehicle System Dynamics Joumal. He is a member of SAE and a Fellow of
the ASME where he has served as Chairman of the Dynamic System and
Controls Division.

357