IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO.

9, SEPTEMBER 2012

2203

Optimal Control of Vehicular Formations

With Nearest Neighbor Interactions
Fu Lin, Student Member, IEEE, Makan Fardad, Member, IEEE, and Mihailo R. Jovanovic, Member, IEEE

AbstractWe consider the design of optimal localized feedback

gains for one-dimensional formations in which vehicles only
use information from their immediate neighbors. The control
objective is to enhance coherence of the formation by making it
behave like a rigid lattice. For the single-integrator model with
symmetric gains, we establish convexity, implying that the globally
optimal controller can be computed efficiently. We also identify a
class of convex problems for double-integrators by restricting the
controller to symmetric position and uniform diagonal velocity
gains. To obtain the optimal non-symmetric gains for both the
single- and the double-integrator models, we solve a parameterized family of optimal control problems ranging from an easily
solvable problem to the problem of interest as the underlying
parameter increases. When this parameter is kept small, we
employ perturbation analysis to decouple the matrix equations
that result from the optimality conditions, thereby rendering the
unique optimal feedback gain. This solution is used to initialize a
homotopy-based Newtons method to find the optimal localized
gain. To investigate the performance of localized controllers, we
examine how the coherence of large-scale stochastically forced
formations scales with the number of vehicles. We establish several
explicit scaling relationships and show that the best performance
is achieved by a localized controller that is both non-symmetric
and spatially-varying.
Index TermsConvex optimization, formation coherence, homotopy, Newtons method, optimal localized control, perturbation
analysis, structured sparse feedback gains, vehicular formations.

I. INTRODUCTION
A. Background
HE control of vehicular platoons has attracted considerable attention since the mid sixties [1][3]. Recent technological advances in developing vehicles with communication
and computation capabilities have spurred renewed interest in
this area [4][12]. The simplest control objective for the one-dimensional (1D) formation shown in Fig. 1 is to maintain a desired cruising velocity and to keep a pre-specified constant dis-

Manuscript received November 15, 2010; revised June 27, 2011; accepted
October 20, 2011. Date of publication December 26, 2011; date of current version August 24, 2012. This work was supported by the National Science Foundation under CAREER Award CMMI-06-44793 and under Awards CMMI-0927720 and CMMI-09-27509. Recommended by Associate Editor M. Egerstedt.
F. Lin and M. R. Jovanovic are with the Department of Electrical and
Computer Engineering, University of Minnesota, Minneapolis, MN 55455
USA (e-mail: [email protected]; [email protected]).
M. Fardad is with the Department of Electrical Engineering and Computer
Science, Syracuse University, Syracuse, NY 13244 USA (e-mail: makan@syr.
edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAC.2011.2181790

Fig. 1. One-dimensional formation of vehicles.

tance between neighboring vehicles. This problem is emblematic of a wide range of technologically relevant applications including the control of automated highways, unmanned aerial vehicles, swarms of robotic agents, and satellite constellations.
Recent work in this area has focused on fundamental performance limitations of both centralized and decentralized controllers for large-scale formations [5], [7], [9][12]. For centralized linear quadratic optimal control formulations based on penalizing relative position errors it was shown in [7] that stabilizability and detectability deteriorate as formation size increases.
In [9], it was shown that merge and split maneuvers can exhibit
poor convergence rates even upon inclusion of absolute position
errors in cost functionals. In [5], it was shown that sensitivity of
spacing errors to disturbances increases with the number of vehicles for formations with localized symmetric controllers that
utilize relative position errors between neighboring vehicles. In
[11], the analysis of [5] was expanded to include heterogeneous
vehicles, non-zero time headway, and limited communication
range within the formation.
The motivation for the current study comes from two recent
papers, [12] and [10]. In [12], fundamental performance limitations of localized symmetric feedback for spatially invariant consensus and formation problems were examined. It was shown
that, in 1D, it is impossible to have coherent large formations
that behave like rigid lattice. This was done by exhibiting linear
scaling, with the number of vehicles, of the formation-size-nornorm from disturbances to an appropriately demalized
fined macroscopic performance measure. In 2D this measure increases logarithmically, and in 3D it remains bounded irrespective of the system size. These scalings were derived by imposing
uniform bounds on control energy at each vehicle.
For formations on a one-dimensional lattice, it was shown in
[10] that the decay rate (with the number of vehicles) of the least
damped mode of the closed-loop system can be improved by introducing a small amount of mistuning to the spatially uniform
symmetric feedback gains. A large formation was modeled as a
diffusive PDE, and an optimal small-in-norm perturbation profile that destroys the spatial symmetry and renders the system
more stable was designed. Numerical computations were also
used to demonstrate that the spatially-varying feedback gains
norm. The
have beneficial influence on the closed-loop

0018-9286/$26.00 2011 IEEE

2204

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

PDE approaches have also been found useful in the deployment

of multi-agents [13], [14] and in coordination algorithms [15].
Even though traditional optimal control does not facilitate incorporation of structural constraints and leads to centralized architectures, the optimal feedback gain matrix for both spatially
invariant systems [16] and systems on graphs [17] have off-diagonal decay. Several recent efforts have focused on identification
of classes of convex distributed control problems. For spatially
invariant controllers in which information propagates at least as
fast as in the plant, convexity was established in [18], [19]. Similar algebraic characterization for a broader class of systems was
introduced in [20], and convexity was shown for problems with
quadratically invariant constraint sets. Since these problems are
convex in the impulse response parameters they are in general
infinite dimensional. In [21], a state-space description of systems in which information propagates at most one unit in space
for every unit in time was provided and relaxations were used
to obtain suboptimal controllers. In [22], the optimal control
problem for switched autonomous systems was studied and optimality conditions for decentralization of multi-agent motions
were derived. In [23], convexity of the symmetric edge weight
design for minimization of the mean-square deviation in distributed average consensus was shown.
While references [18][21] focus on the design of optimal
dynamic distributed controllers, we develop tools for the design of optimal static feedback gains with pre-specified structure. Even though the framework of [18][21] does not apply to
our setup, we identify a class of convex problems which can be
cast as a semi-definite program (SDP). Furthermore, we show
that the necessary conditions for optimality are given by coupled matrix equations, which can be solved by a combination
of perturbation analysis and homotopy-based Newtons method.
We consider the design of both symmetric and non-symmetric
feedback gains and show that departure from optimal symmetric
design can significantly improve the coherence of large-scale
formations.
B. Preview of Key Results
We consider the design of optimal localized feedback gains
for one-dimensional formations in which each vehicle only uses
relative distances from its immediate neighbors and its own velocity. This nearest neighbor interaction imposes structural constraints on the feedback gains. We formulate the structured optimal control problem for both the single- and the double-integrator models. For single-integrators, we show that the structured optimal control problem is convex when we restrict the
feedback gain to be a symmetric positive definite matrix. In this
case, the global minimizer can be computed efficiently, and even
analytical expressions can be derived. For double-integrators,
we also identify a class of convex problems by restricting the
controller to symmetric position and uniform diagonal velocity
gains.
We then remove this symmetric restriction for both the singleand the double-integrator models and begin the design process
with a spatially uniform controller. We develop a homotopybased Newtons method that traces a continuous solution path
from this controller to the optimal localized gain. Along this

TABLE I
SUMMARY OF ASYMPTOTIC SCALINGS WITH THE NUMBER OF VEHICLES
FOR THE OPTIMAL SYMMETRIC AND NON-SYMMETRIC POSITION GAINS.
, IN THE QUADRATIC
THE -INDEPENDENT CONTROL PENALTY,
PERFORMANCE OBJECTIVE LEADS TO SIMILAR GROWTH WITH
OF
FORMATION COHERENCE AND CONTROL ENERGY (PER VEHICLE). ON THE
OTHER HAND, THE -DEPENDENT CONTROL PENALTY THAT PROVIDES
BOUNDED CONTROL ENERGY YIELDS LESS FAVORABLE COHERENCE

R=r I

homotopy path, we solve a parameterized family of the structured optimal control problems and obtain analytical solutions
when the homotopy parameter is small. We employ perturbation
analysis to decouple the matrix equations that result from optimality conditions, thereby rendering the unique optimal structured gain. This solution is used to warm-start Newtons method
in order to efficiently compute the desired optimal gains as the
homotopy parameter is gradually increased.
In the second part of the paper, we examine how the performance of the optimally-controlled formation scales with the
number of vehicles. We consider both macroscopic and microscopic performance measures based on whether attention is paid
to the absolute position error of each vehicle or the relative
position error between neighboring vehicles. We note that the
macroscopic performance measure quantifies the resemblance
of the formation to a rigid lattice, i.e., it determines the coherence of the formation. As shown in [12], even when local
positions are well-regulated, an accordion-like motion of the
formation can arise from poor scaling of the macroscopic performance measure (formation coherence) with the number of
vehicles . Our objective is thus to enhance formation coherence by means of optimal localized feedback design. In situations for which the control penalty in the quadratic performance
objective is formation-size-independent we show that the optimal symmetric and non-symmetric controllers asymptotically
and
scalings of formation coherence.
provide
However, this introduces similar growth of the control energy
(per vehicle) with . We show that bounded control energy can
be obtained by judicious selection of an -dependent control
and
scalings of formation
penalty, leading to
coherence for the optimal symmetric and non-symmetric controllers, respectively. These results are summarized in Table I
and they hold for both single- and double-integrators for formations in which each vehicle has access to its own velocity; see
Sections V and VI for additional details.
In addition to designing optimal localized controllers, we also
provide an example of a spatially uniform non-symmetric controller that yields better scaling trends than the optimal spatially
varying controller obtained by restricting design to symmetric
gains. This indicates that departure from symmetry can improve
coherence of large-scale formations and that the controller structure may play a more important role than the optimal selection
of the feedback gains. On the other hand, our results also show

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

that the optimal localized controller that achieves the best performance is both non-symmetric and spatially-varying.
If each vehicle has access to its own velocity and to relative distances from its nearest neighbors, we show similarity
between the optimal position gains and performance scaling
trends for single- and double-integrators. The latter observation is in agreement with analytical results obtained for spatially invariant formations [12]. We note that performance of
controllers that rely on relative measurements or unidirectional
position exchange can differ significantly for these two models.
For spatially-invariant formations with relative position and velocity measurements, it was shown in [12] that the global perforfor double-integrators and as
for
mance scales as
single-integrators. In Section V-B, we show that spatially uniscaling of the global
form look-ahead strategy provides
performance for the single-integrator model. On the other hand,
a look-ahead strategy that is not carefully designed can introduce unfavorable propagation of disturbances through formation of double-integrators [3], [5].
The paper is organized as follows. We formulate the structured optimal control problem in Section II, and show convexity
of the symmetric gain design for the single-integrator model in
Section III. For non-symmetric gains, we develop the homotopy-based Newtons method in Section IV. We examine performance of localized controllers for the single- and the double-integrator models in Sections V and VI, respectively, where we
provide several explicit scaling relations. We conclude the paper
in Section VII with a brief summary of our contributions.
II. PROBLEM FORMULATION
A system of identical vehicles moving along a straight line
is shown in Fig. 1. All vehicles are equipped with ranging devices that allow them to measure relative distances with respect
to their immediate neighbors. The objective is to design an optimal controller that uses only local information (i.e., relative
distances between the neighboring vehicles) to keep each vehicle at its global position on a grid of regularly spaced points
moving with a constant velocity.
We consider both the single- and the double-integrator
models of the vehicles. The double-integrators are employed
in many studies of vehicular formations; for example, see
[1][3], [5], [7], [9], [10], [12], [24]. On the other hand, the
single-integrator (i.e., kinematic) model is simpler and perhaps
more revealing in understanding the role of network topologies
[4], [23], [25][28]. As we show in Section VI, the singleand the double-integrator models exhibit similar performance
for formations in which each vehiclein addition to relative
positions with respect to its immediate neighborshas an
access to its own velocity. In the remainder of this section,
we formulate the localized optimal control problem for both
single- and double-integrators.
A. Single- and Double-Integrator Models
We first consider the kinematic model in which control input
directly affects the velocity,

2205

Fig. 2. Formation of vehicles with localized (a) non-symmetric; and (b) symmetric gains.

is the position of the th vehicle and

is the diswhere
turbance. The desired position of the th vehicle is given by
, where is the desired cruising velocity and
is the desired distance between the neighboring vehicles. Every
vehicle is assumed to have access to both and . In addition,
we confine our attention to formations with a known number of
vehicles and leave issue of adaptation, merging, and splitting for
future study.
The localized controller utilizes relative position errors between nearest neighbors,

where the design parameters

and denote the forward and
backward feedback gains of the th vehicle. In deviation vari, the singleables,
integrator model with nearest neighbor interactions is given by
(1a)
(1b)
where the relative position errors
and
can
be obtained by ranging devices.
As illustrated in Fig. 2(a), fictitious lead and follow vehicles,
, are added to the formation.
respectively indexed by 0 and
These two vehicles are assumed to move along their desired
, and they are not
trajectories, implying that
considered to belong to the formation. Hence, the controls for
the 1st and the th vehicles are given by

In other words, the first and the last vehicles have access to their
own global position errors and , which can be obtained by
equipping them with GPS devices.
For the double-integrator model,

we consider the controller that has an access to the relative position errors between the neighboring vehicles and the absolute
velocity errors,

2206

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

where
denotes the velocity feedback gain. In deviation vari,
ables,
the double-integrator model is given by

sition errors between the neighboring vehicles and the absolute

velocity errors, and

(2a)
(2b)
(VP2)

In matrix form, control laws (1b) and (2b) can be written as,

denote the zero and identity matrices, and

are defined in (3).
Upon closing the loop, we have

Here,

where , , and denote the position error, the velocity error,

and the control input vectors, e.g.,
. Furtherdiagonal feedback gains are determined by
more, the

and
is a sparse Toeplitz matrix with 1 on the main diagonal
,
and 1 on the first lower sub-diagonal. For example, for

and

where encompasses the penalty on both the state and the control. Here, is a symmetric positive semi-definite matrix and
is a positive scalar. The objective is to design the structured
feedback gain such that the influence of the white stochastic
disturbance , with zero mean and unit variance, on the performance output is minimized (in the
sense). This control
problem can be formulated as [29], [30]

(SH2)
(3)
determines the vector of the relative position errors
between each vehicle and the one in front of it; simdetermines the vector of the relative position errors
ilarly,
between each vehicle and the one behind it.
We will also consider formations with no fictitious followers.
In this case, the th vehicle only uses relative position error
th vehicle, i.e.,
implying
with respect to the
for the single-integrator model
that
and
for the double-integrator
model.
Thus,

B. Structured

Problem

The state-space representation of the vehicular formation is

given by

(SS)
,
For the single-integrator model (1), the state vector is
the measured output is given by the relative position errors
between the neighboring vehicles, and

(VP1)
For the double-integrator model (2), the state vector is
, the measured output is given by the relative po-

where denotes the structural subspace that belongs to.

As shown in [29], the necessary conditions for optimality of
(SH2) are given by the set of coupled matrix equations in , ,
and

(NC1)
(NC2)
(NC3)
Here, and are the closed-loop observability and controllability Gramians, denotes the entry-wise multiplication of two
matrices, and the matrix in (NC3) denotes the structural identity of the subspace under the entry-wise multiplication, i.e.,
, with
for the single-integrator model
and
for the double-integrator model. (For example,
.) In the absence of the fictitious folis imposed in (SH2) and
lower, an additional constraint
thus, the structural identity for the single- and the double-inand
, respectively.
tegrator models are given by
is a diagonal matrix with its main diagonal given by
Here,
.
Remark 1: Throughout the paper, the structured optimal feed. This
back gain is obtained by solving (SH2) with
choice of is motivated by our desire to enhance formation coherence, i.e., to keep the global position and velocity errors
and
small using localized feedback. Since the methods developed in the paper can be applied to other choices of , we

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

will describe them for general

computational results.

and set

when presenting

2207

Similarly, for the double-integrator model, we use the following performance weights
;

Macroscopic (global),
C. Performance of Optimal Localized Controller
Microscopic (local),
To evaluate the performance of the optimal localized con, we consider
troller , obtained by solving (SH2) with
the closed-loop system

(4)
where is the global or local performance output and is the
control input. Motivated by [12], we examine two state performance weights for the single-integrator model
;
Macroscopic (global):
,
Microscopic (local):
symmetric Toeplitz matrix with its first
where is an
row given by
. For example, for
,

(5)

The macroscopic performance weight

global (absolute) position errors,

penalizes the

and the microscopic performance weight

local (relative) position errors,

penalizes the

with
. These state weights induce the macroscopic and microscopic performance measures [12] determined
norm
by the formation-size-normalized

where
is the transfer function of (4) from to . The macroquantifies the resemblance of
scopic performance measure
the formation to a rigid lattice, i.e., it determines the coherence
of the formation [12]. On the other hand, the microscopic perforquantifies how well regulated the distances
mance measure
between the neighboring vehicles are. We will also examine
the formation-size-normalized control energy (variance) of the
closed-loop system (4),

which is determined by the

from to
.

D. Closed-Loop Stability: The Role of Fictitious Vehicles

We next show that at least one fictitious vehicle is needed in
order to achieve closed-loop stability. This is because the absence of GPS devices in the formation prevents vehicles from
tracking their absolute desired trajectories.
For the single-integrator model, the state-feedback gain
is a structured tridiagonal matrix satwhere is the vector of
isfying
all 1s. If neither the 1st nor the th vehicle has access to
, then
has a
its own global position, i.e.,
zero eigenvalue with corresponding eigenvector . Hence, the
closed-loop system is not asymptotically stable regardless of
and
. In the
the choice of the feedback gains
presence of stochastic disturbances, the average-mode (associated with the eigenvector ) undergoes a random walk and the
steady-state variance of the deviation from the absolute desired
trajectory becomes unbounded [12], [23], [28]. In this case,
other performance measures that render this average-mode
unobservable can be considered [12].
For the double-integrator model, the action of
on
is given by

where
then

is the -vector of all 0s. Thus, if

has a zero eigenvalue with corresponding eigenvector
. Therefore, for both the single- and the double-integrator models, we need at least one vehicle with access to its
global position in order to achieve closed-loop stability.

III. DESIGN OF SYMMETRIC GAINS FOR THE

SINGLE-INTEGRATOR MODEL: A CONVEX PROBLEM
In this section, we design the optimal symmetric feedback
gains for the single-integrator model; see Fig. 2(b). This is a special case of the localized design, obtained by restricting the forward and the backward gains between the neighboring vehicles
for
.
to be equal to each other, i.e.,
Under this assumption, we show that (SH2) is a convex optimization problem for the single-integrator model. This implies
that the global minimum can be computed efficiently. Furthermore, in the absence of the fictitious follower, we provide analytical expressions for the optimal feedback gains.
and
and let
Let us denote

norm of the transfer function

(6)

2208

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

For the single-integrator model, the structured gain becomes

a symmetric tridiagonal matrix

..

..

..

(7)

is Hurwitz if and only if is positive

Consequently,
definite, in which case the Lyapunov equation in (SH2) simplifies to

The application of [31, Lemma 1] transforms the problem (SH2)

of optimal symmetric design for the single-integrator model to

Fig. 3. Optimal symmetric gains for formations with follower () and without
= 50,
= , and
= 1. (2) are obtained by evalufollower (2) for
ating formula (9) and () are computed using the gradient method described in
Appendix A.

and the th entry of

in (8) simplifies to
for
. Consequently, the unique minimum of

(SG)
where
is a linear structural constraint given by (7).
(Specifically,
is a symmetric tridiagonal
matrix with the linear constraint (6).) By introducing an auxil, we can formulate
iary variable
(SG) as an SDP in and

is attained for

(9)
We also note that
which can be solved using available SDP solvers. Here, we have
used the Schur complement [32, Appendix A.5.5] in conjuncto express
as an LMI.
tion with
Next, we exploit the structure of to express in (SG) with
in terms of the feedback gains
between the
neighboring vehicles. Since the inverse of the symmetric tridican be determined analytically [33, Theorem
agonal matrix
is given by
2.3], the th entry of
(8)
yielding the following expression for

The above expression for is well-defined for

that
in (7); this is because the
guarantee positive definiteness of
. The global
closed-loop -matrix is determined by
minimizer of can be computed using the gradient method; see
Appendix A.
For the formations without the fictitious follower, we next derive explicit analytical expression for the global symmetric minof (SG) with
. In this case
imizer

(10)
where the third equality follows from (9). This result is used
to examine the performance of large-scale formations in
Section V-C.
Fig. 3 shows the optimal symmetric gains for a formation with
vehicles,
, and
. Since the fictitious leader
and the follower always move along their desired trajectories,
the vehicles that are close to them have larger gains than the
other vehicles. When no fictitious follower is present, the gains
decrease monotonically from the first to the last vehicle; see
in Fig. 3. In other words, the farther away the vehicle is from
the fictitious leader the less weight it places on the information
coming from its neighbors. This is because uncorrelated disturbances that act on the vehicles corrupt the information about the
absolute desired trajectory as it propagates from the fictitious
leader down the formation (via relative information exchange
between the vehicles). When both the fictitious leader and the
follower are present, the gains decrease as one moves from the
in Fig. 3. This
boundary to the center of the formation; see
can be attributed to the fact that the information about the absolute desired trajectories becomes noisier as it propagates from
the fictitious vehicles to the center of the formation.

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

IV. HOMOTOPY-BASED NEWTONS METHOD

the performance weight

uniform symmetric gain

In this section, we remove the symmetric feedback gain

restriction and utilize a homotopy-based Newtons method
to solve (SH2). In [29], Newtons method for general strucproblems is developed. For (SH2) with the specific
tured
problem data (VP1) and (VP2), it is possible to employ a
homotopy-based approach to solve a parameterized family of
problems, which ranges between an easily solvable problem
and the problem of interest.
In particular, we consider

2209

with respect to which the spatially

is inversely optimal. The problem of inverse optimality amounts

for which an a priori
to finding the performance weight
is the corresponding optimal state-feedback gain
specified
[34], [35]. From linear quadratic regulator theory, the optimal
where
is
state-feedback gain is given by
the positive definite solution of

(11)
is the initial weight to be selected,
is the dewhere
is the homotopy parameter. Note
sired weight, and
that
for
, and
for
. The homotopy-based Newtons method consists of three steps: (i) For
, we find the initial weight
with respect to which a
is inversely optimal. This is equivspatially uniform gain
alent to solving problem (SH2) analytically with the perfor. (ii) For
, we employ perturmance weight
bation analysis to determine the first few terms in the expan. (iii) For larger values of , we use
sion
Newtons method for structured
design [29] to solve (SH2).
We gradually increase and use the structured optimal gain obtained for the previous value of to initialize the next round of
iterations. This process is repeated until the desired value
is reached.
In the remainder of this section, we focus on the single-integrator model. In Section VI, we solve problem (SH2) for the
double-integrator model.

A. Spatially Uniform Symmetric Gain: Inverse Optimality for

One of the simplest localized strategies is to use spatially uniand

are diagonal matrices with
form gain, where
and
for all and some positive and . In particular, for
it is easy to show closed-loop stability and to find

..
.

..
.

For the kinematic model (VP1),

and
, with
, we have
and
. Thereguarantees inverse opfore, the state penalty
. The above
timality of the spatially uniform symmetric gain
can be applied to any structured gain
procedure of finding
that yields a symmetric positive definite , e.g., the optimal
symmetric gain of Section III.
B. Perturbation Analysis for
We next utilize perturbation analysis to solve (SH2) with
given by (11) for
. For small , by representing ,
, and as

substituting in (NC1)(NC3), and collecting same-order terms

in , we obtain the set of equations (PA), shown at the bottom
. Note that these equations
of the page, with
are conveniently coupled in one direction, in the sense that for
,
equations depend only on the solutions of
any
equations for
. In particular, it is easy to verify
are satisfied with
that the first and the third equations of
and with
identified in
can be obtained by solving
Section IV-A. Thus, the matrix
the second equation of
, and the matrices , , and
can be obtained by solving the first, the third, and the second

(PA)

2210

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

N=

r
"

Fig. 4. Formation with fictitious follower,

50, = 1,
sinusoidal shape (cf. analytical expression in (12)) at = 10
at = 1.

"

= T , and Q = I . (a) Normalized optimal forward gain f~(")=kf~(")k changes from an almost
to an almost piecewise linear shape at " = 1. (b) Optimal forward () and backward (+) gains

equations of
, respectively. The higher order terms , ,
can be determined in a similar fashion. The matrix
and
found by this procedure is the unique optimal solution of the
. This is because the equations
control problem (SH2) for
(PA), under the assumption of convergence for small , give a
.
unique matrix
We next provide analytical expressions for
obtained by solving the
equations in (PA) with
,
, and
. When a fictitious follower is present,
we have (derivations are omitted for brevity)

(12)
where
and

and
denote the th diagonal entries of
. From (12) it follows that
for
. When a fictitious follower is not present, we have

and choosing the step-size can

The details of computing
be found in [29].
obFor small , we initialize Newtons method using
tained from the perturbation expansion up to the first order in
,
. We then increase slightly and use the
optimal structured gain resulting from Newtons method at the
previous to initialize the next round of iterations. We continue
is reached,
increasing gradually until desired value
that is, until the optimal structured gain for the desired
is obtained.
Since the homotopy-based Newtons method solves a family
of optimization problems parameterized by , the optimal feedback gain is a function of
. To see the incremental
change relative to the spatially uniform gain , we consider
and the
the difference between the optimal forward gain
,
uniform gain

C. Newtons Method for Larger Values of

Fig. 4(a) shows the normalized profile

for a for,
,
, and
mation with fictitious follower,
. The values of are determined by 20 logarithmically
and 1. As increases, the normalspaced points between
ized profile changes from an almost sinusoidal shape (cf. anato an almost piecewise
lytical expression in (12)) at
linear shape at
. Note that the homotopy-based Newtons
when
method converges to the same feedback gains at
it is initialized by the optimal symmetric controller obtained in
Section III.
Since the underlying path-graph exhibits symmetry between
and
, the optimal forthe edge pairs associated with
ward and backward gains satisfy a central symmetry property,

In this section, we employ Newtons method developed

in [29] to solve the necessary conditions for optimality
is gradually increased to 1. Newtons
(NC1)(NC3) as
method is an iterative descent algorithm for finding local
minima in optimization problems [32]. Specifically, given an
, a decreasing sequence
initial stabilizing structured gain
is generated by updating
of the objective function
according to
. Here,
is the Newton direction that satisfies the structural constraint and is the step-size.

for all
; see Fig. 4(b) for
. We note that the
first vehicle has a larger forward gain than other vehicles; this
is because it neighbors the fictitious leader. The forward gains
decrease as one moves away from the fictitious leader; this is
because information about the absolute desired trajectory of the
fictitious leader becomes less accurate as it propagates down the
formation. Similar interpretation can be given to the optimal

To compute the optimal structured feedback gain for larger

obtained from perturbation analysis to
values of , we use
initialize Newtons method, as described in Section IV-C.

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

N = 50, r = 1, Q

Fig. 5. Formation without fictitious follower,

forward () and backward (+) gains at = 1.

"

, and

backward gains, which monotonically increase as one moves

towards the fictitious follower.
Since the 1st vehicle has a negative backward gain (see Fig.
4(b)), if the distance between the 1st and the 2nd vehicles is
greater than the desired value , then the 1st vehicle distances
itself even further from the 2nd vehicle. On the other hand, if
the distance is less than , then the 1st vehicle pulls itself even
closer to the 2nd vehicle. This negative backward gain of the 1st
vehicle can be interpreted as follows: Since the 1st vehicle has
access to its global position, it aims to correct the absolute positions of other vehicles in order to enhance formation coherence.
If the 2nd vehicle is too close to the 1st vehicle, then the 1st vehicle moves towards the 2nd vehicle to push it back; this in turn
pushes other vehicles back. If the 2nd vehicles is too far from the
1st vehicle, then the 1st vehicle moves away from the 2nd vehicle to pull it forward; this in turn pulls other vehicles forward.
Similar interpretation can be given to the negative forward gain
of the th vehicle that neighbors the fictitious follower. Also
note that the forward gain of the th vehicle becomes positive
when the fictitious follower is removed from the formation; see
Fig. 5(c). This perhaps suggests that negative feedback gains of
the 1st and the th vehicles are a consequence of the fact that
both of them have access to their own global positions.
As shown in Fig. 5(a) and (b), the normalized optimal gains
for the formation without the fictitious follower also change continuously as increases to 1. In this case, however, the optimal
forward and backward gains do not satisfy the central symmetry;
see Fig. 5(c). Since the optimal controller puts more emphasis
on the vehicles ahead when the fictitious follower is not present,

2211

= I . Normalized optimal (a) forward and (b) backward gains. (c) Optimal

the forward gains have larger magnitudes than the backward

gains. As in the formations with the fictitious follower, the optimal forward gains decrease monotonically as one moves away
from the fictitious leader. On the other hand, the optimal backward gains at first increase as one moves away from the 1st vehicle and then decrease as one approaches the th vehicle in
.
order to satisfy the constraint
V. PERFORMANCE VERSUS SIZE FOR THE
SINGLE-INTEGRATOR MODEL
In this section, we study the performance of the optimal symmetric and non-symmetric gains obtained in Sections III and
IV-C. This is done by examining the dependence on the formaintroduced
tion size of performance measures , , and
in Section II-C. Our results highlight the role of non-symmetry
and spatial variations on the scaling trends in large-scale formations. They also illustrate performance improvement achieved
by the optimal controllers relative to spatially uniform symmetric and non-symmetric feedback gains.
For the spatially uniform symmetric gain with
, we show analytically that
is an affine function of .
This implies that the formation coherence scales linearly with
irrespective of the value of . We also analytically establish that the spatially uniform non-symmetric gain with
(look-ahead strategy) provides a square-root
asymptotic dependence of
on . Thus, symmetry breaking
between the forward and backward gains may improve coherence of large-scale formations. Note that the forward-backward
asymmetry also provides more favorable scaling trends of the

2212

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

TABLE II
ASYMPTOTIC DEPENDENCE OF 5 , 5 , AND 5 ON THE FORMATION SIZE
FOR UNIFORM SYMMETRIC, UNIFORM NON-SYMMETRIC (LOOK-AHEAD
STRATEGY), AND OPTIMAL SYMMETRIC AND NON-SYMMETRIC GAINS OF
SECTIONS III AND IV-C WITH = AND = 1. THE SCALINGS DISPLAYED
IN RED ARE DETERMINED Analytically; OTHER SCALINGS ARE ESTIMATED
BASED ON NUMERICAL COMPUTATIONS

Q I

least damped mode of the closed-loop system [10]. We then

investigate how spatially varying optimal feedback gains, introduced in Sections III and IV-C, influence coherence of the
formation. We show that the optimal symmetric gain provides
on
and that the optimal
a square-root dependence of
non-symmetric gain provides a fourth-root dependence of
on .
Even though we are primarily interested in asymptotic
, we also exscaling of the global performance measure
amine the local performance measure and the control energy
. From Section II-C we recall that the global and local
performance measures quantify the formation-size-normalized
norm of the transfer function from to of the closed-loop
system,

symmetric and non-symmetric gains, resulting from an -independent control penalty in (SH2), do not provide uniform
. This implies
bounds on the control energy per vehicle,
the trade-off between the formation coherence
and control
in the design of the optimal controllers. It is thus
energy
of interest to examine formation coherence for optimal controllers with bounded control energy per vehicle (see Remark
2). Second, the controller structure (e.g., symmetric or non-symmetric gains) plays an important role in the formation coherence. In particular, departure from symmetry in localized feedback gains can significantly improve coherence of large-scale
formations (see Remark 3).
A. Spatially Uniform Symmetric Gain
For the spatially uniform symmetric controller with
, we next show that
is an affine function of
and that, in the limit of an infinite number of vehicles, both
and
become formation-size-independent. These results
hold irrespective of the presence of the fictitious follower.
For the single-integrator model with the fictitious follower we
(see (5) for the definition of ), and
have
solves the Lyapunov equation (14) [31, Lemma
is determined by (cf.
1]. Since the th diagonal entry of
(8))

from (13) we conclude that the global performance measure

is an affine function of , and that both
and
are formation-size-independent,

and that
is the formation-size-normalized
norm of the
transfer function from to . These can be determined from

(13)
where

For the formation without the fictitious follower, the following

expressions

denotes the closed-loop controllability Gramian,

(14)

, , and
, for
The asymptotic scaling properties of
the above mentioned spatially uniform controllers and the spatially varying optimal controllers, obtained by solving (SH2)
and
, are summarized in Table II. For both spawith
tially uniform symmetric and look-ahead strategies, we analytically determine the dependence of these performance measures
on the formation size in Sections V-A and V-B. Furthermore,
for the formation without the fictitious follower subject to the
optimal symmetric gains, we provide analytical results in Section V-C. For the optimal symmetric and non-symmetric gains
in the presence of fictitious followers, the scaling trends are obtained with the aid of numerical computations in Section V-C.
Several comments about the results in Table II are given next.
First, in contrast to the spatially uniform controllers, the optimal

imply that, for the spatially uniform symmetric controller, the

asymptotic scaling trends do not depend on the presence of the
fictitious follower (derivations omitted for brevity).
B. Spatially Uniform Non-Symmetric Gain (Look-Ahead
Strategy)
We next examine the asymptotic scaling of the performance
measures for the spatially uniform non-symmetric gain with
. We establish the square-root scaling of
with
and the formation-size-independent scaling of .
Furthermore, in the limit of an infinite number of vehicles, we
becomes -independent.
show that

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

2213

Fig. 6. (a) Square-root scaling of 5 (3) using optimal

psymmetric gain of Section III, 0:2784 N + 0:0375 (curve); and (b) Fourth-root scaling of 5 () using
optimal non-symmetric gain of Section IV-C, 0:4459 N 0 0:0866 (curve). The optimal controllers are obtained by solving (SH2) with Q = I and r = 1 for
the formation with the fictitious follower.

For the single-integrator model with

(see
), the solution of the Lyapunov (14)
(3) for the definition of
is given by

(15)

gains, this constraint on control energy (variance) increases the

on
to linear and square-root,
asymptotic dependence of
respectively.
For the formation without the fictitious follower, the optimal symmetric gains are given by (9). As shown in (10),
, thereby yielding

As shown in Appendix B, the inverse Laplace transform of

can be used to determine the analytical expression for
, yielding the following formulae:

(16)
In the limit of an infinite number of vehicles,

with
denoting the Gamma function. These are used in Appendix B to show that, in the limit of an infinite number of vehicles, a look-ahead strategy for the single-integrator model proon
and the formavides the square-root dependence of
tion-size-independent
and
.
C. Optimal Symmetric and Non-Symmetric Controllers
We next examine the asymptotic scaling of the performance
measures for the optimal symmetric and non-symmetric gains
of Sections III and IV-C. For the formation without the fictitious follower, we analytically establish that the optimal sym,
, and
metric gains asymptotically provide
scalings of
, , and
, respectively. We then
use numerical computations to (i) confirm these scaling trends
for the optimal symmetric gains in the presence of the fictitious
and
follower; and to (ii) show a fourth-root dependence of
on and an
dependence of
for the optimal
non-symmetric gains. All these scalings are obtained by solving
(SH2) with the formation-size-independent control penalty
and
. We also demonstrate that uniform control variance (per vehicle) can be obtained by judicious selection of an
-dependent . For the optimal symmetric and non-symmetric

which, for an -independent , leads to an asymptotic squareand

on ,
root dependence of

(17)
asymptotic
Similar calculation can be used to obtain
scaling of .
We next use numerical computations to study the scaling
trends for the optimal symmetric and non-symmetric gains in
the presence of fictitious followers. The optimal symmetric
in Fig. 3) provides a square-root scaling of
gain (cf.
with ; see Fig. 6(a). On the other hand, the optimal non-symmetric gain (cf. Fig. 4(b)) leads to a fourth-root scaling of
with ; see Fig. 6(b). The local performance measure
decreases monotonically with
for both controllers, with
scaling as
for the optimal symmetric gain and as
for the optimal non-symmetric gain; see Fig. 7. For
both the optimal symmetric and non-symmetric controllers, our
computations indicate equivalence between the control energy

2214

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

Fig. 7. (a) 5 (3) using

pthe optimal symmetric gain of Section III, 1:8570= N + 0:0042 (curve); and (b) 5 () using the optimal non-symmetric gain of
Section IV-C, 1:4738= N + 0:0191 (curve). The optimal controllers are obtained by solving (SH2) with Q = I and r = 1 for the formation with the fictitious
follower.

and the global performance measure when

. (For the op,
timal symmetric gain without the fictitious follower and
; see formula (16).)
we have analytically shown that
Therefore, the asymptotic scaling of the formation-size-normalfor the optimal symmetric gain
ized control energy is
for the optimal non-symmetric gain. Finally, for
and
the formations without the fictitious follower, our computations
indicate that the optimal non-symmetric gains also asymptoti,
, and
scalings of ,
cally provide
, and
, respectively.
Remark 2: In contrast to the spatially uniform controllers,
the optimal structured controllers of Sections III and IV-C, resulting from an -independent control penalty in (SH2), do
not provide uniform bounds on the formation-size-normalized
framecontrol energy. These controllers are obtained using
work in which control effort represents a soft constraint. It
is thus of interest to examine formation coherence for optimal
controllers with bounded control energy per vehicle. For formations without the fictitious follower, from (17) we see that the
asymptotioptimal symmetric controller with
cally yields
and
. Similarly,
for formations with followers, the optimal gains that result in
for large
can be obtained by changing control
to
for the optimal sympenalty from
metric gain and to
for the optimal non-symmetric gain.1 These -dependent control penalties provide an
with for the optimal symmetric gain and
affine scaling of
with
for the optimal non-syma square-root scaling of
metric gain; see Fig. 8. The asymptotic scalings for formations
without followers subject to the optimal symmetric gains are obtained analytically (cf. (17)); all other scalings are obtained with
the aid of computations.
Remark 3: Fig. 8 illustrates the global performance measure
obtained with four aforementioned structured controllers
for formations with fictithat asymptotically yield
tious follower. Note that the simple look-ahead strategy outpervs.
scaling.
forms the optimal symmetric gain;
Thus, departure from symmetry in localized feedback gains can
significantly improve coherence of large-scale formations. In
1Both spatially uniform symmetric and look-ahead strategies with
yield 5 = 1 in the limit of an infinite number of vehicles.

=1

Fig. 8. 5 using four structured gains with 5  1 for formations with fictitious follower: spatially uniform symmetric
(5p), N=12+1=6 (blue curve), spap
tially uniform non-symmetric (/), 2 N =(3  ) (green curve), optimal symmetric (p
3), 0:0793N + 0:0493 (black curve), and optimal non-symmetric (),
0:1807 N 0 0:0556 (red curve).

particular, we have provided an example of a spatially uniform

non-symmetric controller that yields better scaling trends than
the optimal spatially varying controller obtained by restricting
design to symmetric gains. Given the extra degrees of freedom
in the optimal symmetric gain this is perhaps a surprising observation, indicating that the network topology may play a more
important role than the optimal selection of the feedback gains
in performance of large-scale interconnected systems. On the
other hand, our results show that the optimal localized controller
that achieves the best performance is both non-symmetric and
spatially-varying.
VI. DOUBLE-INTEGRATOR MODEL
In this section, we solve (SH2) for the double-integrator
model using the homotopy-based Newtons method. We then
discuss the influence of the optimal structured gain on the
asymptotic scaling of the performance measures introduced
in Section II-C. For a formation in which each vehiclein
addition to relative positions with respect to its immediate
neighborshas access to its own velocity, our results highlight
similarity between optimal forward and backward position
gains for the single- and the double-integrator models. We
further show that the performance measures exhibit similar

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

Fig. 9. Double-integrator model with fictitious follower,

velocity gains (5).

2215

N = 50, Q = I and r = 1. (a) The optimal forward () and backward gains (+); (b) the optimal

scaling properties to those found in single-integrators. We also

establish convexity of (SH2) for the double-integrator model
by restricting the controller to symmetric position and uniform
diagonal velocity gains.
The perturbation analysis and the homotopy-based Newtons
method closely follow the procedure described in Sections IV-B
and IV-C, respectively. In particular,
yields
. As shown in [35], for positive and
with
, this spatially uniform structured feedback gain
is stabilizing and inversely optimal with respect to

In what follows, we choose

and
and employ the homotopy-based Newtons method to solve (SH2) for
the double-integrator model. For a formation with fictitious fol,
, and
the optimal forward and
lower,
backward position gains are shown in Fig. 9(a) and the optimal velocity gains are shown in Fig. 9(b). We note remarkable similarity between the optimal position gains for the singleand the double-integrator models; cf. Fig. 9(a) and Fig. 4(b).
For a formation without fictitious follower, the close resemblance between the optimal position gains for both models is
also observed.
As in the single-integrator model, our computations indicate
that the optimal localized controller, obtained by solving (SH2)
and
, provides a fourth-root dependence
with
of the macroscopic performance measure
on ; see Fig.
10(a). Furthermore, the microscopic performance measure and
and
,
control energy asymptotically scale as
respectively; see Fig. 10(b) and (c).
For comparison, we next provide the scaling trends of the
performance measures for both the spatially uniform symmetric
and look-ahead controllers. As in the single-integrator model,
provides
the spatially uniform symmetric gain
linear scaling of
with and the formation-size-independent
and
,

On the other hand, for the double-integrator model the perforheavily

mance of the look-ahead strategy
depends on the choices of and . In particular, for
and
, using similar techniques as in Section V-B, we
obtain

which asymptotically leads to the formation-size-independent

and the square-root scaling of
with , i.e.,
scaling of
. This is in sharp contrast to
which leads to an exponential dependence of
on . Therefore, the design of the look-ahead strategy is much
more subtle for double-integrators than for single-integrators.
Remark 4: For the double-integrator model with
and fixed
we next show convexity of (SH2)
. The Lyapunov equation in
with respect to
(SH2), for the block diagonal state weight with components
and , can be rewritten in terms of the components of
,
(18a)
(18b)
(18c)
Linearity of the trace operator in conjunction with
and (18c) yields

where the last equation is obtained by multiplying (18a) from

and using
.
the left with
, similar argument as in Section III can be used to
For
conclude convexity of with respect to
.
VII. CONCLUSION
We consider the optimal control of one-dimensional formations with nearest neighbor interactions between the vehicles.
We formulate a structured optimal control problem in which

2216

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

pN

p
= N

Q = I r = 1 for formations with the fictitious

 : pN + 0:8830 (curve).

Fig. 10. Double-integrator model with the optimal non-symmetric gain obtained by solving (SH2) with
and
follower: (a) 5 (), 0 0736
+ 0 4900 (curve) (b) 5 ( ), 1 1793
+ 0 0408 (curve); (c) 5 ( ), 0 2742

 :

local information exchange of relative positions between immediate neighbors imposes structural constraints on the feedback
gains. We study the design problem for both the single- and
the double-integrator models and employ a homotopy-based
Newtons method to compute the optimal structured gains. We
also show that design of symmetric gains for the single-integrator model is a convex optimization problem, which we
solve analytically for formations with no fictitious followers.
For double-integrators, we identify a class of convex problems
by restricting the controller to symmetric position and uniform
diagonal velocity gains. Furthermore, we investigate the performance of the optimal controllers by examining the asymptotic
scalings of formation coherence and control energy with the
number of vehicles.
For formations in which all vehicles have access to their own
velocities, the optimal structured position gains for single- and
double-integrators are similar to each other. Since these two
models exhibit the same asymptotic scalings of global, local,
and control performance measures, we conclude that the singleintegrator model, which lends itself more easily to analysis and
design, captures the essential features of the optimal localized
design. We note that the tools developed in this paper can also
be used to design optimal structured controllers for double-integrators with relative position and velocity measurements; this
is a topic of our ongoing research.
As in [10], we employ perturbation analysis to determine the
departure from a stabilizing spatially uniform profile that yields
nominal diffusion dynamics on a one-dimensional lattice; in
contrast to [10], we find the mistuning profile by optimizing

a performance index rather than by performing spectral analysis. We also show how a homotopy-based Newtons method
can be employed to obtain non-infinitesimal variation in feedback gains that minimizes the desired objective function. Furthermore, we establish several explicit scaling relationships and
identify a spatially uniform non-symmetric controller that performs better than the optimal symmetric spatially varying convs.
scaling of coherence with
controller (
trol energy per vehicle). This suggests that departure from symmetry can improve coherence of large-scale formations and that
the controller structure may play a more important role than the
optimal feedback gain design. On the other hand, our results
demonstrate that the best performance is achieved with the optimal localized controller that is both non-symmetric and spatially-varying.
Currently, we are considering the structured feedback design
for formations on general graphs [6], [8], [23], [36], [37] with
the objective of identifying topologies that lead to favorable
system-theoretic properties [28], [38], [39]. Even though this
paper focuses on the optimal local feedback design for one-dimensional formations with path-graph topology, the developed
methods can be applied to multi-agent problems with more general network topologies.
APPENDIX
A. Gradient Method for (SG)
We next describe the gradient method for solving (SG). Let
us denote
. Starting with an initial guess

LIN et al.: OPTIMAL CONTROL OF VEHICULAR FORMATIONS WITH NEAREST NEIGHBOR INTERACTIONS

that guarantees positive definiteness of

, vector is updated
, until the norm of gradient is small
. Here,
is the step-size determined
enough,
by the backtracking line search [32, Section 9.2]: let
and
with
until a sufficient decrease in
repeat
the objective function is achieved,

2217

Now, from (21) and the fact that

we obtain

Similarly,

where
. Note that
is defined as infinity if
in (7) determined by is not positive definite. For
,

Using Stirlings approximation

we have
the entries of the gradient

for large ,

are given by
where we used the fact that
sequently,

. Con. From (19) and (20), it follows

that

and thus,

. We conclude that
asymptotically scales as a
is formation-size-indesquare-root function of and that
pendent as increases to infinity.
ACKNOWLEDGMENT
The authors would like to thank anonymous reviewers and the
associate editor for their valuable comments.

B. Performance of Look-Ahead Strategy

We next derive the analytical expressions for the perforobtained with the look-ahead
mance measures , , and
strategy for the single-integrator model. The solution of the
is determined
Lyapunov equation (14) with
by (15). Since the th entry of the first column of the lower
is
,
triangular Toeplitz matrix
the corresponding entry of the matrix exponential in (15) is
,
determined by the inverse Laplace transform of

Thus, the th element on the main diagonal of the matrix

(15) is given by

in

(19)
thereby yielding

A similar procedure can be used to show that the

entry of is determined

(20)
th
(21)

REFERENCES
[1] W. S. Levine and M. Athans, On the optimal error regulation of a
string of moving vehicles, IEEE Trans. Autom. Control, vol. AC-11,
no. 3, pp. 355361, Jul. 1966.
[2] S. M. Melzer and B. C. Kuo, Optimal regulation of systems described
by a countably infinite number of objects, Automatica, vol. 7, no. 3,
pp. 359366, 1971.
[3] D. Swaroop and J. Hedrick, String stability of interconnected systems, IEEE Trans. Autom. Control, vol. 41, no. 3, pp. 349356, Mar.
1996.
[4] A. Jadbabaie, J. Lin, and A. S. Morse, Coordination of groups of
mobile autonomous agents using nearest neighbor rules, IEEE Trans.
Autom. Control, vol. 48, no. 6, pp. 9881001, Jun. 2003.
[5] P. Seiler, A. Pant, and K. Hedrick, Disturbance propagation in vehicle
strings, IEEE Trans. Autom. Control, vol. 49, no. 10, pp. 18351841,
Oct. 2004.
[6] J. A. Fax and R. M. Murray, Information flow and cooperative control
of vehicle formations, IEEE Trans. Autom. Control, vol. 49, no. 9, pp.
14651476, Sep. 2004.
[7] M. R. Jovanovic and B. Bamieh, On the ill-posedness of certain vehicular platoon control problems, IEEE Trans. Autom. Control, vol.
50, no. 6, pp. 13071321, Jun. 2005.
[8] G. Lafferriere, A. Williams, J. Caughman, and J. J. P. Veerman, Decentralized control of vehicle formations, Syst. Control Lett., vol. 54,
pp. 899910, 2005.
[9] M. R. Jovanovic, J. M. Fowler, B. Bamieh, and R. DAndrea, On the
peaking phenomenon in the control of vehicular platoons, Syst. Control Lett, vol. 57, no. 7, pp. 528537, 2008.
[10] P. Barooah, P. G. Mehta, and J. P. Hespanha, Mistuning-based control
design to improve closed-loop stability margin of vehicular platoons,
IEEE Trans. Autom. Control, vol. 54, no. 9, pp. 21002113, Sep. 2009.
[11] R. H. Middleton and J. H. Braslavsky, String instability in classes
of linear time invariant formation control with limited communication
range, IEEE Trans. Autom. Control, vol. 55, no. 7, pp. 15191530, Jul.
2010.

2218

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 9, SEPTEMBER 2012

[12] B. Bamieh, M. R. Jovanovic, P. Mitra, and S. Patterson, Coherence in

large-scale networks: Dimension dependent limitations of local feedback, IEEE Trans. Autom. Control, 2012, to be published.
[13] P. Frihauf and M. Krstic, Multi-agent deployment to a family of planar
arcs, in Proc. Amer. Control Conf., 2010, pp. 41094114.
[14] P. Frihauf and M. Krstic, Leader-enabled deployment into planar
curves: A PDE-based approach, IEEE Trans. Autom. Control, vol. 56,
no. 8, pp. 17911806, Aug. 2011.
[15] A. Sarlette and R. Sepulchre, A PDE viewpoint on basic properties of
coordination algorithms with symmetries, in Proc. 48th IEEE Conf.
Decision Control, 2009, pp. 51395144.
[16] B. Bamieh, F. Paganini, and M. A. Dahleh, Distributed control of spatially invariant systems, IEEE Trans. Autom. Control, vol. 47, no. 7,
pp. 10911107, Jul. 2002.
[17] N. Motee and A. Jadbabaie, Optimal control of spatially distributed
systems, IEEE Trans. Autom. Control, vol. 53, no. 7, pp. 16161629,
Aug. 2008.
controllers
[18] P. G. Voulgaris, G. Bianchini, and B. Bamieh, Optimal
for spatially invariant systems with delayed communication requirements, Syst. Control Lett., vol. 50, pp. 347361, 2003.
[19] B. Bamieh and P. G. Voulgaris, A convex characterization of distributed control problems in spatially invariant systems with communication constraints, Syst. Control Lett., vol. 54, pp. 575583, 2005.
[20] M. Rotkowitz and S. Lall, A characterization of convex problems in
decentralized control, IEEE Trans. Autom. Control, vol. 51, no. 2, pp.
274286, Feb. 2006.
[21] M. Fardad and M. R. Jovanovic, Design of optimal controllers for spatially invariant systems with finite communication speed, Automatica,
vol. 47, no. 5, pp. 880889, 2011.
[22] P. Twu and M. Egerstedt, Optimal decentralization of multi-agent motions, in Proc. Amer. Control Conf., 2010, pp. 23262331.
[23] L. Xiao, S. Boyd, and S.-J. Kim, Distributed average consensus with
least-mean-square deviation, J. Parallel Distrib. Comput., vol. 67, no.
1, pp. 3346, 2007.
[24] P. Barooah and P. G. Mehta, Optimal mistuning for improved stability of vehicular platoons, in Proc. 48th IEEE Conf. Decision Control, 2009, pp. 30203025.
[25] R. Olfati-Saber and R. M. Murray, Consensus problems in networks of
agents with switching topology and time-delays, IEEE Trans. Autom.
Control, vol. 49, no. 9, pp. 15201533, Sep. 2004.
[26] J. A. Marshall, M. E. Broucke, and B. A. Francis, Formations of vehicles in cyclic pursuit, IEEE Trans. Autom. Control, vol. 49, no. 11,
pp. 19631974, Nov. 2004.
[27] P. Barooah and J. P. Hespanha, Graph effective resistance and distributed control: Spectral properties and applications, in Proc. 45th
IEEE Conf. Decision Control, 2006, pp. 34793485.
[28] D. Zelazo and M. Mesbahi, Edge agreement: Graph-theoretic performance bounds and passivity analysis, IEEE Trans. Autom. Control,
vol. 56, no. 3, pp. 544555, Mar. 2011.
[29] M. Fardad, F. Lin, and M. R. Jovanovic, On the optimal design of
structured feedback gains for interconnected systems, in Proc. 48th
IEEE Conf. Decision Control, 2009, pp. 978983.
[30] F. Lin, M. Fardad, and M. R. Jovanovic, Augmented Lagrangian approach to design of structured optimal state feedback gains, IEEE
Trans. Autom. Control, vol. 56, no. 12, pp. 29232929, Dec. 2011.
[31] B. Bamieh and M. Dahleh, Exact computation of traces and
norms
for a class of infinite dimensional problems, IEEE Trans. Autom. Control, vol. 48, no. 4, pp. 646649, Apr. 2003.
[32] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge,
U.K.: Cambridge Univ. Press, 2004.
[33] G. Meurant, A review of the inverse of symmetric tridiagonal and
block tridiagonal matrices, SIAM J. Matrix Anal. Appl., vol. 13, no.
3, pp. 707728, 1992.
[34] R. E. Kalman, When is a linear control system optimal?, J. Basic
Eng., vol. 86, pp. 5160, 1964.
[35] M. R. Jovanovic, On the optimality of localized distributed controllers, Int. J. Syst., Control Commun., vol. 2, no. 1/2/3, pp. 8299,
2010.

[36] F. Borrelli and T. Keviczky, Distributed LQR design for identical dynamically decoupled systems, IEEE Trans. Autom. Control, vol. 53,
no. 8, pp. 19011912, Sep. 2008.
[37] S. E. Tuna, Conditions for synchronizability in arrays of coupled
linear systems, IEEE Trans. Autom. Control, vol. 54, no. 10, pp.
24162420, Oct. 2009.
[38] G. F. Young, L. Scardovi, and N. E. Leonard, Robustness of noisy consensus dynamics with directed communication, in Proc. Amer. Control Conf., 2010, pp. 63126317.
[39] F. Lin, M. Fardad, and M. R. Jovanovic, On the optimal localized
feedback design for multi-vehicle systems, in Proc. 49th IEEE Conf.
Decision Control, 2010, pp. 57445749.
Fu Lin (S06) received the B.S. degree in instrument
science and engineering from Shanghai Jiaotong
University, Shanghai, China, in 2005 and is currently
pursuing the Ph.D. degree in the Department of
Electrical and Computer Engineering, University of
Minnesota, Minneapolis.
His primary research interests are in the analysis and design of optimal distributed controllers
using tools from convex optimization, compressive
sensing, and graph theory.

Makan Fardad (M08) received the B.S. degree in

electrical engineering from the Sharif University of
Technology, Tehran, Iran, the M.S. degree from the
Iran University of Science and Technology, Tehran,
and the Ph.D. degree in mechanical engineering from
the University of California, Santa Barbara, in 2006.
He was a Postdoctoral Associate at the University
of Minnesota, Minneapolis, before joining the
Department of Electrical Engineering and Computer
Science, Syracuse University, Syracuse, NY, as an
Assistant Professor in August 2008. His research
interests are in modeling, analysis, and optimal control of distributed and
large-scale interconnected systems.

Mihailo R. Jovanovic (S00M05) received the

Dipl. Ing. and M.S. degrees from the University of
Belgrade, Serbia, in 1995 and 1998, respectively, and
the Ph.D. degree from the University of California,
Santa Barbara, in 2004.
Before joining the University of Minnesota, Minneapolis, he was a Visiting Researcher with the Department of Mechanics, the Royal Institute of Technology, Stockholm, Sweden, from September to December 2004. Currently, he is an Associate Professor
of Electrical and Computer Engineering, University
of Minnesota, where he serves as the Director of Graduate Studies in the interdisciplinary Ph.D. program in Control Science and Dynamical Systems. His
expertise is in modeling, dynamics, and control of large-scale and distributed
systems and his current research focuses on sparsity-promoting optimal control,
dynamics and control of fluid flows, and fundamental limitations in the control
of vehicular formations. He has held visiting positions with Stanford University
and the Institute for Mathematics and Its Applications.
Dr. Jovanovic received the CAREER Award from the National Science Foundation in 2007, and the Early Career Award from the University of Minnesota
Initiative for Renewable Energy and the Environment in 2010. He is a member
of APS and SIAM and has served as an Associate Editor of the IEEE Control
Systems Society Conference Editorial Board from July 2006 until December
2010.