116

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 50, NO. 1, JANUARY 2001

A Lyapunov Function Approach to Longitudinal

Control of Vehicles in a Platoon
Tae Soo No, Kil-To Chong, and Do-Hwan Roh

AbstractThe Lyapunov second method is used to derive a control law that can be used to control the spacing between vehicles
in a platoon. A third-order system is adopted to model the vehicle
and powertrain dynamics. In addition, the concept of expected
spacing error is introduced and used to form a Lyapunov function. Then a control law that always decreases the Lyapunov function is selected. A platoon of four vehicles and various scenarios are
used to demonstrate the performance of the proposed control law.

controllers [5][9]. Sheikholeslam and Desoer [12] showed that

tight spacing control is still possible with no communication of
information about the lead vehicle.
In this paper, we propose another form of a longitudinal control law
(2)

Index TermsIntelligent transportation systems, longitudinal

control, platoon, vehicle dynamics and control.

I. INTRODUCTION

T HAS been shown that the concept of operating vehicles

in a platoon can significantly improve the efficiency of
existing road systems. In order to meet such expectations,
it is necessary to develop advanced vehicle control systems
(AVCSs), which form a part of intelligent vehicle/highway
systems (IVHSs) [1][4]. As shown in Fig. 1, one of the
many functions of AVCSs is longitudinal control, in which the
between the predecessor
and the
preassigned spacing
in the platoon is to be tightly maintained.
follower vehicle
has
Most earlier works assume that the follower vehicle
of the platoon lead
access to the velocity and acceleration
and can measure the relative distance, velocity, and
vehicle
acceleration between itself and the vehicle in front of it [5][8].
There are many strategies for designing longitudinal control
laws. To cite a few: proportional-integral-derivative (PID)-type
controllers have been proposed by Shladover [5], [6], Sheikholeslam [7], [8], and Fujioka et al. [9]. Hedrick et al. [10], [11]
have extensively used the sliding mode control methodology.
Readers are referred to [1][4] for the pertinent literature.
In the works cited above, the general structure of the control
may be written as
input to vehicle
(1)
denote the absolute position, velocity, and acwhere
, respectively. It has been reported that
celeration of vehicle
1 m or less spacing can be maintained in a platoon by using such

that completely eliminates the need for a communication link

between the platoons leading vehicle and the follower vehicles
and uses the local link between the predecessor and follower
vehicles, as depicted in Fig. 2. We will show that the procedure
used in this paper for designing control laws somewhat mitigates the ad hoc nature of selecting the variables for feedback
or the structure of the sliding surface. Also, it effectively accommodates wide variations in vehicle dynamics and is suitable for
the spacing control of mixed vehicles.
II. VEHICLE DYNAMICS MODELING
Various models for vehicle dynamics have been used in the
study of longitudinal control of platoons. For a vehicle nominally traveling with a constant speed and direction, we have
adopted the following third-order linear model [5], [6]:
(3)
(4)
(5)
where , the time constant, is a single parameter that represents
the characteristics of vehicle propulsion system, including the
engine, transmission, tires and wheels, and any other internal
controllers. Fig. 3 shows the block diagram of vehicle dynamics.
Jerk and acceleration limits may be included if necessary. Referring to Fig. 1, the error in spacing distance between vehicles
and
can be written as
(6)

Manuscript received October 13, 1999; revised July 26, 2000. This work was
supported by the Korea Science and Engineering Foundation and by the Mechatronics Research Center at Chonbuk National University, Korea.
T. S. No is with the Department of Aerospace Engineering, Chonbuk National
University, Chonju, Korea.
K.-T. Chong is with the Department of Measurement and Control Engineering, Chonbuk National University, Chonju, Korea.
D.-H. Roh is with the Department of Electrical Engineering, Chonbuk National University, Chonju, Korea.
Publisher Item Identifier S 0018-9545(01)01133-1.

where and
respectively.

represent the relative velocity and acceleration,

III. LONGITUDINAL CONTROL LAW DESIGN

The objective of longitudinal control is to maintain the
spacing error below a predetermined level or, if possible, at

00189545/01$10.00 2001 IEEE

NO et al.: LYAPUNOV FUNCTION APPROACH TO LONGITUDINAL CONTROL OF VEHICLES

Fig. 1.

117

Platoon with global link.

Fig. 2. Platoon with local link.

Fig. 3. Block diagram of vehicle dynamics.

zero. For example, the PID-type controller may be written in

its general form as
(7)
where , etc., are controller gains. With this control law, the
control command is determined using the information measured
and/or estimated at the current time. In other words, the controller is always working, as long as there is a spacing error at
the current time. The only time the controller stops working is
when the following condition is satisfied:
(8)

However, if some predictive nature is incorporated into the controller, one can make it behave more intelligently. Even if current is large, if it is expected to be decreasing, no further control action would be necessary. By contrast, some action is required when is expected to increase even though its current
magnitude is below the predetermined level.
With the above reasoning in mind, we introduce the concept
of expected spacing error as follows:
(9)
where

is the time-to-go until the future time

and may be

118

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 50, NO. 1, JANUARY 2001

Fig. 4. Platoon simulation environment.

TABLE I
VEHICLE AND CONTROLLER CHARACTERISTICS

written as
(10)
is zero. As one may
For the sake of brevity, we assume that
easily understand, is the expected spacing error at the future
time if both the predecessor and follower vehicles keep their
respective current accelerations constant for .
The Lyapunov direct method often provides a straightforward
way of obtaining a control law for many dynamic systems. For
this purpose, we need to find an appropriate positive definite
function, which is referred to as a Lyapunov function. Also, a
control law is designed so that it assures the negative definiteness of the derivative of the Lyapunov function [13].
In this paper, we use the expected spacing error (9) to define
a Lyapunov function as follows:

In order to use the Lyapunov direct method, we may select

such that
(15)
where

is a positive constant. Then, (12) may be rewritten as

(16)

which is a negative definite function. This implies that the Lyapunov function is a strictly decreasing function, meaning that
Lyapunov stability is assured. Finally, the controller that satisfies the above argument is
(17)

(11)
If we differentiate the above Lyapunov function, then we simply
obtain
(12)
Also, differentiating (9) yields

(13)
Then, we substitute (5) into (13) to get
(14)

and
are left undetermined as design parameters.
where
While the proposed control law still needs to measure or estimate the spacing error and relative speed/acceleration in computing the expected spacing error, the vehicle does not receive
any information directly from the platoon lead vehicles other
than the first vehicle. But the unidirectional communication link
is required because each vehicle needs to receive a control command from the preceding vehicle. Another feature of the proposed control law is that it systematically accommodates the
dynamic characteristics of the vehicles participating in the platoon, since the vehicle time constant appears in the control
command. This feature will be beneficial if a platoon is composed of vehicles of various kinds. Therefore, due to the cascaded structure of the control laws, a vehicle will have full access to information on the dynamics and control characteristics
of all the preceding vehicles.

NO et al.: LYAPUNOV FUNCTION APPROACH TO LONGITUDINAL CONTROL OF VEHICLES

119

(a)

(b)
Fig. 5. Response to sudden speed change. (a) PID controller. (b) Lyapunov controller.

IV. SIMULATION AND PERFORMANCE ANALYSIS

As shown in Fig. 4, a four-vehicle platoon is chosen and used
in the simulation because it must retain the essential characteristics of a platoon. Individual vehicles are assumed to be identica,l
and
are selected after some nuand the design parameters
merical experiments. Just for comparison purposes, a PID-type
controller given by (7) and a simulation scenario were adopted
from [6]. The relevant numbers used in the simulation are sumis set at zero for ease
marized in Table I. The spacing distance
of presenting the simulation results. Then, the control objective
is that every vehicle line up with the platoon leader vehicle.

A. Response to Abrupt Speed Changes by the Lead Vehicle

In this example, the platoon leader vehicle accelerates at 1
m/s for 2 s to speed up by 2 m/s, and there are no initial
spacing errors between vehicles. As one may see from Fig. 5,
the PID-type controller performs just as human drivers do. In
the early stage, the spacing error between the platoon leader and
the first vehicle is the largest. That is because of the delay in
propulsion or engine dynamics. Although the first vehicle notices the acceleration of the lead vehicle earlier than the other
vehicles behind it, the difference in acceleration and speed are
still great, producing a large spacing error. But the spacing error

120

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 50, NO. 1, JANUARY 2001

(a)

(b)
Fig. 6. Response to initial spacing errors. (a) PID controller. (b) Lyapunov controller.

between the next two vehicles gets smaller because the difference in acceleration and speed have not yet built up enough to
yield noticeable errors. However, as time goes by, the first vehicle catches up with the lead vehicle faster than any of the other
vehicle, and spacing errors disappear after about 6 s.
However, the Lyapunov controller proposed in this paper
shows exceptional performance in the sense that all vehicles
move together and achieve almost identical acceleration and
speed. In this case, the follower vehicle knows what its front
vehicle will do and takes the same control action. Therefore,
the spacing errors remain virtually unchanged.

B. Response to Initial Spacing Errors

In this scenario, all the vehicles are moving with the same
speed as that of the lead vehicle, and there are initial spacing
errors, as shown in Fig. 6. To line up with the lead vehicle, it
would be best if the first vehicle decelerated, the second kept
the current speed (no control action), and the third vehicle accelerated. Unfortunately, the structure of the PID controller does
not allow that sort of intelligent maneuver. Because the first vehicle is ahead of the platoon leader, it will decelerate and the
second vehicle will accelerate because it is behind the first vehicle. This will make the third vehicle accelerate at a higher rate

NO et al.: LYAPUNOV FUNCTION APPROACH TO LONGITUDINAL CONTROL OF VEHICLES

121

(a)

(b)
Fig. 7.

Effects of uncertainty in vehicle dynamics. (a) PID controller. (b) Lyapunov controller.

than the second vehicle. The simulation results clearly support

this expectation.
However, the Lyapunov controller completely eliminates the
need for the second vehicles speed changes, and the first and
the third vehicles adjust their speed in a symmetric pattern using

much lower acceleration/deceleration rates than with the PID

controller. By avoiding any unnecessary acceleration/deceleration, the Lyapunov controller has the effect of saving energy. It
also should be noted that oscillations in response are virtually
removed.

122

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 50, NO. 1, JANUARY 2001

(a)

(b)
Fig. 8. Effects of design parameters

and

(a)

= 1 s. (b) N = 10.

C. Sensitivity to Uncertainty in Vehicle Dynamics

Most longitudinal controllers are designed using some nominal vehicle dynamicseither linear or nonlinear. McMahon
and Hedrick [14] reported that a first-order lag model approximation of vehicle dynamics, including those of the engine and
powertrain, involve a significant amount of simplification and
many assumptions. The time constant of the first-order models
may be a function of many factors, such as vehicle speed and
mass. Therefore, the controller must be designed so that it can
accommodate, or at least be robust to, variations in vehicle dynamics. Shladover [6] has suggested the use of a gain sched-

uling technique. Alternatively, velocity-dependent characteristics in the first-order model may be taken into account during
the design process [15].
The Lyapunov controller proposed in this paper will be ideally robust to any variations in the vehicle time constants if this
information is communicated to the controller. This is because,
as one can see from (5) and (17), the time constant explicitly appears in the acceleration command and is effectively cancelled
out when the command is entered into the vehicle dynamics.
An analytical study of the effects of uncertain vehicle dynamics on the overall performance will warrant further in-depth

NO et al.: LYAPUNOV FUNCTION APPROACH TO LONGITUDINAL CONTROL OF VEHICLES

Fig. 9.

123

Platoon with global and local communication link.

(a)

(b)
Fig. 10.

Response of mixed controllers. (a) Primary: PID; backup: Lyapunov. (b) Primary: Lyapunov; backup: PID.

investigations. Instead, we present some numerical simulation

results to compare the robustness of two control approaches.

Using the same scenario used in the sudden speed change simulation and fixing all relevant gains and parameters for the con-

124

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 50, NO. 1, JANUARY 2001

trollers, the time constants of vehicle models are varied in order

to represent the various kinds of vehicle dynamics. The time
constant ranges from 0.2 to 0.4. In Fig. 7, the spacing errors
are presented. As one can see, the Lyapunov controller seems
to be relatively less sensitive to the uncertainty in vehicle dynamics than the PID controller. The magnitude of spacing errors
is comparatively smaller, and the error propagation is less persistent. One should note that the spacing error propagation becomes more serious when a relatively slower vehicle is placed
in front of other vehicles.
D. Effects of Design Parameters

and

In the proposed controller,

and
are left undetermined
and regarded as design parameters that are to be selected. Although their theoretical effects on the performance of the proposed controller are not treated in this paper, some rough idea of
them may be obtained from (17). Rewriting (17) in its expanded
form using (6) and (9), we obtain

primary controller while the second car used the Lyapunov controller. In the second case, the Lyapunov controller was used as
the primary controller while the second car used the PID controller. As can be noted from Fig. 10, the overall performance
did not seem to be degraded, and the Lyapunov controller outperformed the PID in terms of spacing errors.
V. CONCLUSIONS
In this paper, the concept of expected spacing error was introduced. The Lyapunov second method was applied to the design of control logic used to control intervehicular spacing in
a platoon. The proposed controller uses a local communication
link between the predecessor and follower vehicles. Due to the
cascaded structure of the controller, each vehicle in the platoon
has full access to information on the dynamics and control of
all the vehicles ahead of it. Numerical simulation was used to
compare the performance of the Lyapunov controller with that
of a PID-type controller. The simulation results showed that the
Lyapunov controller is less sensitive to uncertainty in vehicle
dynamics. Also, the effect of design parameters on the performance of the Lyapunov controller was investigated.

(18)
plays the role of overall controller
It may be argued that
acts like a weighting factor between the current
gain and
spacing error and its rate of change . In other words, a small
implies a more tight control of spacing errors. Therefore, it
would be natural to put more weight on the current error. Also, a
means looser control of the current spacing errors, and
large
consequently a longer time will be needed to reduce the errors.
Fig.8showsthesimulationresultsas and vary.Asbefore,
the same scenario was used, and only the spacing error between
the leader and first vehicles are displayed. As one can easily see,
increases. Although any choice
the overshoot is reduced as
and
assures overall stability, too small a value for
of
may lead to the apparent instability of spacing errors between
the lead vehicle and the first vehicle. That is because the dynamic
characteristics of the platoon leader vehicle are not considered in
the design of the controller for the first vehicle, and too small
means almost no control authority. It can be seen that
affects
both the overshoot and the speed of response.
E. Mixed Controllers
In the case of communication link failure, the spacing of the
whole platoon will be jeopardized. Therefore, the systems reliability is very important. While the PID controller requires a
global link in the sense that every follower vehicle should receive information from the lead car, the Lyapunov controller
is wholly dependent on the local links. As suggested in Fig. 9,
using both control approaches will mean that they will serve as
backups to each other because it is easy to switch from one controller to another, regardless of which controller is used as the
primary one. One should note that the mixing of two different
kinds of control logic is possible because although the Lyapunov
controller in the follower vehicle needs to receive the accelerafrom the preceding car, it is not affected by
tion command
is computed.
the way that
To illustrate this, two cases were simulated using the scenario
of sudden speed change. First, the PID controller was used as the

REFERENCES
[1] IEEE Trans. Veh. Technol. (Special Issue on Intelligent Vehicle Highway
Systems), vol. 40, no. 1, 1991.
[2] P. Varaiya, Smart cars on smart roads: Problems of control, IEEE
Trans. Automat. Contr., vol. 38, pp. 195207, Feb. 1993.
[3] J. K. Hedrick, M. Tomizuka, and P. Varaiya, Control issues in automated highway systems, IEEE Contr. Syst., pp. 2132, Dec. 1994.
[4] S. E. Shladover, Review of the state of development of advanced vehicle
control systems, Int. J. Veh. Syst. Dyn., vol. 24, pp. 551595, 1995.
[5]
, Longitudinal control of automated guideway transit vehicles
within platoons, J. Dyn. Syst., Measure., Contr., vol. 100, pp. 302310,
1978.
[6]
, Longitudinal control of automotive vehicles in close-formation
platoons, J. Dyn. Syst., Measure., Contr., vol. 113, pp. 231241, 1991.
[7] S. Sheikholeslam and C. A. Desoer, Longitudinal control of a platoon of
vehicles I: Linear model, PATH, Res. Rep. UCB-ITS-PRR-89-3, Aug.
19, 1989.
, Longitudinal control of a platoon of vehicles III: Nonlinear
[8]
model, PATH, Res. Rep. UCB-ITS-PRR-90-1, Apr. 1990.
[9] T. Fujioka et al., Longitudinal vehicle following control for autonomous driving, in Proc. AVEC96 Int. Symp. Advanced Vehicle
Control, Germany, June 2428, 1996, pp. 12931304.
[10] J. K. Hedrick et al., Longitudinal vehicle controller design for IVHS
systems, in Proc. Amer. Control Conf., 1991, pp. 31073111.
[11] D. H. McMahon et al., Longitudinal vehicle controller design for
IVHS: Theory and experiment, in Proc. Amer. Control Conf., 1992,
pp. 17531757.
[12] S. Sheikholesam and C. A. Desoer, Longitudinal control of a platoon of
vehicles with no communication of lead vehicle information, in Proc.
Amer. Control Conf., 1991, pp. 31023106.
[13] J.-J. E. Slotine and W. Li, Applied Nonlinear Control. Englewood
Cliffs, NJ: Prentice-Hall, 1991.
[14] D. H. McMahon and J. K. Hedrick, Longitudinal model development for automated roadway vehicles, PATH , Res. Rep.
UCB-ITS-PRR-89-5, Oct. 1989.
[15] A. S. Hauksdottir and R. E. Fenton, On the design of a vehicle longitudinal controller, IEEE Trans. Veh. Technol., vol. VT-34, pp. 182187,
Nov. 1985.

Tae Soo No, photograph and biography not available at the time of publication.
Kil-To Chong, photograph and biography not available at the time of publication.
Do-Hwan Roh, photograph and biography not available at the time of publication.