HAL Id: hal-01091154
https://hal.archives-ouvertes.fr/hal-01091154
Submitted on 4 Dec 2014
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Cooperative Adaptive Cruise Control in Real Trac
Situations
Vicente Milanés, Steven E. Shladover, John Spring, Christopher Nowakowski,
Hiroshi Kawazoe, Masahide Nakamura
To cite this version:
Vicente Milanés, Steven E. Shladover, John Spring, Christopher Nowakowski, Hiroshi Kawazoe, et
al.. Cooperative Adaptive Cruise Control in Real Trac Situations. IEEE Transactions on Intelligent
Transportation Systems, IEEE, 2014, 15, pp.296 - 305. �10.1109/TITS.2013.2278494�. �hal-01091154�
1
Cooperative Adaptive Cruise Control in Real Traffic
Situations
Vicente Milanés, Steven E. Shladover, John Spring, Christopher Nowakowski, Hiroshi Kawazoe and
Masahide Nakamura
Abstract—Intelligent vehicle cooperation based on reliable
communication systems contributes not only to reducing traffic
accidents, but also to improving traffic flow. Adaptive Cruise Con-
trol (ACC) systems can gain enhanced performance by adding
vehicle-vehicle wireless communication to provide additional
information to augment range sensor data, leading to Cooperative
ACC (CACC). This paper presents the design, development,
implementation and testing of a CACC system. It consists of
two controllers, one to manage the approaching maneuver to
the leading vehicle and the other to regulate car-following once
the vehicle joins the platoon. The system has been implemented
on four production Infiniti M56s vehicles, and this paper details
the results of experiments to validate the performance of the
controller and its improvements with respect to the commercially
available ACC system.
Index Terms—Cooperative Adaptive Cruise Control (CACC),
adaptive cruise control (ACC), intelligent driving, cooperative
vehicles, connected vehicles, intelligent transportation systems
(ITS)
I. INTRODUCTION
Significant developments in Advanced Driver Assistance
Systems (ADAS) have been achieved during the last decade.
Intelligent systems based on on-board perception/detection
devices have contributed to improve road safety [1]. The next
step in the development of ADAS points toward vehicle-
to-vehicle (V2V) communications to obtain more extensive
and reliable information about vehicles in the surrounding
area, representing a cooperative ITS system. Using wireless
communication, potential risk situations can be detected earlier
to help avoid crashes and more extensive information about
other vehicles’ motions can help improve vehicle control
performance. Research projects have been conducted through-
out the world to define the requirements for an appropriate
vehicular communication system and its possible applications
[2].
Although most of the V2V cooperative ITS applications
have been focused on improving collision avoidance and safety
[3], the extension of the commercially available Adaptive
Cruise Control (ACC) system toward the Cooperative ACC
(CACC) system has a high potential to improve traffic flow
V. Mílanés wants to especially thanks to the ME/Fulbright program and the
Center for Automation and Robotics (CAR,UPM-CSIC) for its support in the
development of this work.
V. Milanés, S. Shladover, J. Spring and C. Nowakowski are with
the California PATH Program of the Institute of Transportation
Studies, University of California, Richmond, CA 94804. (e-
mail: vicente.milanes@berkeley.edu, steve@path.berkeley.edu,
jspring@path.berkeley.edu, chrisn@path.berkeley.edu)
H. Kawazoe and M. Nakamura are with the ITS Advanced and Product
Engineering Group, Nissan Motor Co., Ltd. Kanagawa 243-0123 Japan. (e-
mail: kawazoe@mail.nissan.co.jp, n-masa@mail.nissan.co.jp)
capacity and smoothness, reducing congestion on highways.
By introducing V2V communications, the vehicle gets infor-
mation not only from its preceding vehicle –as occurs in ACC–
but also from the vehicles in front of the preceding one. Thanks
to this preview information, oscillations due to speed changes
by preceding vehicles can be drastically reduced. Benefits from
including communications in ACC systems have been widely
studied in recent years [4], [5], [6].
Prior experimental results using vehicle-vehicle cooperation
to improve vehicle following performance were achieved by
the California Partners for Advanced Transit and Highways
(PATH) in 1997 [7], [8] when a platooning maneuver involv-
ing eight fully-automated cars was carried out using wire-
less communication among vehicles –mainly for longitudinal
control– and magnetic markers in the infrastructure –mainly
for lateral control. Based on the idea of a leading vehicle
guiding several followers, the Safe Road Trains for the En-
vironment (SARTRE) European Union project has developed
virtual trains of vehicles in which a leading vehicle with a
professional driver takes responsibility for each platoon [9].
That concept of the professional driver in the first vehicle was
originally developed in the European project called CHAUF-
FEUR [10].
Specifically related to CACC implementations in production
cars, two important projects were recently conducted in the
Netherlands. The Connect & Drive project, funded by the
Dutch Ministry of Economic Affairs, carried out a CACC
demo using six passenger vehicles [11] adopting a constant
time gap spacing policy. For the Grand Cooperative Driving
Challenge (GCDC) competition in 2011, nine heterogeneous
vehicles from different European research institutions tried
to perform a two-lane CACC platoon [12]. This competition
revealed some of the most important problems to be solved
before bringing this technology into production, including the
communication systems reliability. From the control point of
view, most of the implementations were based on proportional,
proportional-derivative feedback/feedforward controllers [13],
[14], [15] or Model Predictive Control (MPC) techniques [16],
[17].
When it comes to designing a CACC system, string stability
plays a key role [18]. The goal is designing a system able
to reduce disturbances propagated from the leading vehicle
to the rest of the vehicles in the platoon. There are two
different approaches to car following gap regulation: one based
on constant spacing or one based on constant time gap. A
comparative study between them, where CACC stability was
discussed, was presented in [19]. Several papers have dealt
with string stability analysis and simulations [20], [21], [22],
2
based on simplified theoretical models of ACC vehicle follow-
ing behavior, and have shown encouraging results. However
real production ACC systems have significant response delays
that have not been represented in the prior theoretical anal-
yses, but which destabilize the vehicle following responses.
Consequently those theoretical analyses have produced unre-
alistically optimistic estimates of the traffic stability impacts
of ACC.
In previous PATH research, a CACC involving two vehicles
was tested with very favorable results [23], [24], [6]. Building
on that previous work, this paper describes a new control
system design and implementation that is integrated in four
production vehicles. A constant time-gap car following strat-
egy was implemented similar to the commercial ACC, but with
the availability of significantly shorter time-gap settings. This
is achievable because V2V communications permit tighter
control of vehicle spacing and guarantee string stability, so
that inter-vehicle time-gap settings significantly shorter than
the production ACC time-gap settings are comfortable and
acceptable to drivers [24]. The whole system was then tested in
real traffic scenarios in order to compare its performance to the
production ACC system installed in the vehicles. Cut-in and
cut-out maneuvers were also tested to evaluate the controller
under regular traffic circumstances, emulating everyday traffic
situations.
The rest of the paper is organized as follows. Section
II presents a brief explanation about the production vehi-
cles used in the experimental phase, the control architecture
implemented on each vehicle and the vehicle model. The
control system that has been implemented based on a gap
closing controller and a gap regulation controller is explained
in Section III. Several experiments to validate the proposed
systems are included in Section IV. Finally, some concluding
remarks are given in Section V.
II. EXPERIMENTAL VEHICLES
Four production Infiniti M56s (see Fig. 1) were used as
the experimental vehicles. These vehicles are rear-wheel drive
with a 420-hp 5.6 liter V8 gasoline engine. They were fac-
tory equipped with lidar-based ACC, lane departure warning
(LDW) and blind spot detection systems. Additionally, a
5.9 GHz Dedicated Short Range Communication (DSRC)
system with a differential Global Positioning System (GPS)
incorporated in a Wireless Safety Unit (WSU) was supplied
by DENSO. Control was implemented through a dSpace
MicroAutoBox which received data from both the WSU and
the production vehicle’s Controller Area Network (CAN). In
particular, lidar data from the first and immediately preceding
vehicles and data from on-board vehicle sensors –speed,
acceleration, and yaw rate– were used by the control computer.
A. Control architecture
As previously stated, a factory installed ACC controller
was available in the vehicles. This controller sends target
speed commands to the vehicle’s actuators. The same control
variable is available for the CACC system, but it needs to
act through the commercial system that controls the throttle
Fig. 1. Experimental M56s vehicles
and brake pedals, i.e. the low-level controller. This constraint
somewhat limits the options when it comes to designing the
controller since the low-level controller could not be modified,
but it still provides adequate dynamic range for controlling
the vehicle under non-emergency transient conditions. Figure
2 shows the block diagram for the vehicle control architecture.
It consists of a classical robotics control architecture divided
into three stages:
• Perception phase where all information from the sensors
installed in the vehicle is received. Two sources can be
distinguished. On one hand, information coming from
the WSU system, where all data communicated by other
vehicles in the platoon –speed, acceleration, distance to
preceding vehicle, current time gap, control activation
and so on– is received and included on the CAN bus
data structure. This data also includes detection and as-
signment of the vehicle position sequence in the platoon,
which is carried out by the WSU using its GPS with
Wider Area Augmentation System (WAAS) differential
corrections. On the other hand, information is obtained
from the on-board sensors, such as factory available
lidar measurements (relative distance to the preceding
vehicle), odometer (current speed) and acceleration, and
flag signals (to get information about the interaction of
the driver with the driver interface such as activation or
deactivation of the system or gap setting selection). The
driver interface buttons, located on the right side of the
steering wheel, are also used for the CACC controller.
• Planning stage includes the high-level controller. Both
controllers, the commercial ACC system and the newly
developed CACC system, are available during the tests
so the driver can switch between them in real time.
For switching, the control code reads the CAN bus
information coming from the transmission mode selection
button–i.e., eco-driving mode, sport mode, standard mode
or snow mode. When the sport mode is chosen, the high-
level controller will take the CACC controller output.
When any of the other modes is chosen ,the production
ACC controller output will be sent to the low-level
controller. The CACC controller code has been developed
in Matlab/Simulink and is loaded in the vehicle using a
dSpace MicroAutoBox which is connected to the vehicle
via the CAN bus, where the target speed commands are
sent. The CACC controller can be deactivated in the
3
Fig. 2. Control architecture block diagram
same way as the commercial ACC, either using the driver
interface buttons or pressing the brake pedal.
• Actuation phase is in charge of executing target ref-
erence commands coming from the planning stage. As
previously stated, this low-level controller is in charge of
converting the target speed commands into throttle and
brake actions, using the factory ACC controller.
B. Vehicle model
The vehicle dynamic model, to be used for evaluating
vehicle performance from the string stability point of view,
is identified based on its step responses to different speed
changes. For these tests, a simple open-loop controller is in
charge of generating speed target commands to the vehicle.
Once the vehicle is stably driving at its current target speed, a
new speed command for an accelerating or braking maneuver
is sent to the vehicle. The Infiniti M56s vehicles are equipped
with a powerful engine that produces fast and strong responses
to changes in the target speed command. Considering this,
a second order response model can be extracted from the
experimental test data, where two different dynamic responses
are clearly evident in the behavior of the vehicle during the
accelerating and braking phases. Overshoot occurs on the
braking response since a high engine braking force – especially
in the sport driving mode, where the CACC controller was
designed– is added to the friction braking action. Two second-
order models with time delay were identified from the test
data, with the following structure
F (s) =
k
s
2
+ 2θω
n
s + ω
2
n
e
−T
d
s
(1)
with k, θ, ω
n
and T
d
being the static gain, damping factor,
natural frequency and time delay respectively. Parameters
for both models –i.e. braking and acceleration responses–
are defined in Table I. They are obtained using the Matlab
System Identification Tool, which permits selection among
different candidate transfer functions. This transfer function
for representing vehicle behavior was chosen as a trade-off
between simplicity (second order model) and goodness of fit
(over 95%). Responses for both models to a speed change are
depicted in Fig. 3. One can appreciate how well the model
fits the response of the real vehicle. This model incorporates
TABLE I
ACCELERATING AND BRAKING MODEL PARAMETERS
k θ ω
n
T
d
Accelerating 0.156 0.661 0.396 0.146
Braking 1.136 0.5 1.067 0.287
0 5 10 15 20 25
23
24
25
26
27
28
29
Speed (m/s)
Acceleration response
Reference
Experimental results
Simulation results
0 5 10 15 20 25
23
24
25
26
27
28
29
Time (s)
Speed (m/s)
Braking response
Reference
Experimental results
Simulation results
Fig. 3. Vehicle’s longitudinal dynamics response for acceleration and braking
maneuvers
both the dynamics of the vehicle and the low-level controller
in charge of managing throttle and brake actions.
III. CONTROL DESIGN
The goal of the CACC controller is maintaining the driver-
desired time gap to the preceding vehicle in any traffic circum-
stance, with both smoothness and accuracy. The ACC driver
interface is used to manage the CACC controller. It includes
buttons for activating and deactivating the ACC controller,
three gap settings and the option of setting, increasing or de-
creasing the cruise control speed, in case no vehicle is detected
in front of the ego-vehicle. For the ACC factory system, the
available gap settings are 1.1, 1.6 and 2.2 seconds. For the
CACC system, the shortest gap is set at 0.6 seconds. This value
has been chosen based on estimates of the separation needed to
enable crash avoidance under emergency conditions, as well
as previous tests of acceptance by drivers from the general
public [24]. The other two available CACC gap settings are 0.9
and 1.1 seconds. There were two limitations when designing
the CACC controller: it is not possible to access or modify
the low-level controller; and acceleration and deceleration are
limited to maximum values (0.1g and 0.28g respectively) by
the low-level controller.
The controller has been divided in two stages. The first
stage occurs when the CACC system is activated and there is
no vehicle in front of it or the ego-vehicle is far away from the
preceding one. In these cases, the vehicle is usually following
its set speed so an approaching maneuver has to be carried
out before switching to a car-following policy. This controller
-the gap closing controller– has to be as smooth as possible
to permit a good transition to the gap regulation controller. It
is also used in case of cut-out maneuvers when a vehicle in
4
K
P
(s)
+D(s)
K
L
(s)
G(s)
P
P
(s)
P
L
(s)
X
i
X
i-1
X
o
U
i
U
i-1
-
-
+
+
Vehicle model
Preceding car-following policy
Leading car-following policy
Preceding gap error controller
Leading gap error controller
Comm delay
Fig. 4. CACC control structure block diagram
the middle of the platoon decides to leave and the following
one has to cover the cut-out vehicle gap.
Once the vehicle has joined its predecessor, the second stage
controller –the CACC gap regulation controller– is in charge
of implementing the car-following policy depending on the
time gap selected by the driver. Three different time gaps
are available, following the production ACC structure. This
controller is also in charge of managing cut-in maneuvers
when a non-equipped vehicle merges into the platoon. Both
controllers have been designed to be consistent with how a
human driver handles each driving situation.
A. Gap regulation controller
Once the vehicle is close enough and the gap closing ma-
neuver has finished, the system switches to the gap regulation
controller, which is the core of the CACC control system.
Commercially available ACC systems try to reduce the gap
error (x
e
) between the ego-vehicle and the preceding one.
Information from the lidar/radar is used to reduce gap error
(x
e
) between desired time gap (t
g
) and relative distance
(x
r
= x
p
− x
f
) –where x
p
is the current position of the
preceding vehicle and x
f
is the position of the ego-vehicle,
following the next expression
x
e
= x
r
− v
f
t
g
(2)
where v
f
is the speed of the ego-vehicle. For this purpose,
a controller, K
P
(s) based on a standard PD-control structure
[25], [26], can be easily adjusted to get stability with respect to
the immediately preceding vehicle but, as previously demon-
strated [27], this is not enough to guarantee string stability.
Exchanging information among vehicles using wireless
communications permits improving the vehicle’s response as
well as significantly reducing time gaps, keeping safety. When
it comes to designing a CACC system, string stability is one
of the main goals. It can be defined as the system’s ability
to reduce perturbations downstream, avoiding that leading
vehicle speed changes cause amplification in the rest of the
vehicles [28]. According to [29], it can be defined as
|SS(s)| =
X
i
(s)
X
i−1
(s)
≤ 1, i ≥ 2 (3)
where i indicates the position of the vehicle in the platoon.
Figure 4 shows the CACC controller block diagram where
G(s) represents the vehicle model; terms P
P
(s) and P
L
(s)
correspond to the car-following policy with respect to the pre-
ceding and the leading vehicle respectively; terms K
P
(s) and
K
L
(s) represent time-gap error regulation controller for the
preceding and the leading vehicle respectively; U
i
and U
i−1
correspond to the control action for the ego-vehicle and the
preceding vehicle (coming from the wireless communication
system) respectively; the term D(s) = e
−δs
represents the
time delay in the wireless communication; and X
i
, X
i−1
and
X
0
represent the positions of the ego-vehicle, the preceding
and the leading vehicle position in the platoon. The controller
is formed by three main terms: one of them is in charge of
keeping the current speed but, instead of using the ego-vehicle
or preceding vehicle speed [23], the preceding vehicle target
speed (U
i−1
) is used as a feedforward term. This permits
improving the vehicle response time to speed changes and
reducing delays in the transition between throttle and brake
actuations. The other two terms try to keep the errors with
respect to the preceding vehicle K
P
(s) and the leading vehicle
K
L
(s) as small as possible. Both terms correspond to a classic
PD structure defined as
K
P
(s) = k
1
s + k
2
(4)
K
L
(s) = k
3
s + k
4
(5)
where car-following policies can be defined as
P
P
(s) = h
P
s + 1 (6)
P
L
(s) = h
L
s + 1 (7)
with h
P
and h
L
being the time-gap target values with respect
to the preceding and leading vehicles respectively.
As previously stated, the vehicle clearly exhibits different
dynamics in the acceleration and braking phases that can be
modeled by second-order functions sG(s) = F (s) so the
position X
i
(s) for a vehicle in the platoon can be determined
as
X
i
(s) = G(s)U
i
(s) (8)
where U
i
(s) is the target speed command for that vehicle.
Assuming the vehicles start from rest and using equation (8),
the relation between the ego-vehicle and the preceding one is
given by
X
i
(s) =
D(s) + G(s)K
P
(s)
1 + G(s) [K
P
(s)P
P
(s) + K
L
(s)P
L
(s)]
X
i−1
(s)
(9)
Following the criteria defined for string stability, the goal
is to keep the Bode magnitude below the unity gain of both
vehicle dynamics, i.e. acceleration and braking responses.
Control gains were firstly modified using tuning tools from
Matlab/Simulink to fulfill the requirements. Final tuning was
carried out in the experimental vehicle in order to not only
get an accurate response from the car-following policy point
of view but also a smooth riding quality from the driver
perspective. Table II shows the final parameters selected for
the preceding and leader car-following policy controllers. The