Content uploaded by Guoyuan Wu
Author content
All content in this area was uploaded by Guoyuan Wu on Aug 04, 2015
Content may be subject to copyright.
Abstract—Transportation is responsible for approximately a
third of greenhouse gases (GHG) and a major source of other
pollutants including hydrocarbons (HC), carbon monoxide
(CO), and oxides of nitrogen (NOx). Intelligent Transportation
System (ITS) technology can be used to lower vehicle emissions
and fuel consumption, in addition to reducing traffic congestion,
smoothing traffic flow, and improving roadway safety. As
wireless communication advances, connected-vehicles-based
Advanced Traffic Management Systems (ATMS) have gained
significant research interest due to their high potential. In this
study, we examine the concept of ATMS for connected vehicles
using a multi-agent systems approach, where both vehicle
agents and an intersection management agent can take
advantage of real-time traffic information exchange. This
dynamic strategy allows an intersection management agent to
receive state information from vehicle agents, reserve the
associated intersection time-space occupancies, and then
provide feedback to the vehicles. The vehicle agents then adjust
their trajectories to meet their assigned time slot. Based on
preliminary simulation experiments, the proposed strategy can
significantly reduce fuel consumption and vehicle emissions
compared to traditional signal control systems.
I. INTRODUCTION
A variety of Intelligent Transportation System (ITS)
techniques have been proposed and applied to not only
improve roadway safety and relieve traffic congestion, but
also reduce fuel consumption and emissions. Many of these
techniques fall into the ITS area of Advanced Traffic
Management System (ATMS) [1].
As wireless communication techniques rapidly advance,
new connected-vehicles-based ATMS concepts are emerging
in the research literature. With the capability to exchange
real-time traffic information by using vehicle-to-vehicle
(V2V) and/or vehicle-infrastructure (V2I/I2V)
communications, more efficient management strategies can
be developed. Example applications include: 1) intersection
approaching warning systems; 2) lane change warnings [2];
3) variable speed limits; 4) adaptive traffic signals [4 - 5]; 5)
automated traffic intersection control [3]; and 6) cooperative
adaptive cruise control.
There has been recent interest in applying Multi-Agent
Systems (MAS) in different traffic scenarios because they
have the potential to solve complex real-world problems [10].
In this research, we have taken a multi-agent approach to
improve intersection efficiency using the capabilities of
connected vehicles. With V2I and I2V communications, a
vehicle agent (VA) can request a reservation for its
space-time occupancy when passing through the intersection
and preplan its trajectory before it enters the intersection in
order to avoid collisions and minimize its waiting time.
Simultaneously, an intersection management agent (IMA)
can make reservations for each VA to optimize its driving
strategies, ranging from adjusting speed changes to selecting
a desired lane. Using connected vehicle technologies, this
multi-agent system can potentially improve intersection
traffic congestion and minimize energy and emissions along
urban arterials by avoiding unnecessary
acceleration/deceleration and idling at intersections as
compared to those with standard traffic signals that are in use
today.
In general, such an intersection management system using
a multi-agent approach consists of at least three major
components: 1) a vehicle behavior planning module; 2) a
dynamic intersection time-space reservation module; and 3)
individual vehicle trajectory planning that could possibly
include vehicle platoon formation. Fig. 1 illustrates an
intersection with these major components. This paper focuses
Figure 1. An intersection with the multi-agent based management system
Advanced Intersection Management for Connected
Vehicles Using a Multi-Agent Systems Approach
Qiu Jin1,2, Guoyuan Wu2, Member, IEEE, Kanok Boriboonsomsin2, and Matthew Barth1,2, Senior
Member, IEEE
1Department of Electrical Engineering, University of California Riverside
Riverside, CA 92507, USA, emails: qjin@ee.ucr.edu, barth@ee.ucr.edu, and
2College of Engineering – Center for Environmental Research and Technology, University of California Riverside
Riverside, CA 92507, USA, email: gywu@cert.ucr.edu, kanok@cert.ucr.edu
2012 Intelligent Vehicles Symposium
Alcalá de Henares, Spain, June 3-7, 2012
978-1-4673-2118-1/$31.00 ©2012 IEEE 932
on the development and evaluation of the dynamic time-space
reservation connected vehicles techniques. The rest of this
paper is organized as follows. Section II provides some
background information about the proposed methodology,
which is described in Section III. Simulation setup and results
are presented in Section IV, and Section V concludes this
paper with further discussion on future work.
II. BACKGROUND
A. Advanced Traffic Management Systems
Advanced Traffic Management Systems (ATMS) have
been studied extensively in the last decade and a large variety
of algorithms have been developed to improve safety, relieve
traffic congestion and reduce fuel consumption and pollutant
emissions on urban highways and local streets through the
deployment of state-of-the-art sensing, communications, and
data processing techniques. To address these problems,
ATMS take advantage of available traffic information
provided by infrastructure-based sensors (e.g., embedded
loop detectors, radar, video cameras) and vehicle sensors that
are coupled with on-board V2V and/or V2I communications.
Real-time management solutions are then being developed
and delivered using wireless communications to improve
traffic system operation.
B. Connected Vehicles
Coupled with other technologies such as smart sensors and
on-board computer processing, the connected vehicles
technique can fulfill tasks such as identifying hazards on the
roadway, providing warnings to drivers, and sharing
information with others over communication networks.
To support numerous applications in safety, mobility, and
sustainability, a vehicle communication network should have:
1) high-speed transactions among vehicles
(vehicle-to-vehicle or V2V), and 2) communications between
vehicles and the infrastructure (V2I and or I2V). V2V
applications rely on sharing information between vehicles
such as safety warnings and traffic information. V2I/I2V
applications typically have roadside units (RSU) and vehicles
as the communicating nodes. The major benefit of these
vehicle communication networks is to improve safety,
mobility and sustainability by tightly integrating
infrastructure and vehicles as a system, rather than relying on
isolated individual vehicles traveling on a roadway.
A number of applications have already been initiated by
using probe data from connected vehicles to modify traffic
control strategies and improve overall system performance.
Such probe data may include vehicle activity (e.g., position
and velocity) and other traffic or roadway information (e.g.,
slick roads). For example, with the knowledge of signal plan
and a collection of priority request arrival times from probe
data, the heuristic algorithm developed in [6] can reduce
average bus delay by 50% without negative impacts on other
vehicles. Rim and Kim [7] proposed a model to estimate
lane-level travel times under a V2I environment. Individual
vehicles’ speeds and positions at every second were used as
the inputs to the proposed model, and the survey showed that
the mean absolute percentage error (MAPE) was reduced by
less than 10% with a 20% penetration rate. Oertel and Wagner
[8] developed a new approach to control traffic signals at
isolated intersections in which individual vehicles’ delays
were estimated and used to adjust the green splits. This
delay-based control strategy outperformed the other
conventional schemes if the penetration rate is above 10%.
C. Traffic Simulation Tools
A variety of traffic simulation software has been developed
to model traffic networks. For example, SimTraffic is a
Synchro-companion program that allows visual simulation of
a surface street traffic network. Other traffic simulation
systems include CORSIM, Paramics, and VISSIM which are
designed to simulate high fidelity traffic on surface streets
and freeways. Many researchers have integrated wireless
communication simulation components to these traffic
simulation models to investigate a variety of ITS applications,
with limited success.
Recently a new tightly integrated traffic
simulation/wireless communication system has been
developed, which is called SUMO (Simulation of Urban
Mobility) [11]. It is a highly portable microscopic road traffic
simulation package developed by the Institute of
Transportation Systems at the German Aerospace Center.
With SUMO, the advance traffic control can be implemented
in Veins (Vehicles in Network Simulation), which is an
Inter-Vehicular Communication (IVC) simulation framework
composed of an event-based network simulator (e.g.,
OMNeT++, NS2) and a microscopic traffic simulation model
(e.g., SUMO) through a Traffic Control Interface (called
TraCI). Because SUMO was developed from the ground up to
handle traffic and communications together, it has the
potential to simulate much more complex scenarios. In this
study, we use SUMO to evaluate the performance of the
proposed intersection management system.
III. METHODOLOGY
In this section, an advanced traffic management system for
connected vehicles using a multi-agent systems approach is
proposed. The block diagram illustrated in Figure 2 outlines
different approaches in context, with the yellow blocks
showing the components used in this study.
In our design, the ATM system consists of two
components: Vehicle Agents (VA) and an Intersection
Management Agent (IMA). In order to maximize traffic
throughput and to better utilize the capacity of the
intersection, the IMA needs to manage the space-time
occupancies based on vehicle dynamics. Figure 3 gives the
time-space occupancies for example vehicles coming from
one direction in one lane. The total time-space is divided into
n x n grids. To avoid collisions, only one vehicle can occupy
one grid. Once Vehicle A, B, and C decide their trajectories,
some grid cells (e.g., the yellow, green and purple in Fig. 3)
have already been reserved based on their future dynamics.
933
Figure 2. Advanced traffic management research structure
Also, the time slots (s1 (t1, t2), s2 (t3, t4), s3 (t5, t6))
represent the occupancies when vehicles travel through the
intersection. For the multi-direction case, the time-space
occupancies of the intersection will be shared by all the
vehicles from all approaches. This would be represented by a
rectangle solid located at the intersection of multiple planes.
As we can see from this diagram, the capacity of the
intersection is limited. The key to improving the traffic
throughput is equivalent to maximizing the total time-space
occupancies. In this study, we focus on maximizing the
time-space occupancies in the intersection cross-area. To
achieve this goal, the two types of agents need to have
real-time communications and work collaboratively. On one
hand, the IMA needs to arrange the vehicle’s arrival times in
order to maximize the reservation number; on the other hand,
each VA has to preplan its own trajectory to avoid collisions
and to arrive at the intersection with its predetermined arrival
dynamics.
Figure 3. Road Time-Space Occupancies for Vehicles in one direction
As mentioned in the first section, due to different control
objectives (safety, mobility, and environment), the
management mechanisms for vehicle agents are developed
separately according to their locations in the intersection. In
this study, the proposed strategy is designed primarily to
address problems when vehicles are approaching the
intersection.
A. Assumptions
For simplicity, several assumptions were made in the
management mechanism.
Each VA should be a fully controllable agent.
All the agents are equipped with (V2I/I2V) wireless
communication devices.
Each VA can get its preceding vehicle’s dynamic
information (speed, position).
Each VA can get its own dynamic information (speed,
acceleration, turning angle, position, road map, and
etc.) from its on-boarding devices.
No message drops in communications and unlimited
communications capacity.
A VA cannot enter the intersection without a
reservation.
B. Dynamic Reservation System
In order to enter the intersection, each VA will keep
sending reservation requests to IMA until it obtains one.
To determine whether a request from the VA can be met or
not, the IMA needs to use some control policies.
According to a vehicle’s arrival lane, turning intention and
priority, three levels of policies are required to be followed
when making reservations:
1) Level 1. Priority-based Policy: Request with higher
priority (here is 1) will be processed first. The request
message received by IMA can be classified as: Request
message and Cancel-and-Reapply (CnR) message.
Request message is used for the VA who doesn’t have
a reservation in current status, and whose priority is
set to 0. CnR is designed for the VA who needs to
reapply a reservation immediately after its current
reservation is canceled, and its priority is set to 1.
2) Level 2. With-Lane-based Policy: If messages have the
same priority, vehicle arrival lane and position will be
examined. If any of the VAs that are traveling ahead of
the request VA along the same lane has not obtained a
reservation, then the request will be rejected. IMA
then sends a reservation-rejected message to the VA.
3) Level 3. First come, first serve (FCFS) Policy: If
messages have the same priority but different arrival
lane and turning intention, the IMA will serve the
earlier message.
After deciding which messages will be processed, the IMA
will check its current Reservation Table, and find any
available slots for these VAs to reserve a space-time
occupancy cell.
C. Multi-Agent Behavior Design
In this Advanced Traffic Management system, in order to
guarantee that agents can operate without any collision,
vehicle agents and intersection management agents need to
follow a set of separate behaviors. Figure 4 and Figure 5 show
the flowchart of those two types of actions. The MAS mode
D: Vehicle Lane
Position
Intersection
Position
Vehicle A Vehicle B
Vehicle C
t
D1
D2
t1 t2 t3 t4 t5 t6
934
and Default mode in the flowcharts are vehicle motion
planning modes that are describe later in the paper.
Figure 4. Vehicle agent actions
Figure 5. Intersection management agent actions
D. Vehicle Motion Planning
One of the keys to reduce traffic congestion and pollutant
emission is to prevent vehicles from unnecessary stops before
entering the intersection. The three-level reservation policy,
by itself, reduces vehicles stop-and-go actions and therefore
allows vehicles entering the intersection at a relatively high
speed most of the time. Although this policy will be effective
in this regard, there is still room for further improvement on
the vehicle itself. Therefore, it is crucial to carry out
appropriate vehicle motion planning before it enters the
intersection. A cooperative approach between VAs and the
IMA may provide appropriate motion estimations.
In this section, two motion-planning modes are designed
according to different status of vehicles: 1) Default Mode
(arriving at maximum speed and in shortest time) and 2) MAS
(maximum arrival speed) Mode. Note that in the simulation
the Krauss Car-Following Model (KCFM) [9] will govern all
the vehicles whenever and wherever they are. In the
following diagrams, it should be noted that for simplicity,
piecewise linear function is designed to satisfy those
constraints to construct the speed profiles.
(1) Default Mode:
Most of the time, a vehicle agent will not hold any
reservation; even when they are in the communication
network. However, to improve traffic flow and reduce total
travel time, vehicles still need to be piloted in a certain mode.
Under these considerations, the default motion mode is
designed for vehicles to arrive at the intersection in the
shortest amount of time. We can illustrate how the estimation
procedure works using a time-velocity diagram as shown in
Figure 6. In this figure, Vc is the current speed of VA, t0 is
current time, D is the distance between VA and IMA, Vmax is
the roadway speed limit, tend and Vend are the arrival time and
speed at intersection, amax is the maximum acceleration, and
amin is the maximum deceleration. Two scenarios need to be
considered according to the current position of the vehicles.
In case 1, the distance between a vehicle and intersection is
long enough for the vehicle to accelerate to the roadway
speed limit. In case 2, a vehicle cannot accelerate to the road
speed limit even using the maximum acceleration because the
distance is too small. Thus, the vehicle still has to accelerate
to a higher speed using amax to shorten the arrival time.
Figure 6. Time-Velocity diagrams for estimation of arrival time and arrival
speed using Default Mode
(2) Maximum Arrival Speed (MAS) Mode:
Once a vehicle agent receives the Available Time Slot
information from IMA, it is guaranteed to have a place in the
intersection reservation queue. In MAS mode, we need to
construct the trajectory with a given ending time and
maximize the arrival speed at the same time. Figure 7 shows
the time-velocity diagram that is used by vehicle to plan its
trajectory with a given arrival time.
After choosing a target time slot, a vehicle using the start
time point of this slot as its arrival time at the intersection, tend.
Then VA makes an acceleration/or deceleration decision
based on Vavg, which is defined as a constant velocity that
vehicle applies to travel distance D in tend, and Vavg=D/ tend.
After Vend is determined by the velocity profiles in Figure 8,
the VA uses an intersection map to simulate its trajectory in
the intersection in order to get its departure time tdep. If the slot
yes
no
yes
no
no
no
yes
yes
no
yes
yes
no
yes
no yes
no
Priority-
based policy
FCFS policy
Within-lane-
based policy
Make reservation
935
(tend, tdep) can be fitted in the selected time slot, the VA sends
it to the IMA and adjusts its motion based on this preplanned
trajectory. Otherwise, it will redo the trajectory planning
using the next earliest time slot.
Figure 7. Time-Velocity diagrams for estimation of arrival time and arrival
speed using MAS mode.
IV. SIMULATION SETUP AND RESULTS
A. Simulation Setup
A virtual advanced traffic management system for
connected vehicles using the multi-agent approach was
created in SUMO, and the performance of this system was
evaluated under different traffic conditions. Additionally, we
compared the results with those from conventional traffic
signal control in terms of average travel time, fuel
consumption and pollutant emissions.
For each experiment, the general simulation setup is:
An isolated intersection with one lane in each
direction
Lane length: each lane is 500 meters (from vehicle
initial point to the center of the intersection)
Speed limit for all lanes is 17.8 m/s (40 miles per
hour)
Maximum acceleration: 2.5 m/s2
Minimum acceleration: -2.5 m/s2
Vehicle type: light duty vehicle
Vehicle length: 2.5 meters
Vehicle safety gap: 2.5 meters
Only one vehicle is allowed in intersection cross area
at a time
Simulation time step: 0.1 seconds
Total simulating steps: 10000 steps
Communication range is 300 meters
We considered vehicles spawned from two directions
(West-to-East and North-to-South) with various traffic
volumes. For both directions, each individual vehicle’s initial
speed is set to be 0 m/s and traffic volume varies from 54 to
1227 vehicles spawned in 1000 seconds.
Two intersection control strategies have been tested for
each experiment: 1) Advanced traffic management using
multi-agent approaches and connected vehicles strategies;
and 2) Fixed timing signal: total cycle is 70 seconds, with
green phase 30 seconds, yellow phase 4 seconds, and all-red
clearance 1 second.
B. Travel Time Analysis
The Advanced Traffic Management (ATM) Approach
significantly outperformed conventional traffic signal control
in terms of average travel time under different traffic
demands. As we can see from Table I, the reduction
percentage of vehicle (from both directions) average travel
time ranges from 45.82% to 87% when traffic volume varies
from 54 vehicles to 1227 vehicles spawned in 1000 seconds.
As illustrated from the table, when traffic gets more
congested, ATM approaches can more efficiently use the
roadway occupancies compared to traditional signal control
method.
TABLE I. VEHICLE AVERAGE TRAVEL TIME FOR TWO-DIRECTION
TRAFFIC FLOW EXPERIMENTS
Vehicles
Spawned
Probability
(Uniform
Distribution
U (0,1))
Vehicles
Spawned
Volume
Within
1000
Seconds
Average
Travel Time
Using
Traditional
Signal
Control (s)
Average
Travel
Time using
ATM
approach
(s)
Reduction
Percentage
0.1
56
114.59
62.09
45.82%
0.3
598
168.39
61.05
63.74%
0.5
814
430.93
61.10
85.82%
0.7
1227
480.59
61.07
87.29%
C. Emissions and Fuel Consumption Analysis
In order to evaluate emissions and fuel consumption, we
used the emission model based on HBEFA (the Handbook of
Emission Factors for Road Transport) which has been
integrated into SUMO. The objective of the emission analysis
is to observe the variations of the evaluation metrics under
different traffic conditions.
The evaluation metrics include emissions of CO2, CO, HC,
NOx and fuel consumption. Cumulative emissions of the four
pollutants and energy consumption for all vehicles were
obtained from the emission evaluation of the simulation runs.
As shown in the results, both emissions and fuel consumption
are significantly reduced by using the proposed system. The
percentage of reduction for ATM control system ranges from
41% to 71% for CO, 65% to 75% for CO2, 55% to 78% for
HC, 63% to 74% for NOx, and 65% to 75% for fuel
consumption when traffic volume varies from 54 to 1227
vehicles spawned in 1000 seconds. Figure 8 shows fuel
consumption reduction and CO2 (major part of greenhouse
gas emissions) reduction under different traffic demands.
It is shown that the proposed system is able to significantly
936
reduce vehicle emissions, as well as fuel consumption. Such
reduction may result from the better utilization on the
intersection’s time-space occupancy of the proposed system
by avoiding unnecessary stop-and-go maneuvers.
Figure 8. CO2 reduction for a range of traffic volume in the two-direction
traffic scenario
V. CONCLUSIONS AND FUTURE WORK
An advanced traffic management system for connected
vehicles using a multi-agent system approach is developed
and evaluated in this study. The overall goal for this advanced
traffic management system is to make the traffic flow
smoothly, increase intersection throughput, and reduce
energy consumption and pollutant emissions. Multi-layer
reservation policies were proposed and interactions between
vehicle agents and the intersection management agent were
presented in detail. In addition, a vehicle motion planning
algorithm was developed based on these policies. The
simulation results showed that the proposed advanced traffic
management system can considerably alleviate traffic
congestion as well as reduce pollutant emissions and fuel
consumption.
In future work, the proposed system will be tested in a
more complicated traffic network (e.g., with turning
movements and multi-lane roadways). A more realistic
communication protocol between vehicle agents and
intersection management agents based on DSRC and SAE
J2735 protocols will also be developed and implemented. We
also plan on doing a detailed analysis on the system capacity
(i.e., the traffic demand level at which the system will break
down).
REFERENCES
[1] Barth, M. (2009) “Environmentally Beneficial Intelligent
Transportation Systems”, 12th IFAC Symposium on Control in
Transportation Systems, Redondo Beach, CA, September 2009.
[2] Park, H., B. L. Smith (2011) “Integrated Traffic/Communication
Simulation Evaluation Environment for IntelliDrive Applications
Using SAE J2735”, Proceedings of the 90th Annual Meeting of the
Transportation Research Board, January 2011.
[3] Fajardo, D., P. Stone (2011) “Automated Intersection Control:
Performance of a Future Innovation Versus Current Traffic Signal
Control”, Austin, Proceedings of the 90th Annual Meeting of the
Transportation Research Board, January 2011.
[4] Crenshaw, P. (2000) “FHWA Adaptive Control Survey,” presented at
the TRB Committee on Traffic Signal Systems Meeting, 2000 Midyear
Meeting, Seattle, Wash., July 9–10, 2000.
[5] Head, L. and P. Mirchandani (1998) “The RHODES Prototype: A
Technical Review,” Adaptive Traffic Signal Control Systems
Workshop, TRB Midyear Meeting, Asilomar, Calif., July 1998.
[6] Head, L. (2011) “Heuristic Algorithm for priority traffic signal
control”, Proceedings of the 90th Annual Meeting of the Transportation
Research Board, January 2011.
[7] Rim, H., S. Kim (2011) Estimation of Lane-Level Travel Times (L2TT)
in V2X-based Traffic Information System”, Proceedings of the 90th
Annual Meeting of the Transportation Research Board, January 2011.
[8] Oertel, R. and P. Wagner (2011) “Delay-Time Actuated Traffic Signal
Control for an Isolated Intersection”, Proceedings of the 90th Annual
Meeting of the Transportation Research Board, January 2011.
[9] Krauss, S., P. Wagner, and C. Gawron (1997) “Metastable states in a
microscopic model of traffic flow”, Phys. Rev. E55, 1997.
[10] Desai, P., Loke, S.W., Desai, P., and Singh, J., (2011) “Multi-agent
based vehicle congestion management”, Proceedings of the IEEE
Intelligent Vehicles Symposium, Baden Baden Germany, 2011.
[11] SUMO: Simulation of Urban Mobility, see:
http://sumo.sourceforge.net/.
82
84
86
88
90
92
94
96
56
598
814
1227
CO2 Reduction
Reduction(%)
937
