Professional Documents
Culture Documents
Abstract - Finding the optimal staging of traffic lights and pollution, accidents, road blockages, economic slumps and
determining the optimal traffic light cycles is one of the crucial parking issues attesting that traffic research area is still open
tasks involved in modelling modern traffic scenarios. In this and need new ventures of solutions to be explored.
paper, we propose two distinct approaches to find optimal
Expanding the infrastructure of urban areas is usually not
traffic light cycles using Multi-agent Systems (MAS) and
possible in most of the cases and hence researchers often
Swarm Intelligence (SI) concepts and compare the efficiency of
these solutions against a default strategy under heterogeneous
agree that optimized staging of traffic lights [I], [2], [3] can
traffic regions using same input parameters. The solutions help to mitigate congestion by enhancing the traffic flow of
obtained are simulated using SUMO (Simulation of Urban vehicles through cities. However, due to the increased
MObility), a well-known microscopic traffic simulator. We number of traffic lights installed in cities already and their
have investigated both approaches with two large and rapid growth, joint programming of cycles becomes more
heterogeneous metropolitan areas with hundreds of traffic complex due to the large number of combinations that can
lights located in the cities of Colombo in Sri Lanka and occur especially in close intersections. This necessitates the
Chennai in India, and our research solutions are shown to
implementation of intelligent systems to optimally program
obtain optimal traffic light cycles in both scenarios. In
the traffic light cycles and their phase durations.
comparison with predefined static cycle programs (using
SUMO's default traffic light cycle generation algorithm), our
solutions achieved quantitative improvements for the two main Solutions to this problem have evolved over time more from
objectives: reducing the waiting time and the journey time of an engineering perspective towards research trends on
vehicles significantly and qualitatively improving smooth constructing automated intelligent solution models [4].
traffic flow over the regions. Accordingly, current research trends in the field of
optimizing traffic control signals can be categorized into two
Keywords - Traffic Simulation, Multi-Agent Systems (MAS), main approaches: on the one hand, intelligent models which
Swarm Intelligence (SI)
adapts the cycle programs of traffic lights and alters the
phase durations throughout the day [5], [6], [7] based on
the solutions are designed very specific to the traffic regions technology becomes unleashed when several software agents
experimented and cannot be generalized to different regions. are put together to operate on the same execution
Finding solutions using intelligent techniques for large and environment. In this case, the group of agents is usually
heterogeneous scenarios is still an open research area. The conceived as a multi-agent system (MAS) [14], as the
issues need to be addressed become complex due to the successful completion of each individual agent's tasks is
greater number of adjacent intersections and the possible dependent on the decisions and actions of other agents. So,
higher interaction combinations between multiple traffic in multi-agent systems, agents are designed to coordinate
lights in any region of practical importance such as a their activities so as to avoid negative interactions with their
metropolitan area of a city. acquaintances and to exploit synergic potentials. Traffic
management issue itself a distributed problem. Thus the
excellent inspiration of decentralized coordination, self
organizing and autonomy makes multi-agent system as a
viable candidate for traffic solution models. So in our
proposed solution under category one, we have used multi
agent system based approach to dynamically control traffic
light cycle programs adapt to different congestion levels
reducing waiting time and the journey time of vehicles and
enhancing even traffic flow in the experimented research
scenarios. Traffic density detectors were placed in the
simulation environment as data collection agents and traffic
lights agents introduced as core goal driven agents
conceptualizing a multi-agent model and the solution was
implemented using JADE [17], SUMO [11] and utility
technologies [16][22].
Fig 2. Depicting initial input parameters for the two research scenarios.
section III, methodology and the experimental setup for our intelligent paradigm for solving optimization problems
proposed solutions are described. Section IV present the which originally took its inspiration from the biological
discussion of results obtained. Conclusions and future work samples of swarming, flocking and herding phenomena in
are given in Section V. animals. Swarm intelligence idea emerged from swarming
behavioural patterns observed in flocks of birds, schools of
II. BACKROUND STUDY fish, swarms of bees and even human social behaviour.
In this section, the basic concepts of multi-agent systems Particle swarm optimization (PSO) [18] is a population
(the core of our solver technique under the first approach), based stochastic optimization technique, which could be
swarm intelligence (the core of our solver technique under implemented and applied easily to solve various functional
the second approach) and the SUMO simulator (simulation optimization problems, or the problems that can be
tool involved in the experiments of solutions) are briefly transformed into functional optimization problems. Under
explained. the second category, we have proposed a new PSO approach
capable of obtaining efficient traffic light cycle programs for
A. Multi-agent Systems (MAS) generated past traffic patterns learned through the simulation
environment. Particle swarm optimization is a popular
The notion of software agents has become increasingly algorithm shown to perform fast convergence and this is a
popular over the last decade. The real potential of this highly desirable feature for an optimal traffic light cycle
20IS International Conference on Advances in ICT for Emerging Regions (TCTer) 24th & 2Stl, August 20lS
222 Traffic Light Optimization Solutions using Multi-modal, Distributed and Adaptive Approaches
finding solution model, where new immediate traffic light railway lines, traffic light locations, buildings, roundabouts,
schedules are required to address updated events occurring etc. These two scenarios were taken for the experiment,
in traffic environment In our solution the optimization since they constitute quite different urban areas with
strategy is composed of an optimization algorithm and a heterogeneous structures and traffic organization. We have
simulation procedure, The optimization segment is carried converted the extracted OpenStreetMaps to SUMO format
out by means of the particle swarm optimizer algorithm and introduced sample vehicles circulating and created
which has been specially designed to find optimal cycle traffic demand by assigning randomly generated routes using
programs by initialization method, solution encoding, fitness SUMO's random route generator module and provisioned
function, velocity calculation and global optimization initial traffic light logic with SUMO's default traffic light
process, The simulation procedure is the way of repeatedly cycle generator to prepare the preliminary experimental
assigning better quantitative fitness value to the solution, setups. Additionally, in order to obtain generalized
thus leading to optimized cycle programs tailored towards a concluding results, both scenarios were distinguished
learned traffic scenario instance, Our solution model was significantly by incorporating different numbers of vehicles
implemented using MALLBA [19] and SUMO [Ill moving through the streets with different numbers of traffic
lights operating within the regions. In Fig. 1, the selected
C Simulation of Urban MObility (SUMO) areas of the two cities are shown with their corresponding
capture views of Google Earth, OpenStreetMap and SUMO.
Simulation of Urban MObility (SUMO) [11], is a well And Fig. 2 depicts the initial input parameters for the two
known traffic simulator that provides an open source, highly scenarios.
portable, and microscopic road traffic simulation tool and
skilful in handling large traffic scenarios. SUMO requires B. Implementation Details
several input files that contain information about the traffic
1) Category One (Dynamic traffic light optimization
scenario and capable of generating various output files
model during the actual traffic flow): We have formulated
which can be used to analyse and compare results and
the solution model for this category based on Multi-Agent
helpful in generating visual graphical illustrations. SUMO
Systems (MAS) concepts. In our solution model, we have
can hold the information about the experimented traffic
introduced traffic light agents to be the core agents, as they
region in a "network" file using nodes, edges, and
need to communicate with the traffic environment and other
connections between them. This network file can be
supporting agents. We have also modelled data collection
generated by importing a map from popular digital mapping
agents - specifically called as 'detector' agents (Depicted in
software like OpenStreetMap [20] or another traffic
Fig 3) to collect various required information from the
management software like PTV-VISUM [21] and converted
environment to assist traffic light agents to make the
to a valid SUMO network. The "routes" file in SUMO is the
decisions. It is necessary to have vehicles circulating in the
traffic demand definition file which contains the information
environment in order to experiment with the solution model,
of all the edges (streets and lanes) each individual vehicle
but these vehicles are not modelled as agents. If we model
will pass through as route definitions. SUMO provides the
vehicles as agents, it would be computationally very
feature of extensibility by providing means of adding
expensive to simulate a huge quantity of vehicles, each one
"additional" files. These additional files can be used to
with driver's decision making and other cogmtlve
define traffic light cycles, define vehicle paths or to generate
behavioural aspects. In a real traffic environment, our
various output files contallllllg different statistical
solution model can be implemented using sensors and
information. SUMO already incorporates a default algorithm
network topologies facilitating communication between
for generating static cycle programs and phase durations for
components. But in order to experiment and analyse
traffic lights and we will be using this to compare with our
different combinations of optimization concepts, we have
both solutions derived results.
chosen the microscopic simulation environment SUMO for
the trial purposes and have used popular multi-agent system
III. EXPERIMENT AL SETUP
framework JADE to implement agent behaviours and actions.
Thus our solution is implemented in three layer architectural
In this section we present the experimental setup we
design - strategic layer is implemented using JADE which
followed to assess the performance of both our solution
includes agents responsible for core decision making and
models and how comparison results were analysed to derive
operational layer using SUMO responsible for infrastructure
conclusions. First, we describe how the two regional traffic
management and traffic flow and tactical layer using utility
scenarios were formulated specifically for this work. Later,
technologies for bridging both SUMO and JADE. The
the implementation details and input parameters for our
communication between the constituents is visually
solutions and how the results comparison was made against
represented using a components diagram in fig 4.
the already available SUMO's default strategy were
explained in detail.
A. Research Scenarios
24'h& 25th August 2015 2015 International Conference on Advances in ICT for Emerging Regions (ICTer)
S.T.Rakkesh, A.R.Weerasinghe, R.A.C.Ranasinghe 223
I
regional scenarios with exact initial configurations to help us
in comparing and analysing the results. Fig 5 depicts the
[' ':.: : :
--
.
. :
.
--- ..
'
basic pseudo code for the algorithm we have used to find the
optimized traffic light cycles.
I -=:::.:.!t.. I�
t
",,".!:I'- '.'''' �.
.
:
� • ...&. MM. .......... . .......
-' ------ The results and the analyses of them are presented in this
section. First, we study the performance of our solution
Fig 4. Three layer architecture design of our MAS based solution.
models in comparison with the default SUMO cycle
programs generator (SCPG) obtained traffic light cycle
In our solution model, traffic light agents are defined as core performance, and our solution models ability to generate
goal driven agents which are responsible for traffic light more optimized cycle programs for the two different regions
semaphore plans characterized by sequence of phases. Each we have experimented. For analysis purpose, we have
phase has duration and a colour scheme (green, yellow and concentrated on the trip duration and waiting times of
red), whose values correspond to every possible maneuver at vehicles for the completed trips during the simulation
the intersection. A traffic light cycle is the execution of the window of 3000 simulation steps. The following are the
phase sequences and has an execution period equal to the brief description of these comparison attributes.
sum of the durations of the phases. At the same time, these
cycle programs have to coordinate traffic lights in adjacent • Trip duration - The time the vehicle needed to
intersections with the aim of improving the global flow of accomplish the destination (Measured in simulation
vehicles circulating adhering to traffic regulations imposed step units)
in the traffic region. In this context, our objective is to find • Waiting Time - The number of micro step units in
optimized cycle programs locally for all the traffic lights which the vehicle speed was below 0.1 m/s. (In our
located in our research scenarios converging with the global simulation, the number of total micro step units are
optimization. In order to obtain this local and global 30000, which is derived by splitting each
optimization effect we have instigated a learning algorithm simulation step unit by 10.)
which incorporates optimization process through
reinforcement learning.
·Trlp dur;rllo:fi
=� �ii�
formulated the solution model for this category based on
�������
tLi.:!
Swarm Intelligence (SI) concepts. In this solution model, we
� �ff!n1iH §!�g
�
employ a particle swarm optimizer algorithm to find optimal
cycle programs for traffic lights and the experiments were
carried out using SUMO traffic simulator environment.
According to the solution model, the pre-generated traffic
I
data will be fed to the solver and the optimized traffic light � � ------
1 1 �: .!..I---------------
:.o.:,.:.U
� �'II
. K:
cycle programs will be derived as a result. ;OC�
a!�D
: ! ]�: I���������
-------
L
J:£ Uti :!':1'b1 •• .......... .:.I. u 14":I"A% i::t&n ttHI 'b;.lt. fib,. .. ..-.1..... (J{HII!:.l .b t�.� .... !:.:I� � � '" B; ;::: ::j � � r':. ii1
1Ij,r.llf.l.n t,lU.lo,wL ,illAI.: u. IJI'J n,. r.au ..... -:"1 "' ... ,-. '" 'lI: ':1 =i � !::: �
ih1=M i:h1 l-UtlnL.!! wt:l! :h. .... h,1It. UI'.n.HI .... , N'I l.n I:.hri :'I ili'lbTriu.it
20IS International Conference on Advances in ICT for Emerging Regions (TCTer) 24'" & 2Stl, August 20lS
224 Traffic Light Optimization Solutions using Multi-modal, Distributed and Adaptive Approaches
.-c :�:.w
�
-,;:L ....... ----
'� "''''
, -------��=��-- ,
! IOC;c.
�
t-------
�::L
� 1�>O
f: �----',I,----r-.-'---
--------
;;
:! '01(<» --------
� ;roo -,-,,--.,,-,-,....,.---
o ,i. L
,�. ..I
,
Fig 7
1. AI, A2 depicts the vehicles waiting time duration with the SUMO's
default traffic light cycle generator for Colombo and Chennai
respectively,
2. BI, B2 depicts the vehicles waiting time duration with our Multi
Agent Systems (MAS) solution model for Colombo and Chennai
Fig 9 - A, B and C depicts the general vehicle tlow near Rajah Mutiah
respectively,
Road, Chennai with the SUMO's default traffic light cycle generator, our
3. CI, C2 depicts the vehicles waiting time duration with our Swarm
Multi-Agent Systems (MAS) approach and Swarm Intelligence(SI)
Intelligence (SI) solution model for Colombo and Chennai approach respectively.
respectively.
24'h& 25th August 2015 2015 International Conference on Advances in ICT for Emerging Regions (ICTer)
S.T.Rakkesh, A.R.Weerasinghe, R.A.C.Ranasinghe 225
v. CONCLUSIONS [9] Garcia-Nieto, Jose, Ana Carolina Olivera, and Enrique Alba. "Optimal
cycle program of traffic lights with particle swarm optimization."
In this paper, we have proposed two solution models based Evolutionary Computation, IEEE Transactions on 17.6 (2013): 823-839.
on Multi-Agent Systems (MAS) concepts and Swarm
Intelligence (SI) concepts that can find optimized traffic [10] K. N. Hewage and 1. Y. Ruwanpura, "Optimization of traflic signal
light timing using simulation," in WSC '04: Proceedings of the 36th
light cycle programs. For the experimental purpose, we have
conference on Winter simulation. Winter Simulation Conference, 2004, pp.
used SUMO, a microscopic traffic simulator and scenarios 1428-1436.
evaluated with two extensive and heterogeneous
metropolitan areas located in the cities of Colombo in Sri [II] D. Krajzewicz, M. Bonert, and P. Wagner, "The open source traffic
simulation package SUMO," RoboCup 2006 Infrastructure Simulation
Lanka and Chennai in India. Based on our analysis results,
Competition, 2006.
in comparison with explicitly defined static cycle programs
(Using SUMO's default traffic light cycle generator), we can [l2] Balmer, M., Nagel, K., and Raney, B. Large-scale multi-agent
conclude that both our approaches achieved quantitative simulations for transportation applications. Journal of Intelligent
Transportation Systems: Technology, Planning, and Operations 8, 4 (2004),
improvements for the two main objectives: reducing the
205-221.
waiting time and journey time of vehicles significantly and
qualitative improvement of smooth traffic flow over the [13] Silva, B. C. d., Junges, R., Oliveira, D., and Bazzan, A. L. C. ITSUMO:
experimented traffic regions. However, further an intelligent transportation system for urban mobility. In Proceedings of
the 5th International Joint Conference on Autonomous Agents and
improvements can be identified for future work in our
Multiagent Systems (AAMAS 2006) - Demonstration Track (May 2006), P.
solution models. In this paper the accidents and unexpected Stone and G. Weiss, Eds., ACM Press, pp. 1471-1472.
road blocks are not taken into consideration explicitly. For
example, in real traffic scenario road blocks and accidents [14] Van Der Hoek, Wiebe, and Michael Wooldridge. "Multi-agent
systems." Foundations of Artificial Intelligence 3 (2008): 887-928.
can occur and the vehicles should be re-routed efficiently in
order to ensure proper traffic flow. In future we will be [IS] Blum, Christian, and Xiaodong Li. Swarm intelligence in optimization.
focusing on how traffic light cycle programs can adapt when Springer Berlin Heidelberg, 2008.
an unexpected event like accident happens and how vehicle
[16] Krumnow, Mario. "Sumo as a Service-Building up a Web Service to
re-routing can be coordinated in the traffic environment.
Interact with SUMO." Simulation of Urban Mobility. Springer Berlin
Heidelberg, 2014. 62-70.
[5] Adam, L, et al. "Adaptive fuzzy logic traffic light management system."
Engineering Technology and Technopreneuship (TCE2T), 2014 4th
International Conference on. IEEE, 2014.
[6] Pescaru, Dan, and Daniel-loan Curiac. "Ensemble based traffic light
control for city zones using a reduced number of sensors." Transportation
Research Part C: Emerging Technologies 46 (2014): 261-273.
[7] McKenney, Dave, and Tony White. "Distributed and adaptive traffic
signal control within a realistic traffic simulation." Engineering
Applications of Artificial Intelligence 26.1 (2013): 574-583.
[8] Abushehab, Rami K., Baker K. Abdalhaq, and Badie Sartawi. "Genetic
vs. particle swarm optimization techniques for traffic light signals timing."
Computer Science and Information Technology (CSIT), 2014 6th
International Conference on. IEEE, 2014.
20IS International Conference on Advances in ICT for Emerging Regions (TCTer) 24th & 25tl' August 2015