an urban mobility overlay for evaluating cellular driven...
TRANSCRIPT
IT 17 085
Examensarbete 30 hpNovember 2017
An Urban Mobility Overlay for Evaluating Cellular Driven Vehicle Teleoperation
Yiqing Wang
Institutionen för informationsteknologiDepartment of Information Technology
Teknisk- naturvetenskaplig fakultet UTH-enheten Besöksadress: Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0 Postadress: Box 536 751 21 Uppsala Telefon: 018 – 471 30 03 Telefax: 018 – 471 30 00 Hemsida: http://www.teknat.uu.se/student
Abstract
An Urban Mobility Overlay for Evaluating CellularDriven Vehicle Teleoperation
Yiqing Wang
Cellular Driven Vehicle Teleoperation is far too expensive to deploy and implementin real life. Most research community use simulation as a solution. Nowadays mostevaluations are based on one simulator either for traffic or network. In order toguarantee the safety and accuracy for operating a vehicle at distance, in this project,we first did a state-of-art study of simulators and developed traffic and networkscenarios using microscopic traffic simulator SUMO (Simulator of Urban MObility)and accurate network simulator NS3 (Network Simulator 3) then integrated themtogether. To have the accurate result the scope of the traffic mobility model need tobe realistic and fulfill several requirements and we have accomplished all the generalrequirements needed such as map size, mobility obstacles (traffic congestion, cardensity, road regulation) and realism. The process of the work is presented in thisreport.
We evaluate the integrated scenario from aspects of feasibility, capacity andperformance and the result shows that it's feasible to use LTE network to teleoperateup to 5 vehicles with granted Quality of Service requirements in area of Kista. Adrawback of this project is the simulation is off-line. Furthermore, to simulate thebehavior of on-line simulation we include traffic congestion avoidance and we run thesimulation multiple times in small time scale.
Tryckt av: Reprocentralen ITCIT 17 085Examinator: Mats DanielsÄmnesgranskare: Edith NgaiHandledare: Aneta Vulgarakis
Popular Scientific Summary in Swedish
Cellulart driven fordonsteleoperation ar alldeles for dyr att distribuera och implementera
i verkligheten. Sa i de flesta forsknings sammanhang ar simulering en losning. I dagens
forskning ar utvarderingen baserad pa en simulering antingen for trafik eller natverk.
For att garantera sakerhet och noggrannhet for ett fordon i rorelse pa ett avstand, i detta
projektet, har vi forst studerat ”the state of art” av simulatorer och utvecklat trafik och
natverk scenarios med hjalp av microscopic traffic simulator SUMO (Simulator of Urban
MObility) och accurate network simulator NS3 (Network Simulator 3 ) sen integrerade
dem tillsammans. For att astadkomma ett noggrant resultat behovde omradet som
man undersokte vara realistiskt och uppfylla flertalet krav och vi hade genomfort alla de
generella kraven som behovdes sa som kartstorlek, mobilitetshinder (trangsel, fordons
tathet, vagreglering) och realism. Processen for arbetet beskrivet ovan ar presenterat i
denna rapport.
Vi utvarderar de integrerade scenariona fran de foljande aspekterna genomforbarhet,
kapacitet och prestanda och resultatet visar att det ar genomforbart att anvanda LTE
network for att utfora teleoperation upp till 5 fordon, med tillaten Quality of Service krav
i Kista omradet. En nackdel av detta projektet ar att det ar en off-line simulering. Vi-
dare, for att simulera beteendet av en on-line simulator inkluderade vi trangsel avvikande
atgarder sa att ”operatoren” andrar fordonets mobilitetsmonster medans simuleringen
ar korande, dessutom kordes multipla simuleringar.
Acknowledgements
This thesis is carried out in Ericsson Research in Kista and the process is quite informa-
tive. During the project work i experienced how research work is done in a real industrial
company.
I would like to sincerely thank my supervisor Aneta Vulgarakis, Senior Researcher at
Ericsson Research, for her great help and support all the way from developing of the
project to the report writing. I benefit a lot from your ideas, suggestions and comments.
I would also like to give thanks to my reviewer Edith Ngai, Associate Professor at
Uppsala University for her valuable help and support.
Also credits go to my co-worker Yifei Jin, Master student in Royal Institute of Technol-
ogy(KTH), for his remarkable contribution building the network scenario in the project.
As well as all the other people from the research group Cognitive Automation Lab:
Elena Fersman, Rafia Inam, Athanasios Karapantelakis and Kevin Wang. Thank you
for the time sharing with me and the support.
Once again Thank you all.
iv
“In a connected world, everything is possible.”
Quote from Ericsson’s site
Contents
Abstract ii
Acknowledgements iv
List of Figures vii
List of Tables viii
1 Introduction 1
1.1 Vehicular Networking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Vehicle-to-Everything . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Research Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Background 5
2.1 Traffic Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 State of Art of Mobility Model . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Comparison of Traffic Simulator . . . . . . . . . . . . . . . . . . . 7
2.2 Vehicles Routing Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Heuristics in Routing: A* algorithm . . . . . . . . . . . . . . . . . 10
3 Implementation 13
3.1 Build Kista Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.1 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.2 Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2.1 Real-time Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.2 Off-line Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4 Evaluation 23
4.1 Simulation Summary of SUMO . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Feasibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5 Conclusion and Future Work 27
v
vi
Bibliography 29
List of Figures
1.1 An example of V2I and V2V scenario use LTE . . . . . . . . . . . . . . . 2
1.2 An example of V2X scenario[1] . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Illustration of Simulation Framework . . . . . . . . . . . . . . . . . . . . . 4
2.1 Level of mobility model[2] . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 General concept map of mobility model generation[3] . . . . . . . . . . . 7
2.3 Example of Manhattan, Diagonal and Euclidean distance . . . . . . . . . 11
3.1 Select large area from OSM . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 JOSM view of Kista OSM file . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 After Filtering Out Irrelevant Nodes . . . . . . . . . . . . . . . . . . . . . 15
3.4 After Filtering Out Irrelevant Nodes and Buildings . . . . . . . . . . . . . 15
3.5 Kista Scenario Road Network . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.6 Kista Scenario Topology with Polygons . . . . . . . . . . . . . . . . . . . 16
3.7 Kista Fading Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.8 Overview of steps to generate SUMO map file[4] . . . . . . . . . . . . . . 17
3.9 Bus Stop Location in Kista Scenario . . . . . . . . . . . . . . . . . . . . . 18
3.10 Example Bus Route No.178 from OSM . . . . . . . . . . . . . . . . . . . 18
3.11 Establishing a connection to SUMO[5] . . . . . . . . . . . . . . . . . . . . 19
3.12 Closing a connection to SUMO[5] . . . . . . . . . . . . . . . . . . . . . . . 20
3.13 VSimRTI architecture[6] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.14 Illustration of real-time simulation work flow between SUMO and NS3 . . 21
3.15 Illustration of off-time simulation work flow between SUMO and NS2[2] . 22
4.1 Traffic Condition in Google Maps at 8:00 am . . . . . . . . . . . . . . . . 24
4.2 Delay for Critical Stream in 10M Background Traffic . . . . . . . . . . . . 24
4.3 Packet Loss Ratio for Critical stream in 10M and 1M Background Traffic 25
4.4 Geometry location of antenna and vehicle route . . . . . . . . . . . . . . . 25
4.5 Packet Loss Ratio for non-Critical stream in 10M and 1M Back-groundTraffic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.6 Capacity for Bus Router per eNB . . . . . . . . . . . . . . . . . . . . . . . 26
vii
List of Tables
2.1 Features of Main Stream Traffic Simulators-1 . . . . . . . . . . . . . . . . 9
2.2 Features of Main Stream Traffic Simulators-2 . . . . . . . . . . . . . . . . 9
2.3 Features of Main Stream Traffic Simulators-3 . . . . . . . . . . . . . . . . 9
3.1 Topology Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Number of Polygons in the Kista Scenario . . . . . . . . . . . . . . . . . . 16
3.3 Summary of Bus in the Kista Scenario . . . . . . . . . . . . . . . . . . . . 18
4.1 SUMO Simulation Summary . . . . . . . . . . . . . . . . . . . . . . . . . 23
viii
Initials ix
CN Core Network
CQI Channel Quality Indicator
eNB Evoloved NodeB
HetVNET Heterogeneous Vehicular NETwork
LTE Long Term Evolution
QCI QoS Class Identifier
QoS Quality of Service
RAN Radio Access Network
RSSI Received Signal Strength Indicator
SUMO Simulation of Urban MOblity
UE User Equipment
VANET Vehicular Ad Hoc NETwork
V2X Vehicle to Xeverything
Chapter 1
Introduction
1.1 Vehicular Networking
As the consequence of rapid development of the wireless communication and the con-
cept of Internet-of-Things, vehicular communication has been gathering more and more
interests among the research community and industry during recent years. Many vehi-
cle manufactures have decided to include radio communication interface in their latest
model of vehicles. The number of cars with wireless transmitter is increasing and enor-
mous data are collected which will be beneficial for the whole society by contributing a
more efficient and safer traffic environment.
1.1.1 Vehicle-to-Everything
There are several vehicular communication types namely Vehicle-to-Vehicle(V2V), Vehicle-
to-Infrastructure(V2I), Vehicle-to-Network(V2N) and Vehicle-to-Pedestrian(V2P) with
focus on different aspects. For example V2V normally refers to the communication be-
tween vehicles with focus on safety. Information can be received by the vehicle nearby
to the event(accident, road congestion point, etc...) and send to the vehicle approaching
the incident to improve traffic efficiency and safety. V2I generally refers to the commu-
nication between vehicles and the network infrastructure with focus on traffic efficiency.
1
Initials 2
Figure 1.1: An example of V2I and V2V scenario use LTE
V2X(Vehicle-to-Everything)[1] contains any types of communication as long as any ve-
hicle is involved, either as source or destination. Different manufactures of vehicles can
communicate by using wireless communication through some standardized protocols and
interfaces so all the vehicles can be connected. V2X can enhance road safety and traffic
efficiency by collecting and providing information to all the participants in the traffic.
For example the road users can send the real time information to the traffic operator
who can use the data to give better control of traffic by sending route information to
other road users so they can make better decisions.
Figure 1.2: An example of V2X scenario[1]
1.2 Research Problem
Wireless technology plays import role in V2X. There are twp main stream standards used
in V2X network: Long-Term Evolution (LTE)[7] and IEEE 802.11P[8]. LTE is the latest
wireless communication technology, can provide peak speeds of downlink at 100Mbps
and uplink at 50Mbps(average user throughput at 5-12Mbps downlink/2-5 Mbps uplink)
permitting latency less than 5ms for small IP packets from User Equipment(UE) to
Radio Access Network(RAN) edge. LTE channel offers four times more bandwidth than
Initials 3
3G system from channel sizes down to 5MHz in increments of 1.5MHz[9]. Another
import feature of LTE is spectrum efficieny is improved which allows more data can be
transmitted in a given bandwidth[10]. IEEE 802.11p is a specific wifi standard using
5.9GHz frequency band. Compare to LTE it can provide direct communication with
very small latency for direct communication among nodes but lack of spatial coverage.
Consider the high data rates, capability and reliability LTE naturally becomes a better
choice for vehicular communication. Another advantage of LTE is the network infras-
tructure. LTE base stations(eNodeB) normally are located in much higher places, this
is helpful for our case by avoiding multi-path loss caused by vertical obstacles.
Every coin has two sides, on the other hand LTE network is suffering highly loaded
background traffic which may lead to some delay sensitive communications miss their
deadlines. Vehicle teleoperation requires a very responsive network between the control
room and the vehicle, the latency shall be in the range of tens of milliseconds to make
sure safety. Furthermore, the video streaming service from the camera in the vehicle
also put a high quality-of-service requirement[11]. LTE lacks of scheduling mechanism in
MAC layer for the QoS Class Identifier(QCI). Detailed descripition about LTE protocol
can be found in my co-worker Y.Jin’s thesis[12].
Before deploying system to the real world, simulation is essential so potential errors
to the system can be detected before implementation thus the system is predictable
and reliable. Only simulating the communication network is not complete to evaluate
the quality, attention to the dynamic vehicle mobility pattern needs to be paid such
as topology of road network and/or the speed change due to specific constrains. Our
research problem focuses on developing a real-time simulation testbed using LTE cellular
network and a microscopic urban mobility model, to evaluate the realization of different
QoS requirements during the teleoperation.
1.3 Thesis Objectives
In a recent publication[11], the authors evaluate the feasibility of vehicle teleoperation
using cellular network in an urban environment but it could not deal with network
changes based on the traffic information. Compare to their setup our improvement
will be including a unique microscopic traffic mobility model in order to simulate a
realistic traffic pattern so the result will be more accurate. The vehicle mobility will
be a combination of a deterministic model based on the city daily routine and a simple
stochastic model to cover the randomness. The traffic and network module will be
integrated together to evaluate the feasibility, performance and capacity. The whole
Initials 4
project is done by two master students one focus on the Network Simulator for full-layer
LTE and the other focus on the traffic simulator for microscopic mobility scenario. This
thesis is focused on the latter. The objectives of the thesis are the following:
• Do a state-of-art research and evaluate different traffic simulators to determine the
suitability in the simulation framework.
• Build a microscopic realistic traffic scenario of a given urban area. Our choice is
Kista, Stockholm.
• Integrate the developed traffic scenario with the LTE network scenario. The traffic
scenario will generate nodes following certain mobility models which are defined
by the real rode map topology, traffic parameters such as speed limit, lane changes
and vehicle density and so on. The integrated traffic and network simulator should
be in an closed feedback loop: Network simulator calculate the network using the
data(position, routes, speed, etc) provided from mobility model in traffic simulator,
the network simulator could give feedback to traffic simulator to change the vehicle
routes.
• Evaluate the network requirements for teleoperated vehicles in our integrated
testbed.
• Evaluate the performance of our integrated testbed for teleoperated vehicles.
• Evaluate the capacity of our integrated testbed for teleoperated vehicles.
Figure 1.3: Illustration of Simulation Framework
Chapter 2
Background
2.1 Traffic Modelling
Mobility model is a set of rules that regulate how the nodes(participants in the traffic)
will be generated and their moving pattern. Using a simple random generated mobility
model is a common practise for V2X research community in earlier years but obviously
such models is not accurate enough because they are ironing many microscopic traffic
factors such as car acceleration/braking, queuing in the intersection, which could affect
network performance greatly. The mobility model should be as realistic as possible to
describe the behavior of the traffic flow to make sure the network evaluation is accurate.
2.1.1 State of Art of Mobility Model
Mobility models are normally divided into either microscopic or macroscopic sometimes
mesoscopic perspectives. Macroscopic models focus on traffic flows and road topology
including constraints could affect the traffic movement such as traffic lights signals. It’s
mainly used by traffic engineers for traffic planning or road topology design. There are
also some research articles about mobility patterns in wireless cellular network[13][14] on
macroscopic scope. Mesoscopic model study the elements in small groups and another
microscopic model with focus on behaviour of individual member of traffic such as their
position , acceleration/deceleration, lane changing and so on. Since we want to build
a realistic scenario and study the communication performance, a microscopic mobility
model will be needed and suitable.
Jerome Harri et al[15] defined that ”a realistic mobility model should include : Accurate
and realistic topological maps;smooth deceleration and acceleration;obstacles; attraction
5
Initials 6
Figure 2.1: Level of mobility model[2]
points;simulation time;non-random distribution of vehicles and intelligent driving pat-
terns”. All these criteria seems relevant to our scenario and will be considered during
the implementation.
The approach to build mobility model varies. Synthetic model is the first and major
model that was used in the early studies. Researchers use mathematical models such as
stochastic models contain purely random motions, traffic stream models which vehicular
mobility as hydrodynamic phenomenon, car following modes that each driver is modeled
based on the vehicles ahead and furthermore queue models, behavior models, etc.[16].
But in reality the behaviour of the human drivers are not always following a single
pattern so a model is only meaningful for a specific behaviour study. Another approach
trace-based model is instead of using complex mathematics models and validating them,
but extracting generic mobility patter from movement traces. An interesting example
would be the mobility model from Tuduce and Gross[17]. They use real wireless LAN
data at ETH Zurich in Switzerland. They divide the area into squares and derive
the transition probabilities of adjacent squares by data from access points and their
analysis shows that ”the WLAN modes maps the real world mobility characteristics to
the abstract world of network simulators with a very small error”[18]. A major drawback
of this model is the data type. For example the motion trace observed for campus can
not be applied to streets so application is limited.
Initials 7
Figure 2.2: General concept map of mobility model generation[3]
We need to consider several microscopic criteria to set up the rules regarding how the cars
following other vehicles, how they overtaking and how they behave in the intersections.
They are all important for generate realistic scenario and have several widely used models
available. For example Krauss Model[19] for car following and Gipps Model[20] for Lane
Changing. After years of development, advanced traffic simulators start to be used such
as The Simulation of Urban MObility(SUMO) [21] and VISSIM(Verkehr In Stadten
SIMulationsmodell)[22] and they can be used to simulate urban mobility in microscopic
scope which is exactly what we need.
2.1.2 Comparison of Traffic Simulator
There are several traffic simulators available today to model vehicular mobility at macro-
scopic or/and microscopic level and each of them has advantages and disadvantages. The
most suitable one need to fulfill several simulator related criteria such as capability for
map formats, algorithms for intersection and routing, visualization for output etc. We
do not have hard requirements for all these criteria but some pre-defined criteria must
be focused for the selection. The traffic simulator for our project should be open-source,
support at least microscopic type, can handle multiple vehicles, has resource cheap GUI,
also suitable for integration with a network simulator.
The CitySim [23] simulator could simulate stochastic queue-based traffic flow in urban
area for city planning purpose with a 3D graphical user interface. It could be used
by authority to simulate comprehensive urban system to evaluate the traffic volumes,
emissions, atmospheric dispersion and noise within cities. It is open-source but it is not
a mainstream traffic simulator so there is not a big community behind it for us to seek
support and it’s not designed for integration with a network simulator.
The CARISMA [24] traffic simulator supports both microscopic and macroscopic models
and has an interesting feature which is support for real-time V2V communication. It
Initials 8
could exchange and synchronize data with a network simulator continuously using a TCP
connection. The limitation is the lane connection management in complex junction and
we have to pay to use the program.
The CANU Mobility Simulation Environment(CanuMobiSim) [25], is a flexible frame-
work for user mobility modeling in a variety of conditions. CanuMobiSim supports
maps in the Geographical Data Files (GDF) format and provides implementations of
several random mobility models. It features at both macroscopic and microscopic levels.
CanuMobiSim has an extension VANETMOBISIM. At macroscopic level, VanetMobiSim
support for multi-lane roads, separate directional flows, differentiated speed constraints
and traffic signs at intersections. At microscopic level, VanetMobiSim provides realistic
V2V and V2I interactions such as Fluid Traffic Model[26]. Moreover it can generate
trace files for network simulator but it’s not possible to control on-going simulation.
SUMO is an open source, highly portable, microscopic road traffic simulation package
designed to handle large road networks, developed by German Aerospace Center(DLR).
SUMO uses Krauss’s car following model [27] and Krajzewicz’s lane-changing model[28],
includes a stochastic traffic assignment by a probabilistic route choice. It can import
road map for large area in many digital formats for example OSM. We could configure
the routes by shortest-path or weighted-path. In microscopic view the detailed behavior
of every individual vehicle such as decreasing velocity before entering into junction could
have impact to the network performance. The road geometry we modelled is Kista center
area with higher complexity however SUMO is capable to simulate the realistic lane-
to-lane movement. Another import factor is that SUMO contains a parser to generate
traces file could be read by a network simulator and we can also control the vehicle during
the simulation in real-time. Accordingly, we can run network and traffic simulator
simultaneously and send instructions from networks simulator to SUMO using some
middle-ware. SUMO also have a efficient 2D GUI so we can present the result. However
there is one limitation when considering the integration with the network simulation:
The map in SUMO is in two-dimensional therefore it doesn’t take heights of the radio
obstacles into account. Luckily in the network simulator selected in this project we
could implement a fading model and calculate the realistic propagation path loss with
consideration of obstacles’ heights. SUMO is used by many research institutions so we
can have necessary supports from them. Consider all the criteria required we conclude
that SUMO is the best option for our simulation framework.
Initials 9
Table 2.1: Features of Main Stream Traffic Simulators-1
Open Source Microscopic Marcoscopic Map Captiable Multi Lane
VanetMobisim yes yes yes high yes
VISSIM[29] no yes yes high yes
TRANSIMS no yes yes high yes
BonnMotion[30] yes no yes high yes
SUMO yes yes yes high yes
CARISMA unknown yes yes high yes
MobiREAL[24] yes no yes high yes
CitySim yes no yes high yes
PARAMICS[31] no yes yes high yes
Table 2.2: Features of Main Stream Traffic Simulators-2
Source/Destination Position Path Computation NS support
VanetMobisim random, attraction Point density/Dijkstra/speed yes
VISSIM unknown unknown yes(cpp API)
TRANSIMS random random no
BonnMotion random, attraction point density/Dijkstra/A*[32]/user define no
SUMO random, attraction point density/Dijkstra/ yes(TraCI[33])
CARISMA random density/Dijkstra/A*/user define yes
MobiREAL random, attraction point random no
CitySim unknown unknown no
PARAMICS unknown unkown no
Table 2.3: Features of Main Stream Traffic Simulators-3
GUI Radio Obstacles Plattform
VanetMobisim yes yes java
VISSIM yes unknown c++
TRANSIMS yes random unknown
BonnMotion yes yes java
SUMO yes yes c++
CARISMA yes yes c++
MobiREAL yes no c++
CitySim yes unknown c++
PARAMICS yes unknown unknown
Initials 10
2.2 Vehicles Routing Algorithms
Routes planning is always an important factor for traffic planning and optimization. How
to reach destination from source point varies in traffic dynamics(road congestion/inci-
dents can happen) and the decision will affect simulation significantly. There are several
routing algorithm existing for example Genetic Algorithms[34], Tabu search[35], A* al-
gorithm and so on. The general process of how we implement dynamic traffic route
routing is as following:
Algorithm 1 Dynamic Traffic Routing
Input: input parameters Start Position, End Position, weight: travel time/speific
routes
Output: Routes to destination
Calculate the initial route according from origin to destination using chosen routing
algorithm.
for each intersection do
if destination reached then
Trip finished
else
Check traffic condition
if Incident happen and will affect the further path then
Remove the old routing table and re-calculate the new routes
else
Follow the existing routing table
return Routes to destination
2.2.1 Heuristics in Routing: A* algorithm
We use heuristics path finding to calculate the shortest path. We did the calculation
iterativly until converges to an equilibrium state. The key to calculate the path is the
equation below:
f (n) = g (n) + h (n) (2.1)
where g(n) is is the total distance and h(n) is the heuristic distance to the destination
node, n is the starting point. When we start the search, we check the nodes adjacent to
n and expand around until reach the destination. From n we create an open list, in the
beginning we only have one node in the list which is n itself, later more nodes will be
added, some nodes is in the path we are looking for, some are not and will be checked
later. The nodes around n will be checked first and the reachable ones will be added to
Initials 11
the open list and n will become their parent node. Then n will be removed from the
open list to the close list. Next step we need to chose one node from the open list, the
one with smallest f(n) value will be chosen and then go through the same process until
the destination is reached. However we can not guarantee this will always give us the
shortest path as it use heuristics to sort the queue in the open list. Normally there are
three heuristics to apply depending on the moving rules:
• Manhattan Distance
h (n) = abs (cur n.x− dest node.x) + abs (cur n.y − dest node.y) (2.2)
This is based on the street geography of Manhattan in New York City. We use
this when that allows 4 directions of movement.
• Diagonal Distance
h (n) = max {abs (cur n.x− dest.node) + abs (current n.y − dest.n)} (2.3)
We use this when that allows 8 directions of movement.
• Euclidean Distance
h (n) = sqrt (abs (cur node.x− goal.x) + abs (cur node.y − dest.y)) (2.4)
We use this when that allows any direction of movement.
Figure 2.3: Example of Manhattan, Diagonal and Euclidean distance
Furthermore we can give weights attributes during routing. A normal choice is use travel
time on each path the vehicle went through.
Initials 12
The algorithm we followed is called A* algorithm and is one of the best and popular
algorithm used in path-finding problems. If we do not use heuristic in path finding and
consider h always zero then this algorithm is also called Dijkstra algorithm[36]. SUMO
support real time interaction with the vehicle routes during the simulation through
TRACI interface and we implemented A* algorithm for path finding. The simulation
result shows it’s possible to change vehicle routes in real-time and that improved traffic
efficiency.
Chapter 3
Implementation
Now we have discussed and chose SUMO as the traffic simulator to build our microscopic
scenario for our testbed. In order to solve our research problems there are certain
requirements need to be fulfilled: First it need to describe an city area including different
road categories but not exceed the coverage of one LTE eNodeB. Second it need to
support complex mobility modelling to represent traffic-follow in urban area. Third
it should guarantee reality in microscopic level. Fourth it should be compatible with
another network simulator for integration. Also the scenario should have the capability
to re-use in some other user cases by the other research community.
3.1 Build Kista Scenario
3.1.1 Topology
The topology of Kista is a typical city urban area: a town center area surrounded by
several neighbourhoods which are divided in different functionalists (Residential/Com-
mercial/Occupation). In SUMO the road network is modeled using XML files. They
can be written by hand, define map primitives such as nodes and edges and then con-
nect them together, or the data can be imported from some other sources like Open-
StreetMap(OSM) directly. OSM is a free user-edit project that provided map over the
whole world. ”OpenStreetMap creates and provides free geographic data such as street
maps to anyone who wants them. The project was started because most maps you think
of as free actually have legal or technical restrictions on their use, holding back people
from using them in creative, productive, or unexpected ways.”[37]. We choose OSM not
only because it can provide road information but OSM also gathers information such as
building shapes, parking-lot,bus stops and many other Point of Interests(POI). These
13
Initials 14
information will be needed for generating traffic base on people’s daily activities. Lo-
cations and the geometries of some POIs can be used as obstacles in wireless signal
prorogation which is crucial concerning to our research problem.
Figure 3.1: Select large area from OSM
The map we downloaded from OSM contains more information than we need. So we
must manually select what we are interested in such as bus stops and work place locations
which can affect the traffic. We do this by using an open source OSM map editor Java
OpenStreetMap(JOSM)[38]. It allows to load and edit OSM data such as nodes, ways
as well as their relations and metadata tags.
Figure 3.2: JOSM view of Kista OSM file
From Figure 3.2 we can see that the original OSM file directly downloaded contains too
many nodes which made the map even difficult for us to observe. It is necessary to clean
up the redundant objects. JOSM[38] provides a command line tool osmfilter for that
propurse. We can filter out objects with specific tags using command for example:
./ osmfilter a.o5m --drop -tags=" oneway= name=" -o=plain_ways.o5m
Initials 15
Figure 3.3: After Filtering Out Irrel-evant Nodes
Figure 3.4: After Filtering Out Irrel-evant Nodes and Buildings
Now the map in JOSM is clear for us to work. The topology data is actually sufficient
for SUMO to run the simulation but the correctness and robustness is not guaranteed
so we need to manually check the following to enhance the map:
• All the streets have tag ”highway” which will be used by SUMO to set the speed
limit and the right of the lane.
• One-way roads are correctly defined.
• Number of lanes and connections in the intersections.
• Warnings from Netconvert for a smoonth simulation for example adjust the angle
of the roads.
It’s an iterative process between JOSM and NetEdit to go through all the steps men-
tationed above. Now after that the road map is good enough for SUMO to run the
simulation without any significant errors that can affect the traffic flows.
Figure 3.5: Kista Scenario Road Network
The figure above shows the road map of Kista Scenario covers about 2.13 km2 in urban
area. SUMO has its own network format for the map (.net.xml) so we need to convert
OSM file into SUMO network file to make SUMO recognize the map.Table 3.1 shows a
brief summary of the selected map topology.
Initials 16
Table 3.1: Topology Information
Total number of nodes 1190
Total number of edges 1123
Total edge length [km] 103.11
Total lane length [km] 122.99
The geometrical objects have been kept tract in the previous steps now we need to
implement them into our map as colored polygons in order for us to use and easier to
see through SUMO-GUI.
A vaild polygon in SUMO is described like this:
<poly id="<POLYGON_ID >" \\
type="<TYPENAME >" \\
color="<COLOR >" \\
fill="<FILL_OPTION >" \\
layer="<LAYER_NO >" \\
shape="<2D-POSITION >[ <2D-POSITION >]*"/ >
We use POLYCONVERT to convert the buildings and parking-lots from OSM file to
SUMO polygon file(.poly.xml). The polygons are stored as an upper layer compared to
the network file.
Table 3.2: Number of Polygons in the Kista Scenario
Buildings 242
Parking 22
Total 264
Figure 3.6: Kista Scenario Topology with Polygons
Initials 17
Figure 3.7: Kista Fading Map
Figure 3.2 summariz the general process to generate a SUMO map file.
Figure 3.8: Overview of steps to generate SUMO map file[4]
3.1.2 Mobility
Now we have built the network and could load it into SUMO but there will be no vehicle
running. In order to have a realistic microscopic traffic scenario the ACTIVITYGEN
tool[39] is used. It generate the traffic demand based on a population description. It
takes into account activities such as work, school and free-time. Transportation tool
can be feet, bus and cars. To be noticed that there are some traffic not covered by
Activitygen such as tourist from outside so we add 20% random traffic following uniform
distribution.
We use data published in the Internet by Swedish authority organization Stockholm
Kommun[40] and Tranfik Analys[33]. Number of vehicles running in the area selected
is estimated to approximately between 200,000 and 300,000 per day based on the pop-
ulation. 38 bus stops are added to the map as well as 11 bus routes. We extract the
location of bus stops and bus routes in OSM map. There is no tool available now to
automatly convert them into SUMO formate so we have to add them manually.
Initials 18
Figure 3.9: Bus Stop Location in Kista Scenario
Figure 3.10: Example Bus Route No.178 from OSM
Table 3.3: Summary of Bus in the Kista Scenario
Total Number of Bus Lines 11
Total Number of Bus Stops 38
Total Number of Buses 594
ACTIVITYGEN will generate staring and destination edges then we use duaRouter[41]
to complete all the edges in between. Moreover , we use a dynamic routing mechanism
for the vehicles to avoid road congestion. That means a vehicle will re-route using
Dijkstra algorithm every 5 minutes. When SUMO simulation is over we could see from
the summary that the dynamic routing mechanism can provide a smooth traffic flow.
3.2 Integration
Now the Kista mobility scenario is ready for the simulation and can be used together
with a network simulator such as NS3 or OMNet++ so we can evaluate the network per-
formance. There are two types of integration: The traditional way to do the integration
Initials 19
is first let SUMO produce a complete trace file and then let network simulator read the
trace file however not possible to control the traffic flow, we call this type of simulation
off-line simulation. The other type is real-time/on-line simulation which allows interac-
tion between SUMO and network simulator during the simulation so the most accurate
and efficient result could be given.
3.2.1 Real-time Simulation
In our testbed SUMO provides vehicle mobility model including nodes positions, speed
and routes as well as geometry road map. NS3 provides LTE cellular network model
and signal fading model during propagation. V2X simulation need to be realistic in
microscopic level(provide accurate information) and run in real-time to be useful(for
example during simulation re-routing if traffic congestion occurs or signal is weak). If
we had a feed-back loop in a live simulation we refer that as the real-time simulation.
The interface in SUMO for external communication TraCI(Traffic Control Interface)
is included in SUMO package. It gives access and pass the data between SUMO and
external applications. We can retrieve the current values of all objects in the simulation
such as vehicles or traffic light, also we can change the state of these objects on-line.
One example user case could be when we want to simulate a bus broke-down half way,
we could change the bus destination back to the bus depot during the simulation. TraCI
uses TCP connection with SUMO where SUMO acts as the server and listing to the
remote port for the coming connections. The figure below describes a basic flow of
establishing a connection to SUMO.
Figure 3.11: Establishing a connection to SUMO[5]
Initials 20
When the client connects to SUMO by a TCP connection through the configured port
it takes control of SUMO. The client needs to send commands to SUMO in every simu-
lation step and the objects in SUMO can be changed according to the instruction sent
from client. SUMO will also send corresponding acknowledge message as well as some
additional result back to the client.
Figure 3.12: Closing a connection to SUMO[5]
The connection can only be shut down by the client by sending a close command.
In order to have on-line interaction between SUMO and NS3 we need to use some middle-
ware for example The V2X Simulation Runtime Infrastructure (VSimRTI) [6]. IEEE
Standard for Modeling and Simulation(MS) High Level Architecture[42] describes stan-
dardized approach to combine simulators. The idea is to have a central management
component to receive, process and send data between different connecting simulators.
We need a such central infrastructure to manage the arbitrary simulators through some
interfaces. There are several projects can provide such kind of service[43]. VisimRTI
uses the concept of ambassador, inspired from HLA, together with each participating
simulators as a federate to build a Federation. Figure 3.13 illustrate the general VSim-
RTI architecture. Simulation run time infrastructure(RTI) is the essential part uses
ambassadors to communicate between SUMO and NS3. The communication is bidirec-
tional. Except interaction run time infrastructure also provides service for simulator
synchronization by a time management service. In RTI three modules are included:
Federation, time and interaction management. To attach a simulator federate we only
need to configure the federate ambassador interface to RTI. Even though each simulator
is independent but during the simulation all the events has time stamp regardless local
or remote. The reason is for synchronization. It enables the substitution of the most
relevant simulators for a realistic presentation of vehicle traffic, wireless communication
and the execution of V2X application. VsimRTI couples different simulators scenarios
Initials 21
and synchronizes them. Also several visualization and analysis tools are prepared for
VSimRTI.
Figure 3.13: VSimRTI architecture[6]
Yet another middle-ware framework project available is Online Vehicular Network Inte-
grated Simulation(OVNIS)[44] platform. It can couple together SUMO and NS3 so the
SUMO scenario is embedded in the mobility model of NS3. Accordingly corresponding
NS3 module could influence the traffic simulation such as reroute of the vehicles in the
mobility model. OVANIS achieved this by using TraCI interfce to create a sub-module
in NS3.
However when we try to implement the integration, we found that both OVNIS and
VSimRTI have only implemented 802.11p protocol stack not support LTE and they do
not consider fading pattern in the urban area neither for the simulation which violated
our research problem.
Figure 3.14: Illustration of real-time simulation work flow between SUMO and NS3
Initials 22
3.2.2 Off-line Simulation
Traffic simulators and network simulators are designed separately so they should be
controlled separately and can not be used directly together. Most network simulators
are able to generate random movements themselves but now it is possible to load mobility
scenario from traffic simulators so the vehicular network evaluation can be more accurate.
Figure 3.15: Illustration of off-time simulation work flow between SUMO and NS2[2]
For example CanuMobiSim[45] is a generic user mobility simulator which can provide
mobility scenarios in different infrastructure. The simulation need to run prior to the
network simulation and need to be parsed to a pre-defined trace format. During the
network simulation no changes to the mobility is allowed. A recent article (2015) by
Vladi.k, Tetsuya Oda and some other authors did a performance evaluation using a
simulation structure described above with SUMO and NS3. SUMO generates the move-
ment pattern of the vehicles which used by NS3 to analysis the performance of different
routing protocols[46].
Chapter 4
Evaluation
For evaluation of the integrated scenario, we focus on three aspects: Feasibility, Capacity
and Performance. When doing feasibility test, we calculate the delay of data flows with
different priority in one fixed route. In order to find the maximum number of busses our
scenario can handle we did capacity test. We sum up the simulation results from SUMO
and analyze. The performance of LTE network can be found in Y.Jin’s publication[12].
4.1 Simulation Summary of SUMO
A moving vehicle will be sent to teleport if some issues such as congestion occur that
stop the vehicle continue to the destination. In the end of simulation we could trace
the simulation detail though log files. Higher percentage of teleports indicate that the
traffic flow can not be moving smoothly. In our scenario it’s only around 3% so the
unexpected issues in the traffic scenario will not have significant influence to the network
performance. We achieve this by a dynamic re-routing mechanism. If a traffic congestion
occurs the vehicles will be automatically re-routed to avoid the congestion situation. This
is not always true in reality so we only configured 70% vehicles with re-routing set-up
and the rest 30% will stick to their routes no matter the traffic condition. We assume
this will not affect the realism of the scenario.
Table 4.1: SUMO Simulation Summary
Total simlation time [s] 86494
Number of vehicles loaded 25038
Teleports 921
23
Initials 24
To verify that the Kista Scenario can behave in a realistic way we use an external tool:
Typical Traffic from Google Maps[47].
Figure 4.1: Traffic Condition in Google Maps at 8:00 am
The green and red color represent the speed of the vehicle driving on that road from
normal speed to low speed. We compare the running simulation and the snapshots from
the typical traffic and the traffic condition is consistent so we can assume that the reality
of Kista scenario can be guaranteed.
4.2 Feasibility
The network requirements we have for teloperation are constant latency less than 50ms,
constant jitter, more than 99% connectivity guarantee with throughput reached GBR.
Figure 4.2: Delay for Critical Stream in 10M Background Traffic
Initials 25
We set priorities to different data flow types. Traffic Command flow has higher priority
than video flow so the traffic command flow(start from 10s) will affect the video trans-
mission. We could observe from figure 4.1 the video flow delay have more fluctuating
which is due to command packet, as well as channel fading and codes performance. But
it still can guarantee a delay less than 50ms and relatively small jitter. We send some
other dummy packets with lowest priority (start from 0.01s) and it has no affect for the
transmission of video flow. CQI also can cause fluctuation in delay as the time passing
by.
Figure 4.3: Packet Loss Ratio for Critical stream in 10M and 1M Background Traffic
It is important that traffic command should be always received by the vehicles in tel-
operation. So we need to make sure the command flow will never lose connectivity.
Observing figure 4.2 we could No matter how we change the background traffic through-
put, the video flow which have higher priority is not affected, only dummy background
traffic lost more packet itself. This is guaranteed by assigning GBR-bearer to the flow
with higher priority and NGBR-bearer to the flow with lower priority. NGBR-bearer
will only be served until GBR-bearer is fully served.
From perspective of the mobility, we have transmitting antenna located at (880,4100)
and moving vehicle has the route from (188,35,4454,56) to (968.36, 4094,65). The vehicle
will be closer to the antenna when approaching destination, which will result in more
guaranteed connectivity, optimized CQI and decreased PLR.
Figure 4.4: Geometry location of antenna and vehicle route
Initials 26
Figure 4.5: Packet Loss Ratio for non-Critical stream in 10M and 1M Back-groundTraffic
Since this is the feasibility test so we only simulated one bus route.
4.3 Capacity
The capacity of the scenario, which is maximum number of busses that the testbed can
handle without violate network requirements is another parameter we need to investi-
gate. We have one eNB in the network scenario and test different number of busses in
traffic scenario.
Figure 4.6: Capacity for Bus Router per eNB
We simulate up to nine buses in nine realistic bus routes in Kista area. ”Always Yes”
admission control strategy is implemented which means all UEs we put in the system
will be subscribed and served. After some test simulation runs we conclude that if we
have more than six busses the entire network will be ruined. Which packets will be drop
is defined by CQA scheduler mechanism. If we have several data flows with the same
pre-defined priority, the one with better CQI and shorter Head-of-line(HOL) delay[48]
will be given higher priority. So the system capacity will be decided by the UEs with
worst HOL.
Chapter 5
Conclusion and Future Work
As we stated the research problems in the beginning of the thesis: simulate and evaluate
the teleoperation of the vehicles using LTE cellular network. We set several objectives to
achieve that in a proper scientific research manner: The first is to do a state-of-art study
and choose the appropriate simulators. To achieve that we conduct a comparison based
on the study of exist research results and choose SUMO to build our traffic scenario.
The second object is to build the traffic scenario for Kista. The scenario we build has
following characters: It is built on real geographical topoogy and included different
areas such as residential and work place, city road and high way. It supports different
mobility modellings (both stochastic and deterministic). The whole scenario simulates
24 hours traffic movements. It is compatible with other traffic simulator for integration.
We emphasize the importance of the mobility modeling in wireless networking. Not
only the movements in geographical map is considered but also we take case as well the
importance of simulation accuracy by generating the movements under daily activities
distribution based on real statistic data. The presented scenario is repeatable that can
be easily applied to other user case for example in another city urban area, or to evaluate
with other wireless network protocol like IEEE 802.11p. Next object is to integrate the
traffic scenario together with network scenario build by my co-worker Yifei JIn. This is
the weakness of the project. We didn’t succeed in implementing the real-time simulation
due to the only suitable integration framework VSimRTI is not open-source and it has
not implemented complete LTE protocol stack so we can not configure and extend that
for our integrated scenario. To implement a new integration frame work would be beyond
the scope of this thesis project so we are forced to simulate the integrated scenario off-
line. However we make use of the emulation features from NS3 to extend the network
topology to remote host and elevate this to have user access network topology in remote
end. A good compensation we could do is to run the simulation many times in small
time scale so we have results very similar to real-time simulation behavior.
27
Initials 28
So far we have discussed the process of building a V2X simulation testbed framework
which is integrated by an accurate LTE network scenario and a realistic microscopic
traffic scenario. The simulation results show that it’s feasible by using LTE network
infrastructure to teleoperate up tp 5 vehicles per eNB with guaranteed QoS in Kista.
However, as mentioned above we have many assumptions, uncertainties and even weak-
nesses.
Future work will focus on developing the real-time simulation testbed and use it to
evaluate more scenarios, in which we could even develop an external control software
to send control command from internet to the LTE network. And we could extend
the scenario and consider some further user cases for example to develop a machine
intelligence traffic control system.
Bibliography
[1] Roberto Baldessari, Bert Bodekker, Matthias Deegener, Andreas Festag, Walter
Franz, C Christopher Kellum, Timo Kosch, Andras Kovacs, Massimiliano Lenardi,
Cornelius Menig, et al. Car-2-car communication consortium-manifesto. 2007.
[2] Matthias Rockl. itetris: The integrated wireless and traffic platform for real-time
road traffic management solutions. 2009.
[3] Marco Fiore, Fethi Filali, and Christian Bonnet. Vehicular mobility simulation with
vanetmobisim. Simulation, 87(4):275–300, 2011.
[4] http://sumo.dlr.de/wiki/Tutorials/Import_from_OpenStreetMap. Accessed:
2017-05-29.
[5] http://sumo.dlr.de/wiki/TraCI/Protocol. Accessed: 2017-05-29.
[6] https://www.dcaiti.tu-berlin.de/research/simulation/. Accessed: 2017-
05-29.
[7] Lte An Introduction. Lte – an introduction. 2014.
[8] Diego Llusia, Rafael Marquez, Juan Francisco Beltran, Catarina Moreira, and Jose
Pedro Do Amaral. Ieee 802.11p: Towards an international standard for wireless
access in vehicular environments. In Vehicular Technology Conference, 2008. Vtc
Spring, pages 2036–2040, 2008.
[9] Amit Kumar, Jyotsna Sengupta, and Yun Fei Liu. 3GPP LTE: The Future of
Mobile Broadband. Kluwer Academic Publishers, 2012.
[10] David Astely, Erik Dahlman, Anders Furuskar, Ylva Jading, Magnus Lindstrom,
and Stefan Parkvall. Lte: the evolution of mobile broadband. IEEE Communica-
tions magazine, 47(4), 2009.
[11] Rafia Inam, Nicolas Schrammar, Keven Wang, Athanasios Karapantelakis, Leonid
Mokrushin, Aneta Vulgarakis Feljan, and Elena Fersman. Feasibility assessment to
realise vehicle teleoperation using cellular networks. In Intelligent Transportation
29
Bibliography 30
Systems (ITSC), 2016 IEEE 19th International Conference on, pages 2254–2260.
IEEE, 2016.
[12] year=In progress Yifei Jin. Master thesis: Cellular network overlay evaluation for
vehicular teleoperation in an urban scenario.
[13] Derek Lam, Donald C Cox, and Jennifer Widom. Teletraffic modeling for personal
communications services. IEEE communications magazine, 35(2):79–87, 1997.
[14] John G Markoulidakis, George L Lyberopoulos, Dimitrios F Tsirkas, and Efs-
tathios D Sykas. Mobility modeling in third-generation mobile telecommunications
systems. IEEE personal communications, 4(4):41–56, 1997.
[15] Jerome Harri, Fethi Filali, and Christian Bonnet. Mobility models for vehicular ad
hoc networks: a survey and taxonomy. IEEE Communications Surveys & Tutorials,
11(4), 2009.
[16] Marco Fiore et al. Mobility models in inter-vehicle communications literature.
Politecnico di Torino, page 147, 2006.
[17] Cristian Tuduce and Thomas Gross. A mobility model based on wlan traces and
its validation. In INFOCOM 2005. 24th Annual Joint Conference of the IEEE
Computer and Communications Societies. Proceedings IEEE, volume 1, pages 664–
674. IEEE, 2005.
[18] Cristian Tuduce and Thomas Gross. A mobility model based on wlan traces and
its validation. In INFOCOM 2005. 24th Annual Joint Conference of the IEEE
Computer and Communications Societies. Proceedings IEEE, volume 1, pages 664–
674. IEEE, 2005.
[19] Venkatesan Kanagaraj, Gowri Asaithambi, C H Naveen Kumar, Karthik K Srini-
vasan, and R. Sivanandan. Evaluation of different vehicle following models under
mixed traffic conditions. In Conference of Transportation Research Group of India,
pages 390–401, 2013.
[20] Ioanna Spyropoulou. Simulation using gipps’ car-following model—an in-depth
analysis. Transportmetrica, 3(3):231–245, 2007.
[21] http://sumo.dlr.de/wiki/Simulation_of_Urban_MObility_-_Wiki. Accessed:
2017-05-29.
[22] http://vision-traffic.ptvgroup.com/en-uk/products/ptv-vissim/. Ac-
cessed: 2017-05-29.
Bibliography 31
[23] Darren Robinson, Frederic Haldi, Philippe Leroux, Diane Perez, Adil Rasheed,
and Urs Wilke. Citysim: Comprehensive micro-simulation of resource flows for
sustainable urban planning. In Building Simulation 2009: Eleventh International
IBPSA Conference, 2009.
[24] Stephan Eichler, Benedikt Ostermaier, Christoph Schroth, and Timo Kosch. Sim-
ulation of car-to-car messaging: Analyzing the impact on road traffic. In IEEE
International Symposium on Modeling, Analysis, and Simulation of Computer and
Telecommunication Systems, pages 507–510, 2005.
[25] http://vanet.eurecom.fr/. Accessed: 2017-05-29.
[26] Gabriella Bretti, Roberto Natalini, and Benedetto Piccoli. A fluid-dynamic traf-
fic model on road networks. Archives of Computational Methods in Engineering,
14(2):139–172, 2007.
[27] S. Krauss, P. Wagner, and C. Gawron. Metastable states in a microscopic model
of traffic flow. Phys.rev.e, 5(1997):5597–5602, 1997.
[28] Daniel Krajzewicz. Kombination von taktischen und strategischen einflussen in
einer mikroskopischen verkehrsflusssimulation. Fahrermodellierung in Wissenschaft
Und Wirtschaft Berliner Fachtagung Fur Fahrermodellierung, pages 104–115, 2009.
[29] http://vision-traffic.ptvgroup.com/en-us/products/ptv-vissim/. Ac-
cessed: 2017-10-09.
[30] http://sys.cs.uos.de/bonnmotion/. Accessed: 2017-10-09.
[31] http://www.sias.com/2013/sp/sparamicshome.htm. Accessed: 2017-10-09.
[32] Florian Diedrich, Klaus Jansen, Ulrich M Schwarz, and Denis Trystram. A sur-
vey on approximation algorithms for scheduling with machine unavailability. In
Algorithmics of Large Complex Networks, pages 50–64, 2013.
[33] http://www.trafa.se/. Accessed: 2017-05-29.
[34] A. E. Eiben, Paul Erik Raue, and Zsofia Ruttkay. Genetic algorithms with multi-
parent recombination. In Parallel Problem Solving from Nature - PPSN III, Interna-
tional Conference on Evolutionary Computation. The Third Conference on Parallel
Problem Solving from Nature, Jerusalem, Israel, October 9-14, 1994, Proceedings,
pages 78–87, 1994.
[35] Fred Glover. Future paths for integer programming and links to ar tifi cial intelli g
en ce. Computers Operations Research, 13(5):533–549, 1986.
Bibliography 32
[36] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein. Section 24.3: Dijkstra’s
algorithm. In Introduction To Algorithms, Second Edition, 2013.
[37] http://www.openstreetmap.org. Accessed: 2017-05-05.
[38] https://josm.openstreetmap.de/. Accessed: 2017-05-29.
[39] http://sumo.dlr.de/wiki/ACTIVITYGEN. Accessed: 2017-05-29.
[40] http://statistik.stockholm.se/omradesfakta/pdf/SDO01_SVE.pdf. Ac-
cessed: 2017-05-29.
[41] http://sumo.dlr.de/wiki/DUAROUTER. Accessed: 2017-06-09.
[42] Judith S Dahmann. High level architecture for simulation. In Distributed Interactive
Simulation and Real Time Applications, 1997., First International Workshop on,
pages 9–14. IEEE, 1997.
[43] https://www.nsnam.org/wiki/GSOC2012HLA. Accessed: 2017-05-29.
[44] https://github.com/agacia/ovnis. Accessed: 2017-05-29.
[45] http://canu.informatik.uni-stuttgart.de/mobisim/. Accessed: 2017-05-05.
[46] Vladi Kolici, Tetsuya Oda, Yuki Sugihara, Evjola Spaho, Makoto Ikeda, and
Leonard Barolli. Performance evaluation of a vanet simulation system using ns-
3 and sumo considering number of vehicles and crossroad scenario. In Innovative
Mobile and Internet Services in Ubiquitous Computing (IMIS), 2015 9th Interna-
tional Conference on, pages 22–27. IEEE, 2015.
[47] https://maps.googleblog.com/2012/04/get-typical-traffic-for-roads-not-just.
html. Accessed: 2017-09-07.
[48] M. J Karol, M. G Hluchyj, and S. P Morgan. Input versus output queueing on a
space-division packet switch. Communications IEEE Transactions on, 35(12):1347–
1356, 1987.