PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian...

68
UNIVERSITY „LUCIAN BLAGA” SIBIU FACULTY OF ENGINEERING PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS VEHICLE NAVIGATION SYSTEMS SCIENTIFIC COORDINATOR, Prof. Dr. -Ing. Carmen SIMION PHD STUDENT, Dipl. -Ing. Răzvan LUCA Sibiu, 2011

Transcript of PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian...

Page 1: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

UNIVERSITY „LUCIAN BLAGA” SIBIU

FACULTY OF ENGINEERING

PHD THESIS

-SUMMARY-

CONTRIBUTIONS ON AUTONOMOUS

VEHICLE NAVIGATION SYSTEMS

SCIENTIFIC COORDINATOR,

Prof. Dr. -Ing. Carmen SIMION

PHD STUDENT,

Dipl. -Ing. Răzvan LUCA

Sibiu, 2011

Page 2: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Universitatea

Lucian Blaga

Sibiu

Invest in people!

PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND

Project ID: 7706

Title: "Increasing the role of doctoral studies and doctoral competitiveness in a united Europe"

University”Lucian Blaga” Sibiu

B-dul Victoriei, nr. 10. Sibiu

PHD THESIS

-SUMMARY-

CONTRIBUTIONS ON AUTONOMOUS

VEHICLE NAVIGATION SYSTEMS

Scientific coordinator,

Prof. Dr. -Ing. Carmen SIMION

Phd student,

Dipl. -Ing. Răzvan LUCA

Sibiu, 2011

Page 3: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Ministry of Education, Research and Youth

Universiy “Lucian Blaga” Sibiu

Rector of the University B-dul Victoriei Nr. 10, 555024 – Sibiu, România

JURISDICTION

Doctoral committee, appointed by Rector

Order of University "Lucian Blaga" Sibiu

No. 102 of September 30

PRESIDENT: Prof. Dr. -Ing. Ioan Bandrea

University „Lucian Blaga” Sibiu

SCIENTIFIC COORDINATOR: Prof. Dr. -Ing. Carmen Simion

University „Lucian Blaga” Sibiu

REVIEWERS: Prof. Dr. Rer. Nat. Fritz Tröster

University Heilbronn / Germany

Prof. Dr. -Ing. Nouraş Barbu Lupulescu

University “Transilvania” Braşov

Prof. Dr. -Ing. Laurean Bogdan

University „Lucian Blaga” Sibiu

Any observations or comments please send to the address of the University "Lucian Blaga"

Sibiu.

Page 4: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

I

FOREWORD

Personal and professional evolution depends on the appreciation of others and the

personal will of each person. Whether it's a word to help develop an idea or an idea itself to be

developed, people interact and confirm their status to one another through communication,

through work, through collegiality. We practically exist through those confirming us.

The scientific research, the realization concept and development of this doctoral thesis

would not have been possible without support from the people involved directly or indirectly,

productive and critical to appreciate the results.

I would like to use this opportunity to thank to my scientific coordinator Prof. Dr. -Ing.

Carmen Simion for the support and for shareing engineering experience and patience, all

toghether defining professionalism within the entire period of my research activity.

For creative advices based on experience that helped me throughout the research I

would like to thank Prof. Dr. -Ing Ioan Bondrea.

I also speak with gratitude and thanks to Prof. Dr. Rer. Nat. Fritz Tröster from

"Heilbronn University", Germany, the mentor who helped me in developing my professional

and personal skills.

Professors in the Department of Manufacturing Engineering Faculty of Engineering in

Sibiu, as well as members of the evaluation and examination I would also like to thank for

recommendations on improving the quality of my research work.

For the opportunity to have a research and discussion collegue regarding various

topics of the project and beyond I would like to thank Dipl. -Ing. Robert Gall.

Family members and friends I would like to thank for their understanding, love and

moral support given during the research, giving me motivation and conditions for

implementation and completion of the thesis.

Dipl. -Ing. Răzvan Luca

Page 5: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

II

CONTENT * / **1

Introduction.......................................................................................................................... 1 1

Objectives of the thesis .................................................................................................... 2 2

Structure of the thesis ...................................................................................................... 4 4

Chapter 1: Current state ................................................................................................ 6 6

1.1. Market research, importance and applicability of autonomous intelligent vehicles in industry .......................................................................................................... 6 6 1.2. Evolution of autonomous vehicles ........................................................................... 11 7 1.3. Operating principles of the most significant specific automated

guided vehicles .............................................................................................................. 13 8 1.3.1. Networked systems with wire guide ......................................................... 15 . 1.3.2. Guidance systems with magnets ............................................................... 16 . 1.3.3. Laser guidance systems............................................................................ 17 . 1.3.4. Hybrid systems ........................................................................................ 18 8

1.4. Application areas of autonomous vehicles ............................................................... 18 9 1.5. Operating strategies of industrial AGV systems ....................................................... 22 9 1.6. Conclusions ............................................................................................................ 25 9

Chapter 2: Infrastructure of autonomous intelligent vehicles–

prototype realisation ........................................................................................................ 26 11 2.1. Non-holonomic autonomous intelligent vehicle ........................................................ 29 11 2.2. Prototype realisation ................................................................................................ 30 .

2.2.1. Running platform and odometer sensors.................................................... 30 11 2.2.2. Detection and guidance sensors ................................................................. 31 .

2.2.2.1. LASER sensor ........................................................................... 31 . 2.2.2.2. Ultrasonic sensors ...................................................................... 32 .

2.2.3. Odometer sensors ..................................................................................... 34 . 2.2.4. Video sensor ............................................................................................. 35 . 2.2.5. Data acquisition, command and control system ......................................... 36 12

2.2.5.1. Communication of the vehicle on the target-host principle ......... 37 14 2.3. Conclusions ............................................................................................................. 42 15

Chapter 3: Mapping and localization as simultaneous proceses

of autonomous vehicles ................................................................................................... 43 17 3.1. The concept of simultaneous localization and mapping (SLAM) .............................. 43 17 3.2. Extracting particular features of the navigation environment..................................... 45 18 3.3. Data association ....................................................................................................... 46 .

3.3.1 Data association using the Nearest Neighbour method ............................... 47 19

3.4. Autonomous vehicle localization ............................................................................ 50 20 3.5. Real-time environmental mapping and navigation .................................................... 51 21 3.6. Building a graphical virtual map. Simulation of the system ...................................... 53 22 3.6.1. The vehicle module „Vehicle” .................................................................. 55 23 3.6.2. The vehicle control module „Vehicle control” ........................................... 56 23 3.6.3. The laser scanner module „Scanner” ......................................................... 56 23 3.6.4. Global transformation „World coordinates” .............................................. 57 23 3.6.5. The map generation module „Map generation”.......................................... 59 24

3.6.5.1. Algorithms in the process of mapping ....................................... 62 25 3.6.6. The “split and merge algorithm” ............................................................... 63 . 3.6.7. Liniar regression ....................................................................................... 63 . 3.6.8. The RANSAC algorithm .......................................................................... 64 . 3.6.9. The Hough transformation ........................................................................ 64 .

1 *no. of page in thesis; ** no. of page in summary

Page 6: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

III

* / **2

3.6.10. Line extraction algorithm based on DCE technique (extDCE) ................. 65 26

3.6.10.1. Data reduction by merging lines with common properties ........ 68 27 3.6.11. Extracting landmarks from the navigation environment

for vehicle localization ....................................................................................... 70 28 3.6.12. Kalman filter model used for odometry correction ................................... 70 . 3.6.13. Using an extended Kalman filter for localization ..................................... 75 29 3.6.14. Trajectory planning module „Trajectory planning” .................................. 77 29 3.6.15. Object avoidance module „Track deviation" ............................................ 80 30

3.6.15.1. Object avoidance using specific curves for trajectory calculation ............................................................................................. 81 31

3.6.15.1.1. Clothoids .................................................................. 81 . 3.6.15.1.2. Bezier splines ........................................................... 82 .

3.7. Probabilistic mapping as an alternative method of navigation .................................. 84 32 3.7.1. Bayes method. General aspects ................................................................. 85 . 3.7.2. Derivation of the Bayes method ................................................................ 85 32

3.8. Conclusions ............................................................................................................. 90 33

Chapter 4: Navigation tests and applications

of the autonomous vehicle ............................................................................................ 91 35 4.1. The communication infrastructure used for tests conduction ..................................... 91 35 4.2. Comparing data from separate odometer systems ..................................................... 92 36 4.3. Mapping the environment ....................................................................................... 94 36 4.4. Implemetation of the Kalman filter for odometer improving while mapping the environment ............................................................................................................. 95 37

4.4.1. Mapping the environment through by extraction of features ................... 100 39

4.4.2. Changing the scanning perspective.......................................................... 102 41 4.5 The visual representation of the navigation environment ........................................ 107 42

4.5.1. Identifying floor markings in specific environments ................................ 107 42 4.5.2. Hough transformation of extracted lines .................................................. 107 . 4.5.3. Transformation of extracted edges and limitation by line attributes.......... 108 43 4.5.4. Perspective transformation using the darkroom principle ......................... 110 . 4.5.5. Representation of the transformed markings ............................................ 110 43

4.6. Analysis of people and objects in the environment of navigation ............................ 111 44 4.7. Conclusions ........................................................................................................... 117 46

Chapter 5: Applications in the flexible manufacturing system

of automotive industry .................................................................................................. 118 47

5.1. Modeling and simulation of the environment of a six cell flexible manufacturing system ..................................................................................... 123 48

5.1.1. Modeling and simulation of the application with occupancy grids .......... 133 51 5.2. Conclusions ........................................................................................................... 137 53

Chapter 6: Conclusions and future research directions ................................ 138 54

Bibliogrphy ........................................................................................................................ 140 58

Index of abbreviations .................................................................................................. 149 62

Appendix ............................................................................................................................. 150 .

2 * no. of page in thesis; ** no. of page in summary

Page 7: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

1

INTRODUCTION

The flexible manufacturing concepts "Flexible Manufacturing" and ecological storage

"Green Warehousing" refer to the methods of manufacture, storage and handling with

minimum consumption for environmental protection. In this way handling automation and

production resources becomes an important factor that involves both strategy and intelligent

vehicles. By automating this process industry gets efficient storage space, reduced spending

on energy used for maneuvers in dedicated spaces and reduced the use of system maintenance

costs. For automated operation of the intelligent guidance systems important administrative

costs are reduced. Also, the use of vehicles powered from alternative energy sources activate

in an unpolluted area with direct influence on the storage environment and the global

environment. Ventilation systems are not a necessity in this sense, the phenomenon of

automation thereby reducing overall energy consumption used by up to 40%. In terms of

production of company strategy, flexibility should be extended by components that the

company already possesses. A modern organizational structure and computer aided

production management enables the extension required by current conditions, in example,

changing market and uncertainty in the short and long term. In this context, flexibility can be

characterized by three attributes of time:

on time

after a time

over time

Thus, the flexibility of machines and flows are flexibilies “on time” , the products,

process and operations are flexible "after a time" and the flexibility in production,

development and volume are "over time" flexibilities. One can say that flexible production

systems flexibility is determined decisively by the total flexibility of transportation systems

that assist the processing system itself, the possibilities of maintenance, control and their

diagnosis. The responsibility for the development of such automated systems are, in addition

to influential factors, a prerequisite that must take into account the social impact factor and

the possibility of sustainable production systems in the factories of the future.

Page 8: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Objectives of the thesis

2

OBJECTIVES OF THE THESIS

“Everything should be made as simple as possible,

but not simpler than that“. (Albert Einstein)

Purposeful practical doctorate is an advantage for both, industry and academic

environment from the perspective of professionals. Bernhard Frey, head of human resources

at the company MAN GmbH, interviewed on this, says that a manifestation of the concept is

theoretically desirable when promoting a technical field, according to current market

demands. The realization of the concepts researched and developed in this work were a

significant factor in motivation and personal satisfaction of making step by step the

integration of existing in an existent physical prototype.

A simple statistical calculation shows that only 1,000 of about 20,000 graduates

annually decide to continue doctoral studies and researching in a technical field. Being one of

those who succeeds this is the first personal motivation. Although the career starts a little later

for those who choose a doctoral study, it is a prerequisite to continue in science as a teacher.

Taking into account the personal development, the PhD was able to sharpen my technical

senses, while learning how to address a subject analytically. The possibility of being in a

national research project motivated me to achieve results and to publish them in scientific

papers that contribute both quantitatively and qualitatively to the national scientific database

trough the University "Lucian Blaga" Sibiu. During my research at the University of

Heilbronn, Germany, partner institution for the research project, I was offered an opportunity

to develop practical implementation skills, and also the chance to develop my skills as a tutor

with the additional task of guiding students in the completion of their research work.

The use of autonomous vehicles in the industry, the investigated domain represents an

advantage in raw material handling and offers flexibility in production and logistics. In this

sense, the development of transport and transfer systems can reduce production costs

continuously according to studies based on manufacturing concepts like "just-in-time" and the

current global production development based on continuous adaptation to customer demand.

Material handling costs represent a significant proportion of production costs. Eynan and

Rosenblatt [30] indicate in their studies that these costs can achieve 30% of the total cost of

production. The subject matter is important in order to maximize flexibility as a feature of

future factory production systems and to minimize production costs. The level of automation

allows the definition of the degree of flexibility and complexity in the organization of

production. Redefining the flow of material requires recalculating routes of the raw material.

Transport systems such as conveyors or rigid AGV's presented critical in the next section

indicates that the reconfiguration of the factory flow is slow and involves a number of

additional costs.

This paper is an evolving concept of vehicles towards simplification of process

design and re-design of production functions by developing an intelligent autonomous

vehicle navigation concept. Contributions include research on autonomous vehicle

navigation systems. A highly debated visionary trend today becomes the fully autonomous

vehicle is moving in a specific environment, inside and / or outside creating a real-time virtual

map using proximity sensors, then it locates itself and operates in the specific environment. In

Page 9: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Objectives of the thesis

3

the literature the concept of Simultaneous Localization and Mapping (SLAM) is an open topic

for a autonomous vehicle navigation optimum. The control and communication between the

central and autonomous vehicles is based on an exchange of information for real-time

processing tasks and also allows multiple uses of existing data, representing a major

advantage for planning and management system and a significant part in developing the

objectives of this thesis. The safety in operation and driving without collisions, avoiding

obstacles and other handling components, namely people and objects in motion is an

important issue and is treated as part of the development in the concept of autonomous

vehicles. In order to achieve the goals of the research following theoretical and practical

objectives were established:

1. studying the current state of research of transfer vehicles in order to improve

felxibility in manufacturing by developing a navigation system for specific vehicles;

2. designing a simulation software for integrating the research into a informational

platform based on sensor models and representing the operating environment of the

vehicle.

3. realization of a prototype platform for rapid deployment and testing of embedded

algorithms into specific intelligent vehicle navigation tehniques;

4. study and evaluation of sensors and available resources for the implementation of

autonomous vehicle navigation systems;

5. software development and implementation of navigation algorithms that allow a

vehicle to navigate autonomously in real time based on the concept of simultaneous

localization and mapping (SLAM);

6. realization of a virtual map by extracting relevant features of the operating

environment in order to allow an autonomous navigation based solely on extracting

information from the sensors of the vehicle;

7. realization of a graphical interface that allows viewing, monitoring and analysis of

vehicle behavior operating in a specific environment;

8. minimize the amount of data used for navigation through effective programming;

9. stuying the posibilities to implement the navigation systems on real vehicles trough

simulated a environment;

10. implementation and testing of the concept of navigation algorithms and develop a

scaled model of the intelligent vehicle.

The final purpose of this thesis is the realization of a prototype for the permanent

development of intelligent autonomous vehicles by creating a platform for integrating the

studied techniques of measurement, command and control, dedicated systems, development of

algorithms, and design concepts of logistics and handling systems of factories of the future.

Page 10: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Structure of the thesis

4

STRUCTURE OF THE THESIS

Having the aim of studying the autonomous vehicle navigation systems operating in

specific environments, this paper presents six chapters describing methods developed and

applied contributions to improve the autonomous navigation of intelligent vehicles.

The first chapter of the thesis presents a state of actual used autonomous vehicles in

the industry. A market study is realised in order to highlight the importance and applicability

of autonomous vehicles in the automotive manufacturing industry. A top of the current

products scoring the importance of smart product development is done, category which

includes the autonomous vehicles. Some statistics show an increasing development of such

products.

Regarding the research and development direction of the field of engineering and

vehicles used in industial handling and manufacturing systems, a parallel bibliographic

summary of the literature and actual publications in the field was made aiming in particular

the period January 2010 - June 2011. Different concepts, prototypes and trends in the field are

presented confirming the approach of top issues in terms of timeliness research directions.

A presentation of the historical evolution of industrial vehicles is realised. The most

important vehicles and automated systems are a critical study describing their principles of

operation and limiting the areas of application and operating strategies.

Chapter two contains informations about the infrastructure of non-holonomic

autonomous vehicles. A prototype of a vehicle scale (1:8) is made in order to integrate the

research technology to develop and test algorithms for navigation. The Infrastructure

integrates different proximity sensors to cover a wide range of detection and maximize

efficiency by implementing a system of manufacturing infrastructure, providing independence

and robustness in operation. A communication concept of two parallel processing units for

command and control and for monitoring the vehicles activity is developed for optimal use of

resources, which enables testing application in real time.

The third chapter analyzes the concept of simultaneous mapping and localization of

autonomous vehicles (SLAM). According to researchers, this concept is the issue of "chicken

and egg" in navigation technology. By programing a simulation software, functions of

navigation sensors and algorithms developed are modeled in order to obtain relevant data,

also described during the implementation of the applications. A limitation of the system is to

be considered in order to achieve real a real time computation by using the Embedded

MATLAB programming platform.

Each modular component of the chapter presents information about the algorithms

implemented in the simulation.

An approach to achieve a virtual map by extracting features (feature-based) similar to

those described in [21] is performed by using a self developed algorithm for data reduction,

based on discrete contour evolution DCE [53].

Specific phenomena such as, for example, the occlusion of the object during the

environment scanning or travel path comparison during calculations are analyzed, presenting

an implementation solution. The vehicle trajectory planning is an issue addressed at the macro

level. This work focuses on preparing the necessary data to generate the trajectory of

movement, more specifically, the approach of integrating SLAM tehniques. The vehicles

Page 11: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Structure of the thesis

5

trajectory is generated by using potential fields. A mechanism for detection and avoidance of

obstacles is presented in the same chapter based on the integration matrix of specific curves

(clothoids and Bezier curves).

For the vehicles localization in the generated map, considered as a simultaneous step,

an implemented model is presented by extracting parts resulting from information extracted

from the environment during the mapping process. A transformation into a quasi Hough

parameter space of a line is made to extract relevant data and reduce them to a minimum. By

using a Kalman filter as an extended filter (EKF) [41] model, marks are recorded and added to

the vehicle odometer data to achieve a more accurate localization and prediction of the

following relevant data needed to support the vehicle during the navigation.

The last subchapter examines an alternative approach to navigation. Using the method

of segmentation and sharing workspace in cells, a calculation is made using probabilistic

occupancy cells.

Chapter four contains applications made to develop algorithms and to analyze the

behavior and implementation of concepts studied. Applications of Kalman filter behavior

modeled in Chapter three gives an overview of the navigation parameters. Further, the

implementation and evaluation of a video capture concept is done in order to identify dynamic

objects and operating environment markers that define specific areas of vehicle navigation

sectors.

Chapter five describes a special application with input from a simulated flexible

manufacturing system which is done in order to highlight the main tasks an autonomous

vehicle has during the navigation in specific environments. Whether it is an interoperable

transfer or handling in a warehouse logistics, autonomous vehicle behavior is evaluated based

on the output of the algorithms implemented.

In chapter six the results are presented in which special emphasis is placed on

highlighting the scientific novelty of the theoretical postulates and experimental results

obtained. Research methods used are presented and arguments that underpin the credibility of

results. There are also proposed future research directions in the field.

The theoretical and experimental research made during the development and

completion of the research program and their achieved results were used by public support

and discussion at national and international scientific conferences. Titles of these works are

included in the bibliography.

This Research has been conducted in the doctoral school of Lucian Blaga University

in Sibiu, POSDRU/6/1.5/S/26/7706 within the project titled "Increasing the role of doctoral

studies and doctoral competitiveness in a united Europe", co-financed from the Social Fund

European Operational Programme Human Resources Development 2007-2013.

Page 12: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Current state

6

CHAPTER 1: CURRENT STATE

„Manufacturing industry is important

for Europe” (Manuel Barroso, 2009)

1.1. Market research, importance and applicability of autonomous intelligent vehicles

in industry

Strategic areas of research in production refer to building an organization's strengths

and defining application-oriented activities. Competitiveness by reducing costs and adding

value end products ensures quality and a permanent innovation process. The European Union

launched its strategic agenda for future plants through "EFFRA-Factory of the Future" with a

budget of 1.2 billion Euro. This program contains both a national sector for development and

a regional industrial group. As part of the European economic recovery plan, the commission

released three public-private partnerships (PPP). They are powerful means to stimulate

research efforts in major industries, construction of automobiles and manufacturing, which

were particularly affected by the economic downturn.

According to statistics of the International Federation of Robotics, 76,600 robots units

totaling $ 13.2 billion, which covers the services industry, have been sold by 2009. Of this 5%

are logistics, and 6% mobile platforms for general use in manufacturing. A trend of increasing

their use for the next period is observed.

Figure 1.1: Statistical value of sales of robots for industrial and professional services in 2009

Page 13: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Current state

7

1.2. Evolution of autonomous vehicles

A graph of the evolution of vehicles used in industry since 1959 until 2000 is shown

below based on a research of the Frauenhofer Institute in Stuttgart / Germany. Mentioned are

the three relevant categories of development over the years.

Figure 1.2: Evolution of autonomous vehicles [116]

Page 14: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Current state

8

1.3. Operating principles of the most significant specific automated guided vehicles

Robocar systems are used to storage parts in containers for interoperational tranfers. In

this case a special focus on access roads is required and paths followed by them, that’s why

this systems are less common in actual production facilities, meeting them more often in large

stores or warehouses.

A summary of the operating principles of the current solutions available in the

industry is carried out using as a source the Egemin automation company specializing in

guided vehicles AGV of industry type.

1.3.1. Networked systems with wire guide

1.3.2. Guidance systems with magnets

1.3.3. Laser guidance systems

1.3.4. Hybrid systems

Hybrid systems are multipurpose vehicles that allow the automatic operation and

handling without human operator intervention. For an automated service, a laser scanner is

guiding the vehicle in the production areas or distribution environments. The figure below

shows a fork lift equipped with additional systems needed for automated handling.

Figure1.3: Hybrid material handling system [109]

1 - safety sensor, 2 - Laser navigation unit, 3 - sensors to detect obstacles side, 4 - button on /

off, 5 - 3D video camera (optional), 6 - RFID reader

Page 15: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Current state

9

1.4. Application areas of autonomous vehicles

The trend in robotics and automation is to coverage many as possible areas with the

aim to simplify the tasks of people, to obtain safety and precision in production, to develop

road assistance systems, namely the development of the service area in general. In all these

areas mobile robots have been developed generous in recent years. The difference is defined

by the criteria of the applications operating in specific environments:

structured internal enviroment

unstructured external environment

1.5. Operating strategies of industrial AGV systems

Based on the expedition rules of the vehicles different tasks are fullfiled. The

performance of analytical models usually depend on the expedition passed. A parallel

distribution of tasks is an advantage in their processing and is a relatively new topic in this

field. Basically, an autonomous vehicle is responsible for other tasks on the way to the task

received from the central control system. Advantages and disadvantages of the guidance

systems are described in the table below with the extension criteria of GPS navigation and

video video guided systems, two components introduced as developments regarding specific

vehicle navigation systems.

Table 1.1: Presenting quality criteria of the guidance systems

System/criteria Wired

guidance

Magnet

guidance

Laser

guidance

Video

guidance

GPS

Indoor/outdoor Ind.&outd. Ind.&outd. Ind. Ind.&outd. Outd.

Dirty

surfaces

++

++

-

/

++

Trust

level

++ + + + +

Instalation

costs

- / ++ ++ ++

Flexible - / ++ ++ ++

Precizie în mm ~2 3-20 ~10 <=50 <=50

++ very good + good / medium - bad

In order to achieve an optimal flexibility in manufacturing systems, it is recommended

to be used the system with the fastest implementation and low-cost investment. Having the

aim of flexibility maximization in handling, the navigation systems using video and laser

senzors bring large benefits.

1.6. Conclusions

The industrial use of autonomous vehicles is an advantage in raw material handling,

flexibility in production and in logistics. In this respect, a continuous development of these

systems can reduce production costs according to studies based on manufacturing concepts

Page 16: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Current state

10

like "just-in-time" or "flexible manufacturing" and the current overall production based on the

continuous development of adaptation.

Material handling costs represent a significant proportion of production costs. In their paper,

Eynan and Rosenblatt [30] indicate that these costs can reach 30% of the total cost of

production. In this sense, the subject matter is important because of flexibility, characteristic

of the production system. It is important to address the interaction between machines, material

handling and information systems.

The automation level allows establishing a communication between the central control

and autonomous vehicles or between autonomous units. The evaluation criteria of the current

transport architectures and new opportunities provide a technical overview of future research

directions.

Whatever the application, the trend is to automate processes by introducing specific

vehicles.

Based on the last fifty years of development of vehicles used in the automotive

industry, we can say that technical news and information have been accumulated during the

time and contributed successfully in the developlent of the industrial transfer.

With the progress of sensor performance, the vehicles developed additional functions,

increasing the flexibility of this work primarily. Products received by integrating artificial

intelligence increasingly high degree adaptivity, which confirmed the direction of top research

and implementation of autonomous vehicles worldwide. In this context the following research

areas were analysed:

a study of specific literature in order to highlight market developments on the current

market and future autonomous vehicles used both in industry and in other areas;

an analysis of the operating principles of the guidance systems used in manufacturing

industry;

a research of a manufacturing layout to identify and highlight possible operating

strategies;

a study of flexibility in handling of different guidance systems.

Page 17: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Infrastructure of autonomous intelligent vehicles-prorotype realisation

11

CHAPTER 2: INFRASTRUCTURE OF AUTONOMOUS INTELLIGENT

VEHICLES – PROTOTYPE REALISATION

„A creative idea remains an idea until you do something with it. You have

to do something, otherwise you are not creative.” (Glen Hoffherr)

Given the criterion of transportation, the infrastructures of vehicles used in industrial

applications differ by their wheel arrangement and are dependent on their geometry. For the

realization of a mobile robot, it should be determined the handling limit, control and stability,

the practical operating procedure. A classification of industrial mobile robots after wheel

configuration is made according to [83].

2.1. Non-holonomic autonomous intelligent vehicle

The objective of constructing a non-holonomic vehicle on a 1:8 scale operating in both

internal and external environments, a simplified Ackermann model is used for the

mathematical description of the system. Each wheel must meet the slip constraint, and such a

vehicle can have only one instanteneous center of rotation. The whheel drive is established on

all four wheels, with a manoeuvrable front axle. Kinematic relations are explicitly adapted.

2.2. Prototype realisation

2.2.1. Running platform and odometer sensors

For the prototype realisation, a model platform from CEN racing was used. The

steering is controlled by a servomotor and the vehicles position is indicated by incremental

sensors mounted on the rear axes of the vehicle. An electric motor moves the vehicle.

Odometer positioning is corrected algorithmically. Component specifications are described in

Appendix 1.

Figure 2.2: Vehcile platform including the servomotor, drive motor and the odometer sensors

Page 18: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Infrastructure of autonomous intelligent vehicles-prorotype realisation

12

2.2.2. Detection and guidance sensors

2.2.2.1. LASER sensor

2.2.2.2. Ultrasonic sensors

Figure 2.6: Ultrasonic sensor positioning scheme for calculating the coordinates and transformation vector

The coordinates of each sensor is described by the vectors through their position and

the reference to the middle point of the rear axle of the vehicle. These vectors are summarized

in a matrix of size 20x2 in the form below and they are the data transmission interface in

Cartesian coordinates to the command and control system:

[

] [

] [

] (2.7)

2.2.3. Odometer sensors

2.2.4. Video sensors

2.2.5. Data aquisition, command and control system

The limitations of the command and control system is defined by a computer

processing unit representing the tower PC-104 format [ISO], used for industrial applications.

This system can be configured depending on the required infrastructure. The following list of

components form the data acquisition system, command and control of autonomous

intelligent vehicle prototype. The specific details of the components are presented in

Appendix 2 of the thesis.

Page 19: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Infrastructure of autonomous intelligent vehicles-prorotype realisation

13

Figure 2.10: PC-104 - Data acquisition unit, computing and algorithmic processing

A parallel running configuration of the data acquisition and applications on a concept

of two systems requires a synchronization of the two PC-104 units. In this sense, the concept

of communication was developed to cover under the premise the needs to achieve a real time

processing of at least 0.01 seconds.This was established in order to achieve a superior level of

data processing necessary for controlling the vehicle, and for higher algorithmic calculation.

The processing system is running independently from Windows or Linux platforms being

superior to most vehicles currently operating in the industry.

Table 2.5: List of the components of data acquisition unit, computing and algorithmic processing PC-104

Element Hardware description Specifications

a Advantech PC104 CPU

Main board PCM-3380

Embedded Intel Pentium M 1.4 GHz, 2GB RAM

b Modul de memorie SD Card

8 Gb

c Power unit

PCM 3910 Advantech

10 to 24 V output

d PCI to ISA Board

PCM 3117 Advantech

e IO Board (Analog)

IO 526 Sensoray senzor card

AD/DA/Encoder Input

f PWM Signal Unit

DM 6804 RTD for servomotors

A vehicle sensor interface was necessary to develop in order to achieve the vehicle

control. Its specifications are described in Appendix 2 of the thesis.

Page 20: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Infrastructure of autonomous intelligent vehicles-prorotype realisation

14

2.2.5.1. Comunication of the vehicle on the target-host principle

The flexibility in use of the PC-104 system offers the possibility of establishing a

communication with other entities or with a monitoring computer for data transmission using

standard protocols. For this, the XPC target package is used, software integrated into

MATLAB. With it is possible to set priorities for a parallel use of communication while

synchronizing information from the various input/output cards available.

Using the target-host communication and the described architecture, the

communication is performed for the autonomous intelligent vehicle (target) communicating

with a control computer (host). A transfer by UDP protocol is established while entities are

configured with static IPs.

Figure 2.13: Target-host communication scheme

A laptop is monitoring through WLAN the activity of the two command and control

computers mounted on autonomous intelligent vehicle. The communication architecture can

have one or two active processing units. In this respect the two systems of PC-104 become

target 1 and target 2. They communicate one with another based on the master-slave concept,

and their timing can be achieved either through a parallel port or using a CAN communication

protocol, configurations tested in the laboratory. The goal is to run on a PC-104 data

acquisition application and the command and control of the vehicle under the master system

and on the slave system the other algorithms under consideration necessary to determine

navigation parameters. Navigation algorithms are treated in detail in the following sections of

the work.

The transfer performances of the two processing units were measured in terms of

transmitting data packets using the UDP protocol and the Ethernet connection between target

1 and target 2. The ideal transferrate TR is calculated.

Three levels of use are referential:

Low utilization;

Mediul utilization;

High utilization.

Received packets are represented in the diagram below by referencing the received

packet number that decreases with greater use of the processing unit.

Page 21: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Infrastructure of autonomous intelligent vehicles-prorotype realisation

15

Figure 2.14: Limits of data transfer between two PC-104 processing units using XPC target

A percentage representation shows that during the high use, 45.92% of all packets

transmitted fully, representing a critical situation. This is a motivation for establishing the

criteria for effective programming and for using prediction algorithms for the compensation in

lack of data, a phenomenon common in data transfer.

For efficiency common data interfaces were established in terms of increasing the

performances of both processing algorithms and the data acquisition.

The computer communication between the control unit target 1 or 2 takes and the host

is made via a wireless router such as D-Link DWL. This monitors the sensors and the path

traveled, receives data from mapping algorithms for map visualisatoin while the autonomous

vehicle is moving. On the host computer Microsoft Windows 7 is running with special

software developed in Matlab and a graphical user interface for easy manipulation by the user.

An extension of this interface had to be made during the development and in this respect; the

advantage of using a programming package that includes object-oriented programming in

Simulink library and facility to create graphical interfaces (GUI) was highlighted. A joystick

connected to the laptop enables rapid intervention, if the target system signals errors. A view

of the interface presented in this section is avialable during the tests in chapter four.

2.3. Conclusions

The making of a scale model of intelligent vehicle offers the following advantages in

terms of behavior researching navigation:

development costs are lower due to a platform that does not require real

components;

the testing of specific environments is also done at low costs, there is no need

to use a real space and approval to use a space of manufacture;

energy consumption is reduced.

Page 22: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Infrastructure of autonomous intelligent vehicles-prorotype realisation

16

The trend in research and applied development is of using models based on scale

implementation as a second step, the first step is the creation of a simulated software

environment, which is presented in this case in chapter three. After validation tests, the model

is proposed or not, depending on the results, to be manufactured in series. If the

improvements and the visualization platform designed have rapid effects, the vehicle has a

wide maximum efficiency.

By establishing a common interface for communication between sensors, both, quick

and easy implementation of the components is realised and the possibility of integrating data

from different sensors to fusion is just a step away.

Regarding the limitation on communications and computing systems, an assessment

was made of data packets that can be transmitted during an application. The concept of

industrial systems using parallel PC-104 that are used to increase performance by setting the

synchronization of the two modules of the vehicle represents a contribution for the future

similar systems.

By using a multi sensory model we are able to differentiate between areas of activity

of the vehicle. For this, the laser sensor is responsible for providing data in a field of up to 10

m around the vehicle. Ultrasonic sensors with limited capabilities are used to identify objects

in the vicinity of the vehicle. To complete the active sensory domain, a video camera provides

the identification of objects whose properties can not be extracted with proximity sensors. In

this respect, guidance in unknown or partially unknown space becomes possible in terms of

safety of human operators. Although there are also other information retrieval algorithms of

dynamic objects idntification using proximity sensors, video sensors provide more accurate

information and the identification becomes more robust.

Following issues have been studied closely, with contributions in implementation:

the autonomous vehicle rolling platform for modeling and implementation of

its infrastructure;

a study of sensors available for purchase and integrate them on a stable

platform for the purpose of autonomous movement;

development of a concept of communication between the processing unit,

command and control to improve the computing power;

establishing a monitoring communication interface between the vehicle and a

process computer.

The vehicle presented integrates sensors and the latest information technologies in the

development of autonomous intelligent vehicles, making the communication infrastructure on

the taget-host principle to a robust non-holonomic transfer platform.

Page 23: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

17

CHAPTER 3: MAPPING AND LOCALIZATION AS SIMULTANEOUS

PROCESSES OF AUTONOMOUS VEHICLES

“Once you have mapped out the road, it„s impossible

not to go ahead” (Antoine de Saint-Exupery)

3.1. The concept of simultaneous localization and mapping (SLAM)

This concept refers to the simultaneous localization and mapping (SLAM) and is a

technique created by the robotics community in order to explore the possibility of an

autonomous vehicle to start from an unknown location and assimilating information about the

operating environment, to incrementally create a map of the activity space. By using sensors

and simultaneously the completion of new generated mapping, the vehicle can locate itself.

Both, map and estimates of the vehicle location after a SLAM activity are tasks that support

future intelligent autonomous vehicle meets them. A practical solution and mapping of its

own location is inestimable in terms of achieving the navigation of an autonomous intelligent

vehicle as described in [84].

Regarding the evolution of the SLAM a series of concepts and approaches have been

developed. It is possible to classify different approaches depending on the method used to

achieve the map representation, which is a key point because it determines the type of

information explicitly expressed in the model. The most common methods of representation

are:

Ocuppacy grids. They describe the vehicle surroundings by dividing the space into

several regular cells and assigning a probabilistic value of the area by setting empty

and occupied criterias, as described in [46] and [82].

Topological maps. An example of such maps is represented by the Voronoi diagrams

[54].

Feature based maps. Basics of environmental representation is achieved by using a

set of geometric primitives such as points, lines and planes above the raw data

extracted from sensors as described in [21], [18].

Figure 3.3: Generating maps by extracting lines from points

Page 24: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

18

Maps generated from successive scans. The raw data of the sensors are aligned

directly through a process of translation and rotation by finding a maximum overlap

with the data contained in the previous scan. Maps are obtained as segments that are

fused to each new data acquisition. It is suitable for use in situations where no simple

geometric representation of the environment can be obtained. So far, it has been

proven successful only for laser scaners, according to studies [22].

Figure 3.4: Maps generated from successive scans [19]

Once the selection of the method of representation is done, the technical difficulty of

SLAM's main achievement is defined by reducing the possible growing uncertainties and

noise reduction, namely the error of sensors. It is essential to find an effective method in

terms of calculation able to cope with uncertainty, to make an accurate representation to

obtain a precise map, and therefore a reliable vehicle location. Other technical difficulties

related to the SLAM problem refers to the extraction of particular features and the association

problem. Obtaining reliable information from noised sensor data, and discernmenting to

identify previous exploration of the same territory leads to the key of the convergence of a

SLAM algorithm.

3.2. Extracting particular features of the navigation environment

Representations depend on the choice of navigation environment and the type of

sensor used. Choices cause large differences in interpretation. For example, a laser sensor data

can be used either directly, by making global coordinates transformation or processing of the

particular features. [89], [23]. Occupancy grids commonly used sensor models in order to

provide information on the volumes occupied. Extraction methods based on particular

features of the navigation environment are mainly used for processing data from sensors in

order to find entities that can be identified repeatedly. These entities are called individual

characteristics and the process of obtaining them is called to extraction of particular features.

By this technique, for obtaining the particular features, the information provided by

various sensors is handled so as to achieve an extraction as feasible and to meet a number of

criteria.

Page 25: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

19

3.3. Data association

3.3.1. Data association using the nearest neighbor method

A quick calculation method on the association of internal sensor observations (the

vehicle) is described by the nearest neighbor method [83, 8]. It starts from a random object to

which it is most closely associated with the object distance in space depending on a parameter

set. In this way, a neighborhood that can be associated is searched. The version of the

algorithm used associates objects within a so-called regions of interest (ROI), which limits the

space of association data. By defining the ROI, the computational effort is reduced

consistently compared to other existing algorithms and data association is maximal.

Figura 3.5: Asocierea datelor în regiunea de interes (ROI)

If there is only one association made by limiting the ROI, this is done. The other rows

and columns of the matrix calculation containing the extracted data are excluded.

To achieve accurate navigation maps, the sensors perceptual values must be

interpreted. Environmental features are extracted and interpreted due to specific scenes in

order to minimize the environmental impact of imperfection of individual sensors in order to

achieve a robust navigation.

Correct association of data is a crucial component in achieving SLAM algorithms that

can influence estimative methods such as extended Kalman filters (EKF). Data association

methods are dependent on the types of data from sensors. Applications that use the raw data

of the sensors as successive scan, chains scan data by matching the observation with the

previous representation of the map. This technique can be classified into two categories:

using raw data representation;

using geometric representations of the scanned environment.

A widely used concept to combine maps is "map matching", which seeks the common

elements of segments generated in maps and overlapps them in a new map.

The overlap method requires significant resources both for representation and for the

final calculation because the processing is done point by point. In this sense, a simple

approach applied in this project is immediately mating features that are extracted in a

predetermined area of interest of the vehicle on the move. The method is described in detail in

Page 26: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

20

order to achieve a fusion in the simulation software environment by merging generated scan

lines.

3.4. Autonomous vehicle localization

Locating an autonomous vehicle is the answer to the question: "Where am I?" In this

context the challenge is to extract features and combining data for a more accurate

localization. Operating a vehicle environment is an important role due to the contact surfaces

influencing the vehicle odometer and the objects identified due to extraction of features. By

applying correction filters, these shortcomings can be removed.

Finding an optimal solution depends on vehicle operating characteristic environment,

different solutions are presented in [57, 59, 20, 29, 45, and 87].

Figure 3.8: General scheme of an autonomous vehicle localization process

An important element is attached to the extraction of features, defining the so-called

observations that represent navigation markers (landmarks). By efficiently using the

landmarks the autonomous vehicle position update is done.

Figure 3.9: Localization using landmaks

An algorithm developed for the vehicle localization refers to extracting parts of the

lines representing edges of the object. Landmarks are the main features extracted, and by their

re-scan the localization is performed during the navigation procedures. Landmarks of the

Page 27: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

21

maps are outlined during construction, so the localization phenomenon is strictly related to the

generation of map. Thus, the two components, mapping and localization, complement each

other.

3.5. Real-time environmental mapping and navigation

Mapping is the problem of integrating information from data acquired from sensors

and their representation in a map. This process can be described by the question "What does

the environment look line in which to operate?" Central issues relate to environmental

representation and interpretation of data from sensors. In contrast, the problem of estimating

the location refers to the robot position relative to a map. In other words, the robot has to

answer the question "Where am I?" A distinction of the track position is made, where the

starting position of the vehicle is known, and the global location of the vehicle, where there is

no known a priori information. The control problem or the effectiveness involves determining

the travel path to guide a vehicle to a point set. The solution would include the answer to the

question: "How to reach effectively a given location?"

The major problem is that the three can not have an independent approach. Before a

robot can respond to the question of how the environment looks like depending on the sets of

observations, it must know the location from which the observations were made. At the same

time is difficult to estimate the position of the vehicle without a map. According to [19]

integrated approaches are solutions for planning, simultaneous localization and mapping.

Finally, it is assumed that the mobile robot has developed an accurate model of environmental

and has determined its own position relative to this model.

Figure 3.10: Tasks required for a mobile robot so that it accurately models the environmental records.

Overlapping areas are areas of common elements in the field.

Page 28: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

22

Key issues in the context of real-time mapping refer to questions such as:

where should the vehicle be guided during a self-exploration?

how to solve the noise problem when estimating the position and the landmarks

observed?

how to interpret the robot's uncertainty in the global model and interpreted data from

sensors?

how to shape the environment variable changes over time?

how to effectively coordinate an autonomous mobile vehicle?

The contributions in this context are solutions of different aspects of learning above

a global map and location of mobile autonomous vehicle in the map generated.

In terms of data acquisition in the field of mobile robots, one important factor is the location

of the robot in an unknown environment. An accurate estimation of the position is the core of

any navigation system including dynamic construction of the map and trajectory planning.

Using data from robot vehicle odometer are not sufficient because they introduce

errors in the positioning system according to [12] and the results of the measurements made

and shown in the prototype implementation section above. Solving the problem is the

complementary use of various sensors (sonar, infrared, laser, video, etc.) and subsequent

merging of data processed.

3.6. Building a graphical virtual map. Simulation of the system

The perception of the navigation environment is performed on a software platform

developed in the programming environment Matlab / Simulink using the limits imposed on

the processing system to run in real time. After testing and evaluating the communication

command and control system presented in the pervious chapter criteria were established for

use strictly Embedded Software Platform. It was thus ruled out the use of operating systems

on the command and control modules of the PC-104 for algorithmic processing. Two aplroach

levels are developed, simulation and hardware implementation. Sensor interfaces are modeled

mathematical in the simulation to match the actual implementation interface. The diagram

below shows the two levels discussed. A fast implementation is feasible given that the

simulation has strict structure. For this a limitation of the programming code is done in order

to use only dedicated library functions available.

Using a control interface and visualization is available directly in the simulated

environment. The implemented level contains only the compiled code of the application,

visualization is transferred through the WLAN module to the host computer, on the basis of

the communication established.

Dimensions have been introduced to scale the intelligent autonomous vehicle used

also for laboratory tests. Handling objects have similar sizes to the vehicle operating with a

simpified rectangular shape.

Page 29: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

23

Figure 3.12: Software platform of the simulation

3.6.1. The vehicle module „Vehicle”

This module contains the vehicle dynamic model with parameters for determining the

position Xv, Yv, Thetav and the speed V.

With a kinematic model of the transporter, the inverse kinematics problem is solved.

More specifically, knowing the trajectory and speed values, the main task is to determine the

time variation of the angle of the wheels and draw it on the map.

3.6.2. The vehicle control module „Vehicle control”

The parameters that were set here are: acceleration, braking and steering of the vehicle

by assigning experimentally determined values. The model also requires consideration of

vehicle dynamics by setting a variable parameter of friction with air.

3.6.3. The laser scanner module „Scanner”

This is shaping laser scanner interface module for data delivery to the mapping and

location and trajectory planning.

3.6.4. Global coordinate transformation „World coordinates”

This module coordinates the transformation performed by the laser sensor purchased

from a local reference in the global coordinate system.

The reference point of the local coordinate system is the rear axle of the autonomous

vehicle. It differentiates between the sensor coordinate system OL, the rest frame and the

reference of the vehicle global OG.

Page 30: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

24

3.6.5. The map generation module „Map generation”

Represents a complex module for map generation. It includes algorithms and features

extraction from scanned environment. A diagram showing the internal structure of this

module is presented below.

Figure 3.16: Internal structure of the mapping and simultaneous localization module

The common interface of sensor data transformed into the global Cartesian

coordinates are also input for the map generation module The first phase provides a

breakdown of points in clusters on the basis of minimum distance between them. For this

purpose the defined region of interest ROI around the vehicle is 10x10 m. Successive scans

are acquired in cycles of 10 scans, the experimentally determined number to process the

relevant data. A vector out of the "cluster scans" contains data grouped; forming an edge by

taking the following lines in the block function to extract objects.

Figure 3.17: Grouping points of intersection with objects in clusters

Measured points not belonging to any cluster are returned to the position of the group

for a possible re-associate the data from t-1 with the current scan at time t. A data reduction

algorithm performed the extraction of lines scanned on an expanded concept of the evolution

of extended contour DCE and is presented in detail. Data are obtained as reductions of up to

80%. Only the ends of the extracted lines are memorized for further processing, achieving a

data filtering irrelevant for the extraction of features. Merging short line segments in a line

graph to define a relevant detected contour is performed in the „merge Objects”.

Page 31: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

25

Schematically a map construction problem is to determined by the next steps:

environmental sensitization by acquiring data from the laser sensor at time t;

processing of data from sensors by extraction lines and particularities;

integration of time t observations determined in the structure of environmental

observations at time t-1.

Below are cases where the merger of line realizes a further reduction in data,

implicitly considered the method described in [75]. The center of gravity of the lines is used

and coordinates are projected onto a new direction resulting by representing the average of

two lines directions.

3.6.5.1. Algorithms in the mapping process

În procesul de cartografiere există o serie de abordări atât de calcul, cât şi de

reprezentări grafice. Cele mai uzuale tehnici se referă la extracţia particularităţiilor din datele

senzorilor, dar şi procesarea directă a datelor neprelucrate ale senzorilor. Abordările care se

referă la extragerea particularităţiilor geometrice din datele iniţiale ale senzorilor au fost

studiate intensiv în domeniul localizării roboţiilor mobili. Acestea sunt descrise în cercetări

precum [18], [29], [87] ca fiind tehnici compacte care necesită mai puţin spaţiu în ceea ce

priveşte prelucrarea, atingând performanţe de un standard relativ înalt. În ceea ce priveşte

eficienţa algoritmilor care au la bază particularităţile geometrice parametrizate, acestea s-au

dovedit a fi preferate în comparaţie cu algoritmi de calcul care au la bază prelucrarea directă a

punctelor de la senzori. De menţionat este însă că în cazuri particulare o abordare combinată

reuşeşte să satisfacă împlinirea sarcinii de realizare a unei hărţi precise de navigare.

Segmentele de linie reprezintă printre primitivele geometrice cel mai simplu element.

Este uşor de descris în aproape orice mediu şi reprezintă totodată şi cel mai comun mod de

abordare în ceea ce priveşte reprezentarea 2D. În lucrarea sa, Castellanos [20] descrie un

algoritm bazat pe o metodă de segmentare inspirată din domeniul prelucrării video. Vandorpe

[98] introduce un algorim de generarea dinamică a unei hărţi bazându-se pe tehnica extragerii

de linii şi cercuri utilizând un scaner laser.

Pe baza comparaţiei de algoritmi descrisă în [37] au fost selectaţi algoritmii descrişi în

ceea ce urmează, pentru testarea performanţelor, cât şi pentru a dezvolta un algoritm propriu

care să satisfacă necesităţiile proiectului. Informaţiile aferente descriu paşii de realizare a

acestora.

3.6.6. The “split and merge” algorithm

3.6.7. Linear regression

3.6.8. The RANSAC algorithm

3.6.9. The Hough transformation

Page 32: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

26

3.6.10. Line extraction algorithm based on DCE technique (extDCE)

Given the development of algorithms that operate in real time and analying the results

and effectiveness based on the previously presented algorithms, an algorithm was developed

to meet the project requirements. This algorithm is based on successive extraction of lines

while the vehicle is moving. Unlike previous algorithms, which are based on data collected

before the application,this algorithm is based on the line extraction technique DCE [53] with

an immediate processing, the collection is defined by a "buffer" to determine the necessary

points to extract lines resulting from the scanning cycle. For this the algorithm is used in the

processing software to generate a real-time virtual map. The algorithm comes from the

graphic domain, where line extraction is a basic technique.

1. Initialisation: collecting a number of n scans (minimum 3 points)

2. Checking the minimum distance between three points defined by a parameter

3. If parameter value is lower, go through step 5

4. Otherwise the points are returned in the cycle of n scans for new associations

5. Check with a fixed angular value the parameter and condition L12 <L1 + L2 for three

consecutive points

6. Processing of n scans applying the algorithm and extract the lines

7. If the value generated for angular lines coincide, they are merged into a new line

The comparisons of mapping algorithms were evaluated based on criteria described in

the table below.

Table 3.1: Algorithm comparison

Algoritm Processing

speed

[Hz]

Nr. of

lines

Corectness Precision

Real

pozition

[%]

False

pozition

[%]

σΔr

[cm]

σΔα

[grade]

Split&merge+cluster 1470 641 86.0 8.9 1.95 0.74

Incremental 344 561 77.8 5.9 2.04 0.72

Incremental+cluster 617 567 79.2 5.1 1.99 0.76

Liniar regresion 364 577 76.4 10.1 1.97 0.80

Liniar regresion +

cluster

384 562 75.8 8.4 1.68 0.79

RANSAC 29 749 75.6 31.5 1.37 0.77

RANSAC+cluster 93 547 70.7 12.2 1.63 0.70

extDCE 377 720 81.3 13.8 1.59 0.72

Page 33: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

27

3.6.10.1. Data reduction by merging lines with common properties

The shape of the obstacles and their orientation and also the location error of the

vehicle cause wrong positioning of the lines representing edges of the identified obstacle. The

figure below shows such a case.

Figure 3.24: The phenomenon of edge extraction

To solve this problem of representation it is considered calculating the Euclidean

distance between two line segments by determining the minimum distance between any

elements of the two lines.

Figure 3.26: Substitution of small line segments belonging to the same cluster with complete long line segments

After calculating the distance between the generated lines, an outline data reduction is

made by merging lines with common parameters and belonging to the same cluster. This

provides a data reduction up to 70% of the extraction system features by merging lines

parameters.

Page 34: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

28

3.6.11. Extracting landmarks from the navigation environment

For extracting parts the reduction of the DCE algorithm designed to represent edges of

navigation is used. Completion is described below, with the end points of lines as

initialization statements.

The identification of the same reference points provides in the phi / rho space a

clustered representation of points in the same area.

Figure 3.29: Extracting landmarks from features (lines)

3.6.12. Kalman filter model used for odometry correction

The Kalman filter consists of a system of mathematical equations based on minimizing

the squared errors and is thus an optimal estimator for autonomous vehicle position

correction.

However the question arises: Why just Kalman filter?

Arguments supporting its use in the automation industry refer to:

effective implementation;

estimating the current and possible future states;

measuring of „hidden” states;

measuring the quality of predictions by variances;

robustness, the filter behaves very well in inaccurate models under specific conditions

and is stable

Operation models are based on the following cyclic scheme:

Page 35: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

29

Figure 3.30: Kalman filter estimation and correction

The first phase includes the temporary update in which the state vector is estimated for

the next step. In the next phase an estimation of the state of its first phase is realised by

adapting the current measurement, so that it can be corrected.

The adjustment difference is estimated so that a prediction of the Kalman filter would

reduce the difference between the theoretical and actual curve to a minimum, the system

becomes more accurate.

The estimation and calculation of travel triectory, respectively of the odometry

improvement of the vehicle is shown in Chapter 4 in an application developed on the scaled

autonomous vehicle.

3.6.13. Using an extended Kalman filter for localization

An extended Kalman filter describing a mathematical model for intelligent

autonomous vehicles for position correction by using odometer data and environment

information such as landmarks is described in [95].

The state vector contains the current position of the vehicle condition (xr, yr, Θr), as

well as parts of absolute position of the landmarks in the configuration (xn, yn). The vector

size is 3 +2 n where n is the number of lanmarks

3.6.14. The trajectory planning module „Trajectory planning”

This module realizes an optimal trajectory planning based on the common interface in

order to generate the map. It contains the trajectory calculation based on Bezier curves and the

potential field method.

Page 36: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

30

A number of factors are taken into account in order to establish a robust model, to

extract relevant information from the environment and its use for interpreting the scene, the

trajectory of movement of autonomous vehicle.

The operating environment of an autonomous vehicle can be determined by a

continuous geometric description or through a map. The first step is the transformation of

possible trajectory planning environments into discrete map models to choose an algorithm

suitable for navigation.

Depending on how decomposition affects the environment, path planners can be

cathegorized in three general strategies:

1. Map of trails: in this respect, a number of possible routes in space are identified.

2. Decomposition in the cells (grids): provides a discrimination between occupied and

blank cells to static environments.

3. Potential field: to travel in a ramdom space a mathematical function is required

An autonomous vehicle moving between a home and target location points should be

made having regard to the optimal path representing a path of maximum security without

collision and quickly calculated. Such a requirement is satisfied by the potential field method.

In this sense, the trajectory calculated depends on the environment navigation, previously

identified obstacles and accurate representation. An optimal travel path is not represented only

by calculating the shortest way to go.

Figure 3.33: Vehicle trajectory using the method of potential fields [34]

There are several suggested methods of dealing with local minimum phenomenon. The

occurrence of this phenomenon causes the vehicle to stop and the inability to move forward.

One idea is to avoid local minimum of the potential field is to inlclude an intelligent scheduler

so that the vehicle uses information derived from sensors, but still planning to conduct

comprehensive [27].

3.6.15. Object avoidance module „Track deviation”

This module contains objects avoidance algorithms for safe navigation without

collisions. A comparison between the calculated curve and the curve actually traveled by

vehicle is made.

Avoiding objects focuses on changing the path of travel with information delivered

from sensors. The robot movement is both a function of current sensor data and the relative

position of the target point. A number of algorithms and techniques to avoid objects are

presented in [42]. Simple algorithms that use only current sensor data to avoid collisions are

Page 37: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

31

preferred in this case because of the simplicity and fast response of detection systems based

on current information of the sensors.

A model for such a mechanism is shown below.

Figure 3.35: Obstacle avoidance mechanism

3.6.15.1. Object avoidance using specific curves for trajectory calculation:

The bezier curve is defined by two control points. The other points define the

beginning and end of the curve. Control points determine the curvature of this function.

The integration of the autonomous vehicles path planning involves setting the control points

as the acceleration of the vehicle, thereby realizing the establishment of the optimal curves to

avoid obstacles.

The integration of these points is made for both clothoids and Bezier spline by using a

VP matrix.

(3.51 )

Considering a move with constant speed during the maneuver, the following segment

is calculated. In this case, the segments are segments of a clothoid, that changes its curvature

proportional to length.

0000

0000

0000

0000

3000

14

13

12

11

0000

dxpR

dxpR

dxpR

dxpR

VP

Page 38: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

32

Figure 3.38: Segmentele de curba pentru stabilirea traiectoriei de evitare

3.7. Probabilistic mapping as an alternative method of navigation

There are several advantages of using occupancy grids in terms of achieving real-time

computation, accuracy and robustness depend on the adjustment parameters, namely the

working method. The most important aspects of intelligent autonomous vehicles probabilistic

mapping are detailed as follows.

3.7.1. Bayes method. General aspects

3.7.2. Derivation of the Bayes method

For the derivation of the relation similar steps as described in [124] were made, while

having implementing own specific defined issues.

In the first step, in order to define the measurement, the probability of detecting

objects is realised by using the sensors detection rank.

Tolerated values and measured values are those that determine the probability of

detecting objects as determined in relations.

Mapping using the probability scale is achieved by defining a grid that contains the

probabilities of occupancy cells. The grid is defined as a stationary object moving in this

direction from cell to cell within the grid.

Due to the size dependence of individual cells that influence the intensity of

calculation, the approach is one that concerns only a limited area (50x30 m).

(3.57)

For initialization, each cell is assigned the value 0.5.

When an object is detected, the probability tends to a cell occupied while the blank value

drops to 0. The precision of the method depends on cell size and the probability of detection

of the sensor.

Page 39: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

33

Figure 3.42: Occupacy grid

In principle the two resulting probabilities are occupied or unoccupied defining the

configurations totaling value.

Occupancy grid update reprezents an inflection point for the computing performance

of the system. Basically, after each point generated by the sensor, the probability scale should

be fully updated. In order to achieve a good real-time mapping, the process depends on the

processing power and size of the computational system defined cells, responsible for

positioning the probability value of the objects identified.

3.8. Conclusions

To be successful in integrating methods of mapping and path planning it is not

sufficient to consider only the navigation architecture; important aspects are also computer

technology, control and memory. Overall architectural design is the innovative area of the

future in achieving intelligent vehicles. Algorithms require special attention because of the

limitation of the system which makes the project different from other current approaches. The

purpose of this implementation is the independence from operating systems to achieve a real-

time processing. The embedded knowledge level offers algorithmic robustness and ease of

processing and implementation.

The following steps were made in the development phase and they are representing

both algorithmic and simulation own contributions:

Page 40: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Mapping and localization as simultaneous processes of autonomous vehicles

34

a software environment has been created, that contains a separate module to

generate a virtual map and calculate trajectories of moving non-holonomic

vehicles due to generated data interfaces;

a data reduction was achieved by filtering and selecting consistent data relevant

data by developing algorithms to extract features from the software sensor models;

a Kalman filter concept was integrated allowing the vehicle to localise itself by

extracting relevant parts (landmarks) from the navigation environment;

regarding the strategy level; an interface was developed that allows specifying the

desired travel path and obtain a data set that are actually used in autonomous

vehicle control;

follow a path without obstacles;

implement an obstacle avoidance technique based on special curved segments

(clothoids and Bezier curves);

an alternative mapping concept to the concept of mapping and localization by

extracting features was developed using probabilistic calculation methods.

Because precision is required in autonomous vehicle navigation, the probabilistic

method is recommended only for small spaces. In this case the precision limitation coincides

with the cell size. This implies also a power-performance computing.

The simulation level can provide a rapid development of the concepts studied. In this

respect, opportunities for practical implementation were seeked parallel to the software

evolution.

Page 41: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

35

CHAPTER 4: NAVIGATION TESTS AND APPLICATIONS

OF THE AUTONOMOUS VEHICLE

„Practice is everything“ (Periander)

4.1. The communication infrastructure used for tests conduction

Given the need to optimize the process of mapping and to locate the VIA in a system

the following evaluation was performed in two stages:

1. Simulation using Embedded Matlab and Simulink

2. Experimentation with the 1:8 scaled vehicle

In simulation models of the sensors were used to evaluate methods and system

limitations in terms of achieving further performance of the VIA during the software transfer.

The previously tested algorithms in the simulated environments have been described

previously and were evaluated during the same scenario representing a corridor environment

with objects.

Algorithms running are using a 3.3 GHz processor for the computer simulation

environment on a Windows 7 operating system. The software transfer is done by compiling

the program with functions performed by the Embedded PC of the VIA as a target system.

Communication is a standard wireless 802.11 and of 2.4 GHz. The target is a PC 104

industrial embedded computation unit with the ability to run real-time applications because of

the XPC software from MATLAB.

Figure 4.1: Communication system of the host-target system

WLAN

Host PC Target PC

Scaner laser / senzori

ultrasonici

Embedded Software Emergency

control

Page 42: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

36

Data sharing allows viewing the results in a graphical interface where sensor data are

displayed as points and odometer data of the system in the x and y directions and also the

travel speed and distance.

The virtual map is generated in real-time, therefore algorithms and vehicle control of

the VIA are realized on the target unit while the host computer draws the navigation map

incrementally.

4.2. Comparing data from separate odometer systems

To evaluate data from odometer sensors, measurements were made by comparing two

systems.

Values indicate that the odometer is more accurate trough special sensors. The value

of the inertia sensors contained in GIGABOX considers any inclination of the vehicle during

the navigation.

Figure 4.3: Comparative values of the odometer sensors

4.3. Mapping the environment

The tests with the scaled vehicle were made under laboratory conditions. Obstacles

placed along the route have different sizes, but are similar in overall dimensions. After

conducting several tests in the same scene the sample time was modified from 0.01sec to 0.05

sec to allow a discharge of the process on the XPC computer processor. The odometer module

can provide data with an accuracy of about 500mm to 10m of navigation, without correction.

For speeds above 3 m / s the odometer system errors are correlated with the

performance of the laser scanning wich results in a positioning error of up to 10%.

Page 43: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

37

Figure 4.8: Real time environment mapping in the laboratory environment

4.4. Implementation of the Kalman filter for odometer improving while mapping the

environment

The description of the model in the previous chapter is a general and theoretical one,

based on the concept of autonomous vehicle position prediction. A parameterization of the

quantities used is described in what follows. The vehicles operation is considered in a

cartesian coordinate system with the state vector (x, y, v, Θ) where the positon is given by x,y

and Θ, and the speed vector v is indicated.

Input variables are defined:

dt=0.1 the sample-time coincides with the model adopted in the software

amax=30 experimentally determined

thetamax= 1,57 experimentally determined

State vector is initialised by a zero vector:

[

] (4.1)

The covariance matrix is initialized:

[

] (4.2)

The system matrix is given by:

Page 44: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

38

[

( ) ( )

] (4.3)

Modelul măsurărilor este dat de:

[

] (4.4)

The measurement model is given by:

[ ( (

) ( ))

( (

) ( ))

( ) ]

(4.5)

The measurement noise is:

[

] (4.6)

In the first stage following calculations are made:

Calculation of the predicted state:

xpred = A•xest (4.7)

The covariance matrix becomes:

ppred = A•pest•A‟+Q (4.8)

The measurement update phase is made by these calculations:

The Kalman gain is calculated as:

( ) (4.9)

The estimated state is updated:

xest = xprd+Klmgain• ( z - H• xprd) (4.10)

The above implementation shows defining the next test where the actual trajectory is

callculated in order to close a navigation loop.

With this model the following characteristic problem are solved:

errors in the four-wheel drive model;

odometer errors during wheel slipping;

Page 45: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

39

measurement errors due to sensors measuring the pozition angle;

temperature dependence of the sensors.

Figure 4.10: The difference between the calculated trajectory and real trajectory of the vehicle

The tendency of swinging around the odometer values in order to minimize

positioning error and equaling of the path is observed. To include features, the model is

extended with coordinates measured. Speed and acceleration values are also indicated for the

linearity analysis during the navigation. Differences are observed due to influence of the

electric motor and the control.

4.4.1. Mapping the environment through by extraction of features

The next event is defined for mapping. Extracted features are analyzed by applying the

algorithm described in chapter three.

Figure 4.14: Environment mapping scene (laboratory testing)

Page 46: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

40

The graphical representation of the navigation situation indicates objects identified by

extracted lines.

Figure 4.15: GUI representation of the mapping data

There is a doubled line representation, a phenomenon that appears due to imprecision

of the sensors and their reflections, but also due to the dynamic movement of the vehicle and

thus reviewing these objects while moving. Below are the relevant parameters for the

extraction of features represented in the test environment.

Tabele 4.3: Evaluation parameters for the extraction of features

Algoritm extDCE Measurement unit

Driven distance 4,10 m

Extracted objects 3

No of objects 3

Extracted lines 26

Single points 15

Deviation in x direction 0,05 m

Deviation in y direction 0,02 m

Maximum speed 0.62 m/s

Maximum velocity 17 m/s2

Maximum gyroscopic rate 1,8 °/s

The figures below show the parameter values recorded during tests. Important are

deviations in x and y directions and values that estimate the vehicle position by using the

Kalman filter.

Vehicle speed and acceleration values are also indicated in order to analyse the motion

of the vehicle. There is a signal drop in the calculation of the difference in regulation.

Page 47: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

41

4.4.2. Changing the scanning perspective

A test that is based on the change of perspective representation of objects in space

navigation refers to the situation below. In this, the trajectory is calculated using Bezier

curves and the movement is performed using a system matrix “VP”. This application aims to

extract edges represented in the form of the "L" letter for a correct representation of the edges

of objects as used in the testing scene.

A limitation of the representation is the description of objects through simple lines.

For complex geometries it is possible to overlay contours of objects in order to extract the

specific shape. For the association of lines for objects represented only a part of the object is

required at a time. Similar approaches do not take into account this phenomenon, on the

premise of a full scan of the object within a single step, which implies offline data processing.

To generate lines the extraction algorithm criteria must be met. At least three points are

needed to extract a line, and if they are double associated within the first phase a data

reduction is done while checking common line characteristics in order to fusionate the

segments of the extracted lines.

Figure 4.19: Representation of the maneuver while changeing the scanning perspective

Page 48: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

42

Values recorded during performance of the maneuver are represented below. It is

noted in this case a deviation in x and y directions while changing direction of movement.

Speed and acceleration has an inflection while the transition is made, then their value remains

constant.

Table 4.4: Evaluation parameters for the extraction of features

Algoritm extDCE Measurement unit

Driven distance 4,6 m

Extracted objects 2

No of objects 9

Extracted lines 18

Single points 0,04 m

Deviation in x direction 0,09 m

Deviation in y direction 0.61 m/s

Maximum speed 16 m/s2

Maximum velocity 31 °/s

Maximum gyroscopic rate 50 °

4.5. The visual reprezentation of the navigation environment

4.5.1. Identifying floor markings in specific environments

To initiate a guidance process in the handling space

the following criteria must be met:

Correct identification of the space;

verifying the achievement of the maneuvers in the guidance space;

verifying the availability of instantaneous space;

identification of any objects in the (static / dynamic) area of handling.

Video sensors are an optimal solution in order to detect guided markings for

identifying handled objects. Below the concept of extracting the necessary data is presented

by processing image data (video) for guiding autonomous intelligent vehicle into specific

manipulation spaces. A transformation of the representation in the perspective of the so-called

"bird view" is made. This involves using a transformation model based on the darkroom

operation principle.

By using Matlab / Simulink a live capture of the video

is performed and then converted into gray tones to focus strictly on edge extracting of the

necessary information.

Page 49: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

43

Figure 4.23: Transformation of the gray tones (left) and the extraction of edges (right)

The image matrix is modified by assigning the property to identified edges to 0 or 1,

so edges rezulted from the background extraction is performed on the differential color

criterion (Sobel).

4.5.2. Hough transformation of extracted lines

4.5.3. Transformation of extracted edges and limitation by line atributes

By applying the maximum local function a space is searched for dual Hough

transformation after the start and end points of the domain that is found. This information is

stored in a matrix with theta and rho values. The following figures simulating a navigation

space highlight the calculation.

Figure 4.27: Extracting lines from the hough space and limit definition

Generated points are used by the module interface which realises the navigation based

on the potential field theory. Minimal information is used to achieve a representative real-time

navigation.

4.5.4. Perspective transformation using the darkroom principle

4.5.5. Representation of the transformed markings

Capturile imaginilor au fost realizate în mediul de laborator reprezentând o suprafaţă

netedă cu marcaje predefinite. Camera video a fost montată pe un suport cu 400 [mm]

înălţime cu o înclinare de 35° faţă de suprafaţa de rulare a vehiculului autonom.

Camera perspective is highlighted below. To note is that the floor has sections that

reflect light, which limits the identification of system for continuity, a general problem

discussed in the video processing community. However, defining the system limits, we obtain

a robust possibility of using intelligent autonomous vehicles guided by markings. The

Page 50: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

44

transformation model provides results with errors of measurement in the area [± 4 mm]. The

need to implement a detection system taking in account the angles during acceleration and

braking is important in order to outline the correction of the position during navigation.

Figure 4.31: Perspective transformation of the extracted floor markings

4.6. Analysis of people and objects in the navigation environment

To analyze dynamic objects and people, different approaches are described in [17] or

using dynamic Bayesian networks as described in [83]. Extracting accurate data is

complicated by factors such as occlusion, noise or confusion arising in the background. By

using the optical flow method an apparent motion and model of objects is determined, where

surfaces and margins (edges) in a scene have a relative motion between the observer (camera)

and scene. Specific data indicate dynamic vectors during the the change of direction which are

observed depending on the movement of objects to identify dynamic motion vectors values

different from the rest of the captured scene (background). A distinction is to be made

between the optical flow and the motion field.

Extracting features from the background is performed by using the Lucas-Kanadae

algorithm described in [17]. Dynamic objects extraction is shown below:

The diagram below shows the information flow of the video algorithm:

Figure 4.32: The flow of information necessary for identifying and classifying objects

Page 51: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

45

It is not necessarily to achieve a classification of the objects in motion, a simple

identification of them is sufficient to cause the vehicle to take safety measures regarding the

movement in its active space.

Figure 4.32: Detection of people by background substraction

By background extraction, the computation is performed in two phases. Image noise

plays an important role in this respect and is minimized by filters.

Figure 4.34: Relevant angles for identification

To determine the direction of movement of the border for dynamic calculation, the

difference in directions x and y position it the times t and t-1 are needed.

Height is determined relatively accurately, while the width of the object remains

uncertain. However, the algorithm implemented is fast and robust.

Figure 4.38: Error values of deviation from the real position of the object

0

0,2

0,4

0,6

0,8

1

average position error

distance

height

width

Page 52: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Navigation tests and applications of the autonomous vehicle

46

4.7. Conclusions

The bidimensional (2D) mapping of the environment is a simplified method of

representation of the concepts used for autonomous navigation of intelligent vehicles. On the

specific navigation environment, where objects are identified and represented as standard

shapes, a 2D mapping concept is sufficient to detect obstacles and handle containers. On the

other hand, in terms of autonomous vehicles as independent systems, they have to determine

their own location as a process of simultaneously 2D mapping the environment. For a better

representation of the navigation environment and to have greater control over the tasks which

have to be fulfilled by a vehicle it is necessary to implement additional systems such as video

cameras to detect moving objects. The basic problem is the representation of the environment,

fusioning fata from different sensors to improve the navigation system completely in terms of

accurate mapping and location, where odometer data is not sufficient for effective location.

The data reduction algorithms using simple geometry extraction based on particularities such

as lines is an advantage for specific environments such as manufacturing systems. In this

respect, the development of a 3D scanning system would benefit of graphics views, but not

necessary aplicable in real-time processing tasks of autonomous intelligent vehicles of the

current tehnologies. This problem is directly dependent on computing resources and the

control processing unit. Moving can be improved by identifying markings on the floor and

follow them with visual sensor systems such as the system developed.

The test conducted to the following contributions and implementation steps:

evaluation by comparison of the odometer vehicle sensors available;

improving and correction of the vehicle position by implementation and

parameterization of the programmed Kalman filter;

environment representation by using navigation scenes in the laboratory;

evaluating and identifying features extracted and objects with proximity sensors;

analysis of parameter values on the speed of navigation, positioning and giration rate;

implementation and evaluation of the concept by identifying floor markings using the

video sensor;

implementation and evaluation of a concept for identification of dynamic objects and

people from the navigation environment using the video camera.

Environmental complexity in navigation techniques combines different representation

techniques. Given the task of simplifying the navigation environment as possible, its

representation contains only relevant information for a correct navigation.

Page 53: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

47

CHAPTER 5: APPLICATIONS IN THE FLEXIBLE

MANUFACTURING SYSTEM OF AUTOMOTIVE INDUSTRY

„The value of an idea lies in the fact

that is shoud be valued.” (T.A. Edison)

Within the CAM sistem (Computer Aided Manufacturing) the flexible manufacturing

system (FMS - Flexible Manufacturing System) has a special place. This is defined differently

from country to country, but in essence a production unit is capable of manufacturing a wide

(family) of discrete products with minimal manual intervention. It includes workstations

equipped with high production capacity (numerical control machine tools or other equipment

assembly or treatment) linked by a material handling system for the purpose of moving parts

from one workstation to another and act as an integrated system with fully programmable

control.

PP&C

Capacity

planning and

resource

CAM product manufacturing

CNC/FMS programming

CAQ

Quality-

related

activities stock transport manufacturing assembly

material flow Figure 5.2: The role of the transport in the manufacturing system

Handling flexibility, can be achieved by using an automatic transfer systems and / or

robotic of types: AGVs, orientation and transport systems using automated and robotic

manipulators, automatic transfer lines, warehouses, automated deposits associated with

computer-aided management, capable of finding on new paths for a blocked workstation.

Flexibility in the transfer of a given system is expressed as the ratio between the number of

trajectories that the system can achieve and the number of feasible paths realized by a

universal system.

(5.1)

By program flexibility, virtual factory operates unattended for a long time. This

reduces the total processing time by reducing the time in order to reorganize, introduce some

procedures from which to obtain high quality products and precision and efficiency that leads

to an increased production capacity of the system.

Following are realised in this regard:

a simultaneous improvement of productivity and quality;

a reduction of total time with increasing effective capacity in order to produce

unattended.

An evaluation of the flexibility program can be based on the time gained by

eliminating reorganizations reported to the total processing time.

(5.2)

Page 54: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

48

Compared to product flexibility, production flexibility requires considerable

organization, but not necessarily additional capital investment.

Production flexibility is resulting from the aggregation ability and flexibility of the system of

transfer machines, plus the flexibility of the informational and control system.

5.1. Modeling and simulation of the environment of a six cell flexible manufacturing

system

To study the behavior of the flexible manufacturing system operated by a single

guided vehicle the following case is considered with the scheme associated and the

components described below. The configuration is a generalized model in an automated

system for manufacturing operations, existing in the "University Heilbronn"

Figure 5.8: Manufacturing system layout with six work cells served by VIA

Each cell served by the mobile robot is equipped with a transfer belt that connects the

processing machine tool and the mobile robot platform via the transfer shuttles. Setting the

flow of material is made depending on the production schedule adopted. The autonomous

vehicle is moving on a line with a virtual set priority. The necessary material exchanged

during the stops

Table 5.1: Components of the flexible manufacturing system

Cells operated

by VIA

1 2 3 4 5 6

Automa

tic

deposit

Lathe

cell

Robotic

arm cell

Milling

center

Pressing

cell

Injection cell

Cells operated

by human

7 8 9

CAD

laboratory

CAQ

laboratory

Montaje cell

Page 55: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

49

The navigation is performed by using a laser and ultrasonic sensor system mounted on

the vehicle, without any other outside navigation sources.

Download and delivery of the goods transported is handled by the presence sensor at the end

of the cell lines of working transfers. VIA ensure the functioning of each individual

workstation.

Figure 5.11: Possible tehnologial links

Once slowing the navigatin, objects in the environment are reviewed, and associations

can be made to reduce data. There is an increase in the number of lines reduced while the

travel speeds is low.

Figure 5.15: Representation of the navigation environment at different speeds

speed >10 m/s speed <3 m/s

single points 72 7

reduced lines 4 15

extracted objects 18 44

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Navigation environment at different speeds

Page 56: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

50

The study of the VIA behavior problems in navigation is made considering a 6-cell

system described in an active (CL).

There is a substantial reduction of data while items are reviewd during the navigation.

Basically ~ 20% of the extracted objects are reduced lines. The number of unrelated items is

determined either by the sensitivity of the region of interest in that the vehicle scans, or its

displacement effect caused by the nonlinear dynamic.

Figure 5.18: VIA operating the active cells

The particularities extracted from the environment are measured differently from each

transfer made to highlight associations, namely the number of unassigned points in the

scanned environment.

Figure 5.19: Environment representation for navigation on the transfer application

The observed increase in the number of points is unrelated to a maximum during

transfer three parts corresponding to the acquisition of the injection cell in order to transfer

them in the automatically deposit. There is also a large number of lines reduced due

reobservation of the edges by performing specific maneuvers.

transfer1 transfer2 transfer3 transfer4 transfer5 transfer6

single points 12 18 25 34 29 22

reduced lines 4 7 16 12 8 5

extracted objects 15 12 32 28 18 21

0%

20%

40%

60%

80%

100%

Navigaton environment

Page 57: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

51

Navigation times are represented comparatively below.

Figure 5.20: Transfer times

For the calculation of the operating time following technological parameters are taken

into account, valid for the six cells activities.

Figure 5.21: Operarion times

Transfer time is approximately 43% of the time required to achieve the complete

operation. To improve this time of transfer, improvements are needed at the strategic level.

5.1.1. Modeling and simulation of the application with occupacy grids

An alternative approach to represent the navigation environment is studied while using

occupacy grids for modeling the environment after the concept developed in section 3.7.

The cells are defined as having dimensions of 1x1 m. The detection probability is

analysed. The figure below shows the occupancy grid in global coordinates. A cell can

navigation time(s)

020

40

60

80

transfer1transfer2 transfer3 transfer4transfer5

transfer6

transfer1 transfer2 transfer3 transfer4 transfer5 transfer6

navigation time(s) 42 34 51 46 36 61

Navigation time during transfers

0

500

1000

time in s

time in s

operation time of the VIA 270

total operation times 627

Operation time of the VIA

Page 58: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

52

contain multiple points resulting from detection of objects by the laser scanner. In this data

reduction is achieved by establishing the cell size. It also influences the accuracy of detection

of objects regarding their location in space.

Figure 5.22: Environment representation using grids

Given the size of cells, they can be processed faster when their dimensions are large

defined, a division of cells in small cells often require an intensive computation beyond the

computer system.

For this case we have a 50x30 cell field, which takes 1500 calculations for each

measurement that requires an update.

Environment has the following characteristics:

measuring points of the laser scanner defined by the scanning area;

occupied cells containing points of measurement and distances;

updated free cells.

Figure 5.23: Navigation features of the environment

A relatively low occupancy rate is observed, only 10.8% of the cells are used. The

tolerance range and precision of the laser scanner assigns values to neighboring cells. The

0

500

1000

1500

environment cells

spare cells 1338

occupied cells 162

measured values 972

Navigation characteristics

Page 59: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Applications in the flexible manufacturing system of automotive industry

53

representation error for determining the minimum distance between the vehicle and an

occupied cell is possible to determine only by calculating the route on free cells.

5.2. Conclusions

The benefits of flexibility are sought in each production system, this is the current

manufacturing strategy. But its feasibility is proved only under complete control of

technology.

Plants and processes are too expensive to be rebuilt every time a change is required for

the production design. In this respect they are required to be flexible and easily configurable.

For the reconfiguration processes a direct control over the elements is necessary. For this it is

recommended the configuration to have a strong and stable communication network covering

the needs of operating machines and operators. Centralized production is component of past,

current attempts are providing reliable breakdown of activities of individual systems while the

processing category also includes transfer vehicles.

The application developed is detailing key points of the researched and implemented

sistem with contribution in:

analysis of the flow of raw material handling and transport of a Flexible

Manufacturing System;

modeling and integration of the manufacturing software system developed to analyze

the concept of navigation;

representation of the navigation environment during displacement while analysing

features extracted in order to identify the correctness of objects placements in specific

environments;

analysis of transfer times between active work cells;

analysis of specific phenomena that occur in the navigation of the flexible

manufacturing system;

studying the alternative abordation for representation of the environment with

occupancy cells presenting advantages and disadvantages of use.

The representation and interpretation of extracted data from the navigation

environment is an important component for a vehicle to reach its autonomy. Although there is

a limitation in terms of space navigation, extracted features provide real-time navigation

through simple algorithmic processing of features extracted from the environment.

Based on this concept, the environment and vehicle operating in these components are

interconnected throughout the navigation software, without a visible physical connection,

talking from this point of view of a flexible autonomous transfer system.

Page 60: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Conclusions and further research

54

CHAPTER 6: CONCLUSIONS AND

FURTHER RESEARCH

The possibility of a vehicle to move independently in a structured or unstructured

environment either internal or external, to achieve a virtual map and to to locate itself in it are

important aspects of the development of navigation systems used in the manufacturing

industry and in other areas involving the transport.

The research within the project highlighted in this thesis the implemented concept of

autonomous vehicle navigation in two stages:

1. Creating a software concept that allows modeling and simulation of algorithms

developed using the sensor behavior models;

2. Realisation of a scaled autonomous vehicle, allowing testing and implementation

of the developed concepts.

The first phase of the research includes an analysis of the possibilities of implementing

the concept of mapping and self localization (SLAM). Algorithms have been developed,

particularry extraction algorithms by extending the existing DCE concept to simplify the

representation during the vehicles navigation. The necessity of localization led to examination

the existing possibilities and to the development of a concept using a quasi Hough space for

extraction and analysis of identified landmarks. The alternative mapping approach and

alanysis of the grid cells revealed the possibility of implementing probabilistic calculation

methods. With the prerequisite for achieving a real-time processing, computer algorithms

were programmed in effective code, thereby maximizing the opportunities for their

implementation on dedicated computer systems (embedded). Through this the development of

an independent programming and computing platform was ensured.By using vehicle systems

without an additional operating system form the category Windows / Linux or external data

libraries, the implicit development costs are low. The research is trough this fact different

from many current approaches. The results and contributions of the first development stage

are presented below.

Theoretical contributions

study the possibility of further development of autonomous vehicles based on current

research principles and industrial handling vehicles;

establish a navigation concept based on SLAM tehniques by extracting features for

recognizing the environment both internally and externally;

develop a proper algorithm for extracting lines and data reduction resulting in

implementation of a virtual map;

analysis and benchmarking of algorithms available to validate the concept;;

development of a method for extracting landmarks for the vehicle location

determination using a quasi Hough space;

kalman filter modeling for position correction of an autonomous vehicle and

integrating features while exploring the environment;

Page 61: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Conclusions and further research

55

developing a method to avoid obstacles using a matrix with specific curve calculation

using the avialable process infrastructure;

research and modeling of an alternative mapping method based on probabilistic

methods for mapping;

a study on the applicability and implementation of methods investigated;

development of a concept of people and objects representation in the dynamic

navigation environment;

createing a navigation concept based on identifying markings in dedicated spaces;

behaviour analysis of the vehicle during its navigation in a manufacturing space based

on two concepts of representation of the environment.

In the second stage the theoretical research was performed by the implementations. By

creating a physical prototype, an important step in analyzing the behavior of both concepts

and software developed was made. By establishing a communication concept based on the

target-host technology an important component was developed in terms of use and monitoring

of the vehicle during its operation. As the odometer has proven to be imperfect, a Kalman

filter was implemented to correct the vehicles position during navigation without any

additional systems such as GPS, system that is limited by indoor applications. By equipping

the vehicle with a different architecture of sensors in order to remove environmental

dependencies the research is also representing an original model of the trend in the current

approach to automated systems. Differences between theoretical and practical approaches

showed a clear difference both in terms of development and level of implementation, often

requiring changes in the software, therefore, changes in the first stage of development. In

these assessments practical implementations and own contributions are outlined below:

Practical contributions

evaluation and testing of components necessary to achieve an autonomous intelligent

vehicle infrastructure;

development of a prototype of an autonomous vehicle with a limitated processing

system for embedded programming in order to navigate in real time;

development of a communication concept between the vehicle and a central control

system for active monitoring of processes;

development of communication concept between two PC-104 systems by processing

and synchronizing the flow of information;

programming of a simulation in Matlab/Simulink environment with embedded

integration for modeling and rapid prototyping of navigation algorithms.

integration of the theoretical concept on mapping and localization of autonomous

vehicles in a specific environment;

implementing an obstacle avoidance technique based on special curved segments

(clothoid and Bezier splines);

the optical research level was achieved by creating a graphical representation of the

results of the mapping algorithm in order to create virtual maps;

Page 62: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Conclusions and further research

56

implementation of a Kalman filter for correcting the position of the vehicle during

autonomous navigation;

implementation of a concept that identifies persons and objects for dynamic and safe

navigation trough the environment;

implementation of the concept for floor markings in dedicated spaces using a video

camera, required for navigation maneuvers.

By completing the two steps, a realisation of an optimal configuration for the

development of autonomous navigation systems is achieved. The complexity of the project

had the disadvantage that certain aspects of navigation were addressed only tangentially. By

this it was pointed out that the usefulness of these systems depends on the specific areas of

application.

Each chapter presents the final conclusions and original contributions.

Through modeling and simulating an application in a flexible manufacturing system

certain aspects that can be taken into account in order to increase its total flexibility have been

highlighted. This refers to both, the handling within a manufacturing system, but also the

possibility to that merge area with the logistics field.

Further research directions

Future research directions in this area can rely on the sayings of P. Drucker: „We must

become managers of technology, not just user of it”.

If expectations in recent years have not yet brought the desired smart products on the

market, sooner or later they will appear. The confluence of advanced technology will bring

new possibilities closer to reality with practical features, small, effective and of low cost. The

artificial intelligence, the robotic structures and robustness issues will become indispensable

in any field. Looking closely, the future in the development of the autonomous systems in

flexible manufacturing depends on the applications and technologies used.

A diagnosis by computer will allow estimating possible errors of the machine

guidance system and sustaining teams in detecting faults. International coordination and

control of manufacturing facilities by advanced communication technology is a clear line of

research manufacturing systems. Using robots equipped with different sensors to simplify the

tasks of production, and also to estimate the process in simulations are growing and are also

essential components of future plants.

All forms of technological progress from the invention of new sensors to

microprocessors is growing at a significantly rate and provides new concepts in mobile robot

architectures, so unimaginable tactics will become possible to apply.

With reference to the process-product matrix, one can achieve a high flexibility

through innovative technical level and organizational efforts. The Volvo company for

example is based on a manufacturing process in which machines are assembled on mobile

pallets and there is no assembly line in this respect. The process incorporates flexibility.

At the present time, small versionsof flexible manufacturing systems impose. Due to

relatively small production volumes and large variety of applications, automation technology

developed also a largely used segment for other market segments. Innovation comes rather

Page 63: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Conclusions and further research

57

from the existing applications in different areas than form a totally unexplored concept. Using

mapping technologies for mobile robot navigation in specific environments for the transfer in

the field of industrial automation becomes a real-time application of complex adaptive

systems.

Page 64: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Selective bibliography

58

SELECTIVE BIBLIOGRAPHY

[8] Becker J., Fusion der objekterkennenden Sensoren eines autonomes

Fahrzeugs,Workshop Multisensorsysteme für die Exploration natürlicher

Umgebungen im Rahmen der Deutschen Jahrestagung für Künstliche Intelligenz

und des 21. Symposiums für Mustererkennung, Univestität Bonn, 1999

[12] Borenstein J., Experimental results from internal odometry error correction with

the OmniMate mobile robot, IEEE Transactions on Robotics and Automation, vol.

14, pp. 963–969, 1998.

[17] Bouguet J.Y., Pyramidal Implementation of the Lucas Kanade Feature Tracker –

Description of the algorithm, Intel Corporation – Microprocessor Research Labs.

[18] Carpenter R. N., Concurrent mapping and localization with FLS, Proceedings of

the Workshop on Autonomous Underwater Vehicles, pp. 133–148, Cambridge,

MA, USA, 1998.

[19] Castellanos J. A., Tardos J.D., Mobile Robot Localization and Map Building, A

Multisensor Fusion Approach, Kluwer Academic Publishers, Boston, 1999.

[20] Castellanos J.A., Tardos J.D., Schmidt G., Building a Global Map of the

Environment of a Mobile Robot: The importance of corellations, Proceedings of

the IEEE Conference on Robotics and Automation, Albuquerque, NM, 1997.

[21] Cemgil A.T., Ben W. Z., Krose J. A., A Hybrid Graphical Model for Robust

Feature Extraction from Video, CVPR 2005

[22] Choset H., Nagatani K., Topological simultaneous localization and mapping

(SLAM): toward exact localization without explicit localization, IEEE Transactions

on Robotic and Automation, pp. 125–137, 2001.

[23] Cox I.J., Blanche , An experiment in guidance and navigation of an autonomous

robot vehicle, IEEE Transactions on Robotics and Automation, pp. 193–204,

1991.

[27] Egbelu P. J. , Tanchoco J. M. A., Potentials for bi-directional guided-path for AGV

based systems, International Journal of Production Research 24, pp. 1075-1097,

1986.

[28] Escalera A. de la, Moreno L., Salichs M.A., Armigol J.M., Continuous mobile

robot localization using structured light and a geometric map. International Journal

of Systems Science, 1996.

[29] Elfes A., Sonar-based real-world mapping and navigation, IEEE Journal of

Robotics and Automation, pp. 249–265, 1987

[30] Eynan A., Rosenblatt M. J, An interleaving policy in automated storage/retrieval

systems, International Journal of Production Research, pp. 1-18, 2003.

[31] Falcone E., Gockley R., Porter E., Nourkbash. I.,The personal Rover Project: The

comprehensive Design of a domestic Personal Robot, Robotics and Autonomous

Systems, Special Issue on Socially Interactive Robots pp. 245-258, 2003

[32] Fischler M.A., Bolles R.C., Random sample consensus: a paradigm for model

fitting with applications to image analysis and automated cartography.

Communications of the ACM, pp. 381–395, 1981.

[33] Gall R., Tröster F., Luca R., On the development of an embedded system for an

autonomous mobile robot, Proceedings of 2010 IEEE International Conference on

Embedded Systems, Las Vegas, USA, ISBN: 574-550-3345-0589-1, 2010.

Page 65: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Selective bibliography

59

[34] Gall R., Tröster F., Luca, R., Mogan, G., Building an experimental car-like mobile

robot, Proceedings of 2010 IEEE International Conference on Automation,

Shanghai, China, 2010.

[41] Hostetler L.D., Andreas R.D., Nonlinear Kalman ltering techniques for terrain-

aided navigation. IEEE Transactions on Automatic Control, March 1983.

[42] Honey S.K., White M.S., Cartographic databases. In I.J. Cox and G.T. Wilfon,

editors, Autonomous Robot Vehicles, pp 250-258. Springer-Verlag, 1990.

[43] Hu M.K., Visual pattern recognition by moment invariants, IEEE Transactions on

Information Theory, vol. 8, no. 2, pp. 179–187, 1962.

[44] Jensfelt P., Approaches to Mobile Robot Localization in Indoor Environments.

PhD thesis, Signal, Sensors and Systems (S3), Royal Institute of Technology,

Stockholm, Sweden, 2001.

[45] Jennings J., Kirkwood-Watts C, Tanis C., Distributed Map-making and Navigation

in Dynamic Environments, Proceedings of the 1998 IEEE/RSJ International

Conference of Intelligent Robots and Systems(IROS 98), Victoria, B.C., Canada,

October 1998.

[46] Karan M., Gupta, B.E., Monte Carlo Localization for robots using dynamically

expanding occupancy grids, Tech University Texas, 2005.

[47] Khatib O., Real-time Obstacle Avoidance for Manipulators and Mobile Robots,

Proceedings of the IEEE International Conference on Robotics & Automation,

pp.500-505, 1985.

[53] Latecki L. J., Lakämper R., Convexity rule for shape decomposition based on

discrete contour evolution, Computer Vision and Image Understanding Vol. 73,

No. 3, March, pp. 441–454, 1999.

[54] Latombe J-C., Robot motion planning. Norwood, MA, Kluwer Academic

Publishers, 1991.

[57] Lee D., The Map Building and Exploration Strategies of a simple Sonar Equipped

Mobile Robot, Cambridge, UK, Cambridge University Press, 1996

[58] Leonard J.E., Durrant-Whyte H., Mobile robot localization by tracking geometric

beacons. IEEE Transactions on Robotics and Automation, 1991.

[59] Leonard J.E., Durrant-Whyte H, Directed Sonar Sensing for Mobile Robot

Navigation. Norwood, MA, Kluwer Academic Publishers, 1992

[63] Luca R., Troester F., Simion C., Gall R., Data merging and sorting method based

on Discrete Contour Evolution with application on SLAM, Annals of DAAAM for

2009 & Proceedings of the 20th Symposium “Intelligent Manufacturing &

Automation: Focus on Theory, Practice and Education, ISBN 978-3-901509-70-4,

pp 253-254 ,Vienna, Austria, 2009.

[64] Luca R., Tröster F., Simion, C., Gall R., Research on autonomous vehicle

systems, Proceedings of the 4th

International Conference on Manufacturing Science

and Education, vol.1, ISSN 1843-2522 , (pp 51-54) Sibiu, Romania, 2009.

[65] Luca R., Tröster F., Gall R., Simion C., Environment mapping for autonomous

driving into parking lots“, Proceedings of 2010 IEEE International Conference on

Automation, Quality and Testing, Robotics (AQTR 2010), Cluj-Napoca, Romania,

ISBN: 978-1-4244-6722-8, pp.153-158, 2010.

[66] Luca R., Tröster F., Gall R., Simion C. Feature based mapping procedure with

application on Simultaneous Localization and Mapping (SLAM), Robotics and

Automation Systems, Cluj-Napoca, Romania ISBN-13 978-3-908451-88-4, Solid

State Phenomena Vols. 166-167 (2010), pp. 265-270, 2010.

Page 66: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Selective bibliography

60

[67] Luca R., Tröster F., Gall R., Simion C., Autonomous parking procedures using

ultrasonic sensors“, The 21st DAAAM International Symposium Intelligent

Manufacturing & Automation, Annals of DAAAM for 2010 & Proceedings, Zadar,

Croatia, ISBN 978-3-901509-73-5, pp. 691-692, 2010.

[68] Luca R., Tröster F., Gall R., Environment exploration for autonomous driving into

parking lots, Research and Education in Mechatronics, Heilbronn, Germany, ISBN

978-3-00-031548-0, pp.43-48, 2010.

[69] Luca R., Simion C., Tröster F., Gall R., Rapid prototyping and evaluation of

vehicle platform for specific environment navigation. Academic journal of

manufacturing engineering, Timişoara, Romania ISSN, 1583-7904, 2011.

[75] Pfister S.T., Roumeliotis S.I., Burdick J.W., Weighted line fitting algoritms for

mobile robot map building and efficient data representation, Robotics and

Automation, 2003. Proceedings. ICRA '03, IEEE International Conference, 2003.

[76] Poovendran R., Speigle S., Srinivasan S., Raghavan S., Chellappa R.. Qualitative

landmark recognition using visual cues, Proceedings of SPIE - The International

Society for Optical Engineering, pages 74-83, Orlando, FL, April 1997.

[77] Pozna C., Troester F., Autovehiculul autonom - studiu de caz, Editura Universităţii

Transilvania Brasov, 2006.

[78] Ribeiro M. I.; Lima P., Ocuppancy Grid Maps, Institute for Systems and Robotics,

2008.

[79] Ribas D., Towards Simultaneous Localization & Mapping for an AUV using an

Imaging Sonar, Girona, 2005.

[80] Richter B., Identifizierung und Klassifizierung dynamischer objekte auf ein

Parkplatzgelţnde anhand von Videosensoren, Heilbronn University, 2011.

[81] Russel S., Norvig P., Artificial Intelligence, a Modern Approach, Prentice Hall

International 1995.

[82] Sebastian T., Learning Occupancy Grid Maps With Forward Sensor Models,

School of Computer Science, Carnegie Mellon University, 2003

[83] Siegwart R., Nourbakhsh I.R., Introduction to Autonomous Mobile Robots, The

MIT Press. Massachusetts, ISBN 978-0-262-19502-7, 2004.

[84] Siegwart R., Nourbakhsh I.R., Scaramuzza D., Introduction to Autonomous Mobile

Robots, Second Edition, The MIT Press. Massachusetts, ISBN 978-0-262-01535-6,

2011.

[85] Siefert R., Woerner S., Erkennung von Bodenmarkierungen, Heilbronn University,

2010.

[86] Squires M.D., Whalen M., Moody G., Jacobus C., Taylor M., Real-time landmark

based optical vehicle self-location. In Proceedings of SPIE - The International

Society for Optical Engineering, pp. 187-197, Orlando, 1996.

[87] Simhon, S., Dudek. G., A global Topological Map Formed by Local Metric Maps,

„Proceedings of the 1998 IEEE/RSJ International Conference of Intelligent Robots

and Systems, Victoria, B.C., Canada, 1998.

[88] Spinelli J. J., The effects of load/unload times and networkingzoning on an AGV

system, Master's Thesis, Department of Industrial and Management Systems

Engineering, Pennsylvania State University, 1997.

[89] Stachniss C., Robotic Mapping and Exploration, Springer Verlag, ISBN-13

97836420110965, 2009.

[98] Vandorpe J., Brussel H. Xu, H. V., Aertbelien E., Positioning of the mobile robot

LIAS using natural landmarks and a 2d rangefnder, Proceedings of the IEEE

International Conference on Multisensor Fusion and Integration for Intelligent

Systems, pp. 257-264,Washington, DC, 1996.

Page 67: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

Selective bibliography

61

[103] Zeiger F., Schilling K., Design of an User Interface for the Coordination of a

Group of Mobile Robots, The 17th

International Symposium on Robot and Human

Interactive Communication, Munich, Germany 2008.

INTERENT PAGES

[104] ***http://www.nauticexpo.de/cat/handelshafen-krane-portalkrane

portalhubwagen/containerterminals-fahrerlose-transportfahrzeuge-ftf-BB-

1245.html

[105] ***http://www.frog.nl/dui/companyinfo/inside/background/agv/agv.html

[106] ***http://www.darpa.mil/grandchallenge/index.asp

[107] ***http://manufuture.de

[108] ***http://www.vdi.de

[109] ***http://www.egenimusa.com

[115] *** http://www.simsol.co.uk/factoryFLOW_manufacturing.php

[116] *** http://www.fraunhofer.de/en/

[117] *** http://www.kaercher.de

[118] ***http://www.solovatsoft.com/case_study_inventory%20control_system.html

[119] *** http://www.proplanner.com/index.cfm?nodeID=25751&audienceID=1

[120] *** http://distrinet.cs.kuleuven.be/software/agentwise/agvsimulator/#gui

[121] ***http://www.microsonic.de

[122] ***http://www.gigatronik.de

[123] ***http://www.microsoft.com

[124] ***http://en.wikipedia.org/wiki/Bayes'_theorem

[125] ***http://mobilerobots.org

Page 68: PHD THESIS -SUMMARY- CONTRIBUTIONS ON AUTONOMOUS …€¦ · Sibiu, 2011 . Universitatea Lucian Blaga Sibiu Invest in people! PROJECT FINANCED BY THE EUROPEAN SOCIAL FUND Project

INDEX OF ABBREVIATIONS

AGV EN: automated guided vehicle

RO: vehicul ghidat automat

MAP

EN: Manufacturing Automation Protocol

RO: protocolul fabricaţiei automate

ASC

EN: writing/reading device

RO: aparat scriere/citire

ASM

EN: initialization module

RO: modul de initializare

SPS

EN: programable logic controller

RO: modul logic de programare

ROI

EN: region of interest

RO: regiune de interes

CAN EN: control area network

RO: reţea de control

SLAM EN: simultaneous localization and mapping

RO: cartografiere şi localizare simultană

CNC

EN: numerical comand control

RO:control cu comandă numerică

TR

EN: transfer rate

RO: rata de transfer

DARPA

EN: Defense advanced research project agency

RO: agenţia de proiect de cercetare în apărare

avansată

UDP

EN:user datagram protocol

RO:protocol de pachete

DCE

EN: discrete contour evolution

RO: evoluţia discretă a conturului

VIA

EN: intelligent autonomous vehicle

RO: vehicul inteligent autonom

DARPA

EN: Defense advanced research project agency

RO: agenţia de proiect de cercetare în apărare

avansată

VP

EN: control matrix

RO: matricea de control

extDCE:

EN: extended discrete contour evolution

RO: evoluţia discrete a conturului extinsă

GUI

EN: graphical user interface

RO: interfaţă grafică de utilizator

IP

EN: internet protocoll

RO: protocol de internet

I/O

EN: input/output

RO: intrări/ieşiri