Outdoor Navigation: Time-critical Motion Planning for Nonholonomic Mobile Robots Mohd Sani Mohamad Hashim

Transcription

1 Oudoor Navigaion: Time-criical Moion Planning for Nonholonomic Mobile Robos Mohd Sani Mohamad Hashim School of Mechanical Engineering The Universiy of Adelaide Souh Ausralia 55 Ausralia A hesis submied in fulfilmen of he requiremens for he degree of Docor of Philosophy in Mechanical Engineering on February 14

2 Table of Conens TABLE OF CONTENTS TABLE OF CONTENTS...i LIST OF FIGURES...iv LIST OF TABLES... x ABSTRACT...xi STATE OF ORIGINALITY...xii PUBLICATIONS...xiii ACKNOWLEDGEMENTS... xv 1. INTRODUCTION MOTIVATION RESEARCH AIMS LAYOUT OF THESIS LITERATURE REVIEW MOTION PLANNING ALGORITHMS Roadmap pah planning Cell decomposiion pah planning Poenial field pah planning Oher pah planning approaches Geomeric approach for rajecory planning.... NAVIGATION ENVIRONMENTS Oudoor navigaion OBSTACLE AVOIDANCE MULTIPLE ROBOTS COORDINATION SUMMARY AND GAP STATEMENT METHODOLOGY STAGE 1: DEVELOPMENT OF ALGORITHMS FOR TIME-CRITICAL MOTION PLANNING STAGE : OBSTACLE AVOIDANCE APPROACH STAGE 3: SIMULATION WORKS STAGE 4: HARDWARE PREPARATION AND EXPERIMENTAL WORKS CONCLUDING REMARKS DEVELOPMENT OF TIME-CRITICAL MOTION PLANNING ALGORITHMS KINEMATIC MODEL OF NONHOLONOMIC MOBILE ROBOT BOUNDARY CONDITIONS i

3 Table of Conens 4.3 COORDINATE-X EQUATION COORDINATE-Y EQUATION ORIENTATION (θ ) EQUATION STEERING ANGLE (φ ) EQUATION ANGULAR VELOCITY ( u 1) EQUATION OBSTACLE AVOIDANCE APPROACH Avoiding saic obsacles Avoiding moving obsacles CONCLUDING REMARKS SIMULATION RESULTS AND DISCUSSIONS SIMULATION ARCHITECTURE SIMULATED VEHICLE MATLAB FRAMEWORKS TRAJECTORY OPTIMIZATION Replanning approach SIMULATION RESULTS AND DISCUSSIONS Navigaion in saic and open-space environmens Navigaion in dynamic and open-space environmens Navigaion in he ciy-like environmens CONCLUDING REMARKS DEVELOPMENT OF A NONHOLONOMIC MOBILE ROBOT ROBOT CONTROLLER WHEEL ENCODER DETECTION SENSORS COMMUNICATION CALIBRATION OF STEERING ANGLE AND VELOCITY Seering angle Velociy OBSTACLE DETECTION WIRELESS COMMUNICATION CONCLUDING REMARKS EXPERIMENTAL RESULTS AND DISCUSSIONS EXPERIMENT ARCHITECTURE EXPERIMENT SETUP CASE 1: NAVIGATION IN AN OBSTACLE-FREE ENVIRONMENT CASE : NAVIGATION IN A KNOWN STATIC ENVIRONMENT CASE 3: NAVIGATION IN AN UNKNOWN STATIC ENVIRONMENT ii

4 Table of Conens Scenario 1: One unknown saic obsacle Scenario : Two unknown saic obsacles CASE 4: NAVIGATION IN AN UNKNOWN DYNAMIC ENVIRONMENTS Scenario 1: Opposie direcion of mobile robo Scenario : From lef-hand side of mobile robo Scenario 3: From righ-hand side of mobile robo CONCLUDING REMARKS CONCLUSIONS AND FUTURE WORKS CONTRIBUTIONS FUTURE WORKS REFERENCE APPENDIX A APPENDIX B iii

5 Lis of Figures LIST OF FIGURES Figure.1 Pah generaion (a) Pah consrains made of four required posures (b) Generaed pah (Delingee e al., 1991)....8 Figure. Roadmap approach (a) Visibiliy Graph (Jiang e al., 1997). (b) Voronoi diagram (Siegwar and Nourbakhsh, 4)...9 Figure.3 Cell decomposiion mehod (a) A fixed-resoluion grid. (b) A riangulaion (Ge and Lewis, 6) Figure.4 Simulaion resuls by using (a) rapezoidal decomposiion and (b) riangular decomposiion (Ghia and Kloezer, 1)....1 Figure.5 Poenial field mehod (Safadi, 7)...13 Figure.6 Pah generaed by he navigaion algorihm (Cosio and Casaneda, 4) Figure.7 Implemenaion of he proposed algorihm by Koh and Cho (Koh and Cho, 1999)...17 Figure.8 Resuls from he informaion- based mehod (Mihaylova e al., 3)...18 Figure.9 Cell mapping model (a) wih 3 5 cells. (b) wih 4 5 cells (Li and Wang, 3)...18 Figure.1 (a) Generaed rajecory (b) Velociy profile (c) Acceleraion profile (Prado e al., 3)...19 Figure.11 Neuro-fuzzy approach (Hui e al., 6)... Figure.1 Generaed rajecory wih several conrol poins (Haddad e al., 7)....1 Figure.13 Simulaion resuls in (a) a complex scenario, and (b) a long corridor (Ma e al., 13)...1 Figure.14 Differen ypes of curves used o connec four posures for pah generaion (Shin and Singh, 199)... Figure.15 An opimal pah (a) minimum energy, (b) minimum ravel disance, and (c) minimum ravel ime (Liu and Sun, 11)....5 Figure.16 Oudoor navigaion (a) Pioneer3-AT wih URG and SICK (Chang e al., 9) (b) The Cycab used in he experimenal works (Zhang e al., 6)....6 iv

6 Lis of Figures Figure.17 Plan view of he observer moving in dynamic environmen (a) Exocenric reference frame (b) Egocenric reference frame (Fajen and Warren, 3)....8 Figure.18 Avoiding a dynamic obsacle (Jolly e al., 8)...3 Figure.19 A group of robos in huning operaion (Yamaguchi, 3)...31 Figure. Subasks of consrucion ask (Sroupe e al., 5) Figure.1 Overview of he sysem (Klancar e al., 4)...3 Figure 3.1 Sages for proposed mehodology...35 Figure 3. Generalized seps for avoiding an obsacle...37 Figure 3.3 The modified mobile robo used in he experimenal works...38 Figure 4.1 Flowchar of he proposed algorihms...41 Figure 4. A car-like mobile robo...4 Figure 4.3 Avoiding a deeced saic obsacle which is unknown in priori...5 Figure 4.4 Avoiding a moving obsacle (a) perpendicular direcion o he mobile robo and (b) in opposiion o he mobile robo....5 Figure 4.5 Collision predicion approach (a) before deecion of he obsacle, (b) firs deecion, (c) prediced posiion falls inside he collision radius, and (d) obsacle avoidance approach implemened Figure 5.1 Simulaion process flowchar Figure 5. Geomeric model of a mobile robo...59 Figure 5.3 Simulaed Laser Range Finder Figure 5.4 Simulaion map wih saic and moving obsacles....6 Figure 5.5 The Graphical User Inerface (GUI) for simulaion framework (a) Inpu GUI, (b) Oupu GUI...63 Figure 5.6 Original rajecory plan...64 Figure 5.7 Final resul of he rajecory Figure 5.9 Seering angle profiles (a) Planned seering angle (red line) agains adjused seering angle (red dashed), and (b) adjused seering angle (red dashed) agains acual seering angle (blue line)...66 Figure 5.1 Velociy profiles (a) Planned velociy (red line) agains adjused velociy (red dashed), and (b) adjused velociy (red dashed) agains acual velociy (blue line)...67 Figure 5.11 Adjused rajecory (red dashed) agains acual rajecory (blue line)...68 Figure 5.1 Prior map wih wo waypoins connecing he iniial and final poin...7 v

7 Lis of Figures Figure 5.13 Simulaion resuls wih replanning approach Figure 5.14 A complicaed obsruced environmen...73 Figure 5.15 One mobile robo navigaes in he environmen...75 Figure 5.16 Robo 1: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) locaion...76 Figure 5.17 Two mobile robos navigae in he environmen...78 Figure 5.18 Robo : Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion...79 Figure 5.19 Three mobile robos navigae in he environmen...81 Figure 5. Robo 3: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion...8 Figure 5.1 Simulaed environmen for Case Figure 5. One mobile robo navigaes in a dynamic environmen...85 Figure 5.3 Robo 1: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion...86 Figure 5.4 Simulaed environmen for Case Figure 5.5 Two mobile robos navigae in a dynamic environmen Figure 5.6 Robo : Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion...89 Figure 5.7 Simulaed environmen for Case Figure 5.8 Three mobile robos navigae in a dynamic environmen...91 Figure 5.9 Robo 3: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion...9 Figure 5.3 (a) A simplified ciy-like map, (b) Muliple waypoins rajecory planning...95 Figure 5.31 Iniial rajecories in a ciy-like map...96 Figure 5.3 (a) Before deecing an obsacle. (b) Obsacle deeced a he 9 h second. (c) Sars o move along new rajecory. (d) Reaches he firs waypoin a he 3 h second...97 Figure 5.33 (a) Before deecing an obsacle. (b) Obsacle deeced a he 67 h sec. (c) Sars o move along new rajecory. (d) Passes hrough moving obsacle safely...98 vi

8 Lis of Figures Figure 5.34 (a) Before deecing an obsacle. (b) Obsacle deeced a he 68 h sec. (c) Sars o move along new rajecory. (d) Passes hrough moving obsacle safely...99 Figure 5.35 Final resul a he 1 h second...1 Figure 5.36 Second scenario wih wo mobile robos and one moving obsacle...11 Figure 5.37 Final resul a he 1 h second for second scenario...1 Figure 5.38 Third scenario wih hree mobile robos and wo moving obsacles...14 Figure 5.39 Final resul a 1 h second for hird scenario Figure 6.1 The modified car-like robo used in experimenal works...18 Figure 6. Mobile robo plaform Figure 6.3 Sensor plaform...19 Figure 6.4 Sensor plaform aached o he mobile robo plaform...11 Figure 6.5 Robo conroller Figure 6.6 (a) Magnes mouning aached a he wheel (b) Hall Effec sensors aached a he rear axle...11 Figure 6.7 Hall effec sensor...11 Figure 6.8 Locaion of he wheel encoder Figure 6.9 Magne mouning of encoder Figure 6.1 (a) Ulrasonic range sensors (b) Sensor aached o he sensor base Figure 6.11 Wireless communicaion (a) Rouer (b) Coordinaor Figure 6.1 Calibraion work for esablishmen of seering angle Figure 6.13 Relaion beween PWM values and seering angle Figure 6.14 Calibraion work for esablishmen of velociy Figure 6.15 Relaion beween PWM values and speed Figure 6.16 Obsacle deecion range for experimenal works Figure 6.17 Wireless communicaion beween he operaor and he rouer (robo)....1 Figure 7.1 Experimenal work flow...13 Figure 7. Tesing arena...14 Figure 7.4 Mobile robo navigaed in an obsacle-free environmen (simulaion)...16 Figure 7.5 Mobile robo navigaed in an obsacle-free environmen (experimen)...17 Figure 7.6 Case 1: Trajecory planning wihou an obsacle...18 Figure 7.7 Experimenal seup for Case...13 Figure 7.8 Mobile robo navigaed in a known saic environmen (simulaion) vii

9 Lis of Figures Figure 7.9 Mobile robo navigaed in a known saic environmen (experimen) Figure 7.1 (a) Case : Trajecory planning wih a known saic obsacle, (b) Experimenal resuls Figure 7.11 (a) Plan view (b) Acual experimenal seup for Scenario Figure 7.1 Iniial collision-free rajecory for Case Figure 7.13 Mobile robo navigaes in he unknown saic environmen (simulaion) Figure 7.14 Mobile robo navigaes in he unknown saic environmen (experimen) Figure 7.15 Theoreical and acual rajecory for Case Figure 7.16 Experimenal seup for Scenario...14 Figure 7.17 Mobile robo navigaes hrough wo unknown obsacles (simulaion)...14 Figure 7.18 Mobile robo navigaes hrough wo unknown obsacles (experimen) Figure 7.19 Theoreical and acual rajecory for Case Figure 7. Moving obsacle coming from he opposie direcion of he mobile robo Figure 7.1 Scenario 1: Moving obsacle from he opposie direcion of he mobile robo (simulaion) Figure 7. Scenario 1: Moving obsacle from he opposie direcion of he mobile robo (experimen) Figure 7.3 Theoreical and acual rajecory for scenario Figure 7.4 Moving obsacle coming from lef-hand side of he mobile robo Figure 7.5 Scenario : Moving obsacle from he lef-hand side of he mobile robo (simulaion)...15 Figure 7.6 Scenario : Moving obsacle from he lef-hand side of he mobile robo (experimen) Figure 7.7 Theoreical and acual rajecory for scenario...15 Figure 7.8 Moving obsacle coming from righ-hand side of he mobile robo Figure 7.9 Scenario 3: Moving obsacle from he righ-hand side of he mobile robo (simulaion) Figure 7.3 Scenario 3: Moving obsacle from he righ-hand side of he mobile robo (experimen) viii

10 Lis of Figures Figure 7.31 Theoreical and acual rajecory for scenario ix

11 Lis of Tables LIST OF TABLES Table.1 Inrinsic splines family (Delingee e al., 1991)...4 Table 5.1 Inpu daa for replanning approach scenario...69 Table 5. Acual colleced daa of simulaion wihou replanning approach...7 Table 5.3 Acual colleced daa wih replanning approach...7 Table 5.4 Inpu daa for simulaion Case Table 5.5 Acual daa colleced a he final poin for Case Table 5.6 Inpu daa for simulaion Case...77 Table 5.7 Acual daa colleced a he final poin for Case...79 Table 5.8 Inpu daa for simulaion Case Table 5.9 Acual daa colleced a he final poin for Case Table 5.1 Inpu daa for simulaion Case Table 5.11 Acual daa colleced a he final poin for Case Table 5.1 Inpu daa for simulaion Case Table 5.13 Acual daa colleced a he final poin for Case Table 5.14 Inpu daa for simulaion Case Table 5.15 Acual daa colleced a he final poin for Case Table 5.16 Parameers for he firs mobile robo (R1)...94 Table 5.17 Parameers for he second mobile robo (R)...94 Table 5.18 Table 3 Errors for Case 1 a final poin....1 Table 5.19 Parameers for second simulaion case...1 Table 5. Errors for Case a he final poin...13 Table 5.1 Parameers for hird simulaion case...14 Table 5. Errors for Case 3 a he final poin...15 Table 6.1 Seering angles under differen PWM values Table 6. Velociies under differen PWM values Table 7.1 Acual iniial and final posiions for Case Table 7. Acual iniial and final posiions for Case Table 7.3 Acual iniial and final posiions for Case Table 7.4 Acual iniial and final posiions for Case x

12 Absrac ABSTRACT The quesion of iming in mobile robo navigaion sill remains an area of research no horoughly invesigaed. In cerain siuaions, a mobile robo may need no only o reach a desired locaion safely, bu o arrive a ha locaion a a specified ime. Such a siuaion may have significan ramificaions for applicaions o which a robo is asked, for example parolling large areas, delivering goods or coordinaing muliple mobile robos. Thus, i is imporan for a mobile robo o be able o plan is rajecories and movemens in order o navigae from iniial locaion o a final desinaion whils considering iming, orienaion and velociy. Furhermore, i should also be able o deec and avoid any obsacles encounered in is pah during navigaing hrough he environmen. The aim of his research is herefore o develop a ime-criical moion planning algorihm, which includes planning he rajecory, posiion and orienaion of a mobile robo, wih obsacle avoidance capabiliy for a single or muliple nonholonomic mobile robos. In addiion, he mobile robo should be able o replan is original rajecories in order o make up any loss of ime caused by avoiding obsacles. An Ackermann car-like robo has been considered specifically during he developmen sage, wih consideraion given o he kinemaic and dynamic consrains of nonholonomic mobile robo in general. The resulan algorihm is based on he geomeric approach. In achieving he research objecives, his sudy is conduced in four sages. The firs sage deals wih he developmen of a new algorihm for ime-criical moion planning in order o navigae safely in an environmen, o reach he specified locaion a he specified ime, wih he required orienaion, velociy and wih he consideraion of he kinemaic and dynamic consrains of he mobile robo. In he second sage, he algorihm should have he capabiliy o avoid any unknown saic and dynamic obsacles when he mobile robo sars o move from is iniial poin. The algorihm should have he abiliy o replan is original rajecory o compensae for ime loss due o avoiding obsacles. Prior o experimenal works, he simulaions will be carried ou o ascerain he effeciveness of he algorihm. In he final sage, experimenal works will be underaken o validae he algorihms uilising an Ackermann car-like robo. xi

13 Sae of Originaliy STATE OF ORIGINALITY To he bes of my knowledge, excep where oherwise referenced and cied, everyhing ha is presened in his hesis is my own original work and has no been presened previously for he award of any oher degree or diploma in any universiy. If acceped for he award of he degree of Docor of Philosophy in Mechanical Engineering, I consen ha his hesis be made available for loan and phoocopying. Mohd Sani Mohamad Hashim Dae xii

14 Publicaions PUBLICATIONS Conference papers (Main auhor) 1. Mohd Sani Mohamad Hashim and Tien-Fu Lu, Time-dependen moion planning for nonholonomic mobile robo, The 9h Inernaional IFAC Symposium on Robo Conrol (SYROCO 9), Gifu, Japan, 9-1 Sepember 9.. Mohd Sani Mohamad Hashim and Tien-Fu Lu, Muliple waypoins rajecory planning wih specific posiion, orienaion, velociy and ime using geomeric approach for a car-like robo, 9 Ausralasian Conference on Roboics and Auomaion (ACRA 9), Sydney, Ausralia, -4 December Sani Hashim and Tien-Fu Lu, A new sraegy in dynamic ime-dependen moion planning for nonholonomic mobile robos, 9 IEEE Inernaional Conference on Roboics and Biomimeics (ROBIO 9), Guilin, China, 18- December Mohd Sani Mohamad Hashim and Tien-Fu Lu, Performance of a ime-dependen moion planning for a car-like robo in saic environmens, 1 Inernaional Conference on Man Machine Sysem (ICoMMS 1), Penang, Malaysia, 7-8 February Mohd Sani Mohamad Hashim, Tien-Fu Lu and Hassrizal Hassan Basri, Dynamic Obsacle Avoidance Approach for Car-like Robos in Dynamic Environmens, 1 IEEE Symposium on Compuer Applicaions and Indusrial Elecronics (ISCAIE 1), Koa Kinabalu, Sabah, Malaysia, 3-4 December Mohd Sani Mohamad Hashim and Tien-Fu Lu, Time-criical Trajecory Planning for a Car-like Robo in Unknown Environmens, IEEE Business, Engineering and Indusrial Applicaions Colloquium 13 (BEIAC 13), Langkawi, Malaysia, 8-9 April 13. Conference papers (Co-auhor) 1. Zhiyong Zhang, Dongjian He, Tien-Fu Lu and Sani Hashim, Sudy on Seering Acuaor Transfer Funcion of Picking Mobile Robo, 1 Inernaional Conference on Communicaions and Mobile Compuing (CMC 1), Shenzhen, China, 1-14 April 1. Journal papers 1. Mohd Sani Mohamad Hashim and Tien-Fu Lu, Real-ime Conrol of Time-Criical Trajecory Planning for a Car-like Robo in Unknown Environmens, Inernaional Journal of Engineering Research and Technology (IJERT), ISSN: , February - 13 (Vol., Issue ), 13. xiii

15 Publicaions. Mohd Sani Mohamad Hashim and Tien-Fu Lu, Time-dependen Moion Planning for a Car-like Robo in Dynamic Environmens using Geomeric Approach, Inernaional Journal of Imaging and Roboics (IJIR), Revised, 1. xiv

16 Acknowledgemens ACKNOWLEDGEMENTS This work would no have been possible wihou he assisan of a number of people. I would iniially like o hank my faher, Mohamad Hashim and my moher, Norriah Salleh as well as my family who have coninuous suppor and moivae me hroughou my PhD sudy in Ausralia. I wish o hank my principal supervisor Dr. Tien-Fu Lu for his help, paien guidance and encouragemen hroughou he period of his projec. I would also like o hank my co-supervisor Dr Lei Chen for his valuable suggesions. Special hanks also go o my colleagues; ZhenZhang, Tommie, Xinrui, Gunur, Kuan and Sukri. The help and suppor from people in School of Mechanical Engineering have also been invaluable. I would like o hank colleague from he elecronics and mechanical workshop; Philip Schmid, Norio Isumi and Billy Consanine. I would also like o hank Ms. Karen Adams for her helpful language suppor research and o who involve direcly and indirecly hroughou my PhD. xv

17 Inroducion 1. INTRODUCTION Mobile roboics has become a significan research field over he pas few decades. This field has experienced a major evoluion in design, conrol, applicaion and oher aspecs which make he mobile robo useful for human aciviies. Mobile robos come in many shapes and ypes such as car-like robos, wo- or hree-wheel robos, omnidirecional robos and mobile manipulaors. One of he mus-have basic capabiliies of mobile robos is navigaion. Wih a decen navigaion sysem, a mobile robo is able o move and explore he environmen auonomously, accuraely and safely. Furhermore, he mobile robo is also able o go o any seleced places wihou human inervenion. There are wo ypes of environmens for mobile robo navigaion, which are indoor environmens and oudoor environmens. Indoor environmen mosly deals wih navigaion inside buildings, while oudoor environmen deals wih navigaion ouside buildings. Oudoor navigaion can become more complex and sophisicaed han indoor navigaion due o he unknowns of he environmens and dynamicallychanged ambience of oudoor environmens. So far, mos of he researches have aemped o develop he mos reliable navigaion sysems ha mee cerain crieria such as abiliy o choose he shores pah, minimum ime pah or minimum energy 1

18 Inroducion usage. Thus he variaion of crieria will reflec he selecion of navigaion sraegies and approaches for he mobile robo. 1.1 Moivaion Nowadays, here are many ypes of mobile robos which have been developed o assis and o ease human workforce in real-world environmens. These mobile robos are used in indoor or oudoor environmens for varieies of asks and applicaions such as facory auomaion, underground mining, miliary surveillance and even space exploraion. In mos cases, he mobile robos ofen work in unknown and dynamic environmens. Thus i is needed o ensure ha he mobile robos are able o navigae and execue he asks safely and successfully. Furhermore, hey should also be able o reac reasonably o he environmens in he presence of obsacles. One of he fundamenal issues in mobile robo navigaion is moion planning. Moion planning can be undersood as how a mobile robo plans and chooses is pah and moves along ha pah. In moion planning, he main problem is o deermine a collision-free and smooh pah in order o reach he final locaion from is iniial locaion. In general, moion planning can be divided ino wo seps (Delingee e al., 1991; Tounsi and Corre, 1996). The firs sep is pah planning, which is defined as a sep o generae a geomeric curvaure o connec he iniial and final posiion of he mobile robo. Once he pah is known, he second sep is o deermine he moion conrol. Moion conrol is defined as a sep o deermine he velociy of he mobile robo by using linear velociy law, wih which he mobile robo will follow he pah. As menioned earlier, mobile robos have been used in a wide range of applicaions in various working environmens. For oudoor environmens, i is very common for he mobile robos o face unexpeced condiions such as uneven errain, unknown and dynamic obsacles as well as pollued air (dus and smoke). These condiions may cause rouble o he mobile robo s conrol and sensor sysems as he sysem error may increase. Furhermore, he mobile robos may also accumulae errors from wihin he robo sysems hemselves such as fricion and wheel slippage. In conras, indoor environmens are more ideal condiions wih some informaion may be known beforehand, which serves as prior informaion. Prior informaion does no necessarily

19 Inroducion give an accurae knowledge of he environmens bu sufficien knowledge will ensure he mobile robo is able o navigae effecively. For oudoor environmens, prior informaion may also be available such as opological map which can be useful for mobile robos. Alhough prior informaion for oudoor environmens is no as accurae or as exensive compared o prior informaion for indoor environmens, he mobile robos need o ake advanage of his prior informaion in order o navigae safely and o reduce he uncerainies and errors in oudoor environmens. Projecs such as miliary surveillance and social securiy parol are useful o monior and mainain safey in he privae areas such as ciies and buildings. Such projecs can preven or reduce he rae of criminal aciviies, monior social aciviies and raffics. Mos surveillance sysems are using cameras, which are insalled a specific locaions and hese cameras are moniored by auomaed compuer programs. For aerial surveillance sysem, unmanned aerial vehicles (UAVs) are usually being used. The UAVs will capure images or videos of he covered area and he capured visual will be processed and inerpreed o gaher informaion abou he area. Likewise, unmanned ground vehicles (UGVs) are used for acions engaged on he ground. The capabiliy of UAVs and UGVs usually depend on he sensors used and heir abiliy o move around. This sudy focuses only on UGVs for a ground-based surveillance. Thus a basic ask for his applicaion is o ensure he UGV is being able o navigae auonomously from one poin o anoher poin in is environmen wih capabiliy of avoiding he obsacles. Furhermore, in cerain siuaions such as large area parol and goods delivery, iming is crucial as he mobile robo needs o arrive a he desired place wih no only o he righ locaion bu also o he righ orienaion, exacly a he specified ime. In a large area parol, usually he mobile robo needs o arrive a every checkpoin wih he correc orienaion exacly a he desired ime, o ensure he whole parolling area can be covered wihin he specified ime. In such case, he mobile robo should be able o plan is moion and complee parolling he whole area wihin he specified ime and also cover he angle of views for each checkpoin. For muliple mobile robo applicaions, especially in soccer robo compeiion, if robo iming can be conrolled in addiion o is posiion and orienaion, he soccer robo does no need o wai for is eammae for a long ime in order o receive he ball. If he robo wais a he cerain 3

20 Inroducion locaion for quie some ime, perhaps i has already been deeced by he opponen eam and has been man-marked, which makes i difficul o he robo o receive he ball from is eammae. Furhermore in muliple mobile robo applicaions, wo mobile robos may deliver and exchange goods a a desired meeing poin a he specified ime. If he journey ime can be conrolled for each of he robos, hey do no need o wai for each oher for a long ime a he meeing poin. Boh robos can arrive a he meeing poin a he specified ime, exchange goods and hen coninue heir journey o heir separae final locaions. For he aforemenioned examples, iming is crucial o he mobile robo o achieve is ask. This siuaion is advanageous for a ask-based mission, no only for a single mobile robo bu also for muliple mobile robos which requires he mobile robo reach he final locaion a he specified ime. 1. Research aims In brief, he aim of his sudy is o develop a new moion planning for unmanned ground vehicles. The vehicle is a nonholonomic mobile robo navigaing in a parially known and dynamic D environmen wih kinemaic and dynamic consrains are aken ino accoun during developmen sage. Thus, he primary objecives are: 1. To develop ime-criical moion planning algorihm for nonholonomic mobile robos by associaing new parameers such as posiion, velociy, orienaion and ime. To develop a dynamic obsacle avoidance algorihm ha is able o avoid boh saic and moving obsacles safely. Furhermore, he dynamic obsacle avoidance algorihm needs o be able o cach up he ime los due o he mobile robo avoiding he obsacles in order o reach he final poin a he specified ime and orienaion 3. To incorporae he newly developed ime-criical moion planning algorihm for muliple robos and muliple waypoins planning and 4

21 Inroducion 4. To develop a real auonomous mobile robo using an Ackermann car-like robo and o conduc experimenal works in order o validae he newly developed ime-criical moion planning and obsacle avoidance algorihms. 1.3 Layou of hesis The res of his hesis is organized as follows: Chaper : Lieraure Reviews This chaper inroduces he general background of his sudy. The relaed works on mobile robos, moion planning and obsacle avoidance approach are reviewed. A he end of he chaper, all he findings are summarized and gaps and conribuions from his sudy are poined ou. Chaper 3: Mehodology In order o achieve he primary objecives, his sudy is divided ino four sages. The firs sage deals wih he developmen of ime-criical moion planning algorihm for nonholonomic mobile robo. The second sage deals wih he developmen of dynamic obsacle avoidance algorihm. In he hird sage, an auonomous mobile robo will be developed. Lasly, he newly developed ime-criical moion plannning and obsacle avoidance algorihms will be validaed hrough experimenal works using he developed auonomous mobile robo. Chaper 4: Developmen of Time-criical Moion Planning Algorihms The fundamenals and he deail mahemaics of he algorihms are discussed in his chaper. The developmen of he ime-criical moion planning algorihm is based on geomeric approach wih cubic and quinic polynomials are adoped o generae moion rajecories. Furhermore, deail developmen of dynamic obsacle avoidance algorihm, muliple waypoins planning and muliple robos planning are also presened in his chaper. Chaper 5: Simulaion Resuls and Discussions This chaper presens he developmen of a simulaion framework using Malab. The nonholonomic mobile robo and he developed algorihms are esed 5

22 Inroducion using his simulaion framework. A series of simulaions are conduced o invesigae he effeciveness and pracicaliy of he algorihms. The algorihms are esed in he saic and dynamic enviromens wih a single and muliple mobile robos. Chaper 6: Developmen of a Non-holonomic Mobile Robo In his chaper, he developmen of an auonomous mobile robo is presened. A remoe conrol car are modified o be used for he experimenal works. Furhermore, he developmen of he auonomous mobile robo needs o overcome several issues such as he capabiliy of seering wheels o urn for desired angles and he mobile robo requires o speed up and slow down a specified velociies wihin seconds. Hence he kinemaic and dynamic consrains of he mobile robo such as seering angle and velociy limiaion are also considered during developmen of his mobile robo. In addiion, he calibraion works have been conduced o esablish he PWM-seering angle and PWM-speed relaionships for he mobile robo. Chaper 7: Experimenal Resuls and Discussions The experimenal archiecure and resuls from experimenal works are presened and discussed in his chaper. The developed algorihms are esed hrough a series of experimenal enviromens using he developed auonomous mobile robo. Then he experimen resuls are compared o he simulaion resuls in order o validae he algorihms.. Chaper 8: Conclusions and Fuure Works The findings of his sudy are summarized in his chaper. The recommendaion for he fuure works are also given a he end of his chaper. 6

23 Lieraure Review. LITERATURE REVIEW In his chaper, he main areas of relaed research have been reviewed, which are mobile robos, moion planning, obsacle avoidance and muliple robos coordinaion. All hese reviewed areas of research will conribue o he main objecives of his sudy. A he end of his secion, all he findings are summarised and gap saemen is given..1 Moion planning algorihms For he pas few decades, navigaion problems have been exensively sudied. One of he fundamenal issues for navigaion is o plan he robo s moion in he working environmen wihou human inervenion. This issue is commonly known as moion planning. Earlier works in mobile robo moion planning concenraed on how o deermine he collision-free pah in order o reach he final locaion (Salichs and Moreno, ). One common problem in moion planning for mobile robos is o deermine he conrol inpu which he mobile robo requires o achieve a goal posiion (x, y), pose (x, y, θ) or posure (x, y, θ, κ) (Nagy and Kelly, 1). Figure.1 shows he pah consrains made of four posures, which each posure consiss of posiion in 7

24 Lieraure Review Caresian coordinaes (x, y), orienaion (θ) and curvaure (κ) (Delingee e al., 1991). Generally, an auonomous mobile robo has o be able o exrac informaion from on-board sensors in order o know he environmen and plan is moion. Once he pah has been planned, he mobile robo is expeced o follow he pah whils considering velociy, posiion, orienaion and oher requiremens for he mobile robo o achieve smooh moions. In addiion o such consideraions, i also migh be able o boh deec and avoid he obsacles presened during navigaion. Typically, moion pah is planned based on known obsacles posiions in he environmen in prior (Hui e al., 6). Figure.1 Pah generaion (a) Pah consrains made of four required posures (b) Generaed pah (Delingee e al., 1991). Generally, pah is planned o mee several main requiremens such as shores pah, safe pah and smooh pah. Shores pah could be he shores disance o arrive a he final locaion or he shores ravel ime. While navigaing in he environmen, he robo also needs o consider safey issues. This means he pah needs o be collision free and he robo also needs o be able o deec and avoid he obsacles. Lasly he pah should be smooh in order o saisfy he kinemaic consrains. The pah should no have a sharp urn ha is impossible for he robo o urn in smooh movemen. However, he opimal pah is normally a compromise among he hree requiremens. In a known environmen, here are well known and widely used mehods for pah planning such as roadmap approaches, cell decomposiion mehods and poenial field mehods. 8

25 Lieraure Review.1.1 Roadmap pah planning The roadmap pah planning is based on conneciviy in a nework of robo s free space by using lines. Once he roadmap has been consruced, he pah is deermined by searching he series of road ha are connecing he iniial and final sae. Visibiliy graph (Jiang e al., 1997), Voronoi diagram and Visibiliy-Voronoi diagram are he well known roadmap approaches as shown in Figure.. They have been used o compue he shores collision free pah. In his approach, he obsacles are represened by convex polygons. Then every wo nodes beween iniial sae and goal sae in his free space are conneced by line and his line does no inersec he inerior of he obsacles. Visibiliy graph consis of sraigh lines ha join all he polygons edges including he iniial and final poins. (a) (b) Figure. Roadmap approach (a) Visibiliy Graph (Jiang e al., 1997). (b) Voronoi diagram (Siegwar and Nourbakhsh, 4). Jiang e al. (1997) presened hree sages o solve he ime-opimal problem by using visibiliy graph. Firsly, he reduced visibiliy graph is obained. Then i is convered ino a feasible reduced visibiliy graph accouning he robo size and kinemaic consrains. Lasly, a new algorihm is used o search he feasible reduced visibiliy graph in order o obain a safe, ime-opimal and smooh pah. They have used an A* algorihm o search he shores pah. However, his mehod only considered kinemaic consrains, bu no dynamic consrains such as he velociy of he mobile robo. The dynamic consrains are imporan o be considered as he mobile robo may need o slow down during urning and accelerae as fas as possible during moving a he sraigh line. 9

26 Lieraure Review Sridharan and Priya (4) presened a parallel algorihm for consrucing he reduced visibiliy graph in a convex polygonal environmen. They aimed o reduce he compuaional complexiy and space and implemened he algorihm in FPGA. Their algorihm consiss of wo seps. Firsly, binary code is assigned o he verices of he objecs o deermine supporing segmens beween every pair of polygon. Then he nex sep is o eliminae he supporing segmens ha are hidden by he obsacles in order o obain he final graph. From he resuls, he hardware-based approach is approximaely 1 imes faser han using a PC. However, he major drawback of his visibiliy graph approach is ha he pah is very close o he obsacles and i is no pracically safe in real applicaions. On he oher hand, Voronoi graph is able o overcome he problem caused by visibiliy graph aforemenioned. Nagaani e al. (1) proposed mobile robo navigaion using generalized Voronoi graph (GVG). In he paper, hey inroduced a local smooh pah planning algorihm for car-like mobile robo which is bounded by kinemaic consrains. In addiion, hey used Bezier curve o generae a smooh pah in order o saisfy he limiaion of minimum urning radius. The algorihm is execued hrough simulaions only and he compuaional ime cos higher han he convenional approach. This means i akes more ime o generae he pah and i is no pracical in he real-ime conrol of he mobile robo. Vicorino e al. (1) presened a new mehodology for mobile robo navigaion in unknown environmens. Once he mobile robo sared o move, i also sared o consruc he pah using Voronoi diagram based on he informaion from he on-board sensor. From he resuls, he mobile robo was successfully consruced a map and localized iself. However, hey had no discussed on he ime required o consruc he map and navigae o he goal poin. Furhermore he map consrucion and localizaion is relevan o saic environmens only. Thus heir mehod may no be appropriae o be used for ime-dependen planning and in dynamic environmens. As he combinaion of Visibily graph and Voronoi diagram may gives opimal pah for mobile robos, Wein e al. (7) inroduced a new ype of diagram which is a hybrid beween he visibiliy graph and he Voronoi diagram. The aims were o find he smooh shores pah wihou sharp urns. This mehod was used for planning a 1

27 Lieraure Review pah for robos in an environmen filled wih polygonal obsacles. In order o keep he disance from obsacles opimum, hey used predefined clearance value, c. In addiion, hey used Dijksra search o find he shores pah. However, heir mehod was only implemened for a robo wih wo degrees of moion freedom. Furhermore, Voronoi diagram ends o maximize he disance beween he robo and he obsacles, in order o provide more space and safey o he robo. Roadmap pah planning approaches such as visibiliy graph and Voronoi diagram are effecive o be used o obain a safe pah and he shores pah. The approaches used he informaion from map such as he shape of he obsacles o generae he pah. However he mobile robo ends o make a sharp urn and move very close o he obsacles. These siuaions are no appropriae for a car-like robo ha has a seering angle limiaion..1. Cell decomposiion pah planning In cell decomposiion approach, he robo s free space is divided ino several simple, conneced regions called cells. There are several ypes of grid ha normally used such as fixed-resoluion grid and riangulaion grid in order o consruc he cells as shown in Figure.3. Then he cells conaining he iniial and goal saes are locaed and pah in he conneciviy graph is searched o join he iniial and goal cell. (a) (b) Figure.3 Cell decomposiion mehod (a) A fixed-resoluion grid. (b) A riangulaion (Ge and Lewis, 6). 11

28 Lieraure Review Hazon and Kaminka (8) presened new muli-robo coverage algorihms in heir paper. Their algorihms are based on spanning-ree coverage of approximae cell decomposiion of work-area and have achieved a significan improvemen in coverage ime by improving he efficiency of he algorihms. However, hey have no menioned he ype of robo which has been used in heir simulaion and he algorihms were only esed by simulaion works. Furhermore, he algorihms work efficienly in obaining he opimal coverage ime bu no ime dependen. Figure.4 Simulaion resuls by using (a) rapezoidal decomposiion and (b) riangular decomposiion (Ghia and Kloezer, 1). Ghia and Kloezer (1) proposed a fully auomaic planning and conrol sraegy for a car-like robo based on cell composiions approach. The approach used an absracion of he free environmen and an ieraive procedure o find a feasible pah for he nonholonomic mobile robo. The planning and conrol mehod was developed in Malab and he feasible and smooh pah was obained as shown in Figure.4. However, from he resuls, he generaed pah was closed o obsacles and collision may occur beween he mobile robo and he obsacle..1.3 Poenial field pah planning The mos widely used mehod for collision free pah planning is he poenial fields mehods (Huang e al., 6; Safadi, 7; Huang, 9). I was iniially proposed by Khaib in 1986 for mobile robo pah planning. The main aspecs of his mehod are 1

29 Lieraure Review he mobile robo is reaed as a poin, he obsacle generaes a repulsive force and he goal generaes an aracive force. The aracive force lead he robo o he goal and he repulsive force ensures he robo is away from he obsacles as shown in Figure.5. The generaed repulsive force also increases proporionally wih he disance of he neares obsacles. Thus he combined force should drive he mobile robo owards he goal while avoiding he obsacles. Figure.5 Poenial field mehod (Safadi, 7). Cosio and Casaneda (4) proposed an improved arificial poenial field mehod for auonomous navigaion of a mobile robo. In he paper, hey aemped o overcome he problem ha caused by using a single aracion poin which lead o rap siuaion where he mehod is unable o produce he resulan force needed o avoid he large obsacles. Therefore, hey inroduced muliple auxiliary aracion poins ha allow he robo o avoid large or closely spaced obsacles. The force inensiy parameers of he repulsive and aracive cells have been opimised by using a geneic algorihm. From he simulaion resuls as shown in Figure.6, he generaed pah was no oo smooh and ends o make sharp urns. Furhermore, he algorihms were esed only in Malab and he auhors have no discussed he ime required for a mobile robo o reach he final poin. 13

30 Lieraure Review Figure.6 Pah generaed by he navigaion algorihm (Cosio and Casaneda, 4). The earlier works on pah planning using poenial field mehod only concenraed on saic environmens. In he recen years, dynamic obsacles have also been included in navigaion planning. Ferrara and Rubagoi (9) proposed a dynamic obsacle avoidance sraegy for a mobile robo based on harmonic poenial field mehod. Their approach consiss of wo key elemens which are an online generaor is used o rack he reference signals o reach he goal poin and a he same ime, a poenial field mehod is modified online in order o avoid he moving obsacles wih ime-varying speed. In addiion, hey used a collision cone approach o avoid he moving obsacles. The key idea is o modify he radius of he securiy circle around each obsacle on he basis of he so-called collision cone. However, heir proposed sraegy was o conrol he mobile robo bu no o generae he pah. Furhermore, hey only esed heir approach by simulaion works. Jacob (8) proposed a sensor-based navigaion and obsacle avoidance algorihm for mobile robos in unknown dynamic environmens. The proposed mehod allows a mobile robo o navigae in he environmen wih a large number of saic and dynamic obsacles. The mobile robo will navigae hrough he environmen via he global pah which was generaed based on he updaed map which processed by he global planner. Then he local planner coninuously ries o reach each waypoin on he pah using poenial field. However, heir algorihm only esed by simulaion works and hey have no discussed he ime required o reach he final poin. 14

31 Lieraure Review Furhermore, from heir simulaion works, hey encounered several failures in he simulaion such as he rear-end collision occurred due o he blind spo of he laser scanner. Huang e al. (6) proposed a mehod which combined a single camera and poenial field mehod in order o navigae in real-ime environmen. The camera is used o esimae he ime of impac once he obsacle is deeced which hen can be used o make sure he robo navigaes around he obsacle. Furhermore, Huang (9) has exended he work o deal wih he dynamic obsacles. Using he same mehod poenial fields Huang has applied his mehod for pah and speed planning in order o avoid he moving obsacles. Their approach provides boh he direcion and he speed of he mobile robo, which guaranees ha he mobile robo will able o rack he moving obsacle while avoiding i. However, heir algorihms only esed in he simulaion and hey have no discussed he ime require o avoid he obsacle and reach he final poin. Beside a poenial field mehod, a vecor field mehod is also has been used in robo navigaion. The vecor field uilizes a saisical represenaion of he environmen hrough he hisogram grid and i consiss of aracive forces, goal and repulsive forces. Boh aracive and repulsive forces are usually characerised as poin forces. Hong e al. (7) proposed a mobile robo navigaion using modified flexible vecor field approach wih laser range finder and infrared sensors. The laser range finder is used o generae he map and infrared sensors are used for emergency sop and obsacle avoidance. From he resuls, heir algorihms show a smooh moion of he mobile robo navigaes hrough he environmen. However he proposed mehod only demonsraed in saic environmens and he speed of he mobile robo was se o 7cm/s only which is no opimized for he robo s moion. The mobile robo may need o speed up a he sraigh line and slow down a cornering. Furhermore he auhors have no discussed on he ime require for he mobile robo o reach he final poin. Liddy and Lu (7) proposed waypoin navigaion for an Ackermann seering auonomous vehicle. Their aim is o obain a pah wih posiion and heading conrol of he mobile robo. They have inroduced a complex vecor field mehod by combining 15

32 Lieraure Review vecor field componens such as poin force vecor field, roaional field and line force. The resuls successfully demonsraed he posiion and heading can be conrolled a he goal poin. However, he auhors have no discussed on he ime require for he mobile robo o reach he final poin and he algorihms were only esed by simulaion works. Poenial field mehod is one of he commonly used approaches o generae pah for he mobile robo. The mehod has been uilised for various ypes of mobile robo such as he differenial drive robo and he car-like robo wih Ackermann seering limi. One of he problems in poenial fields mehod is he robo is inended o converge in he local minima (Huang, 9). Furhermore, mos of he research in poenial field approach have no addressed he ime require for he mobile robo o reach he final poin. This parameer is one of he imporan poins for he ime-criical moion planning..1.4 Oher pah planning approaches There are oher approaches which have been developed by researchers in order o obain he opimal collision free pah. The approaches could be a combinaion of wo differen approaches, or sampling-based pah planning. Koh and Cho (1999) presened a pah racking algorihm for a nonholonomic mobile robo in order o obain a smooh moion of he mobile robo. This algorihm is based on ime opimal bangbang conrol considering dynamic consrains of he mobile robo in order o avoid he wheel slippage problem during he mobile robo navigaion. Figure.7 shows he flow char on implemening he proposed algorihm. In heir experimen, hey have used a wo-wheel driven mobile robo o validae heir proposed algorihm. However, heir approaches only focused on obaining a smooh moion wihou he consideraion of avoiding obsacles. 16

33 Lieraure Review Figure.7 Implemenaion of he proposed algorihm by Koh and Cho (Koh and Cho, 1999). Mihaylova e al. (3) presened an informaion-based approach for rajecory opimizaion of a mobile robo by a linear combinaion of sine funcions. The mobile robo was equipped wih a sensor which measures he range and bearing o a beacon locaed a a known coordinae. The informaion acquired from he sensor will hen be used o obain an opimal rajecory based on a known, nominal reference rajecory. The accuracy of his approach depends on he number of beacons available in he environmen. If here are more beacons a he appropriae places, he accuracy can be improved considerably. However, he effeciveness of his approach is only demonsraed by simulaion resuls as shown in Figure.8. An experimen using his approach would be useful o validae he opimizaion effeciveness. 17

34 Lieraure Review Figure.8 Resuls from he informaion- based mehod (Mihaylova e al., 3) A new approach using he cell-mapping mehod was inroduced by Li and Wang (3) as shown in Figure.9. Their aim was o achieve he opimal rajecory in erm of minimum ime, energy and jerk. Firsly, his approach performs a global analysis and reconsrucs he whole sysem ino a cell space model. Then, based on his cell space model, his mehod finds ou he sable region as a se of cells in he cellular sae space afer a number of inegraion processes o generae he opimal rajecory. In heir sudy, hey used a four-wheeled mobile robo wih dynamic consrains such as velociy and acceleraion limiaions. However, his mehod was only esed in simulaion works and he auhors have no discussed on he obsacle avoidance approach. (a) (b) Figure.9 Cell mapping model (a) wih 3 5 cells. (b) wih 4 5 cells (Li and Wang, 3). In order o achieve he ime-opimal planning for he wheeled mobile robo, Prado e al. (3) proposed wo asks ha can be carried ou simulaneously or sequenially. The firs ask is spaial-planning which is o obain he shores feasible geomeric 18

35 Lieraure Review pah. The second ask is emporal-planning which is o obain he fases feasible velociy profile for a homogenous segmen which he segmen is he pah lengh navigaed over ime. They also considered kinemaic and dynamic consrains such as velociy and acceleraion in order o ge he opimal rajecory soluion and o avoid he obsacles in dynamic environmens. To validae heir algorihm, hey used a fourwheeled mobile robo which is known as RAM in heir experimen and he resuls are shown in Figure.1. However, from heir resuls, he mobile robo moved very close o he obsacles and he mobile robo ends o make a sharp urn. Furhermore, he auhors have no discussed he ime require for he mobile robo o reach he final poin. Figure.1 (a) Generaed rajecory (b) Velociy profile (c) Acceleraion profile (Prado e al., 3) Then, Hui e al. (6) presened a ime-opimal, collision-free navigaion of a carlike robo using neuro-fuzzy-based approaches as shown in Figure.11. In heir paper, a fuzzy logic conroller (FLC) was used o conrol he robo. The performance of he conroller was improved by using hree differen neuro-fuzzy-based approaches, which are neuro-fuzzy approach, geneic-neuro-fuzzy approach and GA-uned adapive nework-based fuzzy inference sysem (ANFIS), and hen comparing among hemselves and wih oher approaches such as defaul behaviour, manuallyconsruced FLC and poenial field mehod, hrough compuer simulaion. From heir resuls, even hough he performance using neuro-fuzzy-based approaches is beer 19

36 Lieraure Review han oher approaches, i is dependan on he raining daa. This condiion caused he performance of he neuro-fuzzy-based approaches no o work well, paricularly when he raining scenarios are differen from he real scenarios. Figure.11 Neuro-fuzzy approach (Hui e al., 6). In 7, Haddad e al. (7) presened a random-profile approach in order o opimize he free-rajecory planning problem for non-holonomic wheeled mobile robos in consrained workspaces as shown in Figure.1. This mehod is based on a simulaneous search for he mobile robo pah and also handles he obsacle avoidance issues during navigaion. In heir paper, hey focused on he planning he rajecories for he mobile robo wih he consideraion of geomery, kinemaic and dynamic consrains. However heir resuls are presened using only wo- and hree-wheeled mobile robos. I remains o be seen ha heir works are able o be exended o he four-wheeled mobile robo. Neverheless, he algorihm may require o be modified in order o caer he kinemaic and dynamic consrains of he four-wheeled mobile robo.

37 Lieraure Review (a) Figure.1 Generaed rajecory wih several conrol poins (Haddad e al., 7). (b) Ma e al. (13) presened a pah planning algorihm for a nonholonomic mobile robo using he informaion of he sensors o navigae in complex environmens. The robo moved oward a known arge while avoiding obsacles by choosing appropriae inermediae objecives based on he local sensor informaion. In addiion, by choosing inermediae objecives, a local minima problem can be solved. The efficiency of he approach was assessed via differen simulaed environmens as shown in Figure.13. From he resuls, he robo was able o navigae rough he complex environmens. However, he robo s pah was closed o he obsacles and he robo was likely o make a sharp urn as in Figure.13(b). (a) (b) Figure.13 Simulaion resuls in (a) a complex scenario, and (b) a long corridor (Ma e al., 13). 1

38 Lieraure Review.1.5 Geomeric approach for rajecory planning A rajecory is a pah which is an explici funcion of ime. Iniially a pah can be differeniaed o give a coninuous velociy and acceleraion profiles. One common mehodology for rajecory planning in order o obain a smooh-pah and lenghopimum plan is by assembling he arcs of simple curve. A mobile robo has o follow he pah (curve) wih specific velociy which is dependen on is posiion and is orienaion (Tounsi and Corre, 1996). Basically, he orienaion (θ) is defined as he angen of he poin (x(s), y(s)), which s is he lengh along he curve. The curvaure κ is defined as he derivaive of θ(s) wih respec o s. 1 dx θ ( s) = an, dy dθ ( s) κ ( s) = (.1) ds Tounsi & Le Corre (1996) reviewed and compared several ypes of curves used in pah generaion, which are sraigh lines, circular arcs, polynomial funcions, clohoids (cornu spiral) and cubic spirals. Generally, he pah is generaion by a se of robo s posures, which hese posures depend on he posiion and orienaion of he mobile robo (Shin and Singh, 199). They also discussed he mehods o generae he pah as shown in Figure.14. Figure.14 Differen ypes of curves used o connec four posures for pah generaion (Shin and Singh, 199). The pah generaed by several sraigh lines is he simples mehod in erms of calculaion and requires only he choice of inermediae poins. However, in mos cases, he orienaion is disconinuous and he mobile robo needs o sop and change is direcion in order o move o he nex poin. Similarly in he pah generaion by following circular arcs of radius R, he drawback is ha he pah presens

39 Lieraure Review disconinuous curvaure a juncion poins, which means he speed of each wheel of he mobile robo is no coninuous a hese poins. In order o avoid he disconinuous curvaure, polynomial curves were used. There are hree differen ypes of polynomial curves discussed by Tounsi and Le Corre (1996), which are polar polynomials, Caresian polynomials and Bezier s polynomials. Even hough he polar polynomial mehod gives a coninuous curvaure, he radius R mus be fixed and i is only used for symmeric cases. Boh Caresian and Bezier s polynomials are used o connec non-symmeric posures. However hese curves have a complex curvaure profile which is no necessarily smooh and makes hem difficul o follow (Delingee e al., 1991). The oher ype polynomial curvaure is known as polynomial spiral. There are wo commonly used ypes of spiral curves which are clohoid curves and cubic spiral curves. In general, he polynomial spirals are useful for pah generaion because hey provide an easy-o-rack polynomial curvaure profile (Liang e al., 5). In a review by Delingee e al. (1991), he original work by Kanayama (Kelly, 3) on clohoid curves has inroduced he idea of using coninuous piecewise linear curvaure funcion ha was hen exended by Shin and Singh (Kanayama and Miyake, 1986) in order o eliminae disconinuiy a he juncion poins. However, he problems wih his mehod are difficul o choose he coefficien of he curvaure (k) (Tounsi and Corre, 1996), difficul o compue (Delingee e al., 1991) and i sill resuls in a disconinuiy in he derivaive of he curvaure (Nagy and Kelly, 1). Thus, a sudy by Pin and Vasseur (199) considered he problems of complexiy and lenghy pah using clohoid curves by generaing deerminisic and providing rajecories joining all he pairs of configuraions of he mobile robo. Their aim was o deermine he shores pah wih reverse mode capabiliies while he mobile robo is manoeuvring by considering non-holonomic and seering angle consrains. 3

40 Lieraure Review Table.1 Inrinsic splines family (Delingee e al., 1991) Mos sudies have used cubic spiral curve (Nagy and Kelly, 1; Kelly, 3; Liang e al., 5) in pah generaion because i provides a smooh pah and minimizes he variaion of jerk (Delingee e al., 1991). In addiion, i also has been used due o is simple curvaure profile which is easy o follow. Laer, Delingee e al. (1991) developed a family of rajecory called inrinsic splines of degree n, ISn as shown in Table.1. This family is based on cubic polynomials, bu he end condiions of his family are defined in erm of heading and curvaure insead of firs and second derivaive for cubic polynomial. Nagy and Kelly (1) exended he work done by Delingee e al. (1991). In comparison o Delingee e al., he approach is gained by convering he inegro-differenial sae equaion ino four nonlinear equaions and solving hem simulaneously in order o ge he four unknown consan parameers. Subsequenly, Kelly (3) exended he work done by Nagy and Kelly (1) by inroducing an approach which produced an efficien real-ime algorihm o join arbirary poins. However, mos of he researchers have swiched he specificaion of he rajecories in erm of ime o disance, which suis mos of applicaion bu no he ime-criical applicaion argeed in his research. 4

41 Lieraure Review Liu and Sun (11) presened an opimal pah planning of a mobile robo by uilizing Bezeir curves. The objecive of heir approach was o minimize energy consumpion during robo navigaion. The energy consumpion was analysed for boh in geomeric pah planning and smooh pah planning. The effeciveness of he approach has been esed in he simulaion and experimenal works. The resuls of heir works are shown in Figure.15 and he experimen was conduced using wo-wheel mobile robo. The resuls show an opimal pah in erm of minimum energy, minimum ravel disance and minimum ravel ime. This approach can be adoped in his sudy o minimize he energy consumpion and a he same ime o reach he final poin a he specified ravel ime. Figure.15 An opimal pah (a) minimum energy, (b) minimum ravel disance, and (c) minimum ravel ime (Liu and Sun, 11).. Navigaion environmens Mobile robos are being deployed in various ypes of environmens. Some of hem are asked o navigae inside he buildings and ohers ouside he buildings. Oudoor navigaion poses a greaer challenge over ypical indoor navigaion. Oudoor environmens are usually dynamically changed over ime and give uncerainy o he mobile robos. Such environmen, so-called dynamic environmen may consis of saic and moving obsacles. Saic environmens normally have unmoved obsacle wih various shapes and sizes. Thus, saic environmens are no as complicaed as dynamic environmens in erm of planning he pah. In he previous secions, he sandard pah planning approaches, such as roadmap, cell decomposiion and poenial field mehods, have been uilised wheher in saic or dynamic environmens. However hese sandard approaches have no been proven o be effecive in unknown environmens. Due o he uncerainy of he unknown 5

42 Lieraure Review environmens, some approaches such as grid-based or roadmap-based approaches canno generae an opimal pah. Furhermore, local informaion is required o deec and avoid he unexpeced obsacles. Thus some of he researches have developed he alernaive approaches by modifying and improving he exising approaches or wih combining wo or more exising approaches o overcome he limiaions of he exising approaches...1 Oudoor navigaion In oudoor navigaion, he robo will face a new challenge especially due o numerous uncerainies and dynamic changes in he oudoor environmen such as varying errain surface and level, and also lighing condiion as shown in Figure.16. A robus oudoor navigaion sysem will improve he auonomy of he robo and provide a safe and smooh navigaion o reach he final locaion. In order o obain a safe and smooh pah, mos researchers consider he moving obsacle s velociy as known o he sysem. Wih he knowledge of he moving obsacle s velociy, he sysem is able o predic he moving obsacle s moion and probabiliy he collision beween he mobile robo and he moving obsacle. If he mobile robo is indispuably o collide wih he moving obsacle, he mobile robo is able o avoid he moving obsacle by adjusing is pah. However in he real-world, i is difficul o disinguish he velociy of he moving obsacle beforehand. This circumsances may fall shor he sysem. (a) (b) Figure.16 Oudoor navigaion (a) Pioneer3-AT wih URG and SICK (Chang e al., 9) (b) The Cycab used in he experimenal works (Zhang e al., 6). 6

43 Lieraure Review Furhermore, he unknown environmen gives a furher challenge o he sysem. The uncerainy of he informaion in he environmen leads o he needs of a beer deecion and predicion approaches in order o make sure he smooh and safe pah requiremens are me. However in cerain cases, he map of he area ha he mobile robo needs o navigae is available. This map may give some informaion o he robo planner. By uilising his informaion, he planner is able o plan he pah beforehand. Thus a good oudoor navigaion sysem is sill required in order o ensure he mobile robo navigaes and reaches he final locaion safely. Therefore, many sudies are aemped o develop a new and beer navigaion sysem in a dynamic and unknown environmen..3 Obsacle avoidance Avoiding obsacles is one of he problems for he mobile robo o navigae in saic and dynamic environmens. In dynamic environmen, where here are saic and moving obsacles, he ask becomes more complicaed and difficul in comparison o saic environmens. Therefore, many approaches have been inroduced in previous research in order o develop an effecive and reliable obsacle avoidance capabiliy for mobile robos o navigae in saic and dynamic environmens. Fajen and Warren (3) inroduced a new soluion for obsacle avoidance based on observing he human behaviour in dynamic environmens. In heir paper, he aim is o apply he dynamic model o he robo behaviour of seering owards a goal and avoiding he obsacles. Once he se of behaviour variables for seering and obsacle avoidance have been idenified, he general form of he model will be inroduced. The basis of heir work is shown in Figure.17. In Figure.17(a), he auhors considered an observer moving in a simple environmen. The observer moves a a consan speed (s) and a heading direcion (ø) wih respec o fixed verical exocenric reference axis. In Figure Figure.17(b), he goal and obsacle angles can be represened in egocenric reference frame wih respec o he observer s poin of view. In order o model heir approach, hey have used human as paricipans o observe he behaviour during walking from iniial poin o final poin as well as during avoiding he obsacle. The colleced descripive daa were hen being used o develop a model of he behavioural dynamics. This work has been exended by Fajen e al. (3) by using visually- 7

44 Lieraure Review guided locomoion in a dynamic environmen in order o idenify a se of behavioural variables for seering and obsacle avoidance. However, he behavioural approach requires human experimens in prior in order o develop a model of behavioural dynamics. This model is direcly influen by he behaviour of human a he ime of he experimens is conduced ha may lead o inaccuracy of he model. Neverheless, from heir experimen resuls, i was suggesed ha human roue selecion does no require explici planning bu may emerge on-line as a consequence of elemenary behaviours for seering and obsacle avoidance. (a) (b) Figure.17 Plan view of he observer moving in dynamic environmen (a) Exocenric reference frame (b) Egocenric reference frame (Fajen and Warren, 3). The mos commonly used mehod for solving he obsacle avoidance problem is based on he poenial field mehod, firsly proposed by Khaib (1986). Then Huang e al. (6) proposed a vision-guided navigaion approach by adaping Fajen and Warren s work on human behaviour navigaion and his approach was expressed as a poenial field. In heir sudy, he poenial field is used o conrol he angular acceleraion and heading of he robo in order o seer i oward he goals and o avoid he obsacles during robo navigaion. However, his approach has a limiaion since hey used angular widh of he obsacle raher han disance, ye a large obsacle can also has he same angular widh as a smaller obsacle. Furhermore, Hamner e al. (6) also proposed an exension mehod based on Fajen and Warren formulaion. The proposed mehod can learn he parameers of he conrol 8

45 Lieraure Review model auomaically by observing behaviour of he human driver. In addiion, Hamner e al. inroduced a speed conrol funcion based on he obsacle s disance and angle in heir mehod. This speed conrol funcion slows down he vehicle as he obsacles ge closer, which gives ime o he vehicle o urn and avoid he obsacles. However his mehod also allows a sharp urning which has a negaive impac for he vehicle moion. Moreover, heir resuls showed ha he vehicle aemps o follow a far pah while avoiding large obsacles and gave conservaive resuls. The oher mehod o solve he problem of obsacle avoidance was proposed by Brock and Khaib (1999) using global dynamic window approach. In heir paper, he global dynamic window approach used for moion planning is an exension of he dynamic window approach (Fox e al., 1997) by incorporaing a simple and efficien moion planning. This framework allows robus execuion of high-velociy, goal-direced and reacive moion for a mobile robo in unknown and dynamic environmens. However he approach was used for a holonomic mobile robo, no for non-holonomic mobile robo as argeed in his sudy. Casillo e al. (6) proposed an approach ha using sonar deecion for deecing he obsacles. Sonar was used in he research due o i provides a consisen daa and i can simply deecs somehing in he environmen. From heir resuls, he sonar sensor was capable o deec obsacles and ensure he wheelchair as able o navigae safely. However, hey applied his approach only for an auonomous wheelchair, used in an indoor environmen, which can be exended o an oudoor environmen. Recenly, Jolly e al. (8) proposed a mehod for avoiding he dynamic obsacle by modifying he iniial generaed Bezier curve. A he iniial sage, he robo will ravel along he original curve. Once an obsacle is deeced, a new modified Bezier curve will be generaed. This approach is shown in Figure.18. In heir simulaions, a holonomic mobile robo is used bu he idea of he obsacle avoidance approach can be adoped for his sudy regardless he ype of he curve used. 9

46 Lieraure Review Figure.18 Avoiding a dynamic obsacle (Jolly e al., 8)..4 Muliple robos coordinaion There have been many sudies on using muliple robos o achieve a ask given. Using a group of robos insead of single robo in ask-based mission has a few advanages such abiliy o complee he ask faser, more robus, abiliy o locae he goal posiion more accurae and abiliy o complee he ask ha by using a single robo canno be achieved. Some of he applicaions using muliple robos are exploraion of hazardous environmen, search and rescue, auonomous consrucion, huning operaions and soccer robo. Conrolling a group of robos may require a significan conrol law of moion coordinaion. Yamaguchi (3) presened a disribued moion coordinaion sraegy for muliple robos in cooperaive huning operaions as shown in Figure.19. Each robo in his conrol law has is own coordinae sysem and i can sense he arge, oher robos and obsacles. This conrol law is based on formaion vecor sraegy as an inpu. The formaion of each robo is conrollable by he vecors. 3

47 Lieraure Review Figure.19 A group of robos in huning operaion (Yamaguchi, 3). Sroupe e al. (5) presened a behaviour-based muliple robos collaboraion for auonomous consrucion asks. In he paper, wo robos are used o form a eam for he consrucion asks. The consrucion ask consiss of several subasks which are shown in Figure.. Each robo will perform he subasks a every sage in order o achieve he goal. Figure. Subasks of consrucion ask (Sroupe e al., 5). In soccer robo sysem, mos of he cooperaive sraegy is based on vision sysem. The global vision sysem is used o rack he posiion and orienaion of he robo (Klancar e al., 4; Brezak e al., 8). Klancer e al. (4) has used a robo wih colour pach on he robo. In order o esimae he robo posiion, paches and he regions belonging o he ball, opponen eam paches have o idenify. Then he posiion of he robo can be locaed by using image segmenaion and componen labelling. Figure.1 shows he overview of he sysem. Then Brezak e al. (8) 31

48 Lieraure Review used he same approach in heir paper. However hey have used Bayer image forma in order o inerpre he posiion of he robos. Figure.1 Overview of he sysem (Klancar e al., 4). Oher approaches for soccer robo sysem wihou using colour informaion are by using shape informaion (Trepow and Zell, 4), arificial neural neworks (Jolly e al., 7) and reinforcemen learning (Duan e al., 7)..5 Summary and gap saemen From he lieraure, mos research focused on obaining an opimal moion planning in erms of safe navigaion, smoohness pah, shores pah and opimal ime moion plan for he mobile robos. Even hough here are sudies on shores pah and opimal ime moion plan, he focus is only on how o reach he desired locaion as soon as possible. This means he mobile robo will reach he desired locaion in minimum or opimal ime. However, here are siuaions ha iming of reaching he desired locaion can be crucial especially when dealing wih he muliple mobile robos coordinaion. There are many mehods o obain he smooh rajecories. The common approaches such as roadmap approaches, cell decomposiion and poenial field mehod are no he bes approaches o achieve he aims of his sudy. These approaches are usually used for holonomic robos as hey end o require sharp urns. Furhermore inegraion wih ime parameer migh be difficul o be performed due o hese approaches are 3

49 Lieraure Review ypically o generae a pah for he mobile robo. Thus he bes mehod is by using a geomeric approach as his approach can be developed in erm of ime and due o is simpliciy and flexibiliy of geomeric profile. In he geomeric approach, he simples mehod o generae a pah is by assembling he arcs of simple curves. The commonly used ypes of curves are clohoid curves (Kanayama and Miyake, 1986; Pin and Vasseur, 199; Delingee e al., 1991) and cubic spiral curves (Nagy and Kelly, 1; Kelly, 3; Liang e al., 5). However, here are drawbacks using clohoid curves such as i resuls in a disconinuiy in he derivaive of he curvaure. Therefore, cubic spiral curves are adoped insead of clohoid curves in his sudy. This is because he cubic spiral curve provides a smooh pah, minimizes he variaion of jerk and is a simple curvaure profile o follow. In addiion, Tounsi and Le Corre (1996) inroduced a variable velociy funcion in order o minimize he jerk problem and o obain smooh rajecories. However, he reviewed research proposed he algorihm for cubic spiral curves in erms of disance raher han ime. In conras, iming o reach he desired locaion is more imporan raher han disance in cerain siuaion such as for he ask-based missions. The research based on human behaviour observaion in dynamic environmen has been carried ou in order for he robo o avoid he obsacles while navigaing (Fajen and Warren, 3; Fajen e al., 3; Hamner e al., 6; Huang e al., 6). A camera was used by Fajen e al. (3) and Huang e al. (6) as a navigaional aid for robo o avoid he obsacles. Huang e al. used he poenial field approach o conrol he angular acceleraion and heading of he robo. However, his approach gives conservaive resuls as he robo aemped o avoid he obsacles by following he far pah even hough o avoid smaller obsacles. In addiion, Hamner e al. (6) inroduced a speed conrol funcion, which slows down he robo as he obsacle ges closer and gives ime o he robo o urn and avoid i. However, his approach allows he mobile robo o make a sharp while avoiding he obsacles which will give a negaive impac o he robo s moion. Jolly e al. (8) presened a mehod o avoid he obsacles by using Bezier curves. The idea is o ge he conrol poin in order o generae he Bezier curves for he new pah which avoid he obsacles. This idea appears o be useful for his sudy. 33

50 Lieraure Review So far, here is no deph research focuses on ime-criical moion planning wih obsacle avoidance capabiliy for nonholonomic car-like mobile robos and on muliple robos which each robo has a differen mission or objecive in ime criical environmens. Therefore, he purpose of his sudy is o develop a ime-criical moion planning for Ackermann-seering-like nonholonomic mobile robos wih he capabiliy of obsacle avoidance in saic and dynamic environmens. In addiion, he developed algorihm will capable o plan he moion for differen mobile robos from he differen saring poin o accomplish specific missions or objecives simulaneously. A he end of his sudy, i is expeced ha he robo should be able o move from one locaion and reach he nex one wih he specified orienaion, velociy and ime wih consideraion of he kinemaic and dynamic consrains such as maximum urning radius, maximum velociy and acceleraion. Moreover, he robo should have a capabiliy of planning he rajecory wih known obsacles and re-adjus is rajecory while avoiding he deeced obsacles, which are unknown o he mobile robo in order o cach-up wih he ime delayed due o avoiding he obsacles. 34

51 Mehodology 3. METHODOLOGY The mehodology developed for his sudy is driven by he research aims of his sudy. Basically, he mehods are divided ino four sages as shows in Figure 3.1. Each sage will briefly explain in he following subsecions. Moion planning algorihm developmen Obsacle avoidance algorihm developmen Simulaions Hardware preparaion and experimenal works Figure 3.1 Sages for proposed mehodology 35

52 Mehodology 3.1 Sage 1: Developmen of Algorihms for Time-criical Moion Planning A new algorihm is o be developed hrough he use of mahemaics for a ime-criical moion planning wih he consideraion of posiion, orienaion, velociy and iming. Geomeric approach is adoped for generaion of he rajecories. The ypes of curves ha are used for he rajecory planning are cubic and quinic polynomials because hey give smooh rajecories and hey were derive from he kinemaics and dynamic consrains, which will discuss in laer chaper. In addiion, he kinemaic and dynamic consrains which are maximum urning radius and maximum velociy of he mobile robo are aken ino he consideraion during he developmen of his algorihm. The developmen of hese algorihms includes: i. basic rajecory algorihm ii. muliple waypoins planning iii. muliple robos planning 3. Sage : Obsacle Avoidance Approach Once he firs sage has successfully been carried ou, he second sage is o inegrae he obsacle avoidance capabiliy ino he sysem. The seps for his algorihm are shown in Figure 3.. A he beginning, he moion planning algorihm will generae a pah despie he presence of obsacles. Then, when he mobile robo deecs an obsacle, he safey margin and deviaion poin will be generaed. This will give wo opions for he mobile robo, wheher o urn righ or lef, which is depending on he curren posiion and locaion of he obsacle in respec o he final poin. Once he decision has been made, a new pah will be generaed o avoid he obsacle. Generally, during navigaion, he robo will be capable of deecing and avoiding obsacles and re-adjus is original pah once encouners he obsacle in order o cach-up he ime delayed due o avoiding he obsacle. Finally, his algorihm will be exended o deal wih boh he unknown saic and dynamic obsacles. 36

53 Mehodology Deec obsacle Generae safey margin Deermine deviaion poins Which poin o use? Opimizaion Generae new rajecory Figure 3. Generalized seps for avoiding an obsacle 3.3 Sage 3: Simulaion Works Simulaion works will be carried ou a every sage aforemenioned in order o ensure he funcionaliy and he effeciveness of he algorihms developed. The algorihms will be simulaed for several ypes of environmens. Firsly, he environmen is assumed as an obsacle-free oudoor environmen. Secondly, here are known obsacles in he saic environmen. Lasly, here are combinaions of known and unknown obsacles, which make he oudoor environmen more realisic for he mobile robo navigaion. Physical consrains experienced by real robo will be invesigaed and included in he rajecory planning algorihms. The selecion of he sensors also will be carried ou during his sage in order o have a smooh navigaion during experimenal sage. Malab sofware will be used for developmen and conducion of simulaions. 37

54 Mehodology 3.4 Sage 4: Hardware Preparaion and Experimenal Works Once he simulaions have successfully been carried ou a every sage, he algorihm will be validaed by experimens. There will be several experimens o be carried ou based on he environmenal seup as in he simulaion sages. An Ackermann-seeringlike robo will be used in hese experimens wih consideraion of he saic and dynamic consrains of his mobile robo. The mobile robo, which is modified from he sandard remoe conrol car, will be equipped wih sensors and ime-criical conrol sysems o ensure he objecives of his sudy are me. The modified mobile robo used in he experimenal works is shown in Figure 3.3. Figure 3.3 The modified mobile robo used in he experimenal works. The experimenal seup has been divided ino a few asks in order o ease he experimenal works. The asks are: 1. Prepare he mobile robo, which includes upgrading, modifying and calibraion works,. Program he microconroller of he mobile robo, 3. Run he firs es obsacle-free environmen, and 4. Run he second es saic and dynamic environmens. 38

55 Mehodology The main purposes of his experimenal works are o validae he effeciveness of he simulaion framework developed and o verify he pracicaliy of developed algorihms in real-ime applicaions. The experimenal works will be conduced in an open-space area. The mobile robo will communicae wirelessly wih he personal compuer (PC), which will ac as he coordinaor. Then velociy and posiion of he mobile robo a every ime sep will be recorded in PC. These daa will be used o plo he acual rajecory of he mobile robo. The movemen of he mobile robo will also be capured using video camera o observe he behaviour of he mobile robo during navigaing hrough he environmen. These experimenal resuls will hen be used o compare and validae he algorihms agains he respecive simulaion resuls. 3.5 Concluding remarks In his chaper, he mehodology of his sudy was discussed. The work can divided ino four sages which began from he developmen of he algorihm for moion planning unil he verificaion of he algorihm. The algorihm was developed for nonholonomic mobile robo by adoping geomeric approach which includes obsacle avoidance. The algorihm was hen esed by simulaion using Malab. The simulaion works sared from a simple scenario which was he obsacle-free environmen in order o assess he funcionaliy of he algorihm. I was hen furher esed in he more complicaed environmen wih he combinaion of he saic and dynamic obsacles. Once he simulaion works were successfully conduced, he algorihm was esed in he real environmens using a mobile robo. The mobile robo was developmen by modifying a sandard remoe conrol car o become an auonomous nonholonomic mobile robo. The experimenal works were conduced in a series of cases. The saic and dynamic obsacles were considered in he experimenal works in order o mimic he real environmen. The resuls from he experimen were hen being compared o he simulaion works o validae and verify he pracicaliy as well as he effeciveness of he algorihm for he ime-criical moion planning. 39

56 Algorihms for Moion Planning 4. DEVELOPMENT OF TIME-CRITICAL MOTION PLANNING ALGORITHMS In his chaper, he developmen of ime-criical moion planning algorihms and obsacle avoidance algorihm is discussed. The moion planning algorihms are based on he geomeric approach. The developmen of he algorihms begins wih he derivaion of mahemaical funcions and boundary condiions unil he inegraion of moion planning algorihm wih obsacle avoidance algorihm. The proposed algorihms for his sudy are shown in Figure 4.1. The algorihms are divided ino several seps in order o ensure he algorihms will be execued smoohly. Firsly, he planner needs o se he inpu daa for he mobile robo a he iniial poin and final poin. The inpu daa are posiion, orienaion, seering angle, velociy and ravelling ime. Then an iniial rajecory will be generaed based on hese inpus for he mobile robo o move from he iniial poin o he final poin. Parameers such as posiion, velociy, orienaion and seering angle will be deermined a every ime sep. Furhermore he algorihms will check he curren seering angle o ensure his oupu does no exceed he maximum limi. In he case of curren seering angle exceeds he 4

57 Algorihms for Moion Planning maximum limi, he replanning algorihm will be iniiaed. A new seering angle will be used, which is he maximum seering angle and he iniial rajecory will be modified in order o saisfy his limiaion. On he oher hand, if he mobile robo deecs an obsacle, he obsacle avoidance algorihm will be iniiaed. If here is no obsacle and he curren velociy or seering angle does no exceed he maximum limi, he mobile robo will coninue is journey based on he generaed rajecory unil i reaches he final poin. Inpu Daa x, y, Ø, Ө, v, Trajecory generaion Is here any obsacle? Is ø > ø max YES Avoid/Replanning Daa NO Coninue manoeuvre Figure 4.1 Flowchar of he proposed algorihms Regarding he replanning algorihm, if any seering angle exceeds he maximum limi while moving along he pah, he daa a he curren ime sep, which are locaion, velociy and orienaion of he mobile robo will be obained and will be used as he iniial inpu daa. Then he value of he seering angle will be readjused o he specified maximum limi value for he seering angle and a new rajecory will be generaed. The replanning algorihm is required for he proposed moion planning algorihms in order o obain a smooh rajecory in which he mobile robo will be limied o kinemaic and dynamic consrains such as seering angle and velociy. 41

58 Algorihms for Moion Planning 4.1 Kinemaic model of nonholonomic mobile robo Y l Ø y Fron wheel Ө Rear wheel x X Figure 4. A car-like mobile robo In his sudy, a car-like mobile robo is considered. The fron wheels are he seering and he rear wheels are he driving wheels. For his sudy, i is assumed ha boh fron wheels of he mobile robo will have similar seering angle, which is reaed as a single fron wheel as shown in Figure 4.. The disance beween fron wheel and rear wheel axle cenre is l. The midpoin of rear wheel axle is se o be a cenre poin in he space sae, CP. Given he generalized coordinaes is q ] T = [ x, y, θ, φ, v,, wih (x, y) are he Caresian coordinae, θ is he orienaion of mobile robo wih respec o he x- axis in Caresian coordinae, φ is seering angle, v is he velociy and is he required ravel ime. Le ρ be he radius of rear wheel, u 1 be he angular velociy of he driving wheel and u be he seering velociy of seering wheel (Dong and Guo, 5). Then, he sae space ha represens he kinemaic consrains of his mobile robo can be obained from: x& y& & θ & φ cos θ sin θ = u1 an / l ρ φ + u 1 (4.1) 4

59 Algorihms for Moion Planning 43 From he kinemaic model (Equaion 4.1), he range of θ and φ is limied o ), ( π π due o he srucural and mahemaical consrain of he physical mobile robo. From Equaion 4.1, we have: anθ = dx dy, θ φ 3 cos an l dx y d = (4.) 4. Boundary condiions From he kinemaic model (Equaion 4.1), we have se he boundary condiions for he mobile robo as follow:, ],,,,, [ ) (, ],,,,, [ ) ( T f f f f f f f f T o v y x q q v y x q q φ θ φ θ = = = = (4.3) wih v is he velociy of he mobile robo. In his sudy, we have se he iniial velociy as v and final velociy as T v, so ha we can conrol he velociy a boh saes. The generalized velociy funcion for he x- and y-axis is given by: θ θ sin, cos v y v x = = & & (4.4) The deails of boundary condiions a iniial and final sae for x- and y-axis are as follow: ; ) (, ) (, ) (, ) ( f f f f x x x x x x x x & & & & = = = = (4.5) ; cos an, an, ) ( 3 θ φ θ l dx y d dx dy y y = = = = = (4.6)

60 Algorihms for Moion Planning y( dy dx f d y dx ) =, = = an θ f, (4.7) f = f y f an φ f = ; 3 l cos θ f For x equaion, wih consideraion of he boundary condiions (Equaion 4.5), we have chosen a cubic polynomial equaion as: x = a +, (4.8) 3 + a1 + a a3 wih firs and second derivaive as follow: dx d = a1 + a + 3a3, (4.9) d d x = a + 6a (4.1) 3 For y equaion, wih consideraion of he boundary condiions (Equaion 4.6 and Equaion 4.7), we have chosen a quinic polynomial equaion as: y = b +, (4.11) b1 + b + b3 + b4 b5 wih firs and second derivaive as follow: dy d 3 4 = b1 + b + 3b3 + 4b4 + 5b5, (4.1) d d y 3 = b + 6b3 + 1b4 + b5 (4.13) 4.3 Coordinae-x equaion From he boundary condiions (Equaion 4.5) and he cubic polynomial equaions (Equaion 4.8, 4.9 and 4.1), we have: 44

61 Algorihms for Moion Planning ) ( x a a a a x x = = (4.14) ) ( x a a a x d dx & & = + + = (4.15) f f f f f f x a a a a x x = = ) ( (4.16) f f f f f x a a a x d dx & & = + + = ) ( (4.17) Le T a a a a a ],,, [ 3 1 = is he consan vecor and rearrange Equaion 4.14 o Equaion 4.17 as c A a 1 =, we have: = f f x x x x T T T T T a a a a & & (4.18) 4.4 Coordinae-y equaion From he boundary condiions (Equaion 4.6) and he quinic polynomial equaions (Equaion ), we have: ) ( y b b b b b b y y = = (4.19) From Chain Rule, we have:

62 Algorihms for Moion Planning 46 an an ) ( d dx wih b b b b b d dx b b b b b dx dy d dx d dy = = = = = α θ α θ (4.) and cos an an 1 6 cos an an 1 6 ) ( = = + = = = = = d dx d x d wih l b b b b l d dx d x d b b b b d dx dx y d d x d dx dy d y d α α θ φ α θ α θ φ θ (4.1) From he boundary condiions (Equaion 4.17) and he quinic polynomial equaions (Equaion ), we have: f f f f f f f f y b b b b b b y y = = ) ( (4.) From Chain Rule, we have:

63 Algorihms for Moion Planning 47 f f f f f f f f f f f f d dx wih b b b b b d dx b b b b b dx dy d dx d dy = = = = = an an ) ( α θ α θ (4.3) and cos an an 1 6 cos an an 1 6 ) ( = = + = = = = = f f f f f f f f f f f f f f f d dx d x d wih l b b b b l d dx d x d b b b b d dx dx y d d x d dx dy d y d α α θ φ α θ α θ φ θ (4.4) Le T b b b b b b b ],,,,, [ = is he consan vecor and rearrange Equaion 4.19 o Equaion 4.4 as c A b 1 =, we have: + + = f f f f f l y l y T T T T T T T T T T T b b b b b b θ φ α θ α θ α θ φ α θ α θ α cos an an an cos an an an (4.5)

64 Algorihms for Moion Planning Orienaion (θ ) equaion By Equaion 4.1, we have: = = = = an an an an a a a b b b b b a a a b b b b b dx dy d dx d dy θ θ θ θ (4.6) 4.6 Seering angle (φ ) equaion By Equaion 4., we have: = = = + = ) 3 ( ) an 6 ( ) 1 6 ( cos an ) 3 ( ) an 6 ( ) 1 6 ( cos an a a a a a b b b b l a a a a a b b b b l d dx d x d dx dy d y d dx y d d dx dx y d d x d dx dy d y d θ θ φ θ θ φ (4.7)

65 Algorihms for Moion Planning 4.7 Angular velociy ( u 1 ) equaion Le v = ρu1. From Pyhagoras Theorem, we have: ( ρu ) ( ρu ) ρu ρu = ρu = ρu = ( ρu = ( ρu ) cosθ ) + ( ρu cos θ + ( ρu ) cos θ + ρu sin cosθ (cosθ ) + ρu sinθ (sinθ ) dx dy ρu1 = (cosθ ) + (sinθ ) d d [( a1 + a + 3a3 ) cosθ ] + [( b1 u = θ 1 sinθ ) 1 sin θ + b + 3b3 ρ + 4b b 5 4 )sinθ ] (4.8) 4.8 Obsacle avoidance approach Obsacle deecion is fundamenal for a mobile robo o navigae safely in a dynamic environmen. In his sudy, he obsacle avoidance approach deals wih boh saic and moving obsacles in a D workspace using a laser range finder (LRF). The approach is evolved from he dynamic rajecory planning scheme presened (Jolly e al., 8), In a dynamic rajecory planning scheme, he mobile robo will replan and modify is rajecory once i deecs an obsacle and he newly generaed rajecory may differ from he iniially planned rajecory. However, insead of using he Bezier curves, which were used by Jolly e al. (8), polynomial curves have been adoped in his sudy. The reason behind his is o ensure ha he mobile robo will pass hrough all he conrol poins o have a beer conrol for he mobile robo s moion, compared o he Bezier curves, which only pass hrough he firs and las conrol poins (Jolly e al., 8). In his sudy, all he conrol poins are used as inpus o generae he polynomial curves o ensure he generaed curves will pass all he conrol poin. Furhermore, he dynamic rajecory planning scheme is divided ino wo planning schemes, which are uilised o avoid saic obsacles and moving obsacles. 49

66 Algorihms for Moion Planning Avoiding saic obsacles In his sudy, he saic obsacles are divided ino wo caegories: known and unknown. Known saic obsacles are known in advance o he planner during offline planning, while unknown saic obsacles are unknown o he planner and will only be deeced by he sensor during navigaion. For he known saic obsacles, he planner will consider hem in he iniial sage while generaing he rajecory. Thus he generaed rajecory should navigae he mobile robo o be away from he poenially colliding obsacles. Meanwhile, he unknown saic obsacles will only be considered when he mobile robo sars o navigae hrough he environmen. The general view of avoiding an unknown saic obsacle is illusraed in Figure 4.3. Dscan Dcol obsacle iniial rajecory mobile robo Өscan Robs Ddec Өdec Өcol Rsm+w Өdev Ddev Rsm (xdev, ydev) deviaed rajecory w (xsen, ysen) Figure 4.3 Avoiding a deeced saic obsacle which is unknown in priori When he mobile robo sars o navigae along he iniial rajecory, he range finder will also sar o scan he environmen. The maximum scanning range and resoluion is se by D scan and Ө scan, respecively. Once he mobile robo deecs an obsacle, i will check wheher he obsacle is wihin collision region or no. The collision region is defined by collision range (D col ) and collision angle (Ө col ). If he obsacle falls ino his region, a new deviaed poin will be calculaed in order o readjus he iniial rajecory and o ensure he mobile robo avoids he obsacle. The deviaed poin (x dev,y dev ) is deermined by deecion disance (D dec ), deecion angle (Ө dec ), obsacle s size (R obs ), safey margin (R sm ), robo s widh (w) and sensor s posiion (x sen,y sen ). The following equaions are used o obain he deviaed poin: 5

67 Algorihms for Moion Planning ( ) ( ) D = R + w + D + R (4.9) dev sm dec obs 1 R + sm w θdev = θdec + an (4.3) Ddec + Robs x = x + D cosθ (4.31) dev sen dev dev y = y + D sinθ (4.3) dev sen dev dev Once he deviaed poin is obained, a new rajecory (deviaed rajecory) is generaed from he curren poin o he final poin, hrough he deviaed poin. The new rajecory will have o ensure ha i caches up wih he ime los during obsacle avoidance in order o reach he final poin a he specified ime. Noe ha he new rajecory does no necessarily follow he iniial generaed rajecory as he new rajecory is based on he updaed informaion Avoiding moving obsacles The sraegy o avoid a moving obsacle is usually based on prior informaion of he moving obsacle s velociy (Guo e al., 3; Qu e al., 4). However, in his sudy he sraegy is based on he direcion and posiion of he moving obsacle. Furhermore, he moving obsacle s direcion of movemen will influence he selecion of appropriae sraegy o avoid i. For insance, if he moving obsacle is approaching perpendicularly o he mobile robo, he mobile robo will avoid he moving obsacle as illusraed in Figure 4.4(a). On he oher hand, if he moving obsacle is approaching from he opposie direcion of he mobile robo, he moving obsacle is reaed as a saic obsacle and he mobile robo will avoid he obsacle as illusraed in Figure 4.4(b). 51

68 Algorihms for Moion Planning iniial rajecory deviaed rajecory deviaed rajecory obscale s pah moving obsacle previous posiion Rsm previous posiion (xdev, ydev) Өdev Ddev Ddec Ddec iniial rajecory moving obsacle Өdec Өdec Ddev Өdec Ddec (xsen, ysen) obscale s pah mobile robo previous posiion (xsen, ysen) mobile robo previous posiion (a) (b) Figure 4.4 Avoiding a moving obsacle (a) perpendicular direcion o he mobile robo and (b) in opposiion o he mobile robo. Despie he direcion of he moving obsacle, he mobile robo will predic he possibiliy of collision beween he mobile robo and he moving obsacle. As shown in Figure 4.5(b), when he mobile robo firs deecs a moving obsacle, he posiion for boh he mobile robo and he moving obsacle will be regisered ino he regisry. Then, when he nex deecion occurs, he sysem will compare he sored posiion (firs deecion) wih he curren posiion (second deecion) o obain direcion and disance beween hese wo locaions for boh he mobile robo and he moving obsacle. In addiion, he planner will esimae he velociy of he moving obsacle. 5

69 Algorihms for Moion Planning obsacle robo collision radius (a) (b) deviaed rajecory (c) (d) Figure 4.5 Collision predicion approach (a) before deecion of he obsacle, (b) firs deecion, (c) prediced posiion falls inside he collision radius, and (d) obsacle avoidance approach implemened. From his informaion, he sysem can predic he mobile robo s and moving obsacle s posiion for he nex wo seps. If he prediced moving obsacle s posiion falls inside he collision radius of he mobile robo, hen he collision is likely o happen as shown in Figure 4.5(c), he collision poin (x col, y col ) and he deviaion poin (x dev, y dev ) are deermined by using he following equaions: x = x + D cosθ (4.33) col sen dec dec y = y + D cosθ (4.34) col sen dec dec 53

70 Algorihms for Moion Planning D = ( x x ) + ( y y ) (4.35) col movobs col movobs col x = x ( r + w) (4.36) dev col movobs y dev = y (4.37) col where, r movobs = size of a moving obsacle w = widh of he robo. Then a new rajecory which is a deviaed rajecory will be generaed from he mobile robo s curren poin o he final poin, hrough he deviaion poin as shown in Figure 4.5(d). 4.9 Concluding remarks The algorihm for ime-criical moion planning was developed for a nonholonomic mobile robo by geomeric approach. The kinemaic consrains were aken ino consideraion during he developmen of he algorihm. Furhermore, he developmen of he algorihm was also considered he limiaion of he mobile robo such as seering angle and velociy in order o obain a smooh rajecory. In addiion, he obsacle avoidance approach was incorporaed in he algorihm in order for he mobile robo o avoid obsacles. The obsacle avoidance algorihm was divided ino wo caegories which are for saic obsacles and for moving obsacles. The approach uilizes he developed rajecory planning algorihm in order o avoid obsacles. When avoiding obsacles, he algorihm replans is iniial rajecory once he mobile robo deecs an obsacle. The algorihm uses he curren informaion such as locaion of he mobile robo and obsacle, velociy and orienaion o generae new rajecory. Therefore, he mobile robo will able o avoid he obsacle and reach he final poin a he specified ime. In addiion, for moving obsacle, he algorihm will predic he moion of he deeced moving obsacle and he possibiliy ha he collision will occur beween he mobile robo and he moving obsacle. If he collision is likely o happen, he algorihm will 54

71 Algorihms for Moion Planning replan is rajecory based on he direcion of he moving obsacle. Therefore, he mobile robo will able o avoid he obsacle and reach he final poin a he specified ime. 55

72 Simulaion Resuls & Discussions 5. SIMULATION RESULTS AND DISCUSSIONS Simulaion is one of he popular ools o invesigae he effeciveness and capabiliy of a sysem prior o he real experimenal works. A few benefis can be gained by conducing simulaion works such as reducing experimenal maerial s cos and ime. Roboics field also is no excepion in using simulaion. The algorihm can be invesigaed and any adjusmen and modificaion on he algorihm can be done during simulaion works. Then once he algorihm is working well, i can be downloaded ino real robos. Furhermore, simulaion resuls can be used as a guideline or comparison for he experimenal works. In his sudy, he plaform for he simulaion works is conduced in Malab. Malab is one of he developmen ools ha has been widely used in engineering fields. Malab is o develop he simulaion plaform because of is powerful graphics and ease of use. In addiion, i is also suppored by many differen compuer sysems and i comes wih an exensive buil-in library of predefined funcions for mahemaical and echnical soluions. 56

73 Simulaion Resuls & Discussions 5.1 Simulaion archiecure The algorihms inroduced in his sudy consis of an offline and online planning. These algorihms will ensure he mobile robo is able o navigae wih a smooh moion and wihin he limiaion of he kinemaic consrains such as seering angle and i also be able o avoid obsacles. Figure 5.1 shows he scheme of he algorihm seps for he simulaion, which were used in his sudy. A he early sage, he iniial rajecory will be generaed using he inpu daa from he user. Then, he offline planning was execued o deal wih he known or predefined saic obsacles. The algorihm will check wheher here is a known saic obsacle along he way of he iniially generaed rajecory. If here is an obsacle, he new inpu daa a he deecion poin will be used, such as posiion, orienaion, seering angle, velociy and ime, o generae a new rajecory. The process will coninue unil a collision-free rajecory is generaed, wih incorporaion of seering angle limiaion. Once he offline planning has been compleed and he rajecory has been generaed, he new daa from he offline planning will be used for he online planning. The inpus, such as ime, seering angle and velociy, will be used o simulae he mobile robo a every ime seps. While navigaing wihin he environmen, he robo also will check he presence of new obsacle, which is previously unknown. If here is an idenified obsacle, he obsacle avoidance will be execued o avoid he obsacle and once he robo has avoided he obsacle, he rajecory from he deviaed poin o he final poin will be replanned using he acual daa in order o cach-up he ime loss from avoiding he obsacle and o mainain a smooh rajecory. The process will coninue unil he robo reaches he final poin. 57

74 Simulaion Resuls & Discussions Inpu daa Generae pah Any known saic obsacle? YES New inpu daa NO Exceed seering angle limi? YES NO Wireless communicaion Oupu daa OFFLINE PLANNING OBSTACLE REGISTRY New inpu daa Regiser obsacle s daa YES New obsacle? NO YES Simulaion inpu Scan environmen deecion sensors Any unknown saic or moving obsacle? NO Coninue manoeuvre Reach final poin? NO YES End ONLINE PLANNING Figure 5.1 Simulaion process flowchar. 58

75 Simulaion Resuls & Discussions 5. Simulaed vehicle overall lengh Ø l y w Rb CP Seering wheels R Ө x Driving wheels RP Figure 5. Geomeric model of a mobile robo. For simulaion environmen, he mobile robo used for his sudy is an Ackermannseering mobile robo which based on he modified remoe conrol car. I is assumed ha he mobile robo behaves as a ricycle which he rear wheels are he driving wheel and he fron wheel is a seering wheel. The geomery model of he mobile robo is modelled based on work by Liddy and Lu (7). However a few modificaions have been made o sui his sudy. In his projec, he cenre poin (CP) is locaed a he middle of he rear axle insead of cenre of he vehicle and insananeous urning poins (RP) is shown as in Figure 5.. The inpus for his simulaion environmen are velociy (v) and orienaion (θ) which calculaed in Secion 4.1. The following equaions will demonsrae he acual posiion, orienaion and seering angle during online planning. To find he seering angle (φ ): d = u + 1 v (5.1) Rb = d sin( θ / ) (5.) 59

76 Simulaion Resuls & Discussions 1 l φ = an (5.3) Rb where, d = disance beween ime inerval u = iniial velociy Rb = disance beween CP and RP. To find posiion of x and y, he equaions can be reversed by using seering angle and velociy as inpus, as shown by he following equaions: Rb = l anφ (5.4) 1 d / θ = sin (5.5) Rb θ new = θ old + θ (5.6) x y = x d cosθ (5.7) new old + = y d sin θ (5.8) new old Malab frameworks In his secion, he developmen of he simulaion frameworks using Malab is discussed. In order o simulae he mobile robo planner as close as possible o he acual environmen, he Laser Range Finder (LRF) is simulaed as shown in Figure

77 Simulaion Resuls & Discussions Scan angle Angle sep Range sep Deecion range Figure 5.3 Simulaed Laser Range Finder. The LRF is placed in fron of he mobile robo so ha he deecion coverage can be opimised. The coverage of he LRF is defined by he scan angle and he deecion range. And he resoluion of he LRF is defined by he angle sep and he range sep. In he acual LRF, only he angular resoluion is couned. However in order o obain he deecion disance once he mobile robo deecs an obsacle, he range sep is one of he approaches in he simulaion framework ha can be adoped o overcome his issue. The map can be generaed in any graphic edior sofware such as Pain. In his sudy, he obsacles are represened by square and circle as shown in Figure 5.4. The known and unknown saic obsacles are represened by black and green squares, respecively. And he moving obsacle is represened by red circle. The obsacles can be randomly placed in he simulaion map or can be arranged properly o indicae he fixed objecs in he acual environmen such as lamp poss and rees. 61

78 Simulaion Resuls & Discussions Figure 5.4 Simulaion map wih saic and moving obsacles. Furhermore, he Graphical User Inerface (GUI) was developed for he simulaion framework as shown in Figure 5.5. This inerface was developed o ensure he user will be able o run he simulaion wih ease and will give a user-friendly simulaion framework. Wih he GUI, he user will only need o key in he iniial and final sae of he mobile robo. In addiion, he user will also need o key in he physical daa of he mobile robo. (a) 6

79 Simulaion Resuls & Discussions (b) Figure 5.5 The Graphical User Inerface (GUI) for simulaion framework (a) Inpu GUI, (b) Oupu GUI 5.4 Trajecory opimizaion The original rajecory planning may exceed he resricion of he physical limiaions of he mobile robo such as maximum urning radius and maximum velociy or acceleraion. In real-life driving, he lower speed is preferred when he driver is closed o he obsacles or when he driver is making a sharp urn. Thus he mobile robo needs o follow a reasonable velociy profile in order o mimic he acual driving behaviour. In Secion 5.1, he archiecure of he simulaion has been discussed. As we know, he original offline rajecory planning will consider he limiaion of seering angle and velociy of he mobile robo. Furhermore, hese boundary condiions will also be considered in he online planning, meaning ha when he mobile robo sars o navigae he environmen following he original rajecory. An example of he generaed rajecory is shown in Figure

80 Simulaion Resuls & Discussions Figure 5.6 Original rajecory plan. The original rajecory is planned wih wo obsacles known and unknown saic obsacles. The map dimension is 5m x 5m and he ravel ime is se o be 6s. As shown in Figure 5.6, he original rajecory was planned pass hrough he unknown obsacle. This is because he algorihm only considered he known obsacle a he firs place. The unknown obsacle will be considered afer he sensor deecs he obsacle and he obsacle is poenially blocking he pah. Then he algorihm will planned a new rajecory in order o avoid he obsacle. The final resul of he rajecory is shown in Figure

81 Simulaion Resuls & Discussions Figure 5.7 Final resul of he rajecory. The orienaion of he mobile robo was changed dramaically as he mobile robo ries o avoid he obsacle as shown in Figure 5.8(a). If here is no oher obsacle deeced, he mobile robo will ry o follow he original rajecory as close as possible. Once i deeced an unknown obsacle, for example in his case, he mobile robo deeced an unknown obsacle a ime sep 33s, i has urned righ in order o avoid he obsacle. The decision of urning righ or lef is made by observing he locaion or posiion of he obsacle. For example, if he obsacle is a he lef region of he mobile robo respec o he mobile robo s orienaion, i will urn righ. Furhermore, he acual orienaions were given by he red line. The acual orienaions were slighly differen from he adjused orienaion due o he adjusmen made in order o saisfy he acual seering limiaion of he mobile robo during navigaion as shown in Figure 5.8(b). 65

82 Simulaion Resuls & Discussions (a) (b) Figure 5.8 Orienaion profiles (a) Planned orienaion (red line) agains adjused orienaion (red dashed), and (b) adjused orienaion (red dashed) agains acual orienaion (blue line). The seering angle plos are shown in Figure 5.9. The adjusmen on he robo s orienaion will also reflec he seering angle values of he mobile robo. Figure 5.9(a) shows he original planned seering angle values (red line) compared o he adjused on seering angle values (red dashed) once he mobile robo deeced an obsacle. The posiive values indicae he mobile robo is urning righ while he negaive values indicae he mobile robo is urning lef. The acual seering angle values are given by he blue plo as shown in Figure 5.9(b). (a) (b) Figure 5.9 Seering angle profiles (a) Planned seering angle (red line) agains adjused seering angle (red dashed), and (b) adjused seering angle (red dashed) agains acual seering angle (blue line). 66

83 Simulaion Resuls & Discussions One of he inpu parameer is he velociy of he mobile robo. Figure 5.1(a) shows he original planned velociy values which are given by he red plo and he adjused velociy values once he mobile robo deeced an obsacle which is given by he blue plo. Once he mobile robo deeced an obsacle, he algorihm ends o decrease he velociy of he mobile robo in order o avoid he obsacle smoohly. I is also mimic he acual human driver when he encouners an obsacle. The driver will ry o slow down is vehicle once he deecs an obsacle so ha he can seer seering smoohly. Then once he mobile robo was already avoided he obsacle, he mobile robo will speed up in order o pick up he ime los due o avoiding he obsacle. This is he reason why he adjused velociy value is higher han he original velociy afer 39s. In Figure 5.1(b), he acual velociy was mached wih he adjused velociy because he algorihm used he adjused velociy as inpu parameer for he mobile robo o manoeuvre. (a) (b) Figure 5.1 Velociy profiles (a) Planned velociy (red line) agains adjused velociy (red dashed), and (b) adjused velociy (red dashed) agains acual velociy (blue line). The final acual rajecory is shown in Figure The planned rajecory is given by he red dashed plo while he acual rajecory is given by he blue line plo. As we can see in Figure 5.11(b), he acual rajecory is slighly deviaed from he adjused rajecory due o he adjusmen made in order o caer he seering angle limiaion. This proved ha he algorihm is able o incorporae he kinemaic consrains of he mobile robo. 67

84 Simulaion Resuls & Discussions Close-up figure in (b) (a) (b) Figure 5.11 Adjused rajecory (red dashed) agains acual rajecory (blue line). 68

85 Simulaion Resuls & Discussions Replanning approach Replanning approach is inroduced o reduce he errors while he mobile robo navigaes hrough he inermediae waypoins. When he mobile robo reaches he waypoin, he replanning approach will be execued and a new rajecory will be generaed from he curren inermediae waypoin o he nex waypoin. The curren daa a he waypoin such as posiion and velociy will be used o generae he new rajecory. The replanning approach scenario is in Figure 5.1. The black areas represen he walls and/or known obsacles. The waypoins are represened by red circles. Poin 1 and Poin 4 are he iniial and final poins, respecively. Poin and Poin 3 are he desired waypoins. The orienaion a each poin is indicaed by an arrow. The inpu daa for his simulaion are summarized in Table 5.1. The conrol inpus for his simulaion are seering angle and velociy. Table 5.1 Inpu daa for replanning approach scenario Poin (sec) x (m) y (m) θ ( o ) ø ( o ) v (m/s)

86 Simulaion Resuls & Discussions Figure 5.1 Prior map wih wo waypoins connecing he iniial and final poin. The simulaion algorihm consiss of boh offline and online planning componens. The offline planning deals wih he known obsacles and i will be execued a he iniial sage. Then, he online planning will be execued once he mobile robo sars navigaing in he environmen. The online planning is o deec and deal wih unknown obsacles Table 5. summarizes he acual colleced daa a every waypoin wihou replanning approach. In comparison o inpu daa for simulaion in Table 5.1, he errors in posiion and orienaion a he final poin are around.9 meers and.3 degrees, respecively. These errors are quie large, especially for posiion error. Table 5. Acual colleced daa of simulaion wihou replanning approach Poin (sec) x (m) y (m) θ ( o ) ø ( o ) v (m/s)

87 Simulaion Resuls & Discussions In he scenario wihou replanning approach, he mobile robo will follow he iniial rajecory. While in he scenario wih replanning approach, he new rajecory will be generaed once he mobile robo reached he waypoin. The simulaion resuls for replanning approach are shown in Figure 5.13 Simulaion resuls wih replanning approach. 3 (a) A ime sep = 1 sec (b) A ime sep = 1 sec 4 Iniial rajecory 3 New rajecory (c) A ime sep = 41 sec (d) A ime sep = 6 sec Figure 5.13 Simulaion resuls wih replanning approach. Figure 5.13(a) shows he iniial planned rajecory from he iniial poin o he final poin, and pass hrough all he waypoins. Then he mobile robo navigaes along he iniial planned rajecory unil i reaches Poin. Once i reaches Poin, he replanning algorihm is execued. Using he acual daa a Poin, a new rajecory is generaed from Poin o Poin 3, as shown in Figure 5.13(b). The new rajecory is almos idenical o he iniial rajecory because he errors are quie small. Then he 71

88 Simulaion Resuls & Discussions mobile robo coninues is journey along he new rajecory, avoiding he obsacle and reaches Poin 3. A his poin, replanning algorihm once again is execued and a new rajecory is generaed from Poin 3 o he final poin, as shown in Figure 5.13(c). The mobile robo hen coninues is journey and finally reaches he final poin, as shown in Figure 5.13(d). The acual colleced daa a every waypoin are summarized in Table 5.3. Table 5.3 Acual colleced daa wih replanning approach Poin (sec) x (m) y (m) θ ( o ) ø ( o ) v (m/s) From he, he daa a boh waypoins - Poin and Poin 3 - are no much differen from daa in Table 5.. This is because a Poin, he replanning approach was no ye been execued. While a Poin 3, he rajecory was affeced by he obsacle avoidance algorihm. Therefore he significan errors differen can be perceived a he final poin. Wih replannning approach, he errors in posiion and orienaion are around.1 meers and 3. degrees, respecively. Alhough he error in orienaion is greaer han he previous orienaion error, i is considered as sill in saisfacory limis, wih he consideraion of boh posiion and orienaion errors. 5.5 Simulaion resuls and discussions In he previous secion, he simulaion framework used a simple example o explain how he simulaion and algorihms work. The generaed map was consised of wo saic obsacles. In his secion, more complicaed examples are presened. A series of simulaion cases have been seup o invesigae he capabiliy and effeciveness of he algorihms. The scenario of he simulaions will include more mobile robos, obsacles and more complicaed environmens. All he simulaions were conduced in Malab. Seering angle and velociy of he mobile robo have been used as he conrol inpu parameers for hese simulaions. A 7

89 Simulaion Resuls & Discussions few general assumpions have been made for he modelled Ackermann seering carlike robo in hese simulaions: 1. The mobile robo moves on horizonal plane no opological effecs,. Single poin conac of he wheels, 3. The wheels are no deformable, 4. No slipping, skidding or fricion, and 5. The wheels are aached a he rigid chassis Navigaion in saic and open-space environmens The firs se of simulaion cases is conduced in he saic and open-space environmens. In hese scenarios, he here are only saic obsacles and he map is se as an open space are such as a field or large area. The complicaed obsruced environmen is a 1m x 1m region and he obsacles are randomly placed in he map consising known and unknown saic obsacles as shown in Figure The environmen is esed by one, wo and hree robos wih various iniial and final poins. Figure 5.14 A complicaed obsruced environmen. 73

90 Simulaion Resuls & Discussions In Case 1, he inpu daa is abulaed in Table 5.4. In his case, only one mobile robo was used o navigae in he environmen. The mobile robo sared from he boomlef of he map as shown in Figure 5.15(a). The red line is he original planned rajecory wihou considering he known saic obsacles, while he blue line is he pre-planned rajecory wih he consideraion of he known saic obsacles. A 9s, he mobile robo deeced an obsacle and he new rajecory was generaed in order o avoid he obsacle as shown in Figure 5.15(c). The final resul a 6s is shown in Figure 5.15(d). Table 5.4 Inpu daa for simulaion Case 1. Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Iniial 45 Final

91 Simulaion Resuls & Discussions (a) (b) (c) Figure 5.15 One mobile robo navigaes in he environmen. (d) The comparisons beween he planned and acual orienaion, seering angle, velociy and locaion of Robo 1 are shown in Figure Due o he presence of he unknown saic obsacles, he rajecory was adjused in order o avoid he unknown obsacles. From Figure 5.16, he mobile robo ried o follow he planned inpu from he beginning unil i encounered he unknown saic obsacles. The acual posiion of he mobile robo a he final poin is abulaed in Table

92 Simulaion Resuls & Discussions (a) (b) (c) (d) Figure 5.16 Robo 1: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) locaion. Table 5.5 Acual daa colleced a he final poin for Case 1 (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Planned Acual

93 Simulaion Resuls & Discussions In Case, wo mobile robos were used and he inpu daa for each robo are abulaed in Table 5.6. The firs mobile robo sared from he boom-lef of he map and he second mobile robo sared from righ-hand side of he map. Each robo navigaes o he differen final poin as shown in Figure 5.17(a). The ravel ime for boh mobile robos was assigned o 6 seconds. As we can see in Figure 5.17(b) and (c), boh mobile robos were capable o deec and avoid he obsacle. The final resul a 6s is shown in Figure 5.17(d). Table 5.6 Inpu daa for simulaion Case. Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Iniial 45 Final Iniial Final

94 Simulaion Resuls & Discussions (a) (b) (c) Figure 5.17 Two mobile robos navigae in he environmen. (d) The comparisons beween he planned and acual orienaion, seering angle, velociy and locaion for Robo 1 are similar o Case 1 as shown in Figure For Robo, he comparisons are shown in Figure Due o he presence of he unknown saic obsacles, he rajecory was adjused in order o avoid he unknown obsacles. The acual posiion of he mobile robo a he final poin is abulaed in Table

95 Simulaion Resuls & Discussions (a) (b) (c) (d) Figure 5.18 Robo : Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion. Table 5.7 Acual daa colleced a he final poin for Case Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Planned Acual Iniial Final

96 Simulaion Resuls & Discussions In he hird case (Case 3), hree mobile robos were used and he inpu daa for each mobile robo are abulaed in Table 5.8. The firs mobile sared from he boom-lef of he map, he second mobile robo sared from he righ-hand side of he map and he hird mobile robo sared from he lef-hand side of he map as shown in Figure 5.19(a). The ravel ime for he firs and he second mobile robo was assigned as 6 seconds, while he hird mobile robo was 4 seconds. A 4s, he hird mobile robo was already reached he final poin as shown in Figure 5.19(c). The final resul a 6s is shown in Figure 5.19(d). Table 5.8 Inpu daa for simulaion Case 3. Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Robo 3 Iniial 45 Final Iniial Final Iniial 5 Final

97 Simulaion Resuls & Discussions (a) (b) (c) Figure 5.19 Three mobile robos navigae in he environmen. (d) The comparisons beween he planned and acual orienaion, seering angle, velociy and locaion for Robo 1 and Robo are similar o Case 1 and Case, respecively. For Robo 3, he comparisons are shown in Figure 5.. The planned and acual rajecory for Robo 3 is almos idenical because Robo 3 did no encouner any unknown saic obsacles. The acual posiion of he mobile robo a he final poin is abulaed in Table

98 Simulaion Resuls & Discussions (a) (b) (c) (d) Figure 5. Robo 3: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion. Table 5.9 Acual daa colleced a he final poin for Case 3 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Robo 3 Planned Acual Planned Acual Planned Acual From he resuls of Case 1, Case and Case 3, he mobile robos were safely reached he final poins wih he capabiliy o avoid known and unknown saic obsacles. They have ried o follow he planned rajecories closely. However, here is a sligh 8

99 Simulaion Resuls & Discussions differen in orienaion and seering angle of he mobile robos due o limiaion of he seering angle. For example, a he beginning of he journey, seering angle for Robo 3 is 5 o. Thus he algorihm has adjused he seering angle o 15 o so ha he limiaion is no exceeded and he mobile robo can urn smoohly. The acual posiions of all he mobile robos a every ime sep are close o he planned posiion and he acual posiions a he final poin are summarized in Table 5.9. The maximum relaive error for x-posiion is.7 m, y-posiion is.7 m, orienaion is.15 o and seering angle is o Navigaion in dynamic and open-space environmens In his secion, he previous examples are exended o dynamic environmens. Moving obsacles are added ino he environmen and a series of simulaion cases are presened in order o invesigae he capabiliy of he algorihms. In his sudy, he moving obsacle is a mobile robo ha moves along he predefined pah. Figure 5.1 Simulaed environmen for Case 4 The firs scenario (Case 4) discussed in his secion involves one mobile robo and one moving obsacle in a complicaed obsruced environmen as shown in Figure

100 Simulaion Resuls & Discussions The inpu daa for he mobile robo is abulaed in Table 5.1. The moving obsacle sared from he lef-hand side of he map. Table 5.1 Inpu daa for simulaion Case 4 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Iniial 45 Final Figure 5. shows he simulaion resuls for Case 4. The iniial rajecory for he mobile robo and he moving obsacle is represened by red line as in Figure 5.(a). However afer consideraion of known saic obsacles in he environmen, he new iniial rajecory for he mobile robo is generaed and represened by blue line. Once he rajecory generaion is compleed, he mobile robo sars o move along he rajecory. A 17 seconds, he mobile robo deecs a moving obsacle as shown in Figure 5.(b). The algorihm for avoiding he moving obsacle is execued and he new deviaion rajecory is generaed. As in Figure 5.(c) and (d), he mobile robo sars o avoid he moving obsacle and reduce is speed o ensure he moving obsacle avoided before he mobile robo increase is speed o cach up he ime los during avoiding he obsacle. The mobile robo will coninue o navigae along he new generaed rajecory unil i deecs an unknown saic obsacle a 31 seconds as shown in Figure 5.(e). Then he obsacle avoidance algorihm for avoiding a saic obsacle is execued and a new deviaed rajecory is generaed. The mobile robo will coninue o navigae along he new rajecory and reaches he final poin as shown in Figure 5.(f). 84

101 Simulaion Resuls & Discussions (a) (b) (c) (d) (e) Figure 5. One mobile robo navigaes in a dynamic environmen. (f) 85

102 Simulaion Resuls & Discussions The comparisons beween he planned and acual orienaion, seering angle, velociy and locaion for Robo 1 are shown in Figure 5.3. The acual posiion of he mobile robo a he final poin is abulaed in Table (a) (b) (c) (d) Figure 5.3 Robo 1: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion. Table 5.11 Acual daa colleced a he final poin for Case 4 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Planned Acual

103 Simulaion Resuls & Discussions The nex wo cases involve muliple mobile robos and muliple moving obsacles in dynamic environmen. In he second case, wo mobile robos and wo moving obsacle are used as shown in Figure 5.4. The firs mobile robo, R1 sared from he boomlef of he map and he second mobile robo, R sared from he righ-hand side of he map. The firs moving obsacle, M1 sared from he lef-hand side of he map which he iniial poin is (,3) and he second moving obsacle, M sared from he righ hand side of he map which he iniial poin is (1,7). Figure 5.4 Simulaed environmen for Case 5 The inpu daa for each mobile robo are abulaed in Table 5.1. Table 5.1 Inpu daa for simulaion Case 5 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Iniial 45 Final Iniial Final

104 Simulaion Resuls & Discussions Figure 5.5 shows he simulaion resuls for Case 5. As we can see, he iniial rajecory for mobile robo R is represened only by he blue line since here is no obsacle along he original iniial rajecory as shown in Figure 5.5(a). Once he offline planning is compleed, boh mobile robos sar o navigae along heir rajecories. A 15 seconds, mobile robo R deecs a moving obsacle, which is moving obsacle M, as shown in Figure 5.5(b). Then he algorihm for avoiding a moving obsacle is execued and he new deviaed rajecory is generaed. As we can see in Figure 5.5(c), mobile robo R is slowing down in order o ensure he moving obsacle passes. In he case of mobile robo R1, he resul is similar o he previous simulaion resul. Finally, boh mobile robo R1 and R, safely reach he final poin as shown in Figure 5.5(d). (a) (b) (c) Figure 5.5 Two mobile robos navigae in a dynamic environmen. (d) 88

105 Simulaion Resuls & Discussions The comparisons beween he planned and acual orienaion, seering angle, velociy and locaion for Robo 1 are similar o Case 4. For Robo, he comparisons are shown in Figure 5.6. The acual posiion of he mobile robo a he final poin is abulaed in Table (a) (b) (c) (d) Figure 5.6 Robo : Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion. Table 5.13 Acual daa colleced a he final poin for Case 5 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Planned Acual Planned Acual

106 Simulaion Resuls & Discussions In he hird case, here are hree mobile robos and wo moving obsacles in he environmen as shown in Figure 5.7. The iniial poins for boh moving obsacles are se similar o he previous case. However, in his case, wo mobile robos (R and R3) are planned o mee each oher a he final poins a he desired ime. This ype of scenario may have implicaion in he real world such as goods exchange and goods delivery beween robos a he same locaion. Figure 5.7 Simulaed environmen for Case 6. The inpu daa for all he mobile robos are abulaed in Table Table 5.14 Inpu daa for simulaion Case 6 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Robo 3 Iniial 45 Final Iniial Final Iniial 8 Final

107 Simulaion Resuls & Discussions (a) (b) (c) Figure 5.8 Three mobile robos navigae in a dynamic environmen (d) Figure 5.8 shows he simulaion resuls for Case 3. The iniial generaed rajecories for all he mobile robos are shown in Figure 5.8(a), which are represened by blue line. Once he offline planning is compleed, all mobile robos sar o move along heir rajecories. In his case, we only discuss he movemen of mobile robo R3 as he oher wo mobile robos moions are similar o previous cases. A seconds, mobile robo R3 deecs an unknown saic obsacle and a new deviaed rajecory is generaed. As we can see in Figure 5.8(b), here is a known saic obsacle near o he newly generaed rajecory. If he saic obsacle is blocking he rajecory, he offline planning will be execued along wih he obsacle avoidance algorihm. However, in his case, he known saic obsacle is no blocking he way. Thus, he mobile robo R3 will coninue is journey along he new rajecory. A 37 seconds, i deecs a moving 91

108 Simulaion Resuls & Discussions obsacle (M), which is coming from he righ side, as shown in Figure 5.8(c). The algorihm for avoiding a moving obsacle is execued and a new deviaed rajecory is generaed. As we can see, mobile robo R3 reduce is speed in order o make sure he moving obsacle M passes. Afer ha, mobile robo R3 coninues is journey and reaches he final poin, as shown in Figure 5.8(d). The comparisons beween he planned and acual orienaion, seering angle, velociy and locaion for Robo 1 and Robo are similar o Case 4 and Case 5, respecively. For Robo 3, he comparisons are shown Figure 5.9. The acual posiion of he mobile robo a he final poin is abulaed in Table (a) (b) (c) (d) Figure 5.9 Robo 3: Planned (red) agains acual (blue) plo for (a) orienaion, (b) seering angle, (c) velociy, and (d) posiion. 9

109 Simulaion Resuls & Discussions Table 5.15 Acual daa colleced a he final poin for Case 6 Poin (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s) Robo 1 Robo Robo 3 Planned Acual Planned Acual Planned Acual From he resuls of Case 1, Case and Case 3, he mobile robos were safely reached he final poins wih he capabiliy o avoid known and unknown saic obsacles as well as dynamic obsacles. The resuls show ha he algorihms are capable o deec and avoid no only saic obsacles, bu also dynamic obsacles. Furhermore he enire mobile robos were capable o follow he planned rajecories closely. The acual posiions for he mobile robos are summarized in Table The maximum relaive error for x-posiion is. m, y-posiion is.3 m, orienaion is 1.1 o and seering angle is 18.6 o Navigaion in he ciy-like environmens In his secion, he simulaions are based on he muliple waypoins rajecory planning in a ciy-like environmen as shown in Figure 5.3. All he parameers used for he mobile robos; R1 and R, are lised in Table 5.16 and Table 5.17, respecively. Norh direcion of he map is se poining up on he map. As lised in Table 5.16, R1 sars from he boom of he map a poin (1, ) and facing norh. A he firs juncion, i needs o urn righ. The firs and second waypoins are se o ensure he mobile robo can urn a he juncion smoohly. Then, i needs o move along he road unil i reaches he second juncion. I hen needs o urn lef and move unil i reaches he final poin (6, 17) a he 1 h second wih 9 o orienaion. 93

110 Simulaion Resuls & Discussions Table 5.16 Parameers for he firs mobile robo (R1) Poins (sec) x (m) y (m) Ө (o) Ø (o) v (m/s) As lised in Table 5.17, R sars from he righ side of he map a poin (183,6) and facing wes. Then a he juncion, i needs o urn righ and move along he road unil i reaches he final poin (8,17) a he 1 h second wih 6 o orienaion. Noe ha he road is iled a abou 6 o from x-axis. Table 5.17 Parameers for he second mobile robo (R) Poins (sec) x (m) y (m) Ө ( o ) Ø ( o ) v (m/s)

111 Simulaion Resuls & Discussions (a) (b) Figure 5.3 (a) A simplified ciy-like map, (b) Muliple waypoins rajecory planning. 95

112 Simulaion Resuls & Discussions In addiion, here are wo moving obsacles in he map as shown in Figure The firs moving obsacle (M1) sars from he norh of he map and moves sraigh down o he souh of he map. The iniial and final poin for M1 is (61, 18) and (61, 8), respecively. The second moving obsacle sars from eas of he map and finishes a he middle of he map. The iniial and final poin for M is (18, 113) and (9, 113), respecively. Boh moving obsacles move from heir respecive iniial poins and reach heir final poins a he 1 h second. Figure 5.31 Iniial rajecories in a ciy-like map. The iniial rajecories for mobile robos and moving obsacles are shown in Figure Once all he rajecories were generaed, he mobile robos and he moving obsacles were sared o move along heir respecive rajecories. A he 9 h second, R deeced an unknown saic obsacle as shown in Figure 5.3(b). Then a new rajecory was generaed from he deecion poin o he closes waypoin, which was in his case he firs waypoin. R sared o move along he new rajecory as shown in Figure 5.3(c) and reached he firs waypoin a he 3 h second as shown in Figure 5.3(d). 96

113 Simulaion Resuls & Discussions (a) (b) (c) (d) Figure 5.3 (a) Before deecing an obsacle. (b) Obsacle deeced a he 9 h second. (c) Sars o move along new rajecory. (d) Reaches he firs waypoin a he 3 h second. Furhermore, a he 67 h second, R deeced a moving obsacle (M) coming from he righ side of i as shown in Figure 5.33(b). I hen predics wheher i migh collide wih he moving obsacle or no. In his case, collision is expeced o happen and a new rajecory is generaed from deecion poin o he closes waypoin, which is he final poin, based on he obsacle avoidance algorihm of a moving obsacle. Then R sared o move along he new rajecory as shown in Figure 5.33(c). Also as we can see, he mobile robo acually slowed down o cauiously passing hrough he moving obsacle as shown in Figure 5.33(d). 97

114 Simulaion Resuls & Discussions (a) (b) (c) (d) Figure 5.33 (a) Before deecing an obsacle. (b) Obsacle deeced a he 67 h sec. (c) Sars o move along new rajecory. (d) Passes hrough moving obsacle safely. As we can see in Figure 5.31, he iniial rajecory for R1 was already considered known saic obsacles during offline planning. Then R1 sared o move along he iniial rajecory and passes hrough all he waypoins. However a he 68 h second, R1 deeced a moving obsacle (M1) as shown in Figure 5.34(b). Also R1 checked he direcion of moving obsacle and in his case, M1 came from he opposie direcion of R1. Therefore, M1 was reaed as a saic obsacle and a new rajecory was generaed from he curren poin o he final poin, hrough he deviaion poin. Then R1 sared o move along he new rajecory and safely avoided M1 as shown in Figure 5.34(c) and (d). Furhermore, afer avoiding he moving obsacle, R1 deeced an unknown saic obsacle a he 86 h second and successfully avoided i. Figure 5.35 shows he final overall simulaion resuls a he 1 h second. 98

115 Simulaion Resuls & Discussions (a) (b) (c) (d) Figure 5.34 (a) Before deecing an obsacle. (b) Obsacle deeced a he 68 h sec. (c) Sars o move along new rajecory. (d) Passes hrough moving obsacle safely. As we can see, boh mobile robos reached he final poin a he specified ime, posiion and orienaion wih cerain errors as shown in Table The errors are reasonably small as a resul of he online planning approach. A every ime sep, he online planner will use he acual daa o ge o he nex pre-planned posiion of he mobile robo. This means he planner will need o deermine a new seering angle using he acual posiion and orienaion, and he pre-planned velociy of he mobile robo. This pracice will eliminae or a leas reduce he errors a every ime sep. Furhermore, he mobile robos successfully passed hrough all he waypoins and avoided all he saic and moving obsacles. 99

116 Simulaion Resuls & Discussions Table 5.18 Table 3 Errors for Case 1 a final poin. Acual Relaive error x (m) y (m) Ө ( o ) x (m) y (m) Ө ( o ) Robo Robo Figure 5.35 Final resul a he 1 h second. 1

117 Simulaion Resuls & Discussions In addiion, wo more simulaion cases have been conduced o invesigae he effeciveness of he algorihms. Figure 5.36 shows he second simulaion scenario wih wo mobile robos, R1 and R, and one moving obsacle, M1. Figure 5.36 Second scenario wih wo mobile robos and one moving obsacle. Using he similar map seup o he firs case, he inpus for boh mobile robos are abulaed in Table As we can see in Figure 5.36, he pre-planned rajecories are presened by he blue line wih he consideraion of he known saic obsacles. The moving obsacle is se o move along he road from iniial poin (61,17) o final poin (61,8). 11

118 Simulaion Resuls & Discussions Table 5.19 Parameers for second simulaion case Poins (sec) x (m) y (m) Ө (o) Ø (o) v (m/s) Robo Robo Figure 5.37 Final resul a he 1 h second for second scenario. 1

119 Simulaion Resuls & Discussions The final resuls are shown in Figure As we can see in he final resuls, he mobile robos are capable o navigae safey and avoid he moving obsacle as well as he saic obsacles and reach he final poins a he specified ime. The errors a he final poin are abulaed in Table 5.. The errors are reasonable and sill fall wihin he saisfacory limis as he mobile robos do no deviae oo far from he final poins, considering he disance ha he mobile robos have ravelled. The maximum final errors for posiions are.6 m for R1 and.37 m for R, while he maximum orienaion errors are.818 o for R1 and o for R. Table 5. Errors for Case a he final poin. Acual Relaive error x (m) y (m) Ө ( o ) x (m) y (m) Ө ( o ) Robo Robo In he hird case, he scenario is exended wih hree mobile robos and wo moving obsacles are used as shown in Figure The mobile robos sared a he differen iniial poins and moved o he differen final poins as abulaed in Table 5.1. The ravel ime for each mobile robo is se o 1 second. This case is conduced o demonsrae he capabiliy of he algorihms o handle he differen ravel ime and o demonsrae muliple robos coordinaion in he unknown environmen. 13

120 Simulaion Resuls & Discussions Figure 5.38 Third scenario wih hree mobile robos and wo moving obsacles. Table 5.1 Parameers for hird simulaion case Poins (sec) x (m) y (m) Ө (o) Ø (o) v (m/s) Robo Robo Robo

121 Simulaion Resuls & Discussions Figure 5.39 Final resul a 1 h second for hird scenario. From he final resuls are shown in Figure 5.39, all he mobile robos reached he final poins a he specified ime. As we can see he final errors in Table 5., he posiion errors for each robo are accepable. Even hough he orienaion errors for R1 and R are larger han he second case, he resuls are sill wihin he saisfacory limi as he maximum orienaion error only 1.5 o. Table 5. Errors for Case 3 a he final poin. Acual Relaive error x (m) y (m) Ө ( o ) x (m) y (m) Ө ( o ) Robo Robo Robo

122 Simulaion Resuls & Discussions 5.6 Concluding remarks In his chaper, he Malab was adoped for developmen of simulaions and o implemen and es he algorihms for mobile robo navigaion. The algorihm was esed hrough a series of he simulaion seup. The maps for he simulaion works were adoped from open-space environmen and ciy-like environmens which is more complicaed. The simulaion resuls show he mobile robo was able o follow he planned rajecory as close as possible. Furhermore, he mobile robo was able o reach close o he final poin a he desired ime. The algorihm also was able o simulae he differen environmen seup for he mobile robo as well as for muliple robos wih he presence of dynamic obsacles. However, here are errors occurred beween he planned rajecories and acual rajecories due o he acual calculaion of he posiion and seering angle of mobile robos. For example, he acual calculaion of he seering angle is aking ino consideraion of he curren daa a every ime sep o calculae he nex moion of he mobile robo. Thus his cumulaive error caused he sligh differen beween he planned and acual a he final poin. Furhermore, he developed GUI framework ensures he user able o modify he seings of he mobile robo easily. The user only needs o modify he seings a he GUI framework wihou inerfering he conrol funcions of he algorihms. 16

123 Developmen of a Mobile Robo 6. DEVELOPMENT OF A NONHOLONOMIC MOBILE ROBOT The developmen of a mobile robo is based on he applicaion and requiremen of he mobile robo in he environmen. In his sudy, a nonholonomic mobile robo is used o navigae in he oudoor environmen. Thus a car-like robo is preferred as i can be convered from a sandard car and able o ravel in he large oudoor environmens. Therefore, he mobile robo used for he experimenal works was convered from a sandard remoe conrol car as shown in Figure 6.1. The wheelbase lengh and widh of he mobile are 174 mm and 191 mm, respecively. I has a similar srucure o he normal car wih fron seering wheels and rear driving wheels. All four wheels have he same diameer which is 69 mm. The rear wheels are convenional fixed wheels on he rear axle and he fron wheels are cenred urning wheels on he fron axle. The seering wheels are assumed o urn a he same angle and aced as a single wheel locaed a he middle of he fron axle as discussed in Chaper 4. 17

124 Developmen of a Mobile Robo Figure 6.1 The modified car-like robo used in experimenal works. As he focused mobile robo is a nonholonomic mobile robo, an Ackermann seering robo is required as he mobile robo plaform. Furhermore he selecion of he mobile robo plaform should fulfil a few selecion crieria for his sudy such as: Seering should be driven by a digial servomoor. The driving wheel should be driven by a moor. Enough size o hos all he sensors and he microconroller. Easy access o all he componens aached o he car. The basis of he mobile robo plaform is shown in Figure 6.. The mobile robo has a moor ha driven he rear wheels and aced as driving wheel, while he fron wheels are seered by a servo moor and aced as an Ackermann seering wheels. In addiion he RC car is powered by baery pack. This RC car needs o be modified in order o insall all he sensors, microconroller, baery pack and oher accessories. 18

125 Developmen of a Mobile Robo Figure 6. Mobile robo plaform. The sensor plaform is consruced as shown in Figure 6.3. The plaform is designed o be as simple as possible and he maerial used is acrylic. This maerial has a few advanages such as ligh weigh and durable. The plaform is a wo ier plaform which he lower ier is used o insall baery pack and he upper plaform is used o insall sensors, microconroller and oher accessories. Figure 6.3 Sensor plaform 19

126 Developmen of a Mobile Robo The final aachmen of he sensor plaform o he mobile robo plaform is shown in Figure 6.4. (a) (b) (c) Figure 6.4 Sensor plaform aached o he mobile robo plaform. (d) 6.1 Robo conroller A conroller is essenial for an auonomous mobile robo in order o conrol he mobile robo. The robo conroller is used o process he raw daa from he sensors as well as o ransmi he processed daa o he PC. In his sudy, he Oranguan SVP robo conroller is seleced as he main robo conroller as shown in Figure 6.5. This robo conroller is simple and a complee soluion for small and medium-sized robos. 11

127 Developmen of a Mobile Robo Figure 6.5 Robo conroller The feaures in his robo conroller fulfil he requiremens for a nonholonomic mobile robo developed in his sudy.. The module is design based on he powerful Amel microconroller. I has wo moor drivers, a demuliplexer o conrol servo moors, I/O lines ha can be used as analog and digial inpus and also he auxiliary processor ha can read wo quadraure encoders. In addiion, he advanages of his microconroller are easy o program he algorihms as i has exensive sofware libraries for he compiler and i is compaible wih all developmen sofware for Amel s AVR microconroller. Deails funcions and specificaions are given in Appendix B. 6. Wheel encoder The purpose of he encoder is o provide feedback on he speed and ravelling disance of he mobile robo. In his sudy, he magneic encoder is used and aached o he wheel as shown in Figure 6.6(a). Magneic encoder is chosen due o is simpliciy and can provide a beer accuracy for a small mobile robo. In he magneic encoder, he Hall Effec sensors are used as ransducers in which he oupu volage is varied by he changes in magneic field densiy. The Hall Effec sensors physical appearance is shown in Figure

128 Developmen of a Mobile Robo (a) (b) Figure 6.6 (a) Magnes mouning aached a he wheel (b) Hall Effec sensors aached a he rear axle. (a) Figure 6.7 Hall effec sensor (b) The magneic encoder is designed o sui he dimension of he wheel, which he readily available encoder may no be suiable for he specific model of he mobile robo. Deails of he Hall Effec sensor are given in Appendix B. The final aachmen of he magneic encoder o he mobile robo is shown in Figure

129 Developmen of a Mobile Robo (a) (b) Figure 6.8 Locaion of he wheel encoder The magnes mouning has 16 magnes ha has been arranged o cover 36 degrees as shown in Figure 6.9. The calculaion of disance is given by pulse o θ = θ disance = π r 36 (6.1) where, Ө is resoluion of wheel and r is he radius of wheel. The pulse is obained once he Hall Effec sensor passing hrough he magne. The oal number of pulses are couned a every ime sep and hen are used o calculae he disance. magne Ө Figure 6.9 Magne mouning of encoder 113

130 Developmen of a Mobile Robo 6.3 Deecion sensors Sensors for obsacle deecion are essenial in robo navigaion. Infrared, ulrasonic, vision camera and laser range finder are among he sensors ha have been used in obsacle deecion. Laser range finders provide precise and sable range reading, however he drawback is he cos of he LRF is expensive. Using a vision camera is a beer opion o deec he obsacles; however i needs a significan compuaional power in order o process he daa and may be a burden o a small robo. Infrared sensors only provide a single line deecion which more suiable o use as a range sensor. Ulrasonic sensor is he bes soluion o use as a deecion sensor for a small robo. I can provide he similar deecion funcion as he LRF ha adoped in he algorihm, which is o deermine he region of deecion relaes o he posiion of he mobile robo. The region can be on cener, lef or righ of he mobile robo. The ulrasonic sensor adoped in his sudy is he Devanech SRF5 as shown in Figure 6.1. This sensor can cover up o 45o angle and has a deecion range from 3 cm o 4 m. (a) (b) Figure 6.1 (a) Ulrasonic range sensors (b) Sensor aached o he sensor base. 6.4 Communicaion In his sudy, communicaion beween he mobile robo and he PC is required as all he daa need o send o he PC o process. The decision is made based on he daa in he PC and hen he resuls will be sen back o he mobile robo in order o navigae in he environmen. Thus, a wireless communicaion is preferable as he mobile robo will move away from he PC. 114

131 Developmen of a Mobile Robo The wireless communicaion module adoped in his sudy is Xbee/Zigbee RF module as shown in Figure The ransmiing range for his module can be ranged up o 3 m for indoor and 1 m for oudoor environmens. The wireless communicaion consiss of a rouer and a coordinaor. The rouer is aached o he mobile robo, while he coordinaor is plug ino he PC. Deail specificaions are given in Appendix B. (a) (b) Figure 6.11 Wireless communicaion (a) Rouer (b) Coordinaor 6.5 Calibraion of seering angle and velociy Prior o he experimenal work for he mobile robo in he real environmen, he calibraion works are required o esablish he acual seering angle and he speed of he mobile robo. The seering angle and he speed are conrolled by he PWM (Pulse- Widh Modulaion) values. Thus, he relaion beween he PWM values and seering angle as well as he relaion beween he PWM values and velociy of he mobile robo need o be esablished. The calibraion work was conduced in he open-space and fla area Seering angle The calibraion work for he seering angle was conduced by seing he seering angle o he consan PWM. Consan seering angle resuls in a drive along a circle. The radius of he circle was hen been measured and he relaionship beween he seering angle and PWM can be esablished. The daa for calibraion work of he seering angle are abulaed in Table

132 Developmen of a Mobile Robo Table 6.1 Seering angles under differen PWM values PWM (hex) Diameer 1 (cm) Diameer (cm) Ave diameer (cm) Seering angle ( o ) inf inf A A The reference poin for he radius of he circle is shown as in Figure 6.1. Wih radius of circle and disance beween fron and rear wheels, he seering angle can be calculaed from he following equaion ϕ = L/R (6.) where is L = lengh beween fron and rear wheels I was assumed ha he seering angle is a virual middle wheel beween wo fron wheels and disance from cenre poin o ICC is similar o radius of he circle. Seering wheels Mobile robo L Radius ICC Driving wheels Cenre poin Figure 6.1 Calibraion work for esablishmen of seering angle 116

133 Developmen of a Mobile Robo From he daa in Table 6.1, he relaion beween he PWM and he seering angle is shown in Figure The following linear relaions were obained by he Leas Square Mehod: φ =.65 PWM , for urning lef, (6.3) φ =.6717 PWM , for urning righ. (6.4) For zero seering angle, he PWM value of 13 is used. 3 5 Seering angle (degree) urn righ urn lef 6.5. Velociy PWM Figure 6.13 Relaion beween PWM values and seering angle. The speed conrol is crucial o he ime-criical moion planning as i can reflec he moion of he mobile robo. The speed is also conrolled by he PWM value. The calibraion work was conduced by aking ime for 1 m disance from sar o final poin for each PWM value as shown in Figure Sar 1 mm Finish f Figure 6.14 Calibraion work for esablishmen of velociy 117

134 Developmen of a Mobile Robo The daa for calibraion work of velociy are abulaed in Table 6.. Table 6. Velociies under differen PWM values PWM (hex) Time 1 (s) Time (s) Time 3 (s) Ave ime (s) Vel (cm/s) From he daa in Table 6., he relaion beween he PWM and he speed is shown in Figure Wih he Leas Square Mehod, he following equaion was esablished: v =.38 PWM.636. (6.5) Noe ha, he measured speeds were obained during he baery is fully charged. 5 Speed (cm/s) PWM Figure 6.15 Relaion beween PWM values and speed. 118

135 Developmen of a Mobile Robo 6.6 Obsacle deecion In he experimenal works, an ulrasonic sensor was used o deec obsacles during mobile robo navigaion. Ulrasonic sensors have been proven o be effecive o deec an obsacle in he acual environmen. The ulrasonic sensor was placed a he fron of he mobile robo. For he purpose of he experimenal works in his sudy, he deecion range is se o be 1 cm and he deecion angle for an ulrasonic sensor is around 45º as shown in Figure The full specificaions of he ulrasonic sensor used in his sudy can be referred in Appendix A. ~84 cm 1 cm ~45 o Figure 6.16 Obsacle deecion range for experimenal works. 119

136 Developmen of a Mobile Robo 6.7 Wireless communicaion In order o achieve remoe conrol over he mobile robo, he Xbee RF module is used. The wireless connecion configuraion is shown in Figure GHz GPIO USB Rouer Coordinaor Figure 6.17 Wireless communicaion beween he operaor and he rouer (robo). For his ask, wo Xbee RF modules are used as a Coordinaor and a Rouer. Basically, he Coordinaor is conneced o he PC via USB and he Rouer is aached o he mobile robo via he General Purpose Inpu/Oupu (GPIO) por. Once he iniial collision-free rajecory was generaed, he daa oupu daa will be sen o he mobile robo by he Coordinaor. Then he mobile robo will receive he daa as conrol inpu daa by he Rouer. These conrol inpus which are seering angle and velociy will be used o move he mobile robo. The communicaion can be a woway communicaion wih he Rouer will send he daa and he Coordinaor will receive he daa. Such communicaion is necessary when he sensor s reading needs o be processed in he PC. 6.8 Concluding remarks The developmen of a mobile robo is required o invesigae he algorihms is presened in his chaper. In order o implemen he algorihms appropriaely, he mobile robo should has a capabiliy similar o a car and be able o deec he obsacles in fron of i. A range of opions were idenified in he developmen sage and he appropriae soluions were chosen such as he selecion of he of he mobile robo s base, he arrangemen of he sensors and accessories and he sensors selecion. The 1

137 Developmen of a Mobile Robo base of he mobile robo was adoped from a RC car and a wo-ier sensor plaform was assembled and placed on op of he mobile robo s base. The sensors daa will be exraced from he sensors using he microconroller. Furhermore, he mobile robo was also fied wih a wireless communicaion, which allowed he daa ransmission beween he mobile robo and he PC in real ime during esing. 11

138 Experimenal Resuls & Discussions 7. EXPERIMENTAL RESULTS AND DISCUSSIONS In his chaper, a series of experimenal works have been conduced. The aims of hese experimenal works are o validae he algorihms ha have been used o conrol he mobile robo and also o verify he effeciveness of he developed simulaion framework. The mobile robo model was esed in he various condiions of he environmens. The algorihms esed were derived from he algorihms ha were developed in he Chaper 4. In hese experimenal works, he sensor values were ransmied from he mobile robo o he PC in real ime during he esing. The resuls were hen colleced in he PC and ploed in graphs. For each case, he experimenal resuls can be compared o he simulaion resuls for he given scenarios. The experimenal works were divided ino four pars: 1. Mobile robo navigaes in an obsacle-free environmen.. Mobile robo navigaes in a known saic environmen. 3. Mobile robo navigaes in an unknown saic environmen. 4. Mobile robo navigaes in an unknown dynamic environmen. 1

139 Experimenal Resuls & Discussions 7.1 Experimen archiecure The overall view of he sysem archiecure for he mobile robos navigaion is shown in Figure 7.1. In his sudy, he experimen archiecure was designed o caer he known and unknown saic obsacles and i can be expanded o he moving obsacles in he fuure works. Inpu daa from user Generae iniial pah Any known saic obsacle? Yes New inpu daa No Seering angle over limi? Yes No Wireless communicaion Inpu o robo (ϕ, v) Sensor daa Figure 7.1 Experimenal work flow The obsacle-free rajecory will firs be obained from he offline planning afer having inpus from he user. The oupu daa, which are seering angle and velociy, were hen ransmied o he mobile robo via he wireless communicaion and hese daa were used o move he mobile robo for every ime sep. In he same ime, he daa exraced from he sensors; such as daa from deecion sensor; will be sen o he PC o be processed. The decision making process will ake place a his sage and once he decision has been made, he mobile robo will hen reac based on he 13

140 Experimenal Resuls & Discussions decision. For example, if he algorihm decided here is an obsacle in fron of he mobile robo, he obsacle avoidance algorihm will be execued and he new inpu daa will be ransmied o he mobile robo and he mobile robo will reac based on hese new inpus. 7. Experimen seup The esing arena is an open-space and fla errain area as shown in Figure 7.. Iniial poin 4 cm Final poin Figure 7. Tesing arena The firs wo cases have been conduced wih he aim o iniially validae he algorihm. A car-like robo discussed in previous secion has been used. For boh cases, he disance beween he iniial and final poin was 4 cm and he ravelling ime was se as seconds. In he firs case, he algorihm was esed wihou an obsacle and in he second case, he algorihm was esed wih he presence of a known saic obsacle. The experimenal work was hen furher exended o he unknown saic obsacle in he hird case. The algorihm was execued in a PC using MATLAB and he oupu was sen o he mobile robo as he conrol inpu via wireless communicaion. 14

141 Experimenal Resuls & Discussions Furhermore, a marker was locaed a he back of he mobile robo in order o map ou he acual rajecory. When he mobile robo moves from he saring poin, he marker leaves a race of he rajecory on he floor. Then he race of he rajecory was measured manually in order o obain he acual rajecory for each experimen. In addiion, he movemen of he mobile robo was also recorded using a video camera for each experimen in order o race he acual rajecory. 7.3 Case 1: Navigaion in an obsacle-free environmen The purpose of he firs experimen is o verify he conrol sraegy of a car-like robo. The seering angle and velociy are he wo parameers ha need o verify. In his experimen, he iniial sae was se as [,,,,, ] and he final sae was [4,,,,, ]. The mobile robo was moved in a sraigh line for a disance of 4 cm in he environmen wihou an obsacle as shown in Figure 7.3. Sar Finish Figure 7.3 Experimenal seup for Case 1 15

142 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 6 s (c) A ime = 1 s (d) A ime = 14 s (e) A ime = 16 s (f) A ime = s Figure 7.4 Mobile robo navigaed in an obsacle-free environmen (simulaion) 16

143 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 6 s (c) A ime = 1 s (d) A ime = 14 s (e) A ime = 16 s (f) A ime = s Figure 7.5 Mobile robo navigaed in an obsacle-free environmen (experimen) 17

144 Experimenal Resuls & Discussions The resuls for he simulaion works and experimenal works are shown in Figure 7.4 and Figure 7.5, respecively. The mobile robo was placed a he iniial poin as shown in Figure 7.5(a). I was hen sared o move from he iniial poin as shown in Figure 7.5(b). A he 1 h second, he mobile robo was a he half of is rajecory. The mobile robo was approaching he final poin as shown in Figure 7.5(e) and reached he final poin a h second as shown in Figure 7.5(f). The simulaion and experimenal resuls can be compared a he respecive poin hrough he respecive figures as shown in Figure 7.4 and Figure 7.5. From he resuls, he experimenal works demonsrae he mobile robo was able o mach he simulaion resuls in erm of locaion of he mobile robo a he specific ime. In addiion, he experimen was conduced in hree rial runs and he acual rajecory is compared o he planned rajecory as shown in Figure heory rial 1 rial rial 3 y coordinae (cm) x coordinae (cm) Figure 7.6 Case 1: Trajecory planning wihou an obsacle 18

145 Experimenal Resuls & Discussions From he resuls in Figure 7.6, he firs rial is almos idenical o he planned rajecory, bu he mobile robo sopped a 13 cm more han i should be. In he second rial, he mobile robo basically reached he final poin, bu he mobile robo was no moved in a sraigh line as we can see from he resuls. The final rial, he mobile robo moved in a sraigh line, bu is rajecory is iled a abou 1.1 o from he planned x-axis. The posiion errors for each rial are lised in Table 7.1. The ime aken for he mobile robo from iniial poin o he final poin was seconds and mached he ime ha iniially planned. The maximum percenage error a he final poin is 3.3%. As a conclusion, he resul for conrol sraegy is saisfacory as he mobile robo was able o follow he desired rajecory closely. Table 7.1 Acual iniial and final posiions for Case 1 xs (cm) ys (cm) xf (cm) yf (cm) x error (cm) y error (cm) Theory Trial Trial 4 Trial

146 Experimenal Resuls & Discussions 7.4 Case : Navigaion in a known saic environmen The purpose of his experimen is o validae he newly developed algorihms. In his experimen, a known saic obsacle will be placed in he environmen. The planner will generae an iniial collision-free rajecory; which he obsacle is aken ino accoun during generaion of he rajecory and he mobile robo is expeced o follow he rajecory, avoid he obsacle and reach he final poin a he desired ime. In his experimen, he environmen was se as in Figure 7.7. The disance from he saring poin o he finishing poin is 4 cm. A known saic obsacle was placed in he middle of he x-axis wih he obsacle s diameer is cm. The iniial and final saes of he mobile robo were [,,,,, ] and [4,,,,, ], respecively. Sar Robo Obsacle X-axis Finish Figure 7.7 Experimenal seup for Case. 13

147 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 8 s (c) A ime = 1 s (d) A ime = 1 s (e) A ime = 16 s (f) A ime = s Figure 7.8 Mobile robo navigaed in a known saic environmen (simulaion). 131

148 Experimenal Resuls & Discussions Collision-free rajecory (a) A ime = s (b) A ime = 8 s (c) A ime = 1 s (d) A ime = 1 s (e) A ime = 16 s (f) A ime = s Figure 7.9 Mobile robo navigaed in a known saic environmen (experimen). The simulaion and experimenal resuls for Case are shown in Figure 7.8 and Figure 7.9, respecively. The mobile robo and he saic obsacle were placed a he iniial poin and he middle of he rajecory, respecively, as shown in Figure 7.9 (a). The mobile robo was hen sared o move from he iniial poin as shown in Figure 7.9(b). A he 1 h second, he mobile robo was a he half of is way and successfully followed he iniially planned rajecory wih he consideraion of he saic obsacle as 13

149 Experimenal Resuls & Discussions shown in Figure 7.9(c). The mobile robo was approaching he final poin and reached he final poin a h second as shown in Figure 7.9(e) and Figure 7.9(f), repsecively. Iniial collision-free rajecory Obsacle (a) 3 heory rial 1 rial rial 3 y coordinae (cm) x coordinae (cm) (b) Figure 7.1 (a) Case : Trajecory planning wih a known saic obsacle, (b) Experimenal resuls. 133

150 Experimenal Resuls & Discussions The generaed iniial collision-free rajecory is shown in Figure 7.1(a) wih he obsacle is represened by black circle. From he resuls in Figure 7.8 and Figure 7.9, he locaion of he mobile robo a he specific ime in he experimen was mached wih he simulaion works. In addiion, his second experimen was also conduced in hree rial runs and he resuls are compared o he planned rajecory as shown in Figure 7.1(b). From he resuls, he acual rajecory for he firs and second rials is almos idenical o he planned rajecory. And in he final rial, he mobile robo was able o avoid he obsacle, bu he mobile robo was no reached he final poin accuraely. The posiion errors for Case are abulaed in Table 7.. The maximum percenage error a he final poin is.5%. As a conclusion, he resul for conrol sraegy is saisfacory as he mobile robo was able o avoid he obsacle and follow he desired rajecory closely. In addiion he mobile robo was able o reach closed o he final poin a seconds. Table 7. Acual iniial and final posiions for Case xs (cm) ys (cm) xf (cm) yf (cm) x error (cm) y error (cm) Theory Trial Trial Trial

151 Experimenal Resuls & Discussions 7.5 Case 3: Navigaion in an unknown saic environmen The experimen was furher performed on he mobile robo in an unknown saic environmen. The purpose of his experimen is o validae he obsacle avoidance approach for he ime-criical moion planning. In his experimen, unknown saic obsacles will be placed in he environmen. The planner will generae he collisionfree rajecory wihou he knowledge of unknown saic obsacle and i is expeced o deec and avoid he obsacle. Furhermore, he mobile robo is also expeced o reach he final poin a he desired ime. The algorihm is esed hrough a series of scenarios Scenario 1: One unknown saic obsacle In Scenario 1, he mobile robo was required o move from he iniial poin o he final poin as shown in Figure The disance from he iniial poin o he final poin is 3 mm. An unknown saic obsacle was placed randomly beween he iniial poin and final poin wih he obsacle s diameer is 8 cm. The iniial and final saes of he mobile robo were [, 5,,,, ] and [3, 5,,,, ], respecively. Obsacle Sar Finish x Disance = 3 mm (a) 135

152 Experimenal Resuls & Discussions Robo Sar Obsacle Finish (b) Figure 7.11 (a) Plan view (b) Acual experimenal seup for Scenario 1 Using he iniial and final saes of he mobile robo, he planner an iniial collisionfree rajecory which is represened by a blue line is shown in Figure 7.1. Noe ha here is no obsacle in he map as he obsacle is unknown o he planner. (a) Figure 7.1 Iniial collision-free rajecory for Case 3 136

153 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 4 s (c) A ime = 8 s (d) A ime = 1 s (e) A ime = 16 s (f) A ime = s Figure 7.13 Mobile robo navigaes in he unknown saic environmen (simulaion) 137

154 Experimenal Resuls & Discussions Iniial rajecory (a) A ime = s (b) A ime = 4 s New rajecory (c) A ime = 8 s (d) A ime = 1 s (e) A ime = 16 s (f) A ime = s Figure 7.14 Mobile robo navigaes in he unknown saic environmen (experimen) 138

155 Experimenal Resuls & Discussions The simulaion and experimenal resuls for Scenario 1 are shown in Figure 7.13 and Figure 7.14, respecively. Boh simulaion and experimenal resuls can easily compared hrough he respecively figures. A he sar of he experimenal work, he mobile robo was followed he iniial collision-free rajecory unil i deeced he obsacle in fron of i as shown in Figure 7.14(c). Then he dynamic obsacle avoidance approach was execued and a new rajecory was generaed from he poin of deecion o he final poin. The mobile robo was hen followed he new rajecory successfully unil i reached he final poin as shown in Figure 7.14(f). The experimen resuls were hen being compared wih he heory as shown in Figure During he experimen, he mobile robo deeced he obsacle s locaion a (169, 5). The acual measured locaion of he unknown saic obsacle was (163, 5). This was showed ha he ulrasonic sensor was able o deec and locae he obsacle close o he acual locaion. In comparison wih he heory, he mobile robo was able o follow he iniial planned rajecory unil i deeced he obsacle and he new rajecory was generaed in order o avoid he obsacle. From he poin of deecion, he mobile robo s movemen was slighly deviaed from is planned rajecory and sopped before he final poin. The experimen has proven ha he algorihms was pracical o be used in rajecory planning as conrol sraegy for he mobile robo was able o ranslae he inpu o he acual rajecory and he mobile robo was able o follow he planned rajecory as close as possible. y (cm) Deecion poin Deeced obsacle s locaion X X (5,5) (169,5) Theory Acual Iniial X (7,6) x (cm) Figure 7.15 Theoreical and acual rajecory for Case 3 139

156 Experimenal Resuls & Discussions The experimen has been conduced in hree rial runs and he errors in posiions a he final poin were compared wih he heory as abulaed in Table 7.3. The maximum relaive errors recorded in x-axis and y-axis a he final poin for seconds are around 18.3% and %, respecively. Table 7.3 Acual iniial and final posiions for Case 3 xs (cm) ys (cm) xf (cm) yf (cm) x error (cm) y error (cm) Theory Trial Trial Trial Scenario : Two unknown saic obsacles The experimenal works for he unknown saic environmens were furher invesigaed by adding more unknown saic obsacles. In Scenario, wo unknown obsacles were placed in he environmen a he unknown locaion and he disance was increased o 4 mm. The ravelling ime from iniial poin o final poin was se o 3 seconds. The experimenal seup was shown in Figure Obsacles Sar Finish x 1 x Disance = 4 mm Figure 7.16 Experimenal seup for Scenario 14

157 Experimenal Resuls & Discussions The simulaion and experimenal resuls are shown in Figure 7.17 and Figure 7.18, respecively. Boh simulaion and experimenal resuls can be compared hrough he respecively figures. A he beginning of he experimen, he mobile robo was followed he iniial collision-free rajecory unil i deeced he firs obsacle. Then he obsacle avoidance algorihm has been execued and he mobile robo avoided he firs obsacle as shown in Figure 7.18(b). Once he mobile robo has avoided he firs obsacle, i hen coninued i journey unil i deeced and avoided he second obsacle as shown in Figure 7.18(d). Afer successfully avoiding he second obsacle, he mobile robo coninued following he new rajecory unil i reached he final poin as shown in Figure 7.18(f). 141

158 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 6 s (c) A ime = 1 s (d) A ime = 15 s (e) A ime = 1 s (f) A ime = 3 s Figure 7.17 Mobile robo navigaes hrough wo unknown obsacles (simulaion) 14

159 Experimenal Resuls & Discussions Iniial rajecory (a) A ime = s (b) A ime = 6 s (c) A ime = 1 s (d) A ime = 15 s (e) A ime = 1 s (f) A ime = 3 s Figure 7.18 Mobile robo navigaes hrough wo unknown obsacles (experimen) 143

160 Experimenal Resuls & Discussions The experimen has been conduced in hree rial runs and he final resuls were compared o he heory as shown in Figure y (cm) X (45,51) Deecion poin Deecion poin X (141,5) Deeced firs obsacle s locaion X (1,7) x (cm) X (79,77) Deeced second obsacle s locaion Theory Iniial Acual X (41,6) Figure 7.19 Theoreical and acual rajecory for Case 4 The errors in posiions a he final poin were compared wih he heory and abulaed as in Table 7.4. The maximum errors recorded in x-axis and y-axis are around 6.3% and 36%, respecively. Table 7.4 Acual iniial and final posiions for Case 4 xs (cm) ys (cm) xf (cm) yf (cm) x error (cm) y error (cm) Theory Trial Trial Trial

161 Experimenal Resuls & Discussions 7.6 Case 4: Navigaion in an unknown dynamic environmens In his secion, he experimenal works were conduced in order o validae he algorihm for moving obsacles. The dynamic obsacle avoidance approach was used o deec and avoid he moving obsacles as discussed in previous chaper. In his experimen, he remoe conrol car was used as he moving obsacle. The algorihm was esed hrough a series of scenarios Scenario 1: Opposie direcion of mobile robo In he firs scenario, he moving obsacle came from he opposie direcion of he mobile robo as shown in Figure 7.. The moving obsacle was placed randomly in fron of he mobile robo. The disance from iniial poin o final poin was se o 35 cm and he ravelling ime for he mobile robo was se o 3 seconds. Moving obsacle Sar Finish Moving direcion Mobile robo Disance = 35 cm Figure 7. Moving obsacle coming from he opposie direcion of he mobile robo The simulaion and experimenal resuls are shown in Figure 7.1 and Figure 7., respecively. From he experimenal resuls, he planned rajecory is represened by red dash line, while he acual rajecory is represened by solid red line. A he beginning of he experimen, he mobile robo was followed he iniial collision-free rajecory unil i deeced he moving obsacle. Then he obsacle avoidance algorihm was execued and he new deviaed rajecory was generaed as shown in Figure 7.(b). The mobile robo hen avoided he mobile robo and followed new rajecory unil i reached he final poin as shown in Figure 7.(f). 145

162 Experimenal Resuls & Discussions Moving obsacle Mobile robo (a) A ime = s (b) A ime = 13 s (c) A ime = 18 s (d) A ime = s (e) A ime = 6 s (f) A ime = 3 s Figure 7.1 Scenario 1: Moving obsacle from he opposie direcion of he mobile robo (simulaion) 146

163 Experimenal Resuls & Discussions Iniial rajecory Robo direcion Obsacle direcion Deviaed rajecory (a) A ime = s (b) A ime = 13 s (c) A ime = 18 s (d) A ime = s Planned rajecory Acual rajecory (e) A ime = 6 s (f) A ime = 3 s Figure 7. Scenario 1: Moving obsacle from he opposie direcion of he mobile robo (experimen) 147

164 Experimenal Resuls & Discussions From he experimen resuls, he planned (heoreical) rajecory and he acual rajecory can be compared as shown in Figure 7.(f) and Figure 7.3. The planned rajecory was exraced from Malab as he real-ime daa were colleced from he sensors in order o execue he obsacle avoidance algorihm. The mobile robo was able o follow he planned closely i deeced he moving obsacle. I hen sared o deviae from he planned rajecory. A he final ime, he mobile robo sopped a (37, 1) compared o he planned final poin a (35, 1). The errors recorded for x-axis and y-axis for Scenario 1 are around 5.7% and %, respecively. y (cm) Theory Iniial Acual x (cm) Figure 7.3 Theoreical and acual rajecory for scenario 1 148

165 Experimenal Resuls & Discussions 7.6. Scenario : From lef-hand side of mobile robo In he second scenario, he moving obsacle came from he lef-hand side direcion of he mobile robo as shown in Figure 7.4. The moving obsacle was placed randomly a he lef-hand side of he mobile robo and i moves on he sraigh line across he mobile robo from lef o righ. The disance from iniial poin o final poin for he mobile robo was se o 35 cm and he ravelling ime was se o 3 seconds. Moving direcion Moving obsacle Sar Finish Mobile robo Disance = 35 cm Figure 7.4 Moving obsacle coming from lef-hand side of he mobile robo The simulaion and experimenal resuls are shown in Figure 7.5 and Figure 7.6, respecively. A he beginning of he experimen, he mobile robo was followed he iniial collision-free rajecory unil i deeced he moving obsacle and a new deviaed rajecory was generaed as shown in Figure 7.6(b). The mobile robo was hen followed he new deviaed rajecory and i reached he final poin as shown in Figure 7.6(f). 149

166 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 1 s (c) A ime = 1 s (d) A ime = 13 s (e) A ime = 17 s (f) A ime = 3 s Figure 7.5 Scenario : Moving obsacle from he lef-hand side of he mobile robo (simulaion) 15

167 Experimenal Resuls & Discussions Obsacle direcion Iniial rajecory Deviaed rajecory Robo direcion (a) A ime = s (b) A ime = 1 s (c) A ime = 1 s (d) A ime = 13 s Planned rajecory Acual rajecory (e) A ime = 17 s (f) A ime = 3 s Figure 7.6 Scenario : Moving obsacle from he lef-hand side of he mobile robo (experimen) 151

168 Experimenal Resuls & Discussions The comparison beween he planned (heoreical) rajecory and he acual rajecory is shown in Figure 7.6(f) and Figure 7.7. From he figures, he mobile was able o follow he planned rajecory unil i deeced he moving obsacle. I hen sared o deviae from he new deviaed rajecory. However he mobile robo was able o avoid he moving obsacle safely. A he final ime, he mobile robo sopped a (36, 9) compared o he planned final poin a (35, 1). The errors recorded for x-axis and y-axis for scenario are around.9% and 1%, respecively. y (cm) x (cm) Theory Iniial Acual Figure 7.7 Theoreical and acual rajecory for scenario 15

169 Experimenal Resuls & Discussions Scenario 3: From righ-hand side of mobile robo In he hird scenario, he moving obsacle came from he righ-hand side direcion of he mobile robo as shown in Figure 7.8. The moving obsacle was placed randomly a he righ-hand side of he mobile robo and i moves on he sraigh line across he mobile robo from lef o righ. The disance from iniial poin o final poin for he mobile robo was se o 35 cm and he ravelling ime was se o 3 seconds. Sar Finish Mobile robo Moving direcion Disance = 35 cm Moving obsacle Figure 7.8 Moving obsacle coming from righ-hand side of he mobile robo The simulaion and experimenal resuls are shown in Figure 7.9 and Figure 7.3, respecively. A he beginning of he experimen, he mobile robo was followed he iniial collision-free rajecory unil i deeced and avoided he moving obsacle as shown in Figure 7.3(c). Then he new deviaed rajecory was generaed and he mobile robo was followed he new rajecory unil i reached he final poin as shown in Figure 7.3(f). 153

170 Experimenal Resuls & Discussions (a) A ime = s (b) A ime = 11 s (c) A ime = 16 s (d) A ime = 4 (e) A ime = 8 s (f) A ime = 3 s Figure 7.9 Scenario 3: Moving obsacle from he righ-hand side of he mobile robo (simulaion) 154

171 Experimenal Resuls & Discussions Iniial rajecory Obsacle direcion Robo direcion (a) A ime = s (b) A ime = 11 s Deviaed rajecory (c) A ime = 16 s (d) A ime = 4 Acual rajecory Planned rajecory (e) A ime = 8 s (f) A ime = 3 s Figure 7.3 Scenario 3: Moving obsacle from he righ-hand side of he mobile robo (experimen) 155

172 Experimenal Resuls & Discussions The comparison beween he planned (heoreical) rajecory and he acual rajecory is shown in Figure 7.3(f) and Figure From he figures, he mobile robo was able o follow he planned rajecory unil i deeced he moving obsacle. I hen sared o deviae from he new deviaed rajecory. A he final ime, he mobile robo sopped a (3, 11) which is shorer han he planned final poin a (35,1). The errors recorded for x-axis and y-axis for scenario 3 are around 8.6% and 1%, respecively. y (cm) Theory Iniial Acual x (cm) Figure 7.31 Theoreical and acual rajecory for scenario Concluding remarks From he experimenal resuls, i was found ha he mobile robo was capable o follow closely he planned rajecories. The errors for each case were compared beween he planned and acual resuls a he final poin for respecive ravel ime. The posiion errors obained from he experimens show an accepable resul as mos of he rials for each case, he mobile robo has sopped close o he final poin. Furhermore, he conrol sraegy of he nonholonomic mobile robo was applicable o he real mobile robo and he conrol of he mobile robo was ideal wih he calibraion work which was conduced prior o he experimenal works. The algorihm was also showed a good resul in compuaional ime which was shown by he mobile robo when i deeced an obsacle and sared o re-plan is rajecory in order o avoid he obsacle in he real esing arena. Furhermore, he inegraion of he sensors and 156

173 Experimenal Resuls & Discussions algorihm showed ha he mobile robo was capable o deec and avoid he obsacle. This indicaes he dynamic obsacle avoidance approach was pracical o use o avoid he obsacle in he real experimens. However, here is a sligh deviaion beween he planned and n acual rajecories due o inaccuracy of acual seering angle and acual ravelled disance caused by he open-loop sysem. The conrol sraegy for a car-like robo can be developed furher in order o obain a beer resul. For example, he inroducion of close-loop feedback conrol will ensure he speed and seering angle respond beer o he calibraed speed and seering angle. In addiion, he errors occurred due o: 1. The fricion beween he yres and surface caused slippage and reduce he accuracy of velociy and disance recorded for he mobile robo.. The orque of he moor used o conrol he speed was quie small and he mobile robo needs a kick sar o move. Therefore, he newly developed algorihms are applicable and pracical o be used for a car-like robo in real-ime applicaions as demonsraed in he experimenal works. The validaion works for he algorihms and he verificaion of effeciveness of he simulaion frameworks for he mobile robo were successfully conduced hrough a series of experimenal seup. Furhermore he developmen of simulaion framework will be used o furher research and running more scenarios ha are difficul o be conduced by experimen works. 157

174 Conclusions & Fuure Works 8. CONCLUSIONS AND FUTURE WORKS The reviewed lieraure indicaes ha here is no comprehensive research focused on he ime-criical moion planning for a nonholonomic mobile robo. Several simulaion works have been conduced for rajecory planning of he nonholonomic mobile robo o simulae he algorihms, bu here was no experimenal work was conduced o validae he algorihms (Guo e al., 3; Qu e al., 4; Guo e al., 7). This sudy has been carried ou o furher invesigae he ime-criical moion planning for a nonholonomic mobile robo. The geomery approach has been examined and used o generae he ime-criical moion planning for mobile robos. Cubic and quinic polynomials are used o obain a smooh and coninuous rajecory for he mobile robos. The kinemaic consrains of he mobile robo have been aken ino consideraion during he developmen of he algorihms. From he simulaion resuls, all he cases proved ha he algorihms are pracical o be used in moion planning for single and muliple mobile robos. Furhermore he experimenal works validae and verify he algorihms and he conrol sraegies of he mobile robo. 158

175 Conclusions & Fuure Works In his chaper, all he findings will be concluded as a whole and he fuure works for his sudy will be described. Conribuions This sudy had made several conribuions o he curren knowledge and he conribuions are summarized as followings. (1) Developmen of he ime-criical moion planning algorihms for nonholonomic mobile robos. In his sudy, a new ime-criical moion planning algorihm was developed for nonholonomic mobile robos. The movemen of he mobile robo can be analysed hrough his algorihm wih given iniial and final saes, which are x-posiion, y- posiion, orienaion, seering angle, velociy and ime, of he mobile robo. The environmen seup can be easily changed by adoping a required map ino he algorihm. This algorihm can also be uilised for muliple mobile robos planning wih each mobile robo has a differen iniial and final sae. The advanage of his algorihm is i can indicae he iniial and final rajecory a every ime sep for each scenario ha has been se. In addiion he size of he mobile robo can be se simply by changing he parameer of he mobile robo o sui he real mobile robo. Furhermore, using MATLAB as an objec-oriened programming allows a real ime programming which is used in real ime experimens. On he oher hand, he planned rajecory has considered obsacles ha known in prior o he planner before he mobile robo sars o move from he iniial poin. The advanage of his approach is he planner will generaed an iniial free-collision rajecory for he mobile robo. This will ensure he mobile robo will no collide wih any known obsacle while manoeuvring hrough he environmen and will able o reach he final poin a he desired ime. 159

176 Conclusions & Fuure Works Furhermore, he developed D simulaion framework in Malab give a user friendly inerface for he user o se he iniial and final sae of he mobile robo before running he simulaion. () Developmen of he dynamic obsacle avoidance approach In his sudy, a new sraegy in avoiding an obsacle has been is developed for a imecriical moion planning for nonholonomic mobile robos. The obsacle avoidance approach has modified he iniial rajecory in order o avoid he obsacle and a he same ime, he new generaed rajecory will ensure he mobile robo reach he final poin a he desired ime. The advanage of his dynamic obsacle avoidance approach is i can compensae he ime los during avoiding he obsacle. The obsacle avoidance algorihm ensures he mobile robo able o avoid unknown saic and dynamic obsacles which he mobile robo encouners during navigaing hrough he environmens. To he curren knowledge, his dynamic obsacle avoidance approach is he firs obsacle avoidance approach ha considering iming when encounering an obsacle. (3) Validaion and verificaion works hrough series of simulaions and experimens One of he mos imporan pars of his sudy is o validae he algorihm and effeciveness of he simulaion framework wih he real mobile robo. A car-like robo was used o carry ou he simulaed rajecory in he real environmen. A he preparaion sage, he seering angle and speed of he mobile robo were calibraed. Then an acual mobile robo was used o validae and verify he heory. The experimens were conduced hrough a series of cases. Basically, here were four cases have been carried ou for his sudy. The firs case was o verify he conrol sraegy of he car-like robo. The second case was o validae he collision-free imecriical moion planning algorihm. In his case, a known saic obsacle was placed in he environmen and he obsacle was known o he planner prior o rajecory generaion. The hird and forh cases were carried ou in he unknown saic 16

177 Conclusions & Fuure Works environmens o validae he dynamic obsacle avoidance algorihm. In he hird case, a saic obsacle was place in he environmen, bu i was no known o he planner prior o rajecory generaion. While in he forh case, he scenario was exended by placing more unknown saic obsacles. Finally he fifh case was carried ou in he unknown dynamic environmens o es and validae he algorihm for moving obsacles. There were hree scenarios which differeniaed by he moving direcion of he moving obsacle from he mobile robo moving from opposie side, lef-hand side and righ-hand side of he mobile robo. For each case, he experimenal resuls were hen being compared o he heory or simulaion resuls. From he experimen resuls, he mobile robo was able o navigae hrough he environmen and reach he final poin a he desired ime, wih capabiliy o avoid he obsacle along is way. This shows ha he mobile robo can be used for a ask-based mission as he mobile robo can be se o reach a cerain waypoin or he final poin a he desired ime. Furhermore, he seing of parameer for he algorihm is also flexible as he modificaion can only be made in Malab inerface wihou inerfering he onboard conrol algorihm a he mobile robo. Fuure works This sudy can be furher invesigae and improve in he fuure o esablish a robus and opimize algorihms ha can be used for he real applicaions. The possible fuure works for his sudy are briefly described as followings. (1) Opimizaion of he ime-criical moion planning algorihm. Currenly, he algorihm is no considering he overall disance of he rajecory for he mobile robo from he iniial o he final poin. The algorihm can be opimised in order o ensure he mobile robo will ravel using he shores rajecory. In addiion, he usage of baery for mobile robos is one he main concerns. Thus he energy managemen approach can be adoped for he algorihm so ha he mobile robo can ravel for long disance. 161

178 Conclusions & Fuure Works Furhermore, in order o obain a beer driving experience o he mobile robo, he velociy can be opimised. For example, a he sraigh line, he mobile robo can achieve a higher speed han a he curve. This may has impac on dynamic of he mobile robo during cornering. Furhermore he mobile robo migh be also sar a a higher speed from he iniial poin and will slow down when i approaching he final poin. A his momen, he seering angle and speed are se wih an open loop sysem which means here is no feedback conrol. However, he experimenal resuls show he mobile robo was able o perform he driving along he planned rajecory well, bu inroducing a feedback conroller can increase he accuracy resuls in real ime experimens. () Experimenal work for muliple mobile robos in dynamic environmens. A his momen, he experimenal works only conduced in he saic and dynamic environmens wih single mobile robo. In he fuure, he experimen can be conduced in environmens wih he presence of muliple mobile robo and moving obsacles such as oher mobile robo and human. However, prior o he experimen, he mobile robo need o equip wih a beer deecion sensor for moving obsacle such as laser range finder (LRF) or a mobile camera. These sensors will ensure he moving obsacle can be deeced and racked so ha he algorihm will be able o plan he nex sep. Furhermore, he limiaions of he curren car-like robo need o be considered. The mobile robo was modified from he small scale RC car and will no be suiable o carry a large sensor such as LRF. The baery life also needs o be aking ino consideraion as he baery will power all he sensors, moor, servo and microconroller. 16

179 Reference REFERENCE Brezak, M., I. Perovic and E. Ivanjko (8). "Robus and accurae global vision sysem for real ime racking of muliple mobile robos", Roboics and Auonomous Sysems, vol.56, pp Brock, O. and O. Khaib (1999). "High-speed navigaion using he global dynamic window approach", Inernaional Conference on Roboics and Auomaion, Michigan, USA, 1-15 May Casillo, G. D., S. Skaar, A. Cardenas and L. Fehr (6). "A sonar approach o obsacle deecion for a vision-based auonomous wheelchair", Roboics and Auonomous Sysems, vol.54, pp Chang, Y.-C., Y. Y. Lwin and Y. Yamamoo (9). Sensor-based rajecory planning sraegy for non-holonomic mobile robo wih laser range sensors. IEEE Inernaional Symposium on Indusrial Elecronics (ISIE 9). Seoul, Korea. Cosio, F. A. and M. A. P. Casaneda (4). "Auonomous robo navigaion using adapive poenial fields", Mahemaical and Compuer Modelling, vol.4(4), pp Delingee, H., M. Heber and K. Ikeuchi (1991). "Trajecory generaion wih curvaure consrain based on energy minimizaion", Inernaional Conference on Inelligen Roboics Sysems, Osaka, Japan, November Dong, W. and Y. Guo (5). "New rajecory generaion mehods for nonholonomic mobile robos", Inernaional Symposium on Collaboraive Technologies and Sysems, Missouri, USA, 15- May 5. Duan, Y., Q. Liu and X. Xu (7). "Applicaion of reinforcemen learning in robo soccer", Engineering Applicaions of Arificial Inelligence, vol., pp Fajen, B. R. and W. H. Warren (3). "Behavioral dynamics of seering obsacle avoidance and roue selecion", Journal of Experimenal Psychilogy: Human and Percepion and Performance, vol.9(), pp Fajen, B. R., W. H. Warren, S. Temizer and L. P. Kaelbling (3). "A dynamic model of visually-guided seering, obsacle avoidance and roue selecion", Inernaional Journal of Compuer Vision, vol.54(1//3), pp Ferrara, A. and M. Rubagoi (9). A dynamic obsacle avoidance sraegy for a mobile robo based on sliding mode conrol. IEEE Inernaional Conference on Conrol and Applicaions. Sain Peersburg, Russia. Fox, D., W. Burgard and S. Thrun (1997). "The dynamic window approach o collision avoidance", IEEE Roboics and Auomaion Magazine, vol.4(1), pp Ge, S. S. and F. L. Lewis (6). "Auonomous mobile robo: Sensing, conrol, decision making and applicaions", Taylor and Francis Group, 6. Ghia, N. and M. Kloezer (1). "Trajecory planning for a car-like robo by environmen absracion", Roboics and Auonomous Sysems, vol.6, pp Guo, Y., Z. Qu and J. Wang (3). "A New Performance-Based Moion Planner for Nonholonomic Mobile Robos", The 3rd Performance Merics for Inelligen Sysems Workshop, Gaihersburg, MD, USA, Sepember

180 Reference Haddad, M., T. Cheibi, S. Hanchi and H. E. Lehihe (7). "A random-profile approach for he rajecory planning of he wheeled mobile robos", European Journal of Mechanics A/Solids, vol.6, pp Hamner, B., S. Singh and S. Scherer (6). "Learning obsacle avoidance parameers from operaor behavior", Journal of Field Roboics, vol.3(11/1), pp Hazon, N. and G. A. Kaminka (8). "On redundancy, effieciency and robusness in coverage for muliple robos", Roboics and Auonomous Sysems, vol.56, pp Hong, J., Y. Choi and K. Park (7). Mobile robo navigaion using modified flexible Vecor field approach wih laser range finder and IR sensor. Inernaional Conference on Conrol, Auomaion and Sysems. Seoul, Korea. Huang, L. (9). "Velociy planning for a mobile robo o rack a moving arge - a poenial field approach", Roboics and Auonomous Sysems, vol.57, pp Huang, W. H., B. R. Fajen, J. R. Fink and W. H. Warren (6). "Visual navigaion and obsacle avoidance using a seering poenial funcion", Roboics and Auonomous Sysems, vol.54, pp Hui, N. B., V. Mahendar and D. K. Praihar (6). "Time-opimal, collision-free navigaion of a car-like mobile robo using neuro-fuzzy approach", Fuzzy Ses and Sysems, vol.157, pp Jacob, M. (8). Pah planning and obsacle avoidance in unknown dynamic environmens. Jiang, K., L. D. Senevirae and S. W. E. Earles (1997). "TIme-opimal smooh-pah planning for a mobile robo wih he kinemaic consrains", Roboica, vol.15, pp Jolly, K. G., R. S. Kumar and R. Vijayakumar (8). "A Bezier curve based pah planning in a muli-agen robo soccer sysem wihou violaing he acceleraion limis", Roboics and Auonomous Sysems, vol.57, pp Jolly, K. G., K. P. Ravindran, R. Vijayakumar and R. S. Kumar (7). "Inelligen decision making in muli-agen robo soccer sysem hrough compounded arificial neural neworks", Roboics and Auonomous Sysems, vol.55, pp Kanayama, Y. and N. Miyake (1986). "Trajecory generaion for mobile robos", 3rd Symposium on Roboics Research, Gourvieux, France, Kelly, A. (3). "Efficien parameric synhesis of opimal mobile robo rajecories", 11h Inernaional Conference on Advanced Roboics, Universiy of Coimbra, Porugal, 3 June - 3 July 3. Khaib, O. (1986). "Real-ime obsacle avoidance for robo manipulaor and mobile robos", Inernaional Journal of Roboics Research, vol.5(1), pp Klancar, G., M. Krisan, S. Kovacic and O. Orqueda (4). "Robus and efficien vision sysem for group of cooperaing mobile robos wih applicaion o soccer robos", ISA Transacions, vol.43, pp Koh, K. C. and H. S. Cho (1999). "A smooh pah racking algorihm for wheeled mobile robos wih dynamic consrains", Journal of Inelligen and Roboic Sysems, vol.4, pp Li, L. and F.-Y. Wang (3). "Trajecory generaion for driving guidance of fron wheel seering vehicles", IEEE Inelligen Vehicles Symposium, Ohio, USA, 9-11 June

181 Reference Liang, T.-C., J.-S. Liu, G.-T. Hung and Y.-Z. Chang (5). "Pracical and flexible pah planning for a car-like mobile robo using maximal-curvaure cubic spiral", Roboics and Auonomous Sysems, vol.5, pp Liddy, T. J. and T.-F. Lu (7). "Waypoin navigaion wih posiion and heading conrol using complex vecor fields for an Ackermann seering auonomous vehicle", Ausralasian Conference on Roboics and Auomaion, Brisbane, Ausralia, 1-1 December 7. Liu, S. and D. Sun (11). Opimal moion planning of a mobile robo wih minimum energy consumpion. 11 IEEE/ASME Inernaional Conference on Advanced Inelligen Mecharonics (AIM11). Budapes, Hungary. Ma, Y., G. Zheng and W. Perruquei (13). "Real-ime local pah planning for mobile robos", 9h Inernaional Workshop on Robo Moion and Conrol, Wasowo, Poland, 3-5 July 13. Mihaylova, L., J. D. Schuer and H. Bruyninckx (3). "A mulisine approach for rajecory generaion opimizaion based on he informaion gain", Roboics and Auonomous Sysems, vol.43, pp Nagaani, K., Y. Iwai and Y. Tanaka (1). "Sensor based navigaion for car-like mobile robos using generalized Voronoi Graph", IEEE/RSJ Inernaional Conference on Inelligen Robos and Sysems, Hawaii, USA, 9 Oc. - 3 Nov. 1. Nagy, B. and A. Kelly (1). "Trajecory generaion for car-like robos using cubic curvaure polynomials", Inernaional Conference on Field and Service Robos, Helsinki, Finland, June Pin, F. G. and H. A. Vasseur (199). "Auonomous rajecory generaion for mobile robos wih non-holonomic and seering angle consrains", IEEE Inernaional Workshop on Inelligen Moion Conrol, Isanbul, Turkey, - Augus 199. Prado, M., A. Simon, E. Carabias, A. Perez and F. Ezquerro (3). "Opimal velociy planning of wheeled mobile robos on specific pahs in saic and dynamic environmens", Journal of Roboic Sysems, vol.(1), pp Qu, Z., J. Wang and C. E. Plaised (4). "A new analyical soluion o mobile robo rajecory generaion in he presence of moving obsacles", IEEE Transacion on Roboics and Auomaion, vol., pp Safadi, H. (7). Local pah planning using virual poenial field, School of Compuer Science, McGill Universiy. Salichs, M. A. and L. Moreno (). "Navigaion of mobile robos: Open quesion", Roboica, vol.18, pp Shin, D. and S. Singh (199). Pah generaion for robo vehicles using composie clohoid segmens. Pisburgh, Pennsylvania, Carnegie-Mellon Universiy. Siegwar, R. and I. R. Nourbakhsh (4). "Inroducion o auonomous mobile robos", The MIT Press, 4. Sridharan, K. and T. K. Priya (4). "A parallel algorihm for consrucing reduced visibiliy graph and is FPGA implemenaion", Sysems Archiecure, vol.5(4), pp Sroupe, A., T. Hunsberger, A. Okon, H. Aghazarian and M. Robinson (5). "Behavior-based muli-robo collaboraion for auonomous consracion asks", Inernaional Conference on Inelligen Robos and Sysems, Albera, Canada, -6 Augus 5. Tounsi, M. and J. F. L. Corre (1996). "Trajecory generaion for mobile robos", Mahemaics and Compuers in Simulaion, vol.41, pp

182 Reference Trepow, A. and A. Zell (4). "Real-ime objec racking for soccer robos wihou color informaion", Roboics and Auonomous Sysems, vol.48, pp Vicorino, A. C., P. Rives and J.-J. Borrelly (1). "Mobile robo navigaion using a sensor-based conrol sraegy", IEEE Inernaional Conference on Roboics 8 Auomaion, Seoul, Korea, 1-6 May 1. Wein, R., J. P. v. d. Berg and D. Halperin (7). "The Visibiliy-Voronoi complex and is applicaions", Compuaional Geomery, vol.36(7), pp Yamaguchi, H. (3). "A disribued moion coordinaion sraegy for muliple nonholonomic mobile robos in cooperaive huning operaions", Roboics and Auonomous Sysems, vol.43, pp Zhang, S., L. Xie and M. D. Adams (6). "Feaure exracion for oudoor mobile robo navigaion based on a modified Gauss Newon opimizaion approach", Roboics and Auonomous Sysems, vol.54(6), pp

183 Appendix A APPENDIX A DATASHEETS 167

184 Appendix A A1. Robo Conroller (hp:// Specificaions Overall uni dimensions: 3.7"." Inpu volage: V Programmable MHz Amel ATmega34PA AVR microconroller wih 3 KB flash, KB SRAM, and 1 KB EEPROM * Programmable MHz Amel ATmega184P AVR microconroller wih 18 KB flash, 16 KB RAM, and 4 KB EEPROM *(SVP-184 version) Buil-in USB AVR ISP programmer (USB A o mini-b cable bidirecional moor pors ( A coninuous per channel, 6 A maximum 8-oupu demuliplexer ied o one of he AVR s hardware PWMs for easy conrol of up o 8 servos 1 free I/O lines o 17 free I/O lines on he main MCU, of which 8 can be analog inpus o 4 inpu lines on he auxiliary processor, which can be eiher 4 analog inpus or dual quadraure encoder inpus o hardware UARTs Removable 16-characer -line LCD wih backligh Primary 5V swiching regulaor capable of supplying 3 A Secondary adjusable (.5 V 85% of VIN) buck (sep-down) volage regulaor capable of supplying 3 A Buzzer ied o one of he AVR s hardware PWMs 3 user pushbuon swiches user LEDs Power (push-on/push-off) and rese pushbuon swiches Power circui makes i easy o add exra power buons and provides a selfshudown opion Auxiliary processor (conneced via SPI) provides: o Baery volage reading o User rimmer poeniomeer reading o Inegraed USB connecion o In-Sysem-Programming of he main processor o Abiliy o read wo quadraure encoders 168

185 Appendix A A. Ulrasonic Sensor (hp:// SRF5 Range - 1cm o 4m. Power - 5v, 4mA Typ. Frequency - 4KHz. Size - 43mm x mm x 17mm heigh. Two operaional modes are available, Single pin for rig/echo or Pin SRF4 compaible. The inpu Trigger is a 1uS Min. TTL level pulse Echo Pulse is Posiive TTL level signal, wih he widh proporional o he objec range. Mode Single pin for boh Trigger and Echo 169

186 Appendix A A3. Hall Effec Sensor (hp:// Effec-Laches/US aspx) Feaures and Benefis Wide operaing volage range from 3.5V o 4V High magneic sensiiviy Muli-purpose CMOS echnology Chopper-sabilized amplifier sage Low curren consumpion Open drain oupu Thin SOT3 3L and fla TO-9 3L boh RoHS Complian packages Funcional diagram Pin definiions and descripions 17

187 Appendix A A4. Wireless Communicaion (hp:// 171