REMOTE CONTROL OF A SEMI-AUTONOMOUS ROBOT VEHICLE OVER A TIME-DELAYED LINK
A Thesis Submitted to the College of Graduate Studies and Research in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in the Department of Electrical Engineering University of Saskatchewan
MAYNARD R OLSON
Saskatoon, Saskatchewan, Canada FaIl 2001 National Libraiy Bibliothèque nationale du Canada Acquisitions and Acquisitions et ûibliographic Services services bibliographiques 395 WeEinglon SItwt 395. rue Wellington -ON K1AW Ottawa ON K1A ON4 CaMda CaMda
The author has graated a non- L'auteur a accordé une licence non exclusive licence aüowiag the exclusive pemettant à la National Li'brary of Canada to Bibliothèque nationale du Canada de reproduce, loan, distribute or seii reproduire, prêter, distribuer ou copies of this thesis in microform, vendre des copies de cette thèse sous paper or electronic formats, Ia forme de micro fi ch el^ de reproduction sur papier ou sur format électronique.
The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantial extracts from it Ni la thèse ni des extraits substantieIs may be printed or othem'se de ceiie-ci ne doivent être imprimés reproduced without the author's ou auirement reproduits sans son permission. autorisation. PERMISSION TO USE
The auttior has lgntd that the libraries of the University of Saskatchewan may make this thwis fialy available for inspection. Morwverpthe author has agreed that permission for extensive cwngof this thesis for scholarly purpose may be gMted by the professor or proftssors who supuvised the thesis work ttcotded hemn or, in their absence, by the Head of the Department or the Dtan of the College in which the thesis work was done. It is unâastood that due rceogniîiort dlbe given to the author of this thesis and to the University of Saskatchewan in aay use of the materiai in this thesis. Copying or publication or any atba use of the thesis f'or financial gain *out approval by the University of Sadumhmm ad the utk'swntten permission is prohibited.
Request for permission to copy or to make other use of material in this thesis in whole or in part should be addressai to: ACKNOWLEDGEMENTS
The author wishes to sincerely hnkbis supavisor, Dr. H.C. Wood for his assistance and guidance through the variwg challenging situations during the devclopmcnt and hai prepatotion of this thesis.
The author dso wishes to thank his advisory cornmittee mcmbers, Dr. MM Gupta, Dr. G.J. Schanau and Dr. K. Takaya for th& thought-pmvoking questions and duablc comments which have bad signifiant impact on the final form of this thesis.
The author would ab like to acbwledgc the many helpfiil meetings with Mr. T. J. Nelson of Prairie Machine and Parts, Ltd. which were instrumental in fonnulating the design of the system used in this his.
Financial support pmvided by NSERC, Cinadian Space Agency and Prairie Machine and Parts, Ltd. of Saslcaoon W most gmtehlly rcknowldged and appreciated, as is the bancial support fhm the thuis supavisor, Dr. &C. Wood. Remote control of equipmem or systems over communication links baving loop time delays of about 0.5 seconds or more is knom to be a significant problem for a human opcrator in teleoperation mode. "Move and wait " stratcgy is the nod approach employed by an operator. In order to improve upon the inefficiency of this strategy, a number of solutions can be employed, including wc of predictor displayq supeMsory control and Smith wntrol. A pndictor display aids the operator by presenting a non-delayed view of the remote system output. ïhis allows commands to be issued before the actual delayed output is rcceived via feedback. Supervisory control allows the remote system to operete in semi-autonornous fashion by providing autonomous capability that can be directeci by the opemitor using high-bel commanda Smith wntrol provides stable, closeci-loop control of systems with inherent delay by effectively moving the delay out of the loop. This thesis preseiits a robot vehicle wnîrol system that includes each of these techniques in the overall design.
The robot vehicle king contro11ed is intendcd for underground Mning applications. This is a ditIicuit environment hra control system, particulsrly when autonomous operotion is nquind. Movemtnt of a vehicle in such an uncondmd environment coupled with the problem of senaor regdings suggests an excellent application for tiiay control. Remotc conüol of a robot vehicle bssically involves wntrol of spad and stating. A muhivariable, fbqcontrol system bas been developed to accommodate this task A simplified version of Smith control is used to compensate not only for the tirne delay but also for tbe human operator's dynamics. This simplified method dasnot requin the uJual estimate of non-delaycd plant dynamicq oniy a reasonably accurate rneasurc ofthe loop time delay. Semi- autonomous operation provided includes automatic tunnel-traclring and turning into intdngtunnels. Obstaclc-avoidancc, bascd on the we of fuzzy "obstacle firctors" forms an essential part of the control system, A neural nenv~rk-basedpdctor W also included for the operator's assistancestance
System simulation results prow tbst the simplüied Smith control ir#thad can compensate for both the time delay and the human opaator's dynamics to 8 ver'high degnc. Given a stable lincar plant with no dishrrbrnccs, the lcngth of the time dday is essentially irrelevant in maintainhg stable wntrol. ïî is also demonsbated that the predictor allows the operator to control the runote vehicle over the tirne delay by driving a non-delayeci mode1 ofthe vehicle on a console dispiay. Semi-autonomous operation bas not becn tdin animated simulation in a mine environment, but preliminary simulaiions of obstacleavoidmce, automatic tunnel-tracking and automatic intersection-hirning have indicated that the design appears d. TABLE OF CONTENTS
PERMISSION TO USE
TABLE OF CONTENTS
LIST OF FIGURES
LIST OF TABLES
1. ilWRODUCI'ION 1.1 General 1.2 Fundarnentals of Remote-Couüol Systwis 1.3 Time Delay in RemoteControl Systcms 1.4 TirneDelay Compensation 1.5 Underground Mines 1.6 intelligent Control 1.7 Objectives 1.8 OutIine of the Thesis
2. LITERATURE REVIEW 2.1 Time Delay in Remote Manipulation 2.2 Tirne Delay in Remte Dmring 2.3 TiDelayCompensation in Remotd=oatroi Systems 2.3.1 ForcJVeiocity Feedùack 2.3.2 Smith Controllcls 2.3.3 nedietor Displays and ViEnviroaments 2.3.4 Supervisory Control 23 -5 ûther Techniques 2.4 Madehg Human Operators in Con$ol Systcms 2.4.1 Trausfér Function Model 2.42 CrossoverMudei 2.4.3 Precogaitive Control and Preview 2.4.4 ûther Models 2.5 Summq
3. COMPENSATION MIR DELAY IN CONTROL SYSTEMS 3.1 Tune Delay in Coatrol Systems 3 2 Smith Control in Rernote-Control Systems 32.1 Tme-Delay Compensation 3 -2.2 Wuman Operator Compensation 3.3 System Equivalents Using Predictors
4. DESCiUPTION OF PRûPûSED CONTROL SYSTEM 4.1 Introduction 4.2 General 4.3 Local Terminal 4.4 Remote Terminal 4.4.1 General 4.4.2 Obstacl~AvoidaneeUsing Fuzy Logic 4.4.3 Spceâ ControUer Using Fuzty Control 4-4.4 Steaing Controller Using Fuz~yCoatrol 4.4.4.1 Tdeopemtion Mode 4-4-42 Sani-Autonomous Modes 4.4.5 Robot Vebide Dynamics 4.5 Speed and Stœring Control Systcm Diag~ams 4.6 HumanOpaatorCoasiderati~ll~
S. =STEM SIMULATIONS AND RESULTS 5.1 Sinnilation Eiiviroament 52 SystemPerfo~œwahNoT~DdayintheLoop 52.1 No HumOperatot in the Lmp 5.23 HumanOpaatOrintbcLoOp 5.3 System Per£wmuiee with TiDelay in tbe Loop 5.3.1 NoHuman~intheLoop 5.32 EhimanOp«atorhtheLoop 5.4 System Performance Using Predictor 5 -4.1 General 5.4.2 No Predictor in the Loop 5.4.3 Predictor in the Loop 5.5 System Performance with Mismatches 5.5.1 General 5.5.2 Time-Delay Mismatch 5.5.2.1 No Human Opcrator in the bop 5.5.2.2 Human Operator in the hp 5.5.3 Predictor Mismatch 5.6 Obstacle-Avoidance 5.7 Tunnel-Tracking
6. CONCLUSIONS 6.1 Summary and Conclusions 6.2 Unique Contributions 6.3 Future Work
A Matîab/Siaünkhic Systtm Coufigurath A 1 Overail test contiguration A2 Complete system A3 Operasor/Console A4 Openiror SpeedlSteaingErm Response AS Operator SpeedlSteering Response A6. Local Telciuatonomous Contro11cf A7 System Delays A8 Remote Telc~utonomousControfla A9 ObstaclaAvoidance A 10 RTC Speed Control Al1 FuzySpdCa~trol A12 Smith Control A 13 RTC Steering Control A 14 Fuzq Steering Control A 15 Autosteering Control A 16 Robot Vehicle
B MahWSimtiliiOk Alternate System Coniigumtion B. 1 AitmCompIete System B.2 Altanate opartor/Co~w)le B.3 Altemate Local Teleautonomous Contioller B.4 Alternate Remote Teleautonomous ConbolIer LIST OF FIGURES
Figure 2.1. Rirsuit-tracking block diagram: (a) open- and closed-lwp modelq and (b) simplified mode1 (AAer Leslie [441).
Figure 2.2. Major human operator pathways in a manmachine system (Mer McRuer [52D.
Figure 2.3. Schematic view of preview display (After Hess [61].
Figure 3.1. Ede diagnuns of nondelayed and delayed systerns.
Figure 3.2. Smith controI of time-delayed process: (a) Smith wntroller CS(s) in the lwp, and (b) equivalent system with conventional canaaller C(s).
Figure 3.3. Smith control of remote-control system. (a) Smith controller C*(s) in the lwp, and (b) equivaient system with conventional controller C(s).
Figure 3.4. Remote-cmntrol system Using a traditional Smith wntroller.
Figure 3.5. Remotacantrol system using simplitied Smith control.
Figure 3.6.
Figure 3.7. Smithantrolled equivalents of ~cm~tacontrolsystem: (a) time delay at systu~~outpus and (b) tirne delay at system input.
Figure 3.8. Smith control of rernote-coimpl systun including time delays and human operator dynamics: (a) block diagram with Smith controtler, and (b) equivalent system.
Figure 3.9. Remot~.bntrolsystem wing simplified Smith control for bath timadelay and buman operator compmaiio~~.
Figure 3.10. Rcmotaoontrol system with no time delay, but using simplified Smith control to compensate for human opentof dynamics.
Figure 3.1 1. Equhlettt system to Figure 3.10 using predimr Pg.
Figure 3.12. Equivalent system to Figure 3.10 bascd on Figure 3.8@) with no time deiay and using predictorPC. Figure 3.13. Configuraiion mode1 for dnving robot vchicle using predictor PA.
Figure 3.14. Revised configuration mode1 for driving robot vchide using predictor PB.
Figure 3.15. Final configurati*ondel for driving robot vehicie using predictor Pc.
Figure 4.1. Tiewielayed remotecontrol system block diagram.
Figure 4.2. Local terminai block di-
Figure 4.3. Remote terminal block diagram.
Figure 4.4. Spdand stecring cantrol block diagram.
Figure 4.5. Obstacle-avoidance block dim.
Figwe 4.6. Menibership ftnctions for hazard situation fiiay controller: (a) st&ng angle input, (b) angle to obstacie input, and (c) hazard situation output.
Figure 4.7. Manbaship functions for obstacle factor fhqcontroller. (a) hetard situation input, (b) tirne to collision input, Md (c) obstacle fâctor output.
Figure 4.8. Cantrol sudàces for obstacle avoidancc subsystem: (a) hazard situation as a îunction of steaing angle and en& to obstacle, and @) obstacle factor as a Wonof hazard situation ond tirne to collision,
Figure 4.9. Specd am1 bfock diagram. 65
Figure 4.10. Fuay membersbip fbnctions for speed &or aror: (a) speed 67 tàctor input, (b) speed input, and (c) specd -or mroutput.
Figure 4.1 1. Control sudbces for specd control subsystem: (a) spœd fhx 68 enorasafiuietionofspeed~oraadspeed,d(b)spœderror contml as a fiinction ofspad set enof and spced factor arw.
Figure 4.12. Staring con$ol block di- 70
Figure 4.13. Membership fimctions for hybrid constol: (a) autoste#ing 72 ~~ntrorinput, (b) steering set aror input, and (c) hybrid -1output. Figure 4.14. Control surface for hybrid wntrol as a fùnction of autosteering wntrol and steering set error
Figure 4.15. Control surface for intersedion control as a Monof angle to intersection and range to intersection.
Figure 4-16. Membership fiindons for intersection control: (a) angle to intersection input, (b) range to intersection input, and (c) intersection control output.
Figure 4.17. Autosteering control scheme.
Figure 4.18. Speed controI system primary control Ioop showing major gain and dynamics blocks.
Figure 4.19. System speed respouse with no delay or human operator dynamics.
Figure 4.20. Systern steering response with no delay or human operator dynamics.
Figure 4.21. Speed cuntroI system: (a) block diagram, and @) Simulink diagram.
Figure 4.22. Speed eonttoi system: (a) subsystem diagram, and 0)Simuiiink diagram.
Figure 4.23. RTC speed amtrol: (a) block diagram, and (b) Simulink diagram
Figure 4.24. Fuzzy speed control: (a) block diagram, and (b) Simulink diagram.
Figure 4.25. Smitù conîrol: (a) block diagram, and (b) Simulink diagram,
Figure 4.26. Ste&ng control system block diagram.
Figure 4.27. Steering control system Siinkdiagram.
Figure 438. Stœring control system: (a) subsystcm diagram, and (b) SiulinL diagram-
Figure 429. RTC seering control: (a) block dhgmq and @) Simulink diagram
Figure 4.30. Fuzy steering contml: (a) block diagram, and (b) Simul'hk diagram
Figure 5.1. Stc«ing augle response to stepfimction stcaing set inputs with no time delay and no inimui opmbr in the system. Figure 5.2. Maximum speed vs. staring angle for steady-state conditions.
Figure 5.3. Speed response to stepfiinction speed set inputs for various values of steering angle with no time delay and no human operator in the system: (a) unit step input, and (b) step input of 0.5.
Figure 5.4. Speed and steuing angle responses to sinusoidal inputs of speed set with amplitude muge [O 11 and to steering sct with ampiiide range [-1 11: (a) speed set at 0.2 Hz and stecring set atO.l Hz,and(b) speedsetat0.1 Hiandsteeringsetat03Hz
Figure 5.5. Speed responsc to unit stcp-function speed set inputs for various values of steering angle with human opetator but no time delay in the system.
Figure 5.6. Stedng angle response to stepîunction steering set inputs in incnments of 0.1 with human operator but no tirne delay in the
Figure 5 7. Step response with no human operator but with 1 second time delay in the system and no timedelay compensation: (a) speed responsetounitstepspeedsainputwithsteeringsetofzero, and (b) st- angle tesponse to unit step input steering set input withspeedsetofzero.
Figure 5.8. Sinusoida1 nsponse with no human operator but with 1 second thede10 in tbo lyamand no time-delay compensation: (a) speed nsponse to speed set input of 0.1 Hz and amplitude range [O 11 with staring angle of zero, and @) steering angle fesponse to -ng set input of O. 1 Hz and ampiiide range [-1 11 with speed of m.
Figure 5.9. Speed and steering angle nspollses to unit step speed set and steering set inputs with no human opaator: (a) zero time delay, and (b) 2 second time deiay witû timedeIay compensation.
Figure 5.10. Speed re~pollstto 1 Hz sinusoidai speed set input of amplitude m8e [O 11 ad with no human operator. (a) zero time delay, and (b) 2 second time delay with tllnadelay wmpendon.
Figure 5-11, Staring angle nsponse to 1Hi sinusoidal stdng se! input with amplitude range [-1 11 ad no hum~operator: (a) zero time delay, and @) 2 sccond time delay with tirne-delay compensation. Figure 5.12. Step response with human operdor and 1 second time delay 113 in bsystem but with no human dynamics cornpensati011: (a) spad response to unit step speed set inpus with sîeering set of zero, and (b) Jteaing angle response to unit step steering set input with speed set of zero,
Figure 5.13. Responses of speed and steering angle to unit step speed set 114 and steering set inputs with human operator in the loop, and with buman dynamics compensation: (a) zero time delay, and (b) 2 second time delay.
Figure 5.14. Speed repense to 0.15 Hz sinusoida1 speed set input of 115 amplitude range [O 11, with human operator in the hop, and human dynamics compensation: (a) zero tirne delay, and @) 2 second tune delay.
Figure 5.15- Steering angle response to 0.15 Hz sinusoidal aeering set input 116 with ampliaide range [-1 11, with human operator in the lwp, and human dynamics compensation: (a) zero time delay, and (b) 2 secund time delay.
Figure 5.16. Responses of speed and stœring angle to O. 15 Fh sinwoidal 117 spad set and steering set inputs of amplitude raagts [O 11 and [-1 11, nspectively, with human opercrior in the lwp, and humon dynamics compensation: (a) zero time delay, and (b) 2 SeCoad tirne delay.
Figure 5.17. Responses of spœû and steaing ongle to sinusoida1 spad set 118 input of O. 1 Hz with ampiiirange [O 11 and to sinusoida1 steering set input of O. 15 Hz with amplirange [-I 11 wiîh tninienoperatolintheloopdwitfihumand-CS CO~OK(a) zero hedday, @) 5 sec the &y.
Figure 5.18. Stauiy-staîe spœd vs. spœd set for step input hmmenîs of 0.1 120 and fw fixed Jt#riag uigles h oao to +/- 1.0: (a) actud v-4 and @) predi-. Figure 5.19. Stady-state stcuing angie vs. oteering se! for aep inputs at 121 incrcments ofO.1: (a) actd system, and (b) pradictor.
Figure 5.20. Spced nsponse to unit step input for acîuaf systcm and 123 pndieted: (a) aro time deiay, d (b) 5 d time delay. Figure 5.22. Speed response to 0.15 Hz sinusoicial input of amplitude range 125 [O 11 for adsystem and predided: (a) zero time delay, and (b) 5 second time delay.
Figure 5-23. Steering angle response to 0.15 Hz sinusoida1 input of amplitude 126 range 1-1 11 for actuai systern and predicted: (a) zero time delay, and (b) 5 second thedelay,
Figure 5.24. Speed response hrd and pdicted system with 5 second 127 delay and speed set and steaing set amplitude ranges of [O 11 and [-1 11, mpectively: (a) speed set and steering set inputs at 0.15 & and (b) speed set input at 0.1 Hz and steerhg set input at 0.15 ik
Figure 5.25. Actual and predided response of system with predictor in the 129 loop to unit stép-function with 2 second thedelay in the system: (a) spad response, and (b) steering angle response.
Figure 5.26. ActuaI and predicted speed response to unit steptiinctions of 130 speed set and stecring set with 2 second thedelay in the system and with the predictor in the loop,
Figure 5.27. Actual and prcdicted response of system with predictor in the 13 1 loop to 0.15 Hz sinusoida1 inputs and amplide ranges [O 11 and [-1 11 hr spad and stecting set, rwpectively and with 2 second time delay in the mm:(a) speui responsc, Md (b) aeering angle rrsponse.
Figure 5.28. Response to unit step-fion input with 1 second system time 134 delays and 2.02 second loop deloys (ie.l%emr) in the Smith c~ntrollers,and with no human operator in the bop: (a) spœd rewonse, and (b) Jtaring deresponse-
Figure 5.29. Response to 0.15 Hi sinusoihi input with 1 second system tirne 135 delays and 2.02 second Ioop delays (ie.l% enor) in the Smith controllers, with no human opentor in the loop (speed set amprierau* [O 11 8ud acaing set amplitude range 1-1 11): (a) spaed mnse, (b} -ring uigle
Figure 5.30. Rwponse to unit jtep6iinction input with 1second system the 136 delays and 2.002 second lmp ddays Cie, 0.1% aror) in the Smith controllas, and with no Iniman opcrator in the loop: (a) spced respo- and @) steains uigle response. Figure 5.3 1. Respoase to 0.15 Hz sinusoidal input with 1 second system the delays and 2.002 second loop delays (ie. 0.1% error) in the Smith controllers, and with no hwnan operator in the loop (spced set and stdgsa amplitude ranges [O 11 and [Dl 11, respectively): (a) speed response, and (b) sthgangle respom.
Figure 5.32. Response to unit step-fùnction input with 1 second system time delays and 2.02 second Ioop deIays (ie. 1% error) in the Smith controflers, including the human operator in the loop: (a) speed response, and (b) steering angle rcsponse.
Figure 5.33. Response to 0.15 Hz sinusoidal input with 1 second systern time delays and 2.02 second loop delays (ie. 1% mr)in the Smith controllers, including the human operator in the loop (speed set and stcering set amplitude ranges [O 11 and [-1 11, respectively): (a) speed response, and (b) stecring angle response.
Figure 5.34. Response to unit step-fiinction input with 1 second system tirne delays and 2.002 second loop delays (ie. 0.1% mor) in the Smith controllers, including the human operator in the loop: (a) speed response, and (ô) stcering angie response.
Figure 5.35. Steady-me speed vs. speed set for stcp input increments of 0.1 and for fixed stecring angles Erom zcro to +/O1.0 using the 5,5,2 nairan predictor.
Figure 5.37. Response of actuai system and 5,5,2 ncuron predictor to a O. 15 Hz sinwoidal input with amplitude mges [O 11 for spœd set and [-1 11 for steering set: (a) spœd nsponse, and (b) stedng anslensponse- Figure 5.38. Speed response of actual system and 5,5,2 nairon predictor: (a)speedsetandsteaing~iaputsatO.l5Htwithamplitude ranges [O 11 for speed set and 11 for steering set, and @) speed set mput at 0.1 Hz and steering set input at 0.15 Hz wihamplitude ranges [O 11 for spad set and [-1 11 for steering set,
Figure 5.39. Speed response as a fiinciion of obstacle factor for unit step inputsofspeedset.
Figure 5.40. Staring angle nspome to obstrcIe avoid inputs with steaing set input of zero: (a) step-fionrespoase, and (b) sinusoii response to O, 15 Hz input of amplitude muge 1-1 11. Figure 5.41. Steaing angle response to hidalsteering set input of 0.15 Hz and ampliide range [-1 11 with obstacle avoid fixed inputs of: (a) 0.25, and (b) 0.75.
Figure 5 42. Obstacle &or vs. range to obstacle as a fiinction of angle to obstacle with steering angle of zero and speed of 1.
Figure 5 43. Obstade avoid steering adjustment vs range to obstacle as a fiinaion of angle to obstacle with sîeering angle of zero and speed of 1.
Figure 5.44. Steering angle nspansc to tunnel-tracking mor
Figure A 1. Overall system test configuration
Figure A2. Complete system
Figure A3. Operator/console
Figure A4. Operator speed/steering error response
Figure AS. Operator speedlsteering response
Figure Ad. Local teleautonomous controller
Figure A7. System delays
Figure A8. Remote te1eautonomous comr~llu
Figure Ag. Obstacle-avoidance
Figure A 10. RTC speed control Figure A 1 1. Fuzy speed conîrol
Figure A 12. Smith wntrol
Figure A 13. RTC steering coirtrol
Figure A 14. Fuzy stcabg wntrol
Figure A1 5. Autosteehg control
Figure A 16. Robot vehicle
Figure B.1. Ahernate cornpiete system
LIST OF TABLES
Table 2.1. Appmximate maximum dday limits in teleautonomous corn1
Table 4.1. Fuzy associative maÉr9r @AM)for hazard situation
Table 4.2. Fuzzy asochtive rnatrix @AM)for obstacle factor
Table 4.3. Fuzy associative matrix @AM)for speed factor emr
Table 4.4. Fuzzy BSSOCiBtive matrix @AM)for speed error oontrol
Table 4.5. Fuzzy associative matrix (FAM) for hybrid control
Table 4.6. Fuay aSSOCiaSive matrix (FAM) for obWe avoid control
Table 4.7. Fuzy aSSOCjative mrtrix @AM) for intersection contra1 1. INTRODUCTION
Remote wntrol of equipmeut is widwptcad ohd wmmonplace. Consuma products include band-heid trensmitters used to corn1audiohridea equipment in homes. Garagedoor openers dow usets to mmol apeMng and dosing of garage dom Born th& vehicies. Hobbyists wntrof model airplanes, cars and bats using portable tcarismitters. Home automation systems aiiow owners to monitor and eontroi household grstems hmwherever there are suitable communication Wes
Remote wntrol is wed widdy in mmmaciaI and hdwtrial procwses where the operators air physicaüy Wolated hmthe proceas being wntrolled. AircraA, ships and submarinw are controUed rembtcly hmthe cockpit or bridge, although this may not fit the normai perception of remote coirtrol. Power stations, pctro-chernical plants, railtoads and tmning operations use tanote control in th& dsily opentions. Rernotdy- operated (ie. controlled) vchicies (RQVS) arc rodneiy wed for various undcrwater applications. Spaceaaft, including satdlites, are wntroiiai hmthe eartb wing radio communication links. hxwteairtroiied vehicies kwig various levds ofautonomy are used by mi@ forces througbut the worfd on land, sea and in the air hr ddefence purposesuchassurvallaneednco~~iswell~foroniasiveopartions~
Most of these examples have no si@cant deiay between the time îhat contrai sigds are traLlSmitttd and the time that thy Mive at the reeeiver. For these cases, remote coimol is dativdy stnightforward The biggest problern is probably ensuring the integrity of tbe fodand reverse communication cimach. LmesGsight transmission (for radio or optical chan&), diicient transmit powcr and receiver sengtivity, !king of nciae ad distortion are important consideratioas. communication link haMig appdleloop tirne delay 4.5 sec) [l] is laiown to be a ciifficuit task for a human operator. in order to impron systan perfomiance over the "move and waitn ssategy adopted by human operators in the timedehyed remote- control environment, it is usuaüy necessary to provide some degree of autonomy on the remote vehicle. A t#m which bas been introducd to descni the interaction of humans with remote, inteIligent, partiy-autonomous systmis is mteleautonomous" control[2]. This tem ineludes the fiill range of autonomy in rernotacoutrol systems such as manualheleopaation, cornput~~-~stcdhelerobotics,d semi- autonomoudsupcMsoIy controi, although autonomous/fdy automatic wntrol may not strictly fit the definition. These ddptonare not rigid, since thcrc is ohoverlap in function and definition. Teleoperation, for cxampk, is oAen wed in a geaeric sense to rnean that there is a human operator in rd-tirne, on-line wntd of the system. Ifthere is computer-assistance, as will oAen be the case, teierobotics might be a batcf description. in the system dcscfi'bed in this thesis, there is a combination of autonomies, hmtekoperation through supuviso~~control.
Renmtc -01: "Control of an operation hma distance: this imroives a link, usually elcarical, betwan the control device and the apparatus to be opmîd. Note: Ranote control my ùe ovu (A) direct wire (B) otha types ofimercomecting channels such as carrier - aimm or microwave (C) supenrisory control, or (D) mechanid means" [3]. The type ofciirria in (B) shouid be arpanded to include transmission in ohmedia such as opticai or lcoustical triursniissim Anothcr definition of a remotc-coatrol (or tdeoontrol) system is "a closed-lwp system which must consist ofat least the foiiowing demmîs [4]:
1. Sensors of infonuation. 2. A trensmission system to wmmuiiicate infonnaîioa to a ranatacontrol . . 3. A control point wbicû inchides a human or automatic decizaoik.malring systcni. 4. Devices to tnaslate infodoninto appropriate control si@. 5. Links to mnmwicate infomation to rcturrtors a rcmomoatroi points. 6, Acniat~rsopcratedbycomnilsig~I~to~~nsivew~lled opetations Ont of the fûst reported applications of tearote control was the bridge telegraph systan for ships [4] . Conbol cornmandg hmofficers on the btidge were transmi#cd down to the engine room where the appropriate control action was @rmed by humrui opentors. Eventuaiiy, with technologieal ddopment, tiiis process was automatcd JO that the required control action wuid be accomplished directly fbm the bridge.
The deMion of ranote wntroi, as d in this thesis, inchdes the presence of a humari operator. The primary reasons for using remote control are codena, economy/cfEciency andor s&ty, cach of which has specific relevance to the human opecator in the cantrol loop. Garage-ûoor openers are primady a convenimce; centralking control operations a.a remote station is done for rcasons of ecammy and &ciency (and oftm m);satéty is a major motivator Eor runote controI of operations in emrironmcius that are unpleapant unhcalthy or hazardow to humans. These cuviroments include industrial plants (smoke and toxk fumes), nuclear plants (ionking radiation), undersca and outer space (erdnmepnssuns anâ tunperaaite~), dtqtheaters of operation (gunfire and explosions) and underground mifies (fahg rock, poisonous gas, indiutnt oxygq floodhg).
1.3 Time Delay in Remote-Control Systems
"Runote" implics physid distance separating the control and eontroiied sites. Physicaiiy-separatecl sites requin communication Iinlrs to conwy the forward wnîrol signals and to tehvn the statw hfbtrmticm for propa closcâ-loop wntroI. Oepending on the type of system king conbolled, ind the environmens, the comrmmVrtion channels can becorne significauî factors in the ovffoll coatrol system. Informaiim carrying capacity is limited by bandwidth, Trifiocmation rdiddhy is Onied by chamiel distortion and noiseiseBandwidth, diûortion and mise are gaiaaISt not a major problem in rwiott controI since commilM'caîion iinks cm usually be designecl to provide acccptabIe performance. Syscem sadbhy to these f4cton is not COllSidered in this thesis. The most signi6cant fâctor in5roduced by the Gommunidon channei m remote-couüol applications is thne As the total loop thedelay becomes an appteciable Wonof the time cwstant of the ptocess being controiied, the do& loop coatrol systun can beeome uostabie 'iile delay, also calid transportaiion lag in conîrol systems, has been studied foryearsanditseffectserewdknownenddoaimentdinstandardcontrolsystem textbooh. The basic problem with time delay in closed-loop control systems is a decrease in stabiiity. Dependùig on qstem gain (and other &tors), even d delays can render an othenivise stable system as being unstable and unworkable.
Consideration of time dehy in fadback control systems with human operators in the lwp introduces additional complication. Attempîs at modehg human operaton as active deviees with a transfér fiuiction go back to World War II [5] and continue to the present day. It is well known that the human operator is an exûwdy dificuit subject to mode1 due to the complaity of the human informatiomprocwsuig system, inconsistemies in behavior fiom one person to another, or in the behavior of an individual fiom one tirne to another. Reféraces on the effects of tirne delay in human actiMties go back many years. Many of these original studies were performed by psychologists and medical researchers m th& studies on human behavior. In 1960, WMSmith [q reporteci the effeds of a 520 ms fiedback time dehy wben perfonning nine operator tasks including writing letters ofthe alphabet, trachg symbols, writhg words, etc. The delay resulted in oprator hstmtion, increased difkdty and tirne to perform othetwise simple and emy ta&, d-on of lcgi'bility and various kinds of errors. These donsue indicatin of the problems cncountered by hmm operoton in remote-cl,ntrolapplications.
Most lwoarch into timsddayed mote couûol has focused on the control of robotic manipulators as opposai to the control of robotic vehicles. Whüt then arc many dogous féatures conrmon to both types of systans, there arc sigdcant differences. Maaipulators an devices that wdyrequirc a high degrec of spatial precision in three dimensions; thae rnay be many degrces of Wom dependhg on the design and the overail con6guratio~govmbg the motion of the ums or "end efféetors". Vehicle movanent is r_rPait;lihr dedto two (or thnt, m some instances) dimensions and does not usuaüy tequirr comparable pdsion, ManipuIators arc noma@ hed in place and operate in r rrIatively sûudemriromnent; Vetiieles are mobile anci often operate in rn unsrnicbaed cnviroiiment. Tùis acleabates the problem of moddüng both the systan Md the environmuü, wûich is mmdy tequirrd for the design and Miplemcntation ofr traditioaai coatml systan. 1.4 Tiedelay compensation
Various techniques have ban deMsed to compensate for, or ameliorate poor system performance due to time dclay. Som of these techniques are specific to runote manipulation but others are more general adare appiicabie to remote driving as well. Force and velocity feedback in various forms fiom a slave manipulator back to the master terminai is a very common technique with mote control ofrobot manipulators. For remote control of robot vehicles, othcr techniques have been used including Smith contrai, predictor displays, and supavisory control.
Onginally developed fix the control of industriai processes having inhereat tirne delays, Smith control[7] (aiso laiown as Smith's Method or Smith Predictor) continues to attract attention by researchers. Wbile this technique saw limited use in the analog systems of the day, the case of digital Unplemmîation bas tcsulted in its increased use more recently in practicd control systems Es]. Tbe essence of this technique is to replace the conventional controller inside the closed loop with a new design to effectiveiy move the time dclay outside of the hop, thereby maintaining closed-loop control of the non-deiayed poriion of the plaid A traditional Smith controiler nquires an accufate estimate of the nondeiaytd plantlantdyneimcsand an iccurate estimate of the (iiefart) delay for good performance [9,10]. Runate eontrol with timc dclay in the loop is just another example of a closed-loop with the dehy rnd caa be controiied on an end-tdbis by uhga Smith mntroiier. The tirne delay in this systern is essentially compriscd of the communication. . Iinkd&ys,wtuchcMbe measured explicitiy and acamtdy. The plant 6u no inherent delay (presumabiy), w its dynamics are avaiiable for the Smith controk. These fhctors tead to suggest that Smith control may be ideally suited for cl&Ioop control of non-dJayed plants, including direct teleoperatioa or telaabotics, over timedelayed Ws[ll].
Wctor displays provide the haun operator with an estimaîcd (ie. prcdicted) outputofthesystemasifnotiwddaywerepnsent. Tbisisuawilyrccomplishedin one oftwo ways: by a mode1 of the system without timc dday, or by running a mode1 of the delayed systern in fista tirne. A prrdidor display can diow tbe operator to drive the delin sinnilation to obsuve its ctutput in mi tirne. Tb0 samt conîrol signalsusui to dmnthemoddcaa bemtîo tbe nmotesite w îhattheochislsystem outputshdd(iderlEy)beh~t~t&suneutbatoftbeddbutd~in tirne. Obviously,~rthWtoworku~isiatended,tbenioddoftbesystaamustbe vqaccurate. Having access to a mode1 of the systm can aiiow predictor displays ta perfonn okfiinctions in off-line mode such as: vary the speed of the simuiatio~ ficczt the simulation at any point in the tespoasc, make adjustments to moditjr the systern nsponse, etc. A pndictor dispiay cmbe a powerfiil sssistant to a human operator and can provide sigmficant improvernent in the tirne it takes to perform remote-control tasks onr timadelaycd links. hcrebg developmcnt ofcornputer- generated graphics is aiiowing aeation of virtuai environments and virtual reaiity-based teleoperation [12]. This can be wnsidered as being closely associateci with the tùnction of a predictor display, even providing a complementary and enhancing capabiiity.
For long time delays, teleoperation in rdtime is not a viable option even if stability can be maintained, becurse the opcntor requins feedback hmthe remote site in order to detemine appropriate wntrol actions. in this situation, supmisory antrol is necessary. This implies semi-autonomous operation, The only othcr option is My- automatic, totdly-autonomous operation. This option is not of intenst in this pmject since direct operator intervention is required in the cantrol of the remote system. Supervisory wntrol[13] is hi&levd or task-ld control; it is not rd-iime, on-b wntrol. Control wmmands are symbolic, quiring on-board inteiligencc mdhr capabii at the runote terminai for interprebtion and exdon. This is in contt8st to direct teleoperation where dagicwntrol commands are used to directly control the plant in reai-tirne, on-line Mon Sqmhrycontfol usudiy rquires the transmission to the runote site of go& ad sub8oais dong with necessary insüwiions for accomplishing them. Basicaüy, this indudes sending rad updating amputer programs to be wed at the remte site. Sincc supervisory consr01 is not ml-tirne wntrol, it is vitai that some degr# of autonomy be provided on the runote robot vehicle. The reason fbr this is that with aay sigdcaut dday m the systun, a buman operator lacated at a distant site ~uinatrespond quiciriy enough to maice any neassyr correction to the system. Ifa hzrrdous situation should occur at the nmote site, the operator wiii not be aware ofit untii the system status Mformation is recemd ovathe time delay. Aay CO& action initiated by the operator and transmitted back to the remote site will be hiladyddayed. This dwed respotlse could be atmtrophic to the robot vehicie or to auythb&rnyone in the vicinity. Thae may also be r Mnety of less criticai tasks of an automatic aanrre nguired mch as monitoring ad adiwtiag onboard system parameters in a tmiety Mionto midsystem degndrtion or fküm. Thcrcfbre, an tirnad&* rumeanml qstalM Ke llonnilly ptovided with some degree ofnutanomy sufbntto ciram3venS aichproblans. 1.5 Underground Mines
The general area of application fbr tbis tesearch is the tanote control of robotic vehicles operating in underground mina. Intaest in providing robotics and automation in underground mines is high, motivated primarily by cconomics and worker saf' [14-161. This is a very difIiailt environment hrcontrolsystms, compounded by the sensory problems. The walls, cciling ami, to a lwser cxtent, even the floots are uneven and rough, wmposed of varying geologjd dements, nsultmg in complex dections made even more dif~cultif moisturc is present. Tunael openings and intaseetions may be difücult to identify accwateîy. There may be a Mnety of obstacles and hazards - fàllen rock or other debris, mine pasonad, otha maches, etc. Lighting may be low, nsultuig in poor andlor codkhg visii. DrivwMs~rivewheclsy slip or &id. Then rnay be exhaust smoke fiam okvehiclw ind dust in the air. Vibration and vehicle motion is continuaüy pr~~~~lt.Machine sensors such as vida uid sonu do not have an easy taslq with the reflections, scettaing and rbgorption ofwaves. For conditions like these, whasensoiy information mybe vague, ambiguow ad impreaSe, fiizsr control has often been uscd su- [ln. on the universe of discourse. The nile base is a iist of logic Jtetements tJating the input and output linguistic control variables, descn'b'ig the resuiting control action fbr al1 comôiitions. These statments are typidy of the fom:
IF (input Ais X) AND (input B is Y), THEN (output C is Z).
The debase is often represented in table form as a "fiiay rssaciative matriir" or FAM. The iilference engine accepts the system inputs as "fiiaified" linguistic variables and degrees of membership, perfonns the tkySerena proçess and "dcfudies" the result to produce a specific "crispn output due. This rnethod of taking input information in descriptive fonn, applying logical desand inférring the output dtis cIeariy comparable with the human reasoning process, albat at a mdimentary led.
Neural networks are configurations of artifid naitous interconnected in various ways m order to accomplish a desind input-output mapping (211. The most widely used application of ndneâworlcs is in pattern tecognition end they cm accommodate both static and dynamic pa#enw. The input pattern to be rccognized is wnnected to a nurnber of input nodes, usdyha* no associatad weim. These input nodes an then connectai to a second (biddca) layer of namns via connections having adjustable weights. The llcutons usdyhavt an adjusîable biinput as wdl. The weighted inputs arc aU summed togetber in each nanon and the resultimg signal level is appiid to a nonlincar achation fimction (usudly sigrnaidai). The output of thïs fùnction represarts the output of the naa01~The outpub of aii hiddai layer nairons are then usuaüy hi to r subsequent lrya of namms via concdms fiaving adjustable weights. This layer may be motha hiddca lrya or may be the output laycr. Any number of hyas can be uscd, but udiym mon thn tbree in used. The output Iayer newon(s) often contain a linw rctivlsion Monto dktbdy provide ui output which is a linear combiion of its inpits. Som ncurai networb han fecdback connections with delay tmits hmhidden or output laycrs brclr to pmiious layers creating r#xtmat aetworks for modeüng dmcproasses. Many types of networks are posst'ble, depending w the an6guntion and Mdaconncaion ofthe neuroas. Ndnctworks iue capable ofundahg lin#r ud nonkaf fiimaions with grcatacairaeygivtnanappro~~aadd~eatmimbaof~as.By admgthe ~~&gStsushg a mcthod such as bcdt-~fopgdon,the network is tniwd tonalittthetequediuput-outputmr9pm%.Tbislcuningciplbiütyofd Using a neurai network to modd the remo~lsystem without time dday should be dcient to act as a suitable predictor to provide the operator with a non- ddayed eshaîe of the actual system output. Tbis dyedoutput hmthe nad network modd is avaiiabk immediatdy, Wre the actual runote systcm cm nspod due to the time delay. Thdore, it is considercd to be a "predicted" output and the adnetwork mode1 is deda "predictoi'. The predicted output should id- be idddto the ac&ai system output if there was no time de@
Both fiiay logic and neural networks are airrentty being wed to conml cornplex noniinear systems which may be difl6icult or impossible to modd in the traditional mathunatid sense [17=23]. Both fiizy logic and neural uetworks can mode1 nonlineu firnctions of ubitrriiy compIatityatityBoth techMqucs cm work with inputdatathatmaybevague,rmbiguauswimprecise. Bothtcchniquesuerobustm the sensc thaî systan fhilure is gnafiil, not cltlffmphic if a portion should ful, Tbese anaupod~brusùigintdügcnt~~ntrdtediniques. 1.7 Objectives
The objective of this thesis is the design of a system foc the remte controi, over a time-delayed Mc, of a robot vehicle operating in an underground mine enviromnent and incotporating iuteiiigcnt control techniques [24]. induded is a muitivariaùle control systmr based on fiizy controi, a display showing the predicted system output fiom a neural network prcdictor, Smith corn01 for timedday compensation in teleoperation mode and ~u~sorycdntrol hr scmi-nomous operation. Autonomy is provideci by including obstacle-avoidance in the ysiem desigq as weiî as automatic tuinel-tracking and automatic steering into interdg tunnels. System dynamics, often omitted in many robotic applicaîions, are inciuded. Human operator reaction delay and nmmuJcular response are also Uiduded, as well as their compensation in the overall system design.
1.8 Outiine of the Thesis
Chapter 1 introduces the subject of the thesis and briefly disewses the various topics that are hdamental to this project. These topics an ehboratcd on in subsequent chaptcts. Chapter 2 presents a Litcrature miew of devant nseareh. Tbis includes performance resuhs of hunian operators in control of various systerns udh time delay, meth& of compcnsating for tirne delay and pafonnance rdtsof modehg human operators in mntrol applications. Cbapter 3 wvers the basic tbeory of the probiem with time delay in control systei;rs. Smith control theory is Mchided here, and the SimpHed Smith controlier, partiailarfy as it relates to remote wntrol, as well as iîs use in compensating fbr human opem!or dynamics. The ratide for using quivalent circuits in simulations involving predictors is also atplrined hae, Chapta 4 descridm the complete systcm dcsign. AU subsysreais arc descrii dong with hir firnctions. These include the various fhycontroiiq human opaator modelq Smith control and the semi-autonomous modes of operation under supervisory controL Chapter5presentsthesystemsimulati01l~dtbeirrwuhs.Aiisystemfimciionsd opaatingmodesaretestedandtheresubgperf~rmanceWdemonstrîted.Cbripter6 presents the sumamy and conclusions. The conûibuîions of tbis thesis are inciu&d b.Fi, work suggested for the fûture is idddTbis inckidts wotlr origùirls. plamieci and insended for îhis thesis, but time did not permit. 2. LITERATURE REVIEW
2.1 Time Delay in Remote Manipuiation
Intmst in spacc arploration in the 1950's and 1960's dtedm a number of studies involving human operators in control systenis having time dday in the loop. Since much of this space interest was conantrated on hmar exploration, the time delays behg wnsidered had to indude at leut the twbway @op) mmsmbîon dclay to the moon and back of2.6 seconds.
in the eariy l%O'q Sheridan and Fd[25] ind Fd[26] reportal the effccts of the delay in remote manipulation tuks. This was the tesult of a study for NASA 1271 which was interested m tdcopention applicntions in outer space, and particularfy on the mwn. An operator at a master terminai controllcd a slave manipulator through a comunmicrtion iink fnop ddays of O, 1.92.1 d 3.2 sccoads wcnmtroduccdinthelinkbetwcen~~dtheslrntaminals.Twotypesof acperimmtswenpcrformed:theQstwisboPc~~,rtvrri~~~l~of mw, the second wua more complex ram@uMon. As ddry inenued, the toul time to complck the tisk hcmd. In addition, the campletion tune incttued with mcrcasing Idofditti.cuity, as apstaienpeaed,It wu rrportcd tbat tbe opmto~~ atpaiddifKcuityuuiEiÉiguewhendtiayawacptesmtt Anhterdqm was that eacb opastor would automrticnüy rdopt r "move d wait" smteey to accomplish the tssks in the prcscnct of de.Tbis hhnsan opaAoop move followed by a puiod of waiîiag to sce the effect or responsc. 2.2 Time Delay in Remote Drivhg
One of the eariiest experimentai studics on the &iof time dday in teleoperation of a remote vehicle was performed by Adams [28,1] for NASA Its purpose was to investigate performance of systcms wing humrrn operators in the -te eontrol of machines at long traasmission distances - spccificaüy, in remotdy driving a vehicle on the moon &om the earth. There were two phases to the study: Phase I investigaîed the performance of human operaton in simple simulated tracking tests with lcmp delays up to 6 seconds using various ttacking speeds and both vel& and acceleration controls; Phase II iuvolved the teleoperation of an odual vehicle using lwp deIays of O, 112, 1,2 and 3 seconds and Çred velocities fiom 0.12 to 0-82 m/s (0.4 to 2.7 füs).
The tracking experiment used a simulateci vehicle and various ttacks on a vidco scrm. A number of graphs were generated, cleariy iüustrating the degradrition in tracking performance with increased thedelay. For the actual vehicle conmi, ùoth a wntinuous course and an obstacle course were marked out on a parking lot. Twa- wheel and four-wheel drive vehicles wae remoteiy-controlled usiug loop deiays of0, 1/2, 1,2 and 3 scconds at various speeds, A television camera was mounted on the vehicies ta provide visuai feTop speed was 0.82 ds(2.7 fVs) with 0thtest speeds ranghg down to 118 of top se.
Mer a learning period of driving with no delay, operators found a delay of 1/2 second to be "a bit of a ahock", rquiring more conanüation for satisfiiaov driving, One second delay required cxtnmc concentration, both on the ptesent sitwtion (on the display) and the fiiain Won(deciding on stemhg control action to be petforniad). This concmttation was comparable to 'rad physical work". A dday of 2 d (and top spebd) resuitcd in total loss ofcoutrol. Operaton wodd also lose th& vehielc orientation completeiy. By ducing the vehicle's speed, mon dday could be accommodatededSatisfactory control comparable to @ispeed witù no d&y could be sttained at 3 ~axndsdelay ifspeed was reduccd to ln offiill spad (0.12 ds). At 11'3 of fiill speed (0-27 ds) and 3 seconds dm, @ormance was wmparabIe to tbat with fidi spad and 1/2 second dday. In light ofthe rcsuits obtamed, an absohitt macimm spced of 4.5 mls (10 miBir) was recommended firr fanote hraar driving fhm the car& with speed reduced to -0.06ds (-118 milhr) fbr complut terrain nwigaticm Mts of the actual vehide tests produced reasonably good wmlation with the SimulatH( vehidc trawtests. Cormnaüs hmthe operaton were very enügbtening mchrdhg the suggestion for using prediction to idthe operator.
In a more roeent snidy 1291, sevcn opemors took tu- driving a robot vehicic through a maze containmg obstacîes. Tmruns were ht made with no delay in the remoteantrol system. One of thme nuis was performcd with the operatctcs viewing the mbot and maze direetly; the second iun was made immediateiy foliowing the fht using a vida cmera and monitor, with the operaton and monitor in a separate room. The average time oieach nin was 53.1 seconds and 52.4 scwnds, respectiveiy, indicating no degradation in pafonnancc whai ushg a vidm display. A third cun was perfoned on thfoliowing day with a delay of3 seconds added to the control loop. On th& nrst atternpt with deiay in the bop, four of the wen operators aperienced fàiiures mon &er starthg. Ail nuis, succediû and unsucceanfiil, wae timed to completion or Wure, whichever occurred. Althou& ideniScaiion of which nins were successful was not proMded, the average time for al1 was about 245 scconQ, ranging fiom 170 to 350 seconds. The degradation due to time delay is apparent.
2.3 Tim&ehy Compensation in Remote-Contd Systems
Before arPmining runote eontrol of vdiiclea under timc dday, we wiii bricfSr look at some of the ment rescarch on techiques fbr compeasation of tin# dday in îk remote coirtrol of robotic dpuiaîors. The vast of rucarch papen on robotics with time delay invoive ma@hmn dur!ban vehides. While tbis section is applicable ody to mnipukors t is informativt to sec hm much dday cw be accommodated in îhis area and the various tectuiiqucs iwohd lossless transmission line. This guaranteed passivity regardlcsp of the amount of time delay. The authors indicâted that the CO-C~~~OII circuit cenwt destobilize tbe teleoperator, regardles ofthe time delay. Two tests were conducted; fiirst, the slave was commandeci to move until contact was made, then the force was ramped up and dowu; second, backdri* was tested with the master swinging bdyand the slave being hamrnered twice and then being moved sjmisoidally. In di cases, stabiüty was maintained using the passive cuntrol law that was duived.
A variation of this technique, while retaining the principle of passivity, involved transfomiing the system force and velocity variables into wavc variables [3 11. This alxi enabled treating the qtemIike a transmission line which dowed matcbing the end terminations to the "characten'stic Unpedancen to prevent rdectiom and instabiiity. Stable tracking wes achieved with a deîay of 1 second up to the point of contact of the manipulator. At that point, force feedbaclr was d.It was indicated that, in using this type of technique, system p«forrnancc degrades smoathfy with increasing delay.
Shared cornpliant control (SCC) is a recent devdopment hrmmipulator telerobotics [32]. With SCC, a human operator controls a telaobot having a cornpliant telerobot hd. The cornpliant hand sohcollisions and aiiows for contact wing Little force. Ushg SCC, kinesthetic force fiedbaclr ce. sumi through human nerves and tissues) to an operator rcsuhed in stable teimbotic wnt~lat ddrys up to almost 1 second [32]. Inereasing dday up to 4 secor& did not M~LPCmSt(LbilitY if the force feedback was femoved as long as the SCC was rctahed. It was demonstrated that, by using this SCC mahod, tasks wdd be completai with up to 8 Jeconds deiay. Tests included maintaining a constant fora and peg-in-ble insertion at the slsve site. This system effectivdy used two control lwps: an outer loop fine informetion (end-to-end) ami an inncr (lacal) loop at the nmote site controlling compiiana and damping. By mmving thc force fécdbaclt to icbicvt control at delays longer than 4 seconds, the odsystan is o~loopon an end-&-cd lmh despite ha* closcd- loop control at tht slave site.
Another unique procedure mcorporated bth ddayed fime and ddayed position information to provide a "rnodüied krce kdbackn rt the master tamuial[33]. The predictive &ect nsuhcd in stable tdeoperator co~lwith short tm# delays up to 2 or 3 seconds fôr both position and force tncking tasks. Ideal teleopemion would provide the operator with the fiel of directly manipulating the end e&ctor in perfoming a ta& [34]. Sincc the forces F and velocities V at each end are reiatcd by the end impedance Z = FN,perfêct ttansparency requires that the impedance of each end be cquai, Z113%5ter= Zsiave. Lawrence [34] has studied the performance of teleoperaton on the basis of transparency. Four information channels conveyed the forces and velocities between each end. in a manner similar to [30] and [31L a transformation of variables was used to modifj.the architecture so as to ensure passivity (and ihereforc stabiiity) in the prcsence of delays. Transparency vs. stability was studied over a range of impedances for three architectures: position-position, position-force and "passivatedoposition-force. Transparency and stabiüty were hund to be conflicting objectives, requiring traâeofE Transparency with impedana. A delay of 50 ms was used in the study. It was reported that the maximum acceptable delay for this systexn was "probably" around 2 seconds.
Recently, a controlier has been designed for a bilataal, force+reflecting teleoperator using & control and p-synthesis techniques [35]. Using a mode1 of the shuttle remote manipulation systern shdderjoint, srabie control was acbieved at communication delays of 2 seconds in each direetion This was accomplished by considering the tirnt deiay as a pmtu&bn to the sy~tanand desi& the controller to compensate for pcmuktionrotl3 1 seconds. This Tbtntroiier wu sboorn to k unstable at a 3 second deiay.
The previow dtsshow that, fbr most cases, the amimm time delay that can be compensated in the remote contrai of robot manipuiators ushg the techniques described is about 2 seconds. These an esseatidIy passiw systems, with no loap gain, and involving the transmission of force and vdocity iafOrmdion betwœn the master and slave terminais. Thy are not applicable to the tanote control of a robot vehicle.
23.2 Smith Contidlcn
Triantafj,Uou and Grosenbaugh [36] have mxcddy uscd a Smith coatroiicr combined with a cornpcnsatar based on LQG/LTR methodology in the remote mntrol of a tethemi submersile vehicle, ARGO, Lmeaf qurdrPSic gausshn (LQG) methodology with loop traii9fér n#wery (LTR) was Was to eusure suScien! robwtness~with~oventlcoatrolsrJtem~refemdtoururobustSnaith controller". The vehicle was wntrolled through 'dymmic positioning' of the surfrice 4. The surfâce ship's thnistas were usui to provide the contro1 force uid the hydrophones and submerged pingers provided position information. A vchiclt weighing 17000 N in air was used, dong with a 2500 m tekbaving a loap tirne dday of 40 seconds. Human openiton are not able to control systems with such long time constants [3q. Simulation and acnial tests resulted in good (ie. stable) overaü penomiance.
There has ben a considerable amount of research paformed on the subject of Smith control ovcr the years sincc it was introduced. The ICSUlts of some additional research is included in Chapter 3 which disaisses Smith control in daail.
Preâictor displays and predictor instrumentation have been wed in the control of submacines, aircraft and space vehicles [37l. They have assisted huinan operaton in controllhg diving and establishing depth control in submarines [37]. The prediction is based on ruaniag a mode1 of the system in fiist tirne. It is reported that a novice operator with oniy ten minutes of trainhg and using a pndictor can opente a systcm as well or better tfian a skilled operator using codonalinstnimentafioa Typicd predicîion time spans can be quite long: in the nnge of3 to 6 ~ccondsfbr helicoptcn and hm25 to 30 seconds fbr submruines. A number of simple snidies wae donc using this technique to investi- the use ofpredicton in coutroiliag space ship rendezvous, docking and ianding [38]. Experimental dtsushg a detyof people WithdiffCnntbaclrgroundsahodsi~~improvanentwtrenusingapredictor over the casc with no predictor. Other studics involving submarine JiiiPilatioas hive show11 that coiiisiolls using standad inzitnimaitation are ndytbrœ tùms as many u when using predictor instruments [39].
One of the 6rst major studiw iuvohg the use of a predictor in tbt nmotc control of an actual vehicle with signal tMsmwsion delays was by Arnold and Braisted [40]. A humaa opaator was used to stea a robot vdiicle &s& dong a tmck and thai thraugh a mye usiug a communication ünk with idjustable timc dclrys and a ddsyed television display of the iandsap. Supnimposed on the display was r merLer sbowing the pndicted location of the vehicie f9r the aimnt sieuhg commands. Wrttiout tbt ptedictor ad rnarker, an opcrator must mcntaüy eJtimatt the &WC location of the vehicle taking into considcratioa tbe nature of the terrah, implicit dynamics of the vehide anci any control commands made up to that tirne. Having a predietor fines the operator hmthe stress of paformiag such mental caicukions aild nquins ody that the operator steer the predicted marker in nal (ie. operator) tirne. Test resuits showed bat, for a loop delay of 2.6 seconds and vebicie spad of 2-16 m/s (7.1 füs) ,an operator wuld drive the vehicle nearly as well by wing the predictor as one could drive without time delay (and no predictor). The predictor was not hund to be of great vahie during open field driving, but was wofiil when precise maneuvers were needed, such as dtiving through a maze.
Tests of delayed tracking with and without preview (prediction) were conductecl by Stark a ai [41]. Ons-dimensionol tracking of a targct on a vida screen by a human operator was studied ushg delays hmO to 0.5 ~tcomlsin 0.1 second increments. Pr* expMments incluâed the location of the targa fbr the nmt 0.5 seconds. The target motion foiiowed the pseudrandom path gaieraicd by the sum of fourtcen nonharmonically-relrited JimiJoids. Without delay, the gain rwpollst was flet (O dB) to about 1.6 Hz; the phase airve rcveaicd the human respom dday of about 200 m. Preview had virtually no dkî on the delryed gain response, but sigiiificantiy Unproved the delayed phase nspoase by about 2î3 ofîhe ddryed phse.
Conway, VOLEand WiULa [2] have instoduced the tenn ~onomow systcm" to include the whoIe range of huminvohrenient with wntroI systems. They used prediction in conjunction with tmie mi position "chiic&smto overcome the degrading &eets of time dday m tbe amtml of r nmott mipubr. The time ciutch was used to disengage syncbrony bctwcm opaaor spceinertwn time and teierobot manipulation time. This dîowed îhe opartor to phd generiae path segmucb m an off-line manner. The position chacbwas used to dimpge paaition syacbroiiy betwaa the simuiator and manipulrtor pah. This rtlowed control oftbe simulation to test possiile paths in dinerent situisions. Wben the simulation wu in a good position, engaghg the position ciutch provided a short, smootù path to the pwious path, By usingthtsecIutches,taskcompietimtmiewu~~icur3lynducedovasystansu.pinP pure pndictîm whicb, in turn, pravided lowercompletion tims thrn systems bavbg noprediaion. TiieddaysofO,2d4secorwlswausadmttietests. van de Vegte, Milgram ad Kwong 1421 used a modified opWcontrol mode1 of a huiiuui operator to control the amof a sumilated iubmersiile robot in a welding task. Operaton h9d to contend with underwatef twbulence whüe posia'oning a çursor at the end of a robot ann Delays of 2,4 and 6 seconds were used in the wntrol loop. No prediction, Taylor ptediction and "dynamitai" prediction were aü testeci, The prediction toak the fonn of having the cursor at the end of the robot am move relative to the submersl'ble. A two-dimensionaI display of a three-dimensional environment was then presented to the operator. Comparing the mode1 to rdoperaton revealed some difficuities in the modelling ptocess, In some insmces the human operator seerned to perform better than the acpected optimum. It was hypothmbd that the order of the internai model increases as the operasor consinues to gain arpecience. It was also detennined that a human operator wili aiways utilize some predictive behavior in time delay cases, cven when requested not to.
In a paper ammarkhg the stpte-of-w Sheridan 1431 discussed the use of predictor displays. For relativcly short pndictians, a Taylor series extrapolation of cumnt state and time derkaha is satisbtory. For longer prediciious, and ta account for nonlinear dynamics (e.g. saturation), a better method is to use airrerit stott . . and tirne denvatnres plus arpected ne tu-^ control signals to nm in a mode1 much faster than the process beuig controlled. This second type ia bettcr for controlling single-entity (ie. rigid body) deviccs, such as a vdiicie. It is not as goad for telernanipuiators ha- muhiple degrees of fnedom [43].
HirPnger et al 1441 uscd a predictor display in the telerobotic control ofa space-based dpulator over loop ddays up to 7 seconds. Due to the long delay, bilateral wntrol (forces fèd back to the operator) wu not used. A -r~@~~er provided position, touch ad vidm sensor information tiut conthdyupdated a wodd model at the ground-brued control CUW.Usiog pndictive stenographics simulation based on the worid modd, deIrys ofup to 7 seconds were wmpematcd f6r without major problems. A floating object was ru@by the maaipulator despite a 6 sccod round trip delay. It was reported that the system at the runote site and the tirne delays had to be modelied very icnirately fbr suWactory prediction. This would not work so weU for unknown or ullst~cturcdmvimnmaüs. Using data bus speeds up to 6 MBaud the system was nry computati- intensive. Kim and Bejczy [45] dso dunonstrated the rrucccss61 wuof a predictor dispiay in the remote corn01 of a robot manipuiator with fouad-trip dday up to 8 seconds. Thra hed television cameras at the slave taminai transitlitted video imagea baek to the master twminal2500 desaway. Computcr-gawmîd gmphics were used to show predicted donof tbc rnanipuiator by ovtrlay on the video dispiays. The openrtor gmcrated the manipulator motion in simulation, which was visuaiiy checkcd for cofcecttltss b&re adusi ccimmands were sa to the siave terminal. At that site, motion was assistecl by automatic cornpliance eamL The system rquirui four components:
1. High-fideiity gmphics models of the robot arm artd objccts. 2. High-fi- calibration of 3D rnodels to 2D views. 3. Overlay of graphic models ove actual images of am and objtcîs. 4. Conîrol of motion of ~rapbicimegts to be the meas fbr thc real robot am.
Exdait resuits were obtained (positioning emws m orda of 1%).
In gaiera2 pdictor dispiays Jlow opartors to 'Idthe aebal feedbaclr and take larger conî~~Iaciiom with confidarg improving time by up to -5% [43]. A predictordispIay~decreuestbetimecwistrntdtbe~~lltr~Udptocessbyrii amountequivalentîotheprrdiction spm, teduastbtgrinofthehimran~ fiinerionandtfitcontrolrdiondmaei#stfie~~ftbtdosed-loop~yaem [43]. There arc aiso r numbcr of1iniirrt;ins in the use ofpdktm [43]: plant dynamics must be modeiied acairatefy, positioq deand paJpectRre must be cadùiiy calibrated to the vidm displsy; depth pndiction in the direction of the video image plane is not vay uddue to the twa-dhnsiod dispby. Thse limitations are particularly applicable to the remote wntrol of robot manipulators, wbgt precision is essential. The same degree of prrcision is not quirad for the runote control of robot vehicles, but the general advantage of seeiag the road abcad to assist the operator csm~tbe questionad, as reporteci as eariy as 1963 [40].
A recent area of research that is reaiviag indattention is virtuai teality- based teleopdon A cornputer-gencrated Wniai &ment can be considered to be in the same general category as a predictor dispIay, even though the one does not ndyimply nor preciude the other. An accellent didonof the statesf-theart of wtUal reai~ty-based~eopmtion is pnsaited in [12], with parti& emphasis on wping with time delay, and accounting fbr human fàcton such as: enhancing amputer-aided tdeoperation, improving bilataal and shared wntrol, Moping human-antd~chitecanes,and~sensoryf#dbrcktothehumanoperator. The primary application is the wntrol of remte manipukm in gcaerd d hr special applications such as tdcsurgay and micm md nano-& t~pcraîion.Mobile robot tdeoperation is also discussed, but d-time control is limited to the delay-&te case. Examples hstfahg wmI of mobile robota over tiiiLe deiay nnrariably utilize some fonn ofsupaMsory control ushg r viirwl teJity-brssd masta station. Anothcr tum îhaîisusedalmostMterchangeoblywithvirCuilreilityWtdepnsencx. Whilevhd rcaütyatteniptstoprovidethtapartorwithtbe"ilhuio21ofpnsmcenatthc~c site, telcprcscnct attapts ta achieve ui féding of presenccn at the wwte site [12]. From the point of vicw of contmiiiq a physid systan, the latter dcaeriptioa would seari to bc prefQTtd. For the putinilu projeCr dwcniin thig Ms, tkeis no signine~tuseforvahralreaiity(ortdepresena)~otherthnitspossiileusersan enhanced pndietor display. on-line con- albeit deiaycd, and albeit under predicted conditions at times. The oniy way of avoiding such wntinuous control by the operator is to provide the slaved &t with increased autonomy, such that it can assume control ovcr its own actions for a certain period of the. Ifthis period of time is long enou& the destabilizing &ect of the delay can be cimmvented. instead of sending low-levei analogic wntrol signais to the slave teminal, the master tenninal cmsend goals, subgoals or wmplete sequences of subgoals ulPing hi&-level symbolic couuuands. This "supervisory conüol" method was proposed back in 1967 1471. A rcrnote cornputer at the slave site would then proceed to cary out the required wntrol action autonomously. A local wntrol loop at the slave site would provide delay-fie closed loop control at that point. A feedback loop fiom the slave to the master tenninal would provide a (delayeci) sup«visory loop- At the master terminal, a wntrol loop wuld be closed through a locai computer pmviding nondeiayed "quasi-fèedback" for prediction purposes. Sheridan [13] has descni a spectrum of control modes (ie. teleautonomous wntrol) which range hm totd manual control (teleoperation) through computer assistance (telerobotics) and supervisory control (semi-autonomy) to firlly automatic control (autonomy). He provides a summary of the state-of-theart in 1993 in [43].
Xaformation about the Soviet lunokhod program in the uuly 1970's bas nvealed a number of problcms they encountered in remote driving on the moon [48]. Besides visiiiity d&dties caused by BLpn hmthe am, black sbadows lad lack of depth perception, then was operator bondom assoaated with îk2.6 second loop delay. They did not use direct teleoperation. Using mdimentary supervisory control, openiton sent commands which onboard cornputers uscd to steer the vehicie and to control rotation of the wheels. Movment was slow und the hmokhods stopped baween wmmands to avoid gctting into trouble.
Most of the recent rcsauch on tunaddiryed remote-wntrol systans tads to focus on some hm of supuvisory contrai. Lindsay [49] uscd supeniisory control to rcmotdy control a Puma manipulator robot owr a 5 second deiay. Hyùrid fod position control was wed to position in end effcctoi having a campüant, .iaaar-bd wrist. A bdcdei of the sIave aivVonmen~was estabüshed at the riaptcr site hmsensor information at the slave site. TnaMt...;es in this modei, coupled with no* or imp& cownsnds were compumted for at the remote site using the cornpliant wrist and "guarded mves" (described as moves "until some expeded scllsory event occurs"). The humopartor msaacted with the Linanrtic modd in real the and commands were automatidy generated and transmitted to the siave site at least once every second. The wmmand conveyed information on fbrces and motions, tirne durations and co-ordinate hunes which the slave computer used to initiate actions. This system is able to operate semi-autonornously in partiaîiy-known, unstructureci environments over thedelays pater than 20 seconds. The model serves as a predictor for the operator to work in rd-tirne. Since the operator's control commands are based on the local predictor modei, model accuracy is paramouut. This can work oniy in well-stniciured environn~ents.
Stein [SOI extendeci Lindsay's "teleprogrammingn[49] supervisory control technique by incorporating behavior-based wntrol at the siave site as weli as enbanced error diagnosis and mveqtechniques. A Puma -puiator robot, baving six degrees of fieedom was remotdy controiied over a 10 second loop delay. Tasks invoived puncturing and slicing blanket material and boit adraction and reinsertion IndMdualized skilis, which behavior-based control provides, worked weii for the system descriied. However, it was questioncd whether or not this technique would be appropriate for controlling systems having long time delays (e.g. Mars rover control with 8-20 minute one-way delays [5 11). It was also felt that the techniques descrii in this study would not be as effective for tasks involviq little or no force feedback Having an operator available "allows for simplified and robust construction of the remote site semiautonomous controiier." As with Lindsey's technique, this would work oniy in wd-structured cmriroments, wbae rnanipulator robots tend to be 4.
Graves [52] dc3cn'bcd a genemiized architecture for teleautonomous systems based on the object modeling technique (OMT). An ndensive mriew of system- sp&c architecnires uscd in telerobotics and autonomous systms was prescnted to identify the various requiranents for a more generaüzeâ schcme. U&g OMT, he devtloped a design for an action dection mechanian (ASM), which is the "hmtnof his proposed teleautonornous wnt~lsystcm. The ASM int- witù sensors, various command input entities, pIiuincr, controiier and robot. The ASM is a software intensive subsystcn using 12000 lines of ANS1 C code to transbnn a suies of inputs hmdous sources into strcam~of action requests. These te~ucstsam then forwarded to other blacks for aubsequent action. Thc complete system is vqflexible, hierarchicai and supervisory in nature. Simuhtion of the ASM was pdorzlpcd to ver@ its functional performance. Tirne delay in telcauion~mowsystam was not atpliW addnssed or testai. No lwdware systan tests were perfbII13Cd Human supervision, ifnot ditect teleopaation, is mandatory at the preaent state4Etheart of robotic ~ystems. This watt01 could range f?om direct telcopa4Xion ia manual mode to ~Ripervisorycontrol in semi-autonomous de.Bad on tbe cunent state-of-theart as suggested in this section, we can idethe approhaîe maximum tirne dehy that cmbe accommodated using eoch of the four spaSc teleautonomous confrol modes (type admode of con~rol).A table showing these modes and approxixnate dday limits is presented below for cornparison purposes.
Table 2.1. Approximate maximum delay ümits in teleautonomow confroi
Type of Control Control Modç MY -Lii Teleoperation Manual Short -4s Telerobotic Compuîer-Assisted Modente - 10s SupenRsory Sani-AUt0M)rnOuS Long -20 s Fuily Automatic Autonomous VW Lang - 18+ s
2.33 0th- Techniques
"herc is constant nseareh into the gdproblan of compensgtMg fbr time dday or dd-time in ccmtrol systaiiu. Some of the rtant techniques which have ban dcvcloped arc badon the use of %, control[53], nainl aetworks 1541, fkq ~0utt01
Itsboddbed~nedthittootberidiveuciofrtseucbDmvohndwiththc rcm~tecontrol ofboth nmipubnand mobüe robots overtbe Tmerna [57-61]. A -or problem at prcscnî W the uapredidihy ofthe immmhio11 iatcnq (ie. time delay) and tbrwshput making modehg [a]. Whüe nsearch efforts are bQns made in the use of pndictiveaiding tectmiques for direct, oiilline consroi, the emphasis seems to be on providiug idelligenœ (on mobile dots) hr mcreued autonomy [61], ~cenultt~formofbcha~or-buedrrretiveeoII$oIdmsupavisory~. Since the Intemet is a public medium, ho-, it Ù not &dy to be bevrürblt hr commercial purposes such as the nmote wntrol of mining machines, so this parti& medium has not ban consideseci in this thesis.
Modehg human operators in wntrol systems is a vqdifliicult problem, compounded by many Won. Part of the problan is consistency - then is no way of guaranteeing that a parti& operator wüi respond to a wnîrol situation the same way every tirne. Adequate operator training will overcome this problcm to a large degne. Many human operators doing the same task yield results that are similar in terms of reaction delay and response charactecistics, but there is often wnsiderable variation arnong these resuits - a uniforrnity problmi, Them is aiso a strong taskdependem aspect, Human operaton tend to approach ciiffixent ta& in différent manners, hmting atternpts to determine wnvenient transfiér fùnction modtis of the operators.
Human operaton in closed-loop wntrol systans are dyimrohred in pedormhg a tracking hction of some type. Two types of display are commoniy wed for this purpose: a pursuit-type or compeii~atory-type.The pursuit-type of dispiay shows both the "target" (desind output) rrnd the 'Yoilowcf (actuai output), with the operator controlliag the movemcnt of the fbiiower to match the the wmpensatory-type of display shows ody the difhnce bawan the îmget and the foilower as an error signal which the opaator attanpts to minllnize. The target and foilower an not arplicitiy shom
Reports of studies [62] showed that eaciy nsearchenr CO- on identifying and modeling specific charaderistics ofhumaa operators, mcbiding:
1. Gain,K - degrte of amplihide responsc to stimuü. 2. Reaction tirne delay, r -tmiedetym~nspoll~ttostinmili 3. Neut0mwculaftimelag,TN -timecmstantofneuromiianilusystem. 4. Compensation panmeren -phse lad, TLto uxommodste The general transfer function madei, inauding di of the above termg is 1631:
This mode1 has been used by many researetiers over the years [1,28,46,62,64-69] with some variation depending on the partiails shidy. The non-bracketed portion represents the inherent response charaderistics of the human operator, the equalization terms m the brackct represent the dwacmhic nsponse that an operator will tend to autornatically adopt dependhg on the control task.
Adams [1,28] made a compnhQlSive study of the effects oftime delay in the remote control of vehicles by human operaton. Tests mcluded both simulateci driving using pd-tracking and actuai vehide mgunder various timdelayed conditions. Results showed the sigdicmt pmblem that time delay presents to human operaton and system performance (descrii m section 2.2). The simulations used the foliowing transfh fiuiction fbr the human operator/dispiay wmbiio~
This assumed a nominal operator dondelay tirne r of 0.25 seconds rad a neuromwailar lag TN of 0.25 seconds. The ovetalf systan gain K was un* ad lead times TL of O, 0.5,l and 2 seconds wcre tcsted. Resutts of these sinniluions werc compareci with those Ilring a human opcntor (both in sidatut driving tests). Various @s @s were used with both no prcvicw (ie. no hvadvision) and simuiated previcw (unobstructed and unimpoired forward vision). OvaJl rcsults varid dmmaticaily with speed, type ofcontrol (dodyor lccdmthn) and kad tirne. In gentrai, shmhtcd preMcw provided best tracking, as he expumi. At iower opeeds, uid time iqp lus than about 0.5 seconds, the traasrér fiimction modd Mnhtivdy good results, Leslie [62] hund that during pursuit-tracking tasks, human operaton switch back and hrth between open-loop and ciosed-loop coatrol. This was modeleci as shown in Figure 2.1(a). His tests established that, whüe the open-bop mode1 was a necessay part of pursuit-tracking, the most prominent mode is closed-loop control as one would expect. On this basis, the block diagram was simplified as shown in Figure 2.l(b) and using Equation (2.1) by letting TL and TIequal zero and Ietting Kh= K, or
He detennined that human operaton have cutoff tiequencies between - 0.56 Hz and - 0.78 Hz Ge. - 4 to 5 ris) depending on course wmplexity and no time d&y r in the system- The resulting closed-loop respanse is
d=r+z =0.3+t and TN = 0.45.
K, wuthe transfcf funciion for manuai amtmi, ~n~itmod(by Leslie) to bc constant over aii fkqucnts. By plotting the response and camparing it with experimentai results for the case of zen, delay r, perameter ducs were detcfmined thaî yidded the best fit. These values for K, d and TN are show above. For this zero deIay condition, a plot of the kqumcy response showed a artoff ûcquency of 0.53 Hz (3.33 rfs). Beiow îhis fiequency, tracking @ormiince is theorctically puféct and fiills at a rate of 6 dB/octavcabove. Tbis camparcswJIwiththecrossovcrûequen~(disaissedin section 2.4.2) mode1 for humen opaaton m vehiddrivkg or similar tasks whcn crossova fkquency values of 4.75 r/s (0.75 Hz)[70],3.5 r/s [71] and 4 rls (0.64 Hz) [72] have beÉn reporteci. It shouid be noteci that these modd tcsults did wt approxixna!e the experimentally determined aitoff frequcnucs as ciosety as hopad, but the general sbape and characteristics wtre considerai @yMc) to be gwd. -Output
Output
Figure 2. I . Pursuit-- bloc&diagram: (a) open- and &sui-bop modds, and @) simplifieci mode1 (MaLeslie V2D.
Eerlitf studics ofhimisn opaiitaps m control syskms [63] fôdvalues for paranieters to be:
withTN =-0.1 bciugtypicsl. ~~ofvduesofthtsetwopuametasspaathe values fodin many more recent sîudies [28,62,64166,72-781. Reaaion dJays r tmd to range fiom 0.1 sead[w to 0.3 d 162,721 with the mijority m the muge of 0.15 - 0.25 seconds [1,46,64,65,ï3-n]. Experiniental studies of human operaton in wntrol systems have established a fiindamental characteristic response refend to as the "crossover mode4 " [70-72,791. This mode4 is based on singie-input, singleoutput tncbg, on a continuous basis, of random or quasi-random inputs. For this situation, a human operator tends to adopt a response Yh(ja) such that
wkeYp@) is the response of the plant, a, is the aossover hquency, r rcpresents the total time delay in the human information processing systcm, including musailar response. ïhis model has ban found to be valid over a ficqucacy nage of 1 to1.5 decades around the crossover fiequency. This model is consicid to be fiiadamental for human operators and is ohused m studies [ 66,77,80-821.
2.4 Precognitivc Control and Pnview
McRuer [70] bas deveioped r srnichinl modd hypotbeshiq r nurnber of patlnuays in the huxnan paœptadnamnmirnilu iystan to uamt fbr dinerent types ofconttol used by operaton. These types inchde wmpeasatory, pursuit d precognitive control modes. Depending on which pithwiy is used, tbe cuutrol can be open-loop, combination open-loop/closed-loop or totdy dosed-loop to visuai stimuli. Proprioceptive feedback is also inchided hmotha scasory sources such as joint recepton. This is shown in Figun 22. Closcd-loop wmpensrtory control is concemedwitûrcsponsetoerrorsorplrntoutpnonly. Whensysteininputscanbe distinguished hmsystem outputs via îhc display, the open-Ioop pursuit control pathway is ache. Under conditions oftotrl fhüidywith plant dynimics d the environment, a skiüed operator can genente neurodcornmirads *ch are appropriately timed, scald and sequCIlCCd to gaiarte the plant output ercaaly as desired. This opeil..loop behaviour is dedpreeognitivc contmL Pmm@ivc conÉrol is often combiind with wmpwatory control m a dual-mode type ofcoa~rolwhen the precogniîivc corn1compormt hitiatm tbe (ipproprirte) control &an and the wmpmsatory corn1compoxmt cornplaes the control rctiw by mûhbing the error L --.r tir . .. and mamtammg wntrol. Dciving a motof vehicie is an example in which aii tbra of the control pathways are present [70]. This accounts for a sküled opaator's ab'ito make use of preview of the road ahead and howledge of the vehicie's output caponse in generating feedforward control[71]. Aside Grom the initial delayed response to an input, this feedforward wntrol can &ectinty compensate for the operator's don delay.
Tominika and Fujimora [65] studied "extendecl signal quickeninf (ÏmcIudhg derivatives of output signal in the display) for manual wnîrol of pursuit-type tracking by incarporating mdedfiinue refcrtnce trajectory wmponents into the displayed signal. Their mode1 of the human operator was the fàmkform
where K = 5 to 10 and TN= O. 125 to 0.25.
Thq detemined that - 5 seconds of preview of the refércnce trajcctory was sutiiuent to conbol a tripleintegrai plant having a step iuput. Wthout signal quickening, human operaton were not able to controt this phusing &cf compensatory-type or pursuit- type of dispiays. WCexteadai signai qujckcning had its own merit, the &kt of preMew was dramatic. The preview essentisny dows the operator to use prccognitive conîroland&ctivdystiifftheoutputresponserheadmtime. Itsbouid benotadtbat signal quickening (or extendeci) is usefiil oniy when an explicit rdkam m-ectory to be followed, inchidhg sufiicicnt preview, W nnilable. Tbat is not exactly the case fbr the system descn'bed in this thesis., the derence traject~ryis m the operator's mind.
Hess [79] studied models of hum~pilots us& compensatory, pursuit and precognitive controL These modeis were basicaiiy cxîmions ofthe models ddoped by McRuer [70,71] (sa Figure 2.2) and desaibed typicai response charactaistics m the cotrtrol of a sccond-order piant. He devdopd a stnictunl model of an adaptivt human pilot for compauiatory-putsuit-precognÎtivcbehsvior showhg the dousprshs for d type of wnîd He considcred the effect of prrview by using r prcview display as shown m Figure 2.3. This display &OINS the nférena trajcdory as it mm fiom right to left. The objective m this case is to keep the small cirailar traçlÉins symbol on the trajectory came (ie. the Urput) by moving the symbol up and dom vertidy. Ifthe preview and postvicw portions wae not shown, the display would be a pure pursuit display. He uidicates the !otai opdoop time delay of the human ntuf0- muscuiar sysem couid theoreticaüy be completeiy elimuiated using preview. He reported that previous work has sbown that signiscant improvement in tracking using preview over nu preview occurred ciutmg the 6rst 0.4 seconds, whh marginal improvement after that.
Figure 2.3. Schematic view of pmiew display (Aftcr Hess[79]).
Hess and McNaüy [n]studied human opaaton m multiloop maaual control of systems such as helicopters. The multiloop mode1 inciuded an dfédk tirne deliy of 0.3 seconds for the hunuin opmtor. Tbe modd was usai to test respollse to raSe cornmancl, attitude command, velocity conimand ind position commaad. A square wave input cornmand si~nal(h cadt test) was displaycd in prcview tiuhion with the signai moving across the scrœn fiom right to le& Rcsuits indicated no apparent the iag baween the Gndamental components of the input ad the nsponse. This implies wmplete compensation of the hniman nrcsion time dclay with previcw. . . Boer and Kenyon 1831 studied a human opaator trackhg a 0.6 Hz bandltinned random signal with a the-vuying gap betwœn the üackhg ausor rad the initial appearancc ofthe signal. The gap was variai ktweai zero and 1 sccond using a sgpil with a bmdwidth of 0.18 Hz. It was detamined uiat hunian opaators wen capabk of predicting the hxcMy acanaiety up to about of the peciod of the higûest~encycomponentinthcsigd,ortoabout0.2~ndsfOrthispartiaihr expermrent- Other methods bave been wed to delhumiui operaton iu control systems. These include optimal control modds [5,73,80], structurai models [5,79,84], auto-regressive moving average mdels (ARMA)[66,68,78,83] and ftzty madels [5,68]. These have not beaconsiderad m this thesis. Hess 151 has comparui the crossover model, structurai mode1 and optimal control model using a parti& example. Gain and fiequency responses are aii vay sidar, particularly m the region of the crossover frequency vaiidating the various mcthods,
2.5 Summary
This chapter bas pcwcmed results ofrwearch into the problcms of time delay as it affects the remote control of robotic maniputators and the mnote driving of vchicles over a tirne-delayeci liak A mber of sohitions have bœn presented. Remote contra1 of manipulaton is not directiy appücable to the problem of runote driving but the same general problem oftime delay is fiwd. Whüe mmsolutions to the remote control of manipulators involve force end velocity feedbock, m~yof the other soiutions arc ah applicable to mnote Wg. These inchide the use of prediction, suphsory control and Smith control (which will be discwd in more detaü in the nart chaptcr). Each of these techniques offér thcir own unique rdmtages and are inciudd m the system designed for this projcct.
The modehg ofhuman operotors in control systems has been studid by various researchers and has aiso been ineiuded in this cbapter. Transfa Gnction models have been emphasized since thiS is one of the most cornmon approacbes and dm sincc a transfér Monis a good -011 ofthe fiiadamental response ofa human operator in a wntrol system, The 11cu~omusailrirnsponse is vcry basic and can be modelai quite weU Of particular mserest is the aormal rdond&y, wùich cau be cornpensateci for ifsufncicnt preview is provided AD mdicasom suggat îhaî the transfw tiinction modd ofthe human opartoruseû in tbis systeai sùould be muomb1e and that, @venthe @re)vicw ofthe road ahdof the vchicie, thezr shouid be no reaction deIay to compensate for. This assumes tbat îhe opentor is making appmpriat~ use of a predictot display. 3. COMPENSATION OF DELAY IN CONTROL SYSTEMS
The adverse && of bme delay on system sîability is coverai adequotcly in control system tucibookS. Brie& the pmblem is as foiiows. In tbe complcx domain, a tirne delay of t seconds is represented by the Laplace transfonn
Both the gain and phase margpls an obsand to be reduced by the pnsena of delay. This means tbas, ifdday is presais, the ovaJl sy- @wdd hve to k rcduced to miniah the sune mount of mqh(ie. dcgcœ ofsirbiüry) as tbere was without delay. By redueing the @, the uughdt curvt shifts to the left loweriag the crossover~~sndreducingthebtndwidthoftbesystanTfthedelayis sufliciently d that the rtJultin8 margulJ muin hrgc CMW@ for nquircd systun staùii,nofûrtkactionaadbc~Comddddiy-~cantrol~es (eg. PID control) shoufd be sahfhctory. If fbt ddry is hqp, the mar@s may bccome too small or even negative indicating instaùility. Reducing system gain andior sw bandwidth may be a solution.
Baba [85] has examincd the use of- ipproximiiSions m a simple unity- féedbacksystemcontainingafintsrdaplintdtimeddrymtbc~path.For a plua with transfer fimction û(s) = W(TI + l), the dodioop mpllsc U where K is the plant gain, t is the time delay and T is the plant time constant
The first type of approximation wed was a Taylor suies expansion,
For mal1 dues of r and sr << 1 for hquencits of iatacst, the system response becomes
From the Routh-Hurwitt &on thb system is stable iftbe denomirutM dcients are aü positive. This singie-pole appm~cmis tben stable if
Simüariy, a two-pole qpmhti011hs bem sbown [851 to be stable if
The second type of approlcirrmtion used was iPldé appf~xbufion, For positive codcimts in the denominator, the system is stable if
Baidi [85] then provides the foliowing stability criteria for Equation (3 2)based on tbe Nyquist Criterion:
Forr < For r >> T: -1 < K < [1+ (n~lry]~~. (3.12) From the stabi criteria descn'bed in Equations (3.5),(3.6),(3.9),(3.10),(3.11) and (3.12) based on a variety of methods and approiamations, Bahill[85] has indicated that large gains can be used in timedelay systems only ifthe tirne delay r is small compared to the plant the constant T. Wbile lowered values of gain may be dcieat to maimin system stabii, low gain may weil result in unacceptable transient rcsponsc bebavior. A teleoperation system, for example, uuaiiy rcquires high gain for a slave mairipuiaîor to nspond to the cornmanch hma master controiier in a timcly mamer. ûther runote-control systems may similarIy rquire high values of gain, In addition, the type of systein being considerad for ranote control in this projet wüi bave relotively short time constant for good transient rqollst. If the ~ssiondelay betwœn the local and nmote tedsis si-cant (ie. in the order of seconds), as is pmumtd in this projeet, the restriction on gain basad on the above criteria would undoubtediy rtsuit in unsatidâctory pedôrmancc- in the mûreme, a systan may be rddunwor)cob1~. 3.2 Smith Control in Remote-Control Systems design C*(s) mch tbat closed-loop control of the non-deiayed proass dynamics wris acc0mpiishe.d. Given a procas (or plant) having a tirne deiay of r secoads, Smith pposed his new controiier C*(s) to be instaiied withM the closed-toop with the the- delayed process G(s)ets as shown in Figure 3.2(a). The &ect of the Smith controiier C8(s) is to remove the time dclay fiom witbin the loop as shown in Figurc 32@) aad to nalize &&ive closeci-loop control of the nondelayed pracess dynamics G(s) using a coddcontroila C(s). It should be noted that the time de@ may be shown at the hput or output of the closed-loop. (b) Figure 32. Smith control of timadelaycd process: (a) Smith controiler C*(s) in tbe loop, ind (b) @wht systan with conventional controUa C(s). @O Figure 3.3. Smith conml of remoteantrol systan: (a) Smith controIIa C8(s) in the loop, and (b) quivalai3 system with conventional contr011er as). Substihitin8 this controiicr into Figure 3.3 d mmm&g Smith controiier blocks, the system show in Figure 3-4 is obtabd. It must b noted hi the time dday t' d nondelayed dynamics G'(s) tams in the Sniith conîroIIer of Figure 3.4 are aih!es, since these parameters arc imbedded in the pnrcess. This is the typical wd~n for the Smith CWtfOllcr [9,f O] #apt th in thb rrmofbc~ndrolcase tbae is Jday m both forward and reverse paths, so the d@ bloclr in the Smith controller codr 2t' term for the total loop delay. It is quite Qiticai m the usuaî rppficatim of the Smith controiier to provide accwote esthm 9 and GG'(s)of the piauî dtiay r ad the non-ddayed dynamics G(s), mpddy,to miabize mimath probkms [9,lO,3 Il. in tbh applidon, the plant bàiig controiied does mt have my dday so we do not d to mdei G(s) for the Smith controk. We cmn taire the output of the acbial pn~ess G(s) aud recon6gure the deiay block, ~-e-**~.Tbis rcsuits in the sirnpüoed Smith- controiied system shown m Figure 3.5 [Il]. ControUa Plant G(s) Figure 3.4. Remote-control system wing a traditioaal Srnith controller. Lod Tenuinal Deiays Remotc TenninaI ControUer Plant Figure 3 .S. Remotc~controlsystem ushg simplified Smith control. Whüe this "Smithn wntroIIer docs not nquire a delof the nondclayed plant, it stiii requires a good estimate 29 ofthe hop dday 2r. This delay mmtidy represems the sum of d transmission ddrys bezwan tbe lacal and tcnrote tenriinals. This loop delay can be meaauraî aumüdy using any of r numbcr of tccbniques within the communication system. Oepezading on the communiutiotl systan used to c~ythe feedforwafd and kbacksigniils, it is prababiy pntdent to masure the system loop delay on a continuow basis Pnd update the simplifieci Smiîh coa$oUer rccordmgly. Properties of Smith-contr011edsyskms hsve ban studied by various cesearchers over the years. Some of those resuhs are included hem. System stabiity studies with Smith eonîroliers bave identifid semiWy to modeling errors as mentioned above. Since the plant parameters must be modeled very accurately for normal applications of Smith wntrol (ie. control of plants with dead-the delay), adaptive control has been proposed to wntinuousiy adjust the estimated th delay and nondelayed dynarnics [85]. It has also ban estabLished that Smith control cannot stabilize an unstable plant [8q. The robot vebicle plant in this project is stable. For some types of plants, the performance of Smith-controUed systems can be somewhat sluggish and have poor load dishabancc rejection This has initiatd a number of modifications to the basic structure [87-931 for improved puformance. These have not been found to be necessary for this projet% From the point of view of robustness, it bas been established thaî the nomina1 stabiof a timedelayed system wing Smith control is equivalent to that of the corresponding nondelayed system [94]. Robust stability is also the same for both delayed and non-delayed systans, based on gwantœs for additive umtahty, @ut multiplicative uncertaùity and output mula'pIicative uncatainty [94]. Robustncss has not been examineci for the systern proposed fbr this project. Many of the wncerns with Smith wntrol iddedand studied abow are relevant only in the usual applications iavolving phbaving dead-tirne delay, and wherethereismismatchbetwanactuaîsystcinuxiestimwdmodciparcunaers. The remote control applidon &g the simpiiüed Smiîh controiier in this thesis oniy requires an estmiate of the loop dtloy which can be mdammîdy. The &ects of mismatch wiü bc examineci in Chapter 5. An interesting and thougût-provohg statanent reiaîing Smith consr01 with predictor dispiays and supervisory contfol waa made in [12]: "In tcrms of pure coatrol thcory, pndictive displays and 'wntimious' tdepro&nmmine could be kiyseca as rn implemdoq in the fhcof teItopedon, of th wdi-knmva Smith Pndrctor, pmposed by Smith m 1957". Teleptogriunmme (or tdapmgrammina) was disarssed in Section 2.3.4 as part of supeMsory wntrol The opdonof tht Smith-controUed remota«rnttol system can be dcscni simply using the hiiowing tirnedomain explanotion dong with Figun 3.6. The enor signal e&) is applied to the system at the local terminai and Mives at the f~mdte tenninai a€ter the transmission delay r. This delayeci input aror signai e~(t-r)baeamcs the input signal to the closed-loop system at the remote terminai A negative feedback signal s(t-t) provides the dual Wonof ensuring closed-lwp coatrol around the wntroiier c(t) and plant g(t) and providing a delayed output wbich will be wed to cancel the end-to-end closed-loop deiay. This is apparent hmthe hilowhg adysis: Lod Terminal Runote Terminai The input error transmitted to the mate tamiaal W: Note that this indudes an output féedback tam which is ddryed by two theddays (relative to r(t)). Arriving at iâc mate temhd, this siplbecoiiLes: The error signal *ch is spplied to the pîaat (and controiier) is The faedback tenn is the Smith control signal The twa deiaybd output terms y(t-3r) caacef, leaving Sit is a constant delay in ail three terms, it is superfluous and can be removed, so Equation (3.2 1) is the equation tbat would describe a nondelayed closed-lwp system. The achial output fit-t) is of course delayed tdativt to the input ao the overall quivalent system CM be represented as show in Figun 3.7(a). However, sina the input r(t) to the cld-lwp is aciuaüy delayed in a nmotacon$ol system, Figure 3.7(b) isamorcacauatenpnsanrtmd reauitsdirsalyhmEquition(3~O). 3.2.2 Hnmrn Operator Compensation To account for the human operator in the systeng the WuMon mode1 givcn by Equatiou (2.3) bas ban d in the rumtecon$ol system dtscniin this thesis. The efktis to add anothu tnnsfér fiinction block hllowing the 6rst enor detector as shown in Figure 3.8(a). During the dcvdopment of the system, it was fPund that the inclusion of the human opentor dyunmics d toUy unieeoptable pwfbrmancerrrrmeeTbis problem modt it nceessuy to ptoMdc caqmWio11. Sin# Smith scmrolanr~lisldyhdudcd~~de~i~nmcompanirte~atbc~ddr~,it sameci logical to use it for the human operotor dynimics 8s wdi. Specific reguhs of sy~tan response befbre and aftcr Smith conîrol ampendon will k disaissed m seciion 5 but it can be noted thai resultmg succes of the sysian sidations proved that ibis wnceptissound. Theusualrestncti~~~rndsenatMties(iemodelrccuncy)with (b) Figure 3.7. Smith-controUed equivaicnts of remotacontrol system: (a) time delay at system output, and (b) time delsy at system inpu! Humas operator Ddrv Coutrotka Plant W) t Y(s1 Ws) e* c(a G(s) I b (b) Figure 3.8. Smith control of mnoteamüol systan h&&g time ddayJ ad humin operator dynamics: (a) block dhgm wiih Smïîh controk, iad (b) equivalent sntem. Smith control still hold me. The same ad@s as in section 3.4.1 is applicable adis presented here in brie£ If we equate these two configutations in Figure 3.8, the Smith controiler will effectively remove not only the tirne delay but the human operator dynamics fiom within the closed lwp as shown in Figure 3.8@). EquatMg the two systems and simpüfjing &es the resuiting Smith controUer The only clifference Born Equation (3.14) is the pnsence of the human operator dynamics H(s) in the denominator. Thesefore, in wnfOrmance with Figure 3.5, the resuiting system incIuding the simplified Smith coatroller to compensate for tirne delay and human operator dynamics is as shown in Figure 3 -9. Controiier Plant Figure 3.9. Remotaconüol systan using simpbd Smith control for both timedday and human operator cornpensab*oa 3.3 System Eqiiiwlents Using Piedietors The purpose of a predictor is to provide tbe opetator with an wtima;tt of the output of the system as ifthere wae no tirne delays. ifwe nmove the time delays fiom the system shown in Figure 3.9, we are le&wiîh the nondciayed systm shown in Figure 3.10. A prcdidor PB modehg tbe non-ddayed syJtem thaî the operator would see comprises the portion of the system doscdby the dotted lines. This would nsuh in the cIosed-loop grstan shown in Figure 3.1 1. This pndictor was dcveiopai usmg a neural network mode1 as will be deJcribed in section 5.4 ad the system shown in Figure 3 11 was tested. Techm'cai es in sirnularion using MatlaWSimulinlr wge encountered so alternate but equivalent systrm configurations wcre imrcstigated and developed. Figure 3. IO. Remotecoiitrol system with no time d*, but us@ simpsimplificd Smith control to compensate for human opartor dynmk Byusingthe Smith wntrolequivaleat,weknowtbatF~.3.10 caabeshownas in Figure 3 .Il@)without the timt ddq blOCJL: (for the rnn&îqed case). in other words, we are usiag the simpii6cd Smith conboller to removejua the buman opentor dynanii~~outoftheclosed-loopintht~amemannerisitrcm~vedtht~d~outof the loop in Fî3.5. By genaating a new predictor PC to modd tbe closed-1oop Figure 3.1 1. Equivalent system to Figure 3.10 wing pndictor PB. Figure 3.12- Equivdcnt system to Figure 3.10 based on Figure 3.8(b) with no tirne delay and ushg predictor PC. The system con6guraîion that was on- donedfor the use of the predictorbytheoperatorindriving therobotvehicIeisrsdLowninFigure3.13 whae H(S) rcpresents the h~manopastois dyniiilics, q) tbe c~ntronctand G(S) reprcswts the vehicle's dynimics. Tbae is, dcaune, m need fbr Smith control in this configuration sina iî is opebloop on in cmh-end buis. The opentor would issue command U(s) which would drive prcdictorPA in ciosed-ioop control as shown. The same connnand U(s) would ab dcive the robot vehicle once the codreached the remote terminai owr thc timc &y. Boweva, it bas dmdy bcen mentiontci that technical dificultics wcm enCountercd when sirdahg tbb predictor loop. An examination of this system shows thai, at the arar sipiE(s) = 0. But, U*(s) is just the delayed command signai, or and we want the nondelayed signal U(s) to drive predictor PA to give us Y*(s), the nondelayed (ie. predicted) estimate of Y(s), or Y*(s) = Y(s) e*. (3 25) But, fiom Equation (3 .N), U(s) is the non-delayed version of U*(s), or Substituthg Y(s) for U*(s) hmEquation (3.23) rad reunnging, we bave Thedon, fiom Equations (327) and (3.29, a! steady-rrtrtc, U(s) is equal to Y*(s), the nondehyed eJsimate of Y@), Wch is what U winted hmthe predictor, so it wouid appear thai the predictor is inelevllls. This sinusion did not sam logical, and there wen problems with the simuhion as meationed, so tbc coaf~gutationwas chmged to that Jhown in Figure 3.14. The huma opsrrtor di drbi predictor in closeci-lwp but the ptedictor is now PB as in Fi3.1 1 rnd the output of tbis At stdy-staîe, the errot signal El(s) = O and, thercfon For this case, Y*(s) becames the command si@ transmitted to the mnote terminal. The delayed command signal at the remote twmùial then is At steady-state, Q(s) = O and the system output Therefore, by substitutin8 Equation (3.30) into Equation (3.29) and solvhg for Y*(s), we get indicating that F(s) is a tnie praiicted estirnote of Y(s). This cascade systun should work in a merthaî is vidyidenticai to the system shown in Figure 3.13 sincc tbe basic system tirne constant (ie. with no ddaya rad no huriuui operator in the system) is much shorter than that of the human operatois dynamics, which arc wntrolling. Just as with the modd m Figure 3.12, problems were mcountered with system simulations of this systrm, Howent, it it cui be thnt the closui-loop with pndictor PBinFigure3.14isthe~stnicnrreassbo~hFigure3.11.This,mturn,is equivaient to Figure 3-12 witb pndibor Pc. Thdbre, we can modd yct snotha configuration using predictor PC as shown m Figure 3.15. It should be noted tbat, whüe this cofiguration is open-lctop on an end-m-aid bais, îhis is an equivalent of the closed-loop system of Figure 3.14. Ptedictor Pc inciudcs feedbaclÉ, Systmi tesrs wing Figure 3-15. Ficonfiguration modd for robotvebide usiag pndidar Pc. this model should thdonserve to wrify the #luivrlent performance ofthe system model in Figure 3.14. The model m Figure 3.15 will bc wed for testing the system with the predictor in the lwp in Section 5.4.3. 4. DESCRIPTTON OF PROPOSED CONTROL SYSTEM 4.1 Introduction The application wkhhas guided the design and dcwlopment of this project is the remote control, over a time-dekyed linlr; darobot vebide which would operate in a tunnel environment constrainuî by the waiis, gmdtopography and envirom of an underground mine. The primary Monofthis machine wouid be to traverse the mine, follow the ûmdq avoid obs&clts and &appropriate tunui at intcrseaing tunnels. These actMties are to be automatcd to a high dqp. Having a inunan operator in the loop to make every adjwbnent in speed and steahg would dtin a prohiiveiy slow process due to the long time delays in comrcyine system s&tw information fiom the machine back to the opaator and in comnying control cod information fiom the operator to the macfiine. NCYQtheiess, telcoperationitelerobotic capability in reaI time is an important part oftàis systcm. A human opaator must be able to control the hrward and reverse motion and the Jteaing of the machme over the time-deiayaî link. The adeat to which dinct telcopaatiodtderoboticscan be used is obviousiy a Monof the time deiay as indi& in Table 2.1. Thaefire, tbe type of system descn'bed in thW thesis should lceommoc&îe r total loop dclay in the order of a few seconds (-10 seconds or less) fDr a machine travelling up to -10 kdh(-3 mls). This implies that the maxbmdistance allowed for tbe robot vehicle to trwd autonomously is about 30 metns. By the time the operator receives statw Ulformation fiom the remote site, the machine wiü Jnady bnn travcîied up to 15 metres hmthe location where the statu was sent. The opcntor's mdtiqconunand is tbcn based on the machine's speed, steaing augie, pasition d )ulidllin at that previous l0CItio11, Depending on the QlVifOma speed of -3 dswith a loop delay of -10 seconds should be a nasonable objective This dday sbould uxommodatc uiy runote robot vehicle operation on the earth uJins radio or optid comminiication links, mcluding geostationary satellite links. It should iIso rcconimodrte robot vchicie motion conîrol on the maon hmthe earth, ahhough hmar miahg cctMtics mry require inneuing the tirne delay bond 10 sccoads. Depaidine on tk comminncation linits, ddq~betweai Earth and Mowspacc an be qiiite Io- reportediy u long u 18 seconds [49J The generaIized architecture for the runote cotltrol of a robot veincle over a the-dehyed link showing di major dements is &OWII m Figure 4.1. The human operator, using ttie console at the local tenniaal, controis the mavernent of the roôot vehicie in its avironment over the (time) de@ via two major hterhce btctcks, the local and remote tdeautonomous controilers. The operator wes bsystem information provided by the console's display to plan, make decisions and issue commands. The operator's commands are wed in the Idtdeautonomous coatrolia @TC) to generate a predicted response of the vehicie for the operator's assistance. These commands may also be transmitted as wntrol wmmands to the remote site over a (time-delayed) communication ünk, dependin8 on configuration. This link may be a wmplex communication system either dedieated to the femotecontrol system or shared with other users. For the purposes of this thesis, it is simply a ûampamt link conveying the control and statu inknnetion bawan the two terminais at a satidàctocy pedbrmance level. Its only degrading fé9ture king coasidaed for the rernote-control system is its inherent propagation îime dday. At the ranote site, the delayed wntrol wmmands are processed in the runote teleautonomous coatroiler (RTC) wbich generates the neœswy wntrol si@ to the robot vehicle's amuton. Closed-Iwp control is provided on the vehicie for stability purposes ushg onhard sensors. These SC~WI~Salso retum status information about the vehicle d its environment to the operatois dispiay wnsole. Figure 4.1. Time-delayed taiu,t4control pysian block dhpm in addition to the contimious manuai amiml by the opetator in tdcopaation mode, semi-auto1~1mou~modes are ptovided usiq supenRsory control. The the rernote site as normaîiy infined by the tam supenkiry eontrol. In this case, it refcrs to sending appropriate cornmands to open or close switches at varbus locations in the control system. These switches enable end disable the pdcular semi- autonomous mode as required. The two scmi-autonomous modes provided with this system are tunnel-tracking and intcrscction-tuxuing. In d-tracking mode, the robot vehicle simply procads down the tunnel inâcpeuddy, awiiding obstoclw as best it can In the cvait of major difiiailty (e.g unable to avoid an obstacle), the machine will stop. The operator will then intdtto take the necessaq action. h either semi-autonomous mode, the overd1 end-tolend Ioop is retained but the operator is notsequired to drive the system. Closed-loop feu0ack control is pmided by closing Iow-level Ioops at the runote taminal in real time. This avoids any instability problans with the vehicle due to time delay. This wu a cornof one of the originai partaas holved m the definition of the system descriiaed in this thesis. For long Ioop ddays, supavWgr controi is nquind to allow the remote vehicie to operate suni-auto~~)mOuSlywhile ntawlg wntroi âom the locai terminal. This is accommodated by the muhivariable na!ure of the control system. The manuai contrai loops for both spccd and staring an ncva disabled iIthough the operator can set the sp#d to a hed vahie ond remain essarcii9ty "brnds off by l&g the st#ring control device (e.g. joystick) at zero mgle. The mcdng would be contraiied automaticaiiy to eithafoilowthctunndortumiutoui~tuwd.~~wouid automaticelly avoid obstades. In orda ta midexceab speed diaing staring manaivefs (eitha mamai or automatic), îkaystem is Wdcàgaed sa tht the speed is dependent on the staruig aagie. Byond r ninaw range uaund the zero steering angle, the speed wiü automaticaüy be nduad proportionrl to the angle of the wheels. 4.3 Local Terminal operator uses this closed-loop wntrol information to provide appropriate commands to the mnote site by effeetively driving the predictai systan in real tirne. Since the predictor is considercd to be a valid repmumion of the subsequent system without delays (ie. RTC and robot vehicle), the robot vebicle's actual motion should foUow the predicted motion r seconds later, where r is the one-way time delay. This has ban didin Chapters 1 and 3. Since the predictor is a mode1 of the RTC and robot vehicle in a pristine environment, any reaI-world anomaly (e.g. obstacle, wtieel slip) that the actuai robot vehicle experiences.cannot be predicted. Therâore, th- wiii oAcn be an enor between the predicted and actuai robot vehide's position and heading once the statu feedback is ceceiveci hmthe runote taminal. This fadbsck infi,cmation wodd be used to updatc the actuai robot vehicle's position and heading on the console display. 4.4 Remote Terminri Theheutafthercmotetamiarl,dindœdoftkmtinsystemisthenraoto teleautol~~lowCOII$D11ct (RTC) hwnin Fm4.3. The rcrl-time conÉrol ofthe robot vcùide in its emrironmaa is pafWmed by this aibsystem. Ineoming corn1 counnands~mth&locsitaniinilrredemultiplexedtoscpuatetbewipavisary conml cornpontnt Born the tdeoperation wntrol. The latter is firrther sepanitad iato its basic constitucnts - speed and steering commands. Both of thwe ~0mnriinA.Pare processcd using simpl%ed Smith controllas to compensate for the time dclay and for the human operator dynamics More king fed to the respective spad md steering control stages. Rernote Teleautonomous Controller Robot Vehicie Delayed Cornman status I status I I Signals Signals Figure 4.3. Remote tedblock diagram The main subsystan of the RTC is tbe speed rnd sleering control block This subsystem is fiirther comprisad of thra min subsystems: obsticl~voidrace,RTC speed conîml and RTC staring control as show11 m Figure 4.4. Eacti of these thne subsystcllls are in turn compcised of aother lryer of subsystems which are bascd on control. The speed anâ steuing control signais generated M the RTC are used to drive the spad and stœring acbrton on the robot vehicie. Tbe intersection co~&~l, obstacle detection and tncking blocks, whiie sbown u put of tbe RTC, ue sssumed to be part of a viddsonar sensor system considaed to be ouMe afîhe scope of this thesis. The signals pcovided by tbat systan rre the an& d range to intenectim angle and range to obstacles rnd ûmnd-tmckiq error as shown in Fw4.4. A description of a control system should start with its fiindom. This systan supports two major fiinctions - tdwpention @chihgtdembotics) aad supervWory control, The telwperation requirrmens is StniehSfoIWUd Telcopaition over tirne delay is a fundamentai objective oftbis thesis. The human opartor must have the capab~of~interveniagmthecontrolofthevebiclertd~evmiftha controi is delayed. The opartoc's (ddayed) consr01 codshow11 m Figure 4.4 arc dyerror si@ bassd on the diffacncc baween the dePired refamcd iaprt and the ddayed system output fcd badc to the local terminal. Actuai output spdand output steaing Pagle hmthe robot vehicie an also rrquired for proper fiinctiohg of the obstacle awid, speed contro1 and steering control subtystam- Figure 4.4. Speed and stœring contml block and sonar information is also required fbr obstade decection and avoidance. This is vital for a vehide being nmotely-controiled over a time delay, regardlas of whether the wntrol is --autonomous/supervisory or mamaiMeopera!ion. The vehicle must be able to respond to hazardow situations in a timely manner and take appropriate corrective action The siguals nquired to accammodate thcse various feahires as shown in Figure 4.4 are: angle to intersection, range to intersection, angle to obstacle, range to obstade aad tracking mr. Video and sonar sensor fùsionlprocessing is a major research and development area and is not addressed in this projeet. It is assumed that the range and angle to intersections and obstacles, possiily based on fiiay logic, WLU be auailabie. It is darlyassumeci that the tracking enor in the tunnel, also possi'bly bascd on îùzq logic, will be made avaiiable when required. Range and angle to obstacle is combii with the spad ami steaing angle of the vehicle to detdean "obstacle factor" and an "obstacle avoid" signal as shown in Figure 4.5. The obstacle îàctor is a mcasun of the Iilrclihaad of collision d is usai by the speed cantrol system to reduct the vehicie's specû as necessary. The obstacle avoid signal is an adjustmait to the staring control systan proportid to the likelihood oEcoUsion. The obdefactor and obsîacle avoid signais are useci by the speed and staring wntrol systems to ddy navigatc around hazarda or to simply stop the vehicie if there is uo ciear path avdble. Tbe decision-mahg ~rocessto detedethese two fisctors is based on îbzq logic. This was done primarily for two reasons: linguistic variables seern more approprirde in descdiqthe conbol patamaers than conventional "uisp" values, and it seuns appropriate to use hguistic desin determining the extent or sisnifiamce of the teailting M.To say that the range to an obstacle, for exampk, is "close" or W means more (to a humiin) than, Say, 14.23 metres or 56.18 mebcs, respectivdy. It is also derto estabIish a control scheme, since the designer can say, for example, Wthe muge to obsuele is 'close', and the angIe to obstacle is 'smaii ri@, then the hazard situation is 'hi@'". The designer aiao has wntrol ovcr the shape and distniution of the membership fiunaions tu enable sûapmg of the output coatrol db.The ~beconttolsurfas repr#eats the ensire input- outputspaçein~~ot~sbowiagtbtwtputnluefPtdp~Ie wmbiions ofthe two inputs over theif fidi univena ofdiscoutse. Figure 4.5 Obstacle avoidance block diagram The obstacle factor varies hmsafit to caution, waniing and danger. For example, let us assume that the robot vebicle is presensly hirriing right by a d amount. In fiizzy control tenninology, we can caii this steuing agie "dright". If an obstacle is detected which is somcwtiat to the left of our present hding îhis angle to obstacle can be ded"close left". Using Table 4.1, we 6nd that the Mtcrseetion of these two angies is in the third column fiom the right and third row hmthe top. This results in a "srnail leA" hazard situation This is represented in Table 42 by the second coIumn f?om the lefi. This tells us that the obstacle &or wil be classified as "caution" ifthe tirne to collision is "very short" and will be considered "&en hr any longer times to collision This second fiiay controlief couId have ban simpli6ed and given a shorter debase by eliminating twelve combinations on the bais of symmetry- This would require the hazard situation descripton to simply mnove any detence to left or right. This was considered a fàiriy low priority and was not done. The rnembership fiinctions for each input and output variable are shown in Figure 4.6. Seven membership hctions were selected based on a human operator's fidescription of angles, such as: "dead ahead", "medium leW and "hard right". Fm symmetrical triangular fiinctions uniformiy spaced across the [-1 11 universe of discourse plus a bisected triangle at each end are used to define the steering sngit and angle to obstacle. Seven membership fûnctiom also define the output brizprd situation, since this parameter can be wddesaihi by tenns such as: "hi&", "mediumn, %ow* and "zero". The symmetry was discwsed in the previous parrigraph. Four of thest an symmetxicai triangles and three are trapeu,ids. The fiinctions at cach end arc bWected trapezoids with smaü shouiders to indiate the ktthat there is liüie diange at the ends [95]. Similarfy, a trapczoidal fiinction was wed at the centre of the miverse of discourse to minimize the aect of small vaciaîions. This region reprwaits the %gh" hazard situation JO it is prudent to mnintriin an extra degret ofsafkty for the systun Sithe control surfâce in Figure 4.8 wu vrry smooth over much of it, no nining was felt neassary. The same membersbip ttnetions as used for the ùazard Wonoutput are used for the hazard situation input for tbe second fiizy contrdler as sùom in Figure 4.7. The finai input, time to wUisiou, is dchcd by two trianguiar and two bisected trapezoidal mcmbership fiuictions. The non-symmetry over the univaJe of discourse is an attempt to provide higher priority kr sborter times to coiiisi011. The obstacle factor output membership fimctioas were tuned to improvt the dupe and smoothness of the control surface shown in Figure 4.8 Table 4.1 Fuay asdadve maûix @AM) for hzard situation St#ring Angle Angie to kd Obstacie L& Far L& Hish Medium Med L& Eeht Close Small Lefi Ri* Dead Zao Ahead Ksht Close Zen, fi* Medium Zao Right Ri* Far Zen, Ri@ The control surfaces fbr bath the bazarci situation as a fiinction of angle to obstacle and staring angle and hrthe obstacie fàctor as a hction of tirne to collision and the hazard situation are shown in Figure 4.8. The obstacle factor classifial as the "danger" condition is repnseated by the sman ngion when both the fiiay hanird situation and hzqtirne to collision inputs are zero. The obstacle factor bas an gnmediate &ect on the robot vehicle's speed, reducing it to a delevel ifnecessuy. Steering amund obstacles is more ùnrolved. The obstacle factor is tiuidarn- associated with the robot vehicle's steering angle. Regardless of the nwnber and location of obstacles, the vehicle's steering wili be adjwted in the direction of improved obstacle @or whncver poss1ile. This is done by 6rst examinhg the obstacie fàctors rcIating to the steering mgies directiy adjamto fie. to the left and to the rigbt of) the airrent angle of the wheels and then by making any appropriate steering adjustnicnts, In an rctual implmmtation of tûis system, obstacle factors wodd be deterrnined for di obstacies in the forward path of the vehicle. The priniary obstacle fktor is the one having the lowest value @etween O anâ 1) of d the obstacles in fiont of the vehide and Worerepnsmtieig the higtiest daqalevel. For the systcm block diagram and simulations done for this proja it is assumeci that this has alteady bcai dcterniined. Left and nght obstacle factors an th calculateci for stdng mgles 20% above and below the cumû steering ande (2O'% has been arbiiychosen to repfCSQlS a reasoneble incnment bmthe cwrent steering angle to look for a safe parh around or by the obstacle). If the prMary obstacle fàctor is larger (and safir) theach of the adjacenî obstacle fàctorsthen no steering adjustmart wüi be made. However, if* ldt or right obstacle fàctor is higher (and du)than the primary obstacle &%or, then the robot vehicle will be steered in that direaion. This obstaclfi~voidancemethod is expected to be dequate hrall buî the most extreme situations. In such an ment îbobstacle factor would Joon generite a "dauger" condition d the vehide muid stop. For the odsemi-aut~nomous system, this is considacd acqtabIe saice the operator is rhnays dtimatdy adable. The daailcd Matlab/Simulink dguration used in the qstm simuiationr W included in AppaidiK A The RTC speed conml bl& in Figure 4.4 contains the simpii6ed Smith control subsystém and the fiuq spœd control subsystcm, as shown in Figure 4.9. The five inputs are speed commaad, range to intersection, obstacle &or, steehg angie and speed. The spoed cornniand is the control si@ sent fiom the local tennmaL The other four inputs are generated on-site at the remote tuminal. The range to intersection input is used to ensure that the speed is reduced when approaching, crossing and automatidy turning at intetsections. The obstacle borinput is used to slow or stop the robot vehicle in the presence of obstacles. The steering angle input provides for the speed to be reduced appropriately as the angle of the wheels increases to avoid any problem with skidding and to maintain good control during turns. The steeMg angIe, obstacle factor and range to intersection inputs are ail wnsidcred to have qua1 priority. Only when ail thr# permit a "f8st" condition (as defined in the fuzy controilers) will the openitor be able to activate fidi speed on the vdiicle. Any other condition (medium, slow or very slow) will result in a proportional reduction in speed. For example, ifthe obstacle fhornrches the danger kd, the speed control output goes to zero to stop the vehicle. The range to intasdon, obstacle factor d steering angle are ail uscd to restrid the spœd ofthe veûicle in lccorduice with thcir individual requirements. They rn di considaed equrl such tbrt tbt @ailu input with the minimum value wili detumine the "spifâctof. Tbe spœd consr01 philosophy is thai the human operator bas fûü contra1 over the spœd of the *de via the speed wrnmand unless the spced fiusor raptsa lower speed. The fiily speed wnîrol biock dkpmin Figure 4.9 shthe simple mdhod of selecting the minimum ducof the mage to iatcrscdoa, obsîacle fictor and stemhg angle to determint the speed Êactor. This aubsystan uses two 2-input controllers, onetogaieratetbespadEiictoranirudtb4oktogaiaatettiespad crror control signals. The inputs to the spdâctor error bladc ue speed fktor and apeed. The FAM is shown in Table 4.3. This block tm k viewed as an ermr detector with specd bang the positive input and spced haor bang tht ncgatiw. For example, iftht actuai spced of the vehicle is "nd" d the spad Wris iIso 'Lfnst" (ie. no reduction required), îhe speed Gaor arot is "zeran, mdi- no chgtin spdis required. However, ifthe spœd fâctor quima reduced qdof "ald,the spdfâctorett~ris"NMed"(~mcdium)~tbe~spad mrcontrol to slow dom Similady, if if spdwas ''slow" and spad factor was %utn,the speed faaor ermr is "PMcd" (positive medium) tdling the fuay spœd aror control to speed up. Figure 4.9. RTC spœd coatrol block d@m. Table 4.3 Fuzy associative mtrk @AM) for spœd faaor aror Vslow Zero NSd NMed NLu Slow PSd Zao NSmJl NMed Mcd PM PSdI Zao NSdi back fiom the remote tenninal ovcf the timedelayed link The speed error control block is designed to output the minimum of either the speed set =or or the speed Mor =or. This is to ensure that rcgardless of which input re~ucstsa 10- spad, the speed will be lowered accordingiy. Table 4.4 Fuay assoc&ivt nuitrix (FAM) for speed mor wntrol Speed Factor Error Speed Set Ertor NLarNMedNSmallmPSmaUPMedPLat NLar NLar NLar NLar NLar NLar NLar NLar NMed NLar NMed NMed NMed NMed NMed NMed NSrnall NLar NMed NSmall NSmall NSmall NSd NSmall Zero NLar NMed NSmall Zero Zero Zero Zcro PSmall NLar NMed NSmall Zero PSmall PSmall PSd PMed NLar NMed NSd Zero PSd PMed PMed PLar NLar NMed NSd Zero PSd PM4 Pb The memMpfiinctions for the hayspœd @or error block an shown JI Figure 4.10. The speed factor and speed inputs hPve idcnticsl mernbership functioru. Shouidmd irapezoids are again uaed at the extreme ends of the universes of disCsme to dectthe desire for minimum change m these regions [951. Spœd fâctor aror output has 6ve trianguiaf and two shouidncd tnpezoidal membership functiow. Nanawer triangles are uscd toward the centn (zero) to incnaJe the sensitMty in this region. Tbe conîrol surfiice is shown in Figure 4,11(a) rdecring is @cd natute. The membership fiuictions for the speed ertor control block are aii ideatid to the sped factor cttor membersbip îunction d are not shown. The fiil91 conttol wirfaa in Fig. 4.11@) shows the spad emir consrol rs; r fWonof spdfictor error and speed set error. Each of these inputs is san to have m identical &ect on tbe the output. Ebothinputsereatthesameldtotequestaputiculrrvrl~tofoutpit,tbaS valuewiii beaccepted; ifeitùeroneis setto r Iowalenl, ththat lowervihieof Figurc4.11. Con$olaufrccJhrspœd~l~:(a)spdfiiaorarorasa b~onofspadbrndspesd,rnd@)speederrorwmr~lisa fimction of speed set error d spccd fàc&oterror. output will be accepted. Both of thcse inputs bave identical control over the output, such that the lower value can hysreduce the qedof the vehicie. The Ma!laù/Simulink diagram wod for simuMons is uicluded in Appendix A The RTC ateerhg control block in Figure 4.4 contaùis the simpMed Smith control subsystem and the fiizy steering control subsystan blocks, as show11in Figure 4.12. The six inputs are angle to int#section, range to intersection, tracking mr, steering commaad, obstacle avoid and stecring mgle. The first three inputs are assumeci to be provided by an onboard sensor processing subsystem. The steering wmmand is the control signal sent hmthe Idterrninal. The obstacle avoid signal is the adjwtment sent fiom the obstacle-midana subsystern to msure safé navigation around obstacles. The steuing augle is the angle ofthe whaels dérenced to zero (ie. ustraight ahead") and mdcürcctiy. The tint thtoe inputs are used for the suni- autonomous modes of oeon Cintcnaction-turaing and tunnel-tracking). Whiie al1 inputs cm directly control the steering, the hiemrchy bas bccn designai with pnority in rnind as seen in Figwe 4.12. Due to the time d&y in separating the local and remote terminais and the option of usiag &autoaomous modes, oôstaclc-avoidance is the bighest priority. Nat, we want the operator to be able to intervent in the wntrol of the machine whmever requind or desind via the &g comrnand. The order of the other thfa inputs is daenmned by the scrniautonomous Mons&ed with each input. This wiii be discusscd during the description of the autosteering wntroL 4.4.4.1 Tdcoperation Mode Teleoperation mode uses direct opartor control via the steuing wmmd input or steering set aar as it is iilso daL The staring set error is the différence between the staring set desged (or set) by tbe operator at the local terminai and the actuai delayed steering angle féd Mt bmthe ranote taiMnaI over the timadelayed ünk. In teleoperation mode, only the st- set aror ud &deavoid inputs are able to control the stcering. It staoufd be notsd th the steaing set arof has a mge [-2 21 sincc it is based on the dinérence oftwo signaS each of which bas a range [-1 11. The steering set aror aud utost&q coatrol signais are wmbined wing the iÙzzy hybrid wntml bl& Autostariag cmtrolis used fbr scmi-autonomous operation The FAM for the hybrid wntrol block is shown in Table 4.5. It can be seen that, for teleoperation mode (autosteaing -01 signal is zero), the hybnd mntd output has the same linguistic variables as the stating set error input. in the some manner, if a semi-autonomous mode was active, and the steaing set mrinput WPS zero, the hybnd wntrol output would have the same linguistic variables as the autosteering control input. The fkq membership fûnctions are shown in Figure 4.13. The autosteering control and stecring set enor hctions are both identicai bipolar fiinctions having th triangles nad two bisected shouldcrcd traperoids siaiilar to those in the spadcontrol section. There is somt "bunching" of the manbetship fîmctions toward the centre for additional sensitivity around zero. The hybrid wntrol output membership fiinction is similar but the triangies are donnly spaced. The eontrol surface is shown in Figun 4-14. The hybrid coutrol output sipiis then combinai with the obstacle avoid input in the fuzzy obstacle avoid wntrol block. Iust as for the previous fiiay controUer, the FAM shown in Table 4.6 is dcsigned such that the output linguistic variables are identical to the input ifather input is m.The input and output îhy manbcrship fîmctions an identical to those of the fiiay hybrid wntrol block and are not shown. The identical control surface is likmise not bwn. Inta- Sect -0 Range to ccontrol IntCiSCCCion - Tracking Fnor SteuiuRSa, ~~1- -c Control Error SGeaiagc Avoid ControI Obstacle d.Consrd Avoid - Figure 4J2. RTC sieaiag control block dirgnin, TabIe 4.5 Fuzzy associative rnatrix (FAM) for bybrid control Medium Smdl Small Left Left zen> Ri* Small Left Large Medium Sd Zero Ldt Left Ldt Zao Small Right Medium Small Smrll Mediimi Left Zero Right Right Figure 4.14. Control surface for hybrid control as r fi,mction of autosteaing control and steuing set mr. 4.4.4.2 Semi-Autonornom Moôes The two sani-autonomow modes are piimuily concuned with automatic steering. Tu~el-trackingis SelfkpIrniaor~r.tbe robot vebicle pddm the tunnel on the bais of the approxïmatc antrdme. Tbrt is, it uses rn estmrite of the centreiine of the tunnel as the desid tcf#ena, and ury dcviaîion hmit of the vebicle's heading will be wcd u the mck@ arot f9r rnmlpuposes. in order to enable this mode, the operator would send a ntpavisory wntrol wdto close the switch which is MyfoUowing the uitostscring control block shown in Figure 4-12. It wouid then be nodfor the opemtor to rdtlde the steahg wntrol dm-a(e.g. joystick) thereby saiding a stœring colllFUd Ge. stœring set error) of- to the runote tenninnl and the stœring conîml- This tbt opentor hm having to make all of the rd-the control decisim d reapanses, but retUns the capability to intdeat uiy time sincc tkstœci~ rnmmind lwp is stüi comieaed on-line. In tumPd-tnckmg mode, the ~geto intdon W iIsa d Speed is reducad for direasolls in case of -e througit the Monas dWcussed in section 4.4.3. Table 4.6 Fuzzy associbtive matrix @AM) for obstacle avoid control Obstacle Avoid Mediumw Medium Large Law Sud Left Left Zero Large SrnaJi Ldt Zero Ri@ Large Smalt Medium Left Right Right Medium Medium Large Ldt Right Ri* Medium Sd Ldt LeA Sd Left Zero Sd Zero Ri* Table 4.7 Fuay BSSOCiative matrix @AM) for intasection control Adeto Intdon Range to Large Medium Small Srnail Medium Large Intdon Left ~~RinhtRiPbt~ Large Medium Sd Smd Medium Latge Zero Left Left Ldt Zero Ri@ Ri@ Right Srnail Sd Medium Left Zero Zero Zero Zao Zao Rigbt Figure 4.15. ControI surhcc fbr intasectioa comtrol u a fiiiaction of uigle to intersection and nage to intersection. Autosteering Control, A Intersection O 0.1 0.2 0.3 Range to Intersection, R Figure 4.17. Autosteering control scheme. 4.4.5 Robot Vehicle Dpimics The robot vehicle block dhgmm shown in Fig 4.3 contains the speed and steering iactuators. One ofthe orieiapi prrtnen imroived in tbis project, a local companywiîh~experi~~l~~with~~badre~ucsted~high- gain, pmportiod coasroNers driviag tb ictu~tonin sep~tecioscd loops be provided on the vehicle. This became part ofthe dmip ad wuntiined through aii ofits development which was done using bWiaô6- Ui the eariy siaga of îhe wecî, this portion of the gpeed control qs&m inchid4d an rmpîi6er of gain 100 rnd a first- order plant having a time constant of 1 seumd in a chuî loop. The stemhg control system portion ah used an amplifia of @a 100 kit the fint-orda plrra bad r time constant of 0.25 seconds. These pLnt modds wcre ubiarrüy choscn on the basis of thetimeconstams,since IseewiddO.2SsccaaddtobegoodrepresaiuIive vaiuw for the spœd and steaing rrspoase oftbc rchul nbide tht was intaided to be uscd in field testing of the systcm ddqedin this thefis. Fi-order plras modds seemed nasouable for the vehicie. Tbhpiied thmî the tirne to nach stady-state would be - 4-5 seconds for the speed uid - 1-125 ssc~nQfÔr the s&&g nspomes, rqtaidy. Actuai tests oftht test vehicle wat never donc sincc it wcis never wmpIeted While dydevdopnient tuhg tbe drst~rderph modds progressed ~riiywithQtOCnentpaîormiacererPhqit~aiegestedtiutwingahigba- order plant d might be r beücr ehoia ud r W test ofthe sysâem. Even though the primary objective ofthis project was to detumine a solution to overcome problans due to the delay in the remotcantrol systan, this suaestion was acccpteû and van'ow highaadu plant modeIs were testai. The criteria wed in the soldon of suitable models were thrafold: 1. The modeis should be of (at least) second-orda. 2. The damping co&cient ( should be - 0.5 (or grtater) for good nsponse with minimal overshoot. 3. The cutoff fiequency should be at least 10 times that of a human operator so tbat in teltoperation mode, the dynamics ofthe operator would be controiiing. This would mean a cutoff ftequency ofat least 5 Hz (or 30 ris). It should be noted that the nason for this critaion is to awn acaincy of the systan output whcn usllig the predictor. The ptedictor design used m this system provides only the steady-state output of the system, A predictor that can lccommodate the transient nsponse of the system for any speedlsteering agie initiai condition would not need this restriction. basic secondsrder configuration bavuig a proportional gain of 100 to conforni to the original design. The steering model sdeded wss a first-order plant in suies with a proportionai gain of 100 and an integrotor. The wparate mprpmvides the (anguiar) displacenient output for a (anguIar) veiocity input. Both of- modeis in unity-feedback Iwps provide ~ec~nd-ordercld-lwp rqnses. Many models for speed and steering dymmics wcrc inwsb'gaîted The two open-iwp models seleded that met the Cnteria were: 100 Speed dynamics model, G&) = s2 +65s+2 100 Steering dynamics model, Gs(s) = s(s + 100) The speed and steering control systems were testai with their respedive vehicle dynamics models given in Equations 4.1 and 4.2 using th& primary control loops. The prirnary wntrol loop for the speed coatrol system is shown in Figure 4.18. The gend structure of the stemhg control system is identicai, with the exception of tbe mput gain, whicb is unity, and the vehicle dynamics wtiich is G&). The Maîiab/Simulink system used m the simulasions ore shown in Appendor A These sinrulaiions include di system eiumts except tbe deiays and the buman operator dynamics. The spd d steering control systan responscs ushg these models were daermined using Maîhb uid are presented in Figures 4.19 and 4.20, nspectivelynspectively These arc based on hearized -te-space models of the systems, as determMed by Matîab, bmwhich tbe Bode diagrams were pioücd. From these plats, we can sa that the dsrmping we&ht Figure 4.18. Speed wntrol systan paimuy coa~rolloap major ph and dyasmicsMocb. 0.5 since there is a siight overshoot fbr the Jtaring rcspoase, whüe for the spad response, it IooIrs to be more like 1; = 0.6. The cutoff fiequeacy is seen to be - 52 rfs (8.3 Hz) and - 90 rfs (14.3 Hz) for spad ad stcahg mpddy,susugsestiag tht, m general, the criteria have been met. Figure 4.20. System stcuing nsponse with no deiay or tnunsn opaator dynamics. Modeis of automobile spœd dynamics indiate tbat, wbile higher-orders are ptesent, first-order dynamics are domininans [W.Gillespie [971 shows that a vehicle's acceleration is essentiaiiy determined by tht ratio of agine power to vehicle weight or where F is the tmctive force at the drive wheels m is the mass of the vebicle g is the gravidonal constant, 9.8 m& W is the weight of the vehicle HPistheenginehorsepowcr V is the forward spad. The velocity terni in îhe denominator quiruthe acceidon to decrease with increasing sped. Since velocity is the integrai ofthc accddon, it can be dy shown thai the vdocity will be of the fom whaeV,isthemrilrirmnnspeed T is the time-constant. This implies a first-ordu specd respoiwe. A simüar aiiafysis by Woag [98] shows that the speed response of a passenga car foiiows r similac expoiiential rise to fiill speed. anaiysis of a simplistic model by EllW [100] considers osciüatory stem behavior using a second-order modeL The optimum transient response is the fastest response with minimum oscillation [99]. A mon indepth m@is of the vehicie's sîcuing response is not witbin the scope of this thesis, nor has a partesiscuIardesign beai specified. On the basis of the foregoing discwsion, it sams nasonable to we a steering modei restricted to secondsrder. Furthermore, sina oscillation or even overshoot in steering is hi- undesirable, the mode1 must have sufnaent dampin& ifwe examine the closed-loop responses of the unity feedback Jpeed and steering dynamics models, it is apparent that each is strongly overdarnped. This was found to be necessary for overail system stabiiity. While the resuits are not included here, the step-fûnction responses cmbe apprownated by a Grst-order qstun based on the dominant pole of each model. From the step nspo115e9, the timcconstants of the closed-loop, unity-fecdback spetd and stccrùlg dynamics models are approxhtely 0.75 seconda and 1.1 seconds, rwpectiveiy. These vahies are vgr close to the respective time constants, 0.625 seconds and 0.99 seconds, due to the dominant pies. 4.5 Speed and Steering Control System Dirgrrims The subsystern block dkpnschematic -on ofthe remotbcontrol system as descn'bcd in the forego& donsU not the most codent for appreciating the structure hman overall wnttol system point of view. For this reason, the complete spœd and steering control systcm diagmm rn Wuded in Figures 4.21 and 4.26, respectiveiy. The Theoop, multivariable natun ofthe systems is apparent. These systern diagrams show eacâ conîrol system based on thar prirnary control loops. For acample, the obstrclc-avoidauce suôsystem is not Wedhm. In order to relate these diagrams to the fiinctid subsystcm blocb deJcribed in the preceâing paragraphs, the parîicuiar elanaits have ken grouped inso the same subsystem blocks and shown in Figuns 4.22 [email protected] and Figures 4.27 throua 4.30 for each system, respectivdyrespectivdy Only the major blocks are inchided. These subsystems have ban descn'bed previousiy m bischpter. Both the block diagnms and beir MatlaWSi equivalent arc inciuded in each figure ta twist the readcr who may not be familiar with Sumilinlr nomcnchm m making compuisolls. Some of tbcse dbgrams an alm induded m the coqlete MaîiatdSicodpaii01~ diagrams m Ap@ A - Speed + Human Time Set L9-C- 0perator - hlay Robot SPeed \ Range to Inteniection -4 RTC - Vehicle Obstacle Factor . Speed Steering An~le Control , Timc May SCALlNd DEUYS RTC SPEED WN HUMAN SPEED OPERATOR CONTROL ROBOT VEHICLE SPEEO 1 -- Figure 4.22. Speeâ conirol system: (a) subqstem diagram, and @) Simulink diagram. Soead = t Command Smith Control - I .. Range to b Tnt-cin Fuzy Obstacle spad Spad Factor ad Consr01 controi) Steering m. Ade Figure 4.23. RTC speed control: (a) block ad (b) Siaailmk diagram, Obstacle SteeM Figure 4.24. Fury rpad Cornmi: (a) block diypui5 d (b) simihd:dLgim. speed Speed command Sa * Figure 4.25. Smith wntroI: (a) bIock dhgam, ad 0)Sinnilrmt diagram. Figure 4.26. Stdgcontrol system block diagram. Figure 4.29. RTC steering wntr01: (a) block and (b) Smnrlink diagram. Angle to \ Tntednn Inter- sect. Range to + ;Control Intersection ht+ Trackin~ lhlr Hybrid Steering Set Contr01 - Obn. . Enor -P Avoid Obstacle --coatrol Avoid (4 INTERSECTION @ 1 O a------, The operator typicaily uses pursuit-mode traaMg when wing a pursuit-type display, ahhough there is no guaranta of this [79]. It is possible that the operator will visualize the problem in terms of minimizing tâe enor, thereby adopting compensatory tracking. This is not likely to be a major pmblem since a human operator wüI tend to do whatever is required to perform the tracking task by adopting any necessary lead or lag equalization [63]. This adds to the dEculty of modeiiug human operators. While the human operator may be able to compensate for his/her neurophysiological limitations given diciait preview, there an more coniplicating considerations The operatoc's control output is the output of the error detector, ie: the difference between the reference input and the system fadback signal. Given sufncient preview of the rderence signal and the current feedback signal, it appears that the trained human operator can compensate for the inhercnt nanophysiological limitations (primarily reaction delay) to effèctiveiy allow an "ideel" error-detection Mon[TV. The main problem m this case is that the fkuiback sipiis delayed. It is this delay which causes operaton so much problem, leading to 105s of coatrol (ie. instab'i) with rehtively short dciays (4 s) [l]. By pmvidiug simplificd Smith coatroi, howcvct, the &eet of the timc delay can be compcnsatcd ifthe opaitor am provide idd arc^ detection, It is the contention m this thesis that, givea sufncient previcw of the niad ahead, the operator can campensate for Ulbaent hunan rerrction delay and provide the required mer detection. Another scenario involvcs the use of a system pndictor and pndictor display. W~tha predicted system output appachg on the saan showing the output ofthe systern without time deiay, the operator now bas a non-deiayed "féedbeck" signal to use in the nrordaection fiinction, The resulting "non-ddaycdn error wiü now be different than when using the dclayed systan féedbsck. Thdore, this si@ can not be usod as the command signal to the runote tamina dess we opaate in opdoop mode on an end-tdbash aud disable the simplrned Smith controiia. It is an option. This "nondciayuP' etror si@ is d to dtive the pndicted mode1 wbicb providesthenonddayedoutpit. TheoutputofthWmoddcanrlsobeustdtodnve the robot vehide in opIoopon an end-tbend basis. ïhis is motha option. These optiotlswazdi~~~~din~3andwiilbemmidwiththe~onsd resuhs pnsented in Chapta 5. Wbile tbW "opes1oop" contml oftbe robot nhidewill be shown to perform in an ordent mamia, it must be tcco~that fadback to the local terminal is necessaq to at least update the dispiay contiuuousiy. Supervisory, or semi-autonomous control operation is accornmodated in the system by the provision ofa numbcr ofswitches throughout the systcm designecl to be activated by the opaator hmthe lacai terminal. In this case, the operator can exit the contra1 loop by releasing a switch on the control device and just sit back and monitor the system. This switch wouid &&My bypass the opaator and insert an admr detector in the Idterminal and ahptovide a bypass of the human dynamics blocks in the simplifieci Smith controiicfs at the remotc tuminal. The end-td closed-loop control would stiU be maintained wllig the Smith control to compensate for the time delay. The opaator couid stdi intenmre in the systan by rctivating the ssme switch on the control device to activate telaperation mode ud dinctiy enter speed and staring set commds. Besides inserting the opaiitor back into the loop, the switch would provide a supervisory control command to reinsert tâe human dynamics compensation in the simplified Smith wntroilers. 5. SYSTEM SIMULATiONS AND RESULTS 5.1 Simulation Environment Ail simdatiolls were made using The Matbworks, Inc. so&e package Matlab, simdation package Simulink and their various twlboxes as follows: Matlab v. 5 Shuli v. 2 Fuay Logic Toolbox v. 2 NdNetwob Toolbox v. 2 The system simulations were nm using the general system dguraiiom shown in Appendix A and B. Some runs were made on simplified versions for wnvenience, such as removing the system delays for noedelayed system tests. Ail speed set inputs and speed outputs are nonarilized to the range [O 11; steering set inputs and stecring ongle outputs are nomaiid to the nnge [-1 11, mapt in instances where the singlasidal range [O 1] is useci, simply due to l&-ri@ symmetry. Allspeedandsteaingsetstepinputsueunitstepfinctions,e~aptwhere noted; speed set sinusoida1 inputs ùaw a pcak-to-pedt amplitude of un& and ocaipy the range [O 11; stœring set Jinwoidal inputs hvea peak amplitude ofunity and occupy the range [-1 11. 5.2 System Performance with No Timt Delay in tbe Loup 5.2.1 No Human Operator in the Inop The basic criterion f9r appraising system performance onr the tirne delay is a comparison with the system pdbrmance with no delays in the system adno tniinan operator present. The step-bciion response of the stœ@ uigle to Jt#riag set inputs hmO. 1 to 1 in increments of 0.1 is hwnm Figure 5.1- This can be considead as turaing to the right. Ncgdvc inputs fb left ainu dlgive the sme fesponse using even symmetry and is not sbown. The linearity of the steady-s!atc Figure 5.1. Steering angle nspomc to sregfimction scehg set inputs with no time delay md no humm opaitor in the systcm. One oftbc main feràins oftbs conirol system design is the dependence of the speed on the steering qie. This brs ban deraiasd in seaion 4.4.3. Fipce 53 shows this dependcncy cle~rlyfor maximum ipad u a fùnction of steering angle (using fixed increments of O. 1). The gnph Stiows th tht spœd can runain rt maximum until the aa@e tticbes - +/- 0.25. A! ths point, tbc rrmximum spced begins to deatue. Tbreduetion in spdwith stdng WCis nlitinly Iiear until it ruches its minimum vilue of - 0.32 at the maximum @ositivt or ntgative) steeting angle. This is to ensure tba shrp ue made a rpptopriately despceds. 12 - Steering Angle - (Normalized) 03 - 0.3 0.4 0.5 - 0.6 0.7 - 0.8 0.9 1.0 035 (9) Steaing Angle (Normalizcd) 1 (ie. fiil1 speed), the &ect of the staring angle on the speed is lessened. Figure 5.3(b) shows the response of a tep-fundion spdset oommond of 0.5, hdicating dependency on the steering angle only at steuing angles greater than 0.8. The dependencc of the speed on the steering angle in a dynamic environment is iIlustrated in Figure 5.4. While this appears to be derund at htsi& it is just intended to show the required dependtace ofthe spad on the steering angIe. These two examples illustrate the sQeed and staring angle responses when botb the speed set and steering set inputs are varied hmminimum to maximum siausoidally. This would be an exampIe of ex&rune opaating condition. The hgmcies of both signals bave been chosen to be sufIicientIy low to not incur any demase in amplitudes duc to fiequency response efKects. The desind amplitudes, in the absence of@ dependency on steering angle, wodd occupy the fidi nodiranges of[O I] and [-1 I] for speed and steering angie, respectively. The depeadency, however, requires that the speed be decreased with inaeased steefing angle, and this is shown, The specific value tbat the speed is reduœd to is not important, oniy that it be redud sufficiently to dyaccomplish the vehicle's tum. Iî is int#csting to envision the resulting motion of the robot vehicle hmthese eums, showing the spdbeing reduced as the steering angie increasw in both the negative and positive directioas, and reaching higb duwonly when the steaing angle W at rdativtIy lm values. 5.23 Humrn Opurtor in the Loop Figure 5.4. Spad and steerbg augie rqotwcs ta aimisoiciai inputs of speed set with amphde range [O 11 aud to stariag set with uapiitude range 1-1 11: (a) Jpadsetat OîHzind SC- setat0.1&, iad(b)speedsetrtO.l capability to cancel the reaction delay. For this masoh reaction delay has not bcen included in the system simulations even bugùit is rqgy to ampensate for in the sirnplified Smith wntroller. The MatlaWSimulink models of the human operator in the Appendix include the reaction delays but they have bem switched out in the simulations. Initial results of system responae tests with the mode1 of the human operator in the loop but without wmpeasation wen disasimus. Response was ragged and tdy unacceptable. Since simplified Smith control was to be included in the system to compensate for time delays, this concept was W to oyttcorne the problcm cause. by the human operator dynamics. The technique proved to be totally successfiii, By including the same dynamics in the Smith eontrol loop as the loop delay, wmplete compensation was accomplished. The simplified Smith coatrol sysîem wed in this project includes a mode1 of the human (opaator) dynamics. Speed and steering angle rtsponses bave been npeated with the human operator in the loap and human dynamics compeasation in the simplifieci Smith conttollu. Resuits are shown in Figures 5.5 ad 5.6, respactivcly. The spœd rwponse is very similar to the results sbown in Fip53(a). The 04tnnsient nsponse is now controlled by the ncuromuscular time constant of the hurnan operator, raki4 approxirnatcly 1.5 seconds to reach steady jtcite vs. about 0.2 seconds for the case without the operator in the Imp. Tbt speed/sîming angle dependcncy is dl apparent as is the ovcrshoot. Likewise, tbe s&euingangîe ns~oaseis similu to tbat in Fipn 5.1 except for the longer time to nech atauiy-sbtc. Sincs the time constant of the human operator is controiiing, the tronsieat resp~nsesue essentiaiiy identicai for both systems. Itisapparmsthtthaeismdelctaiouse&atottic~usiagabuman opemot 0th- than a change of tirne conslant if the dynamics are compeaMted for, as they are by wing the simplified Smith controller. 5.3 System Petformance with Tirne Dehy in the Loop 5.3.1 No Humrn Opcritoc in the Loop 0.4 - 0.5 0.6 - 0.8 0.9 - 1.O (Normrlized) - - I 1 I O 0.5 1 1b 2 25 rie(s) Figure 5.5. Speed tesponse to imit step-fhction Jpad set inputs fiir wious vJues of steuing angle with buman opmator but no the dclay in the system. 42I , , I I I , p l , 1 O 1 2 3 4 S 6 7 8 0 10 Tome(s) responses of both the specd and staring control systems, nspeaively, wbm a delay of 1 second is introduced into the system with no Smith control compensation. (Al1 derences to time delays in the rmotaconîrol systcm refér ta the owway systcm delay in both forwani and mmse pPths, unless total loop delay is specified). The speed response in Figure 5.7(a) seuns fine unîil the 3 second point. This is when *he output signal which is fed back to the input appears at the output again. From here on, the response deteriorates. The desired output (normalized) value is unity. Figure 5.7(b) shows the stœring augic nsponse xising rapidiy beyond the dtsired output (normaüzed) maximum value of 1. nie simulation was stoqpcd manullly at 1.5 secoads busethe MatlablSimulink simulation began issuing wamings due to going out of range. Figures 5.8 is iaciuded to show spad and steering angle responsa to sinusoidai inputs at 0.1 Hz. nie actuai ~wpons~sshow liale evidence of the expected sinumicial output rcspma. Ona again, the simulations were stopped manualiy wtitn warnings began appeving on the serrai. Clcarly, tune delay is a totally aîasiqhic problem for systms likc this, making mpaisatiori absoluteiy -datary- By including the simplified Smith controuer in a191of the speed d stecring contml systmis, time delay wmpensrtion is provided. Figure 5.9 3howa the step Montcsponses of both spad d s&er@ angle to8dfor timt delays of O seconds and 2 seconds, mpechdy. Due to technid dif&ubes ellCOIMtd with simulations wtien ushg bhtWSa non-zero dday of 0.001 s wu fwnd to be necessary in the system dehy blocks fk the case ofna dryin the system. This has negiigile effect on tbe results. It W obvious bmthese two plots m spite of the time delay, the cespollses are idutîid Tbe non-delryed spcad nspoast is identicil to the response shown in Figure 5.3 wiîh the différent time scaie rrd cldyillusbates the dependency of tbe pedon stcaiag uigle fi,r the murimum nofinilized angle of (+/O)unity. The sinusoida1 nsponses wmiIso testai ushg the aame valus of time delay. The re~ponseof the spœd conbrol system to spœd set input of 1 Hz is shown in Figure 5-10. These plots sbow thrt the mpoms ue dso idential eKapt fix the tirne delay. Similady, the nspoiw ofthe st#rias consr01 systeni to stemiq set inputs of 1 Hz shown in Figure 5.1 1 is undmgai reg8rdlcss of the time May. Tbe horizontal lines at an amplitude of zero in Figuns 5,11(a) rnd @) W the spœd response to th spad set input ofzao. Thme piots dl vcrify t& exœlicnt pafamrana oftk Smith coatrol schcme to ampm~~f8f time &y in ~~~~)tacolltr~lsptems. How wldl thiswiUworLwhma~opatcorisinehidedintbtloopwiUbeaaminednext. -1 2 1 I l 1 I O 2 4 6 8 10 12 Tie(s) (a) 1O 1 0 - 1':7 a- 2e s- - P 4- Di3- 3j 2- 1 - O 1 1 O 0.5 1 1.5 2 25 3 3.5 Tiie (s) (b) 1 0.9 0.8 - 0.7 g 0.6 % 0.5 O 1 0.4 a:.I- 0.1 O O 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Tic(s) FigmlJ. Speedrdntng~e~tounitstepspd~ardaraig~ inpm~mhimin~(i)arotmwdel.y,md@)Z~cc~ndbimt delay with timedehy compeamh O 1 2 3 4 5 6 Time (s) 5.33 Human Opentor in the Loop Before investigating system perfiormana includiag the compensation of human operatoc dynaniics, it is important to ansider the case of having no wmpensaîion The unit step-input respollsts of bath the speed and staring control systems with a time delay of 1 second are shown in Figure 5.12. The ideal (or desirai) response for each is a unit step. The aeuromwcular time constant of the hunian operator model is evident in the transient response. The spad response airve looks excelIent until the 3 second point, similar to Figure 5.7. However, just as in that case, the response deteriorates after 3 seconds. It does remain essentially a! the right level but it is very irregular. The steering angle response is dramatically imgular. We can see the effect of the féedback as it repeats every 2 seconds (ie, the loop dehy time). It shouId be noted that this response was allowed to finish despite MatlaWSimulink wanùngs der the 3 second mark. This is obviowly not acceptable performance. It is time to include the compensation. As soon as the model of the human dynamics is included in the Smith wntroller the responses clear up. Step-hction responses of both speed and steering angle are shown in Figure 5.13 for time delays of O seconds and 2 seconds. These tesQonses show the dependency of spœd on steuing uigle in exactly the ~omemamier and the same steady-state ducs as shown in Figun 5.9. The only différence is the time scale due to the human operator's dynamics. Figure 5.13(a) is the same tesponse as shown in Figures 5.5 and 5.6 for the case when both spd set and staring set are unit stepfunctions. The spad and steering angle nsponses an again seen to be identicai to one anothcr despite any tirne delay in the system. Responses to sinusoida1 inputs are shown in Figures 5-14 and 5.15 for the speed and stecring angle, respectively, with tirne delays of O seconds and 2 seconds. Since the iddspeed and steering angle rcsponses would occupy the full normalid canges of [O 11 rnd 1-1 11, respectively, a smail amount of attdondue to hquency nspom dktscan be seen in the actual responses. This is due to the human operator's dynemia. Agah, the responses are identicai hmont to motber fbr each paramda. FdIy, in Figures 5.16 and 5.17, spad and stecring angle rwpollsts were plotted wdyto show the speed dependency on the stcuing angit by using siausoidai spdset and steering set inputs. Once again, as in Figure 5.4, tht spd is setn to be tcduced to a safk valut as the steering angie inaeases in each dmctioa The smali amount of aümuîion duc to hqucncy response effects can be sœn qahin the responses as in Figures 5.14 and S. 15. Two ambiaasioas of mencies wae d:spad set ad stcaing set inputs cachat0.15EIiandtbenwithspeedsartO.l&~~~rt0.15&.These two tests were plotud at time delays of zero seconds and 5 mmds in Figure 5-16 ad zero seconds end 2 seconds in Figure 5.17. Once agaia, the time delay is seen to hve no &ect on the systcm cespases. Ail simulations involving time &ay include deiay compensation ushg Smith control unless othawise indicated. 5.4 System Performance Using Predictor The predictor is âesigaed to provide the human operator with an immediate estimate of the output of the systern during teleoperation mode as if there was no tirne delay in the loap. This should dow the operator to drive the rabot vehicle in simulation while the same coircrol signals are wmitted to the remte terminal to drive the acaial vehicle. (AlternativeIy, the output of the predictor may be used as control signals to the nmote termina1.a~discwsed in Chapter 3.) Obvioudy, this process wiil woik paféctly only with perfeet modelmg of the vehicle in its enviroment. Neverthtless, the wefulness ofpredictors in tirne-delay control applications ha0 been proven beyond quesbesbonas didin Cbapta 2. The predidor used in tfüs projeet W a 2-inpit, 2outp1saarral network The inputs are speed and steering cornmand signais (ie. speed set d st#rieg set); the outpuis are the predided speed iind steering angle of the robot vehicle. Ideally, the predictor should behcxady as the rd (non-ddayad) sysien~npresaiting the inputoutput mappiag pdkdy, rqardcss of initial amditions or dination of input vahes and including the complete transient response to steady state This was the goal for the nairal network This iddwrs not di.A&r a wnsidcrable e&rt of testing vananousnwal nctwark mod& using varioru configrrriitioiy îhc transient bchavior of the pndictor cwld mt be nalized. This promptcd motha approach Since the time constants of this systeni arc cont-miiedby the dynamics of the human operator, it was decided to modd the neural networlr using the steady-strte conditions of the sys&em. It was slnady bwnthot training tbe neuroI imetwork in this capady was fiasiMe. Tests would pmthe sodmsofthis conCCPt, 12- 1 I I I I 1 i I i 1 - h f 0.8 * - -.- 0.6 - - O - 202 1 - O i 1 I 1 1 I I t I I -0 1 2 3 4 5 6 7 8 9 10 Time (8) Tie(s) 0.9 - I02 : 0.1 * 0 I L I I A O 1 2 3 4 5 6 Tune (s) Fip5.13. Rcrpnrvrofrpodmd~~emima~~ads rie(s) Figurt5.17. Respoll~~~ofspeedandstcaingrngietoSnusoidaispeed#rinputof 0.1Ilzwitb~z;mge@1]~tD~~sbaaing~etinpitof 0.15 Hz with am@itu& muge [-1 11 with humau aptr~oorin tbc lap tim delay. Input pattem and output targct data for training the ndnetwodc were generated using the actuai system with input speai set and steaing set mmmands over their full ranges of [O 11 and [-1 11, respectively, at iucremcremcnts of O. 1. This mhed in a speed/stee~gangle matrix of [ll x 211 or 23 1 points forming the base of the 3- dimensional control surfiice. The neural networlr provides the interpolation (or generdization) between the output data points to create a continuous control surfkce The predictor is a mode1 of the complete input-output system. As test nsults ofthe predictor dlveri@, these points are sufiicient to train the network to a high degree of accuracy over the entire surface. A feed-forward backpropagation network was chosen using an 8,8,2 neuron, firllyannected, three-layer contiguration. Matiab's "n& simulation soffwarc t?om The Mathworks' NdNetwork Toolbox was wed. The fÙst two hyns uscd tan-sigrnoid activation fûnctions with the third iayer king finear. The network was üahd for 1000 epochs and the accuracy attained was a mean-squafed-error of 1.26 x 10-6. The rrrniiting sped and steuing mgit targct rrrponses are shown in Figwa 5.18 and 5.19, respectively, compared with thtir training data pattem. The match is vimially @a. Two system configurations usîng the predictor wae tested. In the fh,or basic coa6guraîion, as shown in Appendix A, the predictor was driva in @el with the acaial system by the spad set and steuing set inputs. Modds of hunwi operator dynamica were d to provide the ncccmry transient system mpoms since the pradictor is based on steady-state &es. The second, or ahmate COIIfjguntion, shown in Appendix B, is based on Figure 3.15 in which the output of ihc predietor was used in series to drive the actuai robot vehicle. This has ban didin Section 33. Appendix B includes only those portions of the system which are changed hmthe system in Appendix A The scaling gain of 0.4 fbr the speed set input Y not tequired in the local lerminal since its effihas ban included in the iiftarl network d. The predicted output duehm the delW tben CO- Hawevat when ushg the predictor to drive the mnote taminai, the gain is stiii nquind in the jpesd set cornanand path, since the pralicted output is naw effcccively rn @ut to the aiginai. system, which requires the gain. For cadence, it is included in the black diagnm ofthe remote teleautonomous controiiu. This conCimon is open-loop on an euci- to-end bis, so Smith conüol is mt required. 1- @ @3- 8 0.6- O O 4 b 000t' a7- .-T DI- - go004 1 000004 OS- e *00000 c 1"- P O oaooo > III OI- O 1 O QI- O O U- 0.4- Qt- O - of 0.1: (a) rctutl systcm, rnd (b) pndictof. 5.4.2 No Mctorin the Loop The predictor was instdled in the local tel~mmouscontrolla as shown in Appendix A for the first (ie. parallel) configuraiion and system nsponse tests were perfonned. Step input rwponse for speed is showu in Figure 5.20 for the cases of zero delay and 5 seconds delay. The expanded defor the zero delay response shows the wry small discrepancy between the adsystem response and the predictd This appears to be less tban 0.05 seconds, obviously of no concern at dl in a practical tel:coperation system. This discrepancy is due to the transient response of the actuai system; the predicted response is based on steady-staîe outputs only. Both actual and predicted responses include the same buman operstor dynamics. For the case of 5 seconds delay, the compdscale dmnot show the 0.05 second diierence. There is oniy a slightly discemiMe disctep811cy as the response nears the steady-state value. Figure 5.21 shows the step input respo119t for sieering angle for zero delay and 5 second delay. Again, ody the expanded defor the zen, delay response shows the small discnpancy of less than 0.05 seconds between the ad system response and the pndicted. At 5 seconds delay, thae is no discernible diierence. These results verify the exceptional accuricy of the predictor. Sinwoidal responses wmalso plotted at zero dâay and 5 secand delay for both the qxai and stCçnng angle. These arc shown in Figures 5.22 rad 5.23, respectively. Once again, the nceuracy of the prcdictor dsody a Jlight différence between the adand predicted nsponscs for the spcui. Tbere is no discernible difîèrence between the actuai and pndicted mponses fiir the sîe@ me, at ktat thescaleshown, Two~testswerenmtottatbedepeaaencyofthespadonthe steering angle using the predictor with r 5 second delay in the ~ystemFigun 524(a) shows the results ofthe spad response when ôoth spœd set aud steering set inputs use a 0.15 Hz sine wave. Fi- 5.24@) shows the jame nspoiwe Wng 0.1 Hz for speed setand0.15Hzforstœringset, Again,the~cybetw#ndsystmaad prcdicted is barely disaraible- - - - - d t I 1 1.S 2 25 (s) 1 - I 0.8 - - 0.8 - - f 0.7 - - - e O>- 0.4 - w O2- - 0.1 - - 1 I I l I 1 t O0 2 4 6 8 10 12 14 16 mmt (s) O 0.5 1 1.S 2 2.5 Tie (s) 1 0.9 - 8 0.8 - 10.7- Ose - - ! :: - -10.3 - 02- d 0.1 - - O 1 1 1 I 1 2 I O 2 4 6 8 10 12 14 16 rime (s) Figure 521. Staringaugiere~po~wtmunit* mputfbraduai sys~emand pdictak (a) zero tirne &y, rnd @) 5 secand time ddry. Time (s) (8) Figure 5.22. Spad nsponse to 0.15 EL hiddinput ofunplmide range [O 11 foraad sysfcmandpdictal:(~)zaotimcdelay,md@) 5 second tim+ delay. 4.1 1 I 1 1 1 I 1 I 1 1 O 2 4 6 8 10 12 14 16 18 20 The (s) (1) Anotha set of simulations was run wing the second (ie.series) configuration as shown in Appendix B with the opmûr driving the pdictor and the output of the predictor dmnng the robot vehicle. Since the test results mimr those of the previous configuration vidyidenticaüy, oniy a srnaIl number of dtsfor the case of 2 second tirne delay are included here. These inchde stepfiuiction and sinusoiciai responses in Figures 525,526 and 527, The slight dismpancies in the step responses of Figures 5.25 and 5.26 are due to the actuai system traasient responses as was descnied in Section 5.4.2. There is also a sdldrop in the steady-state amplitude of the steering angle step response, pbably a frequency response effect. Otherwise, the performance of the system using this configuration is obviously excellent, validating both the accuracy of the predictor and its use in driving the remote system. 5.5 System Performance with Mismatches The excellent pdomunce nsults descriied in the previow sections are based on a number of spacific assumptions: - the loop tirne delay udin tbe simplitied Smith controiicr is the aract time delay in the lwp - the mode1 of the human dynamica used m the Jimplied Smith controUer is a perfect repiica of the hanopemds dywnics - the mode1 of the Iniman dyaamia used in the predidor path is iperféct replia of îhe buman opentor's dynamics - in teleopemion de,the human operator acts as an idcal emr deiector 0.1 Y I , I I 1 O 1 2 3 4 5 8 rune (s) It has ben stated eariier that the loop time delay should be relativcly easy to masure with great acairacy, so the sirnplified Smith controiia Jhauid be valid. Neverthalas, mismatch in the (estimirted) tirne dday is a known sipificant problem in traditional Smith control so this will be examinai. Modeling a human opaator's response dynamics is a well-known problem. The basic model used in this project represents the neuromuscular rcsponse characteristics of a typical hyman operator. b should be a good example. Reaction delay time has ban omitted since sufncicnt preview of the paîh in hntof the robot vehicle is expected to be available via the vida display. Providing a @kt model of the human dynamics for the simplifiai Smith controller and for the path of the prcdictor obviowly cm only be attanpted. However, it is well known that, in non-delayed control environments, a human operator will adopt the necessary equalization chancten'stics requind to ~~~~rnptisha particular wntrol action. Thenfore, the use of models which are ideal replicas of tbe human operator's dynamics should be nasonable. The expectation tbat the operotor will be able to perfom the error detection Wonin a near-ideal capacity given appropriate training has sircady ban didin section 4.6. Al1 real-world systems experience dhmdmœs of one fOrm or llllOthCf. Depending on the type of communication link usecl, thae would likeiy be some signal variation. Assuming adequrte di@ transmission, miae rad distortion Mdmt be a factor, but there ddbe some timing Ji.This hs bot ban modeled, nor bos any der spdic type of di*. ï)hhmshuid be considerai in ury future work on this subjact. The sensitivity of the Smith c01ltcoUuto mismatch aror in cstinutes of non- delayed plant dynamics lad hedtime dehy hu bœn identifid [9,lO,3 11. The simplifieci Smith controller used in this projcct rsqiiirrs ody the loop time delay to be provideci. For runote-coa$ol ryrtems with tumddiyd plrnss, ihis time &hy cm be measured accmtdy. Nevatheiess, it W ofintaestto Cxrmine the pedormrna of the system with minmiteh in the loop thedeliy uscd in the simplifiai Smith CWotrdlcr. Fo1lowMguesystem~platsdebyuSingsystem~ysdlmcod and loop dday mimaths in the Smith amîmUa of2.02 d ind 2.002 ssconds, npnseetingerrorsof1%d0.l~~yyItmustbberaGogMzedtbrt these mors represent 20 ms and 2 mq nspectively. From tbc point of view ofwing communication ranging techniques to measun loop îimc delay, 20 ms or 2 ms should be no problem at dl. In 1983, time accuracy of lps was achievable with ranging (for dep-space communications) wing a 1MHz cef«enœ signal [IO 11. The basic system performance criterion usai througimut this thesis is the cornparison witb a non-delayed system and with no buman operator. Those results have been presented earlier. The following system paforrnance plots present a âirect cornparison of tirne-dehy mismatch vs. no mismatch. 5.5.2.1 No Humrn Operator in the Loop Redts of loop delay mismatch of 1% (2 O2 seconds) with no buman opaator in the loop are shown in Fi- 538 and 5.29. The speed md steering mgle unit step responses in Figure 5.28 are very similar, showiq significant @ches at the 3 second point, but decaying rapidly to zero. This is causeci by the output signals being fed back around the 2 second loop and not being totally compensaid for by the Smith controllers due to the time mismatch. bcrcasing the mis& nsuhs in Iargm glitches, but kresuits are not includai. For exunpl+ at 2% arofthae is dl ringing in the speed response and signifiani, dcclyedtinging in the stemkg aagle response every two seconds beyod the 19 second po9at. The sine wave responses in Figure 5.29 reveai the same glitcbcs as the unit st~nspo1lse9 for spcd but the steering tespon5e shows only the slighsest cffid st the 3 second point, Oace again, increaskg the mismatch to 2% nveals lrvger glitcbes and ringing thst increases every2seconds, Thesensuhsanalsonotsbownbutitisciearthatthesystem(both speed and steering) is unstable at 2% mismatch. Results of loop dday mismatch of 0.1% (2 002 002nds) with no human operator in the loop are shown in Figuns 5.30 ad 53 1. Again, the Jpad and steering angle responses in Figure 5.30 an very similar to eoch other, wiîh somewhat larger glitcbes for steaing, but the of the improved mima!& is apparent wiîh the mucb srnaller giitch as wmpared with Figure SB. The sine wave mpnses show a very slight degrading &kt only fbr the spad nspoase at the 3 sceond point. There is obviousIy neglig'ble degradation in basic systent gdômmœ with loop tirne mismatch of 0.1%. 1 , I 1 1 1 I 1 I I O 2 4 6 8 10 12 14 16 18 20 Tme (s) (8) Figure 5.28. Response to unit st~~oninput with 1 second system time delrys and 2.02 second loop ddiys (ie.194 mr)in the Smith controiias, tnd withno~opaitorinthelaop:(~)spœdre~ponae,d@)~ 12 1 1 1 - v I - 1 0.8 & 0.6 - al .I f 0.4 - iCD 'C 02- - iz8 0- - 02 I I O 5 10 15 Time (s) 5.5.2.2 Humrn Operator in the Loop Resuits of loop delay mismatch of 1% (2.02 seconds) and including the human operator in the loop are shown in Figures 5.32 and 5.33. The speed and steering angle unit step responses in Figure 5.32 are Whially identicai, showing a small dip starting at the 3 second point. While not included here, increasing the inismaîch to 2% results in an increase in the severity of the dip. The sine wave responses in Figure 5.29 rcveal a very slight degradhg effect only at the 3 second point for the speed response. This effect is more pronounced with iarger mismaîches, but those results are not included here. Results of lwp delay misrnatch of O. 1% (2 002 seconds) and includig the human operator in the loop are shown in Figure 5.34. Again, the speed and steering angle responses are very sirniiar ta each der, but tbe effect of the improved mismatch is apparent with the vay minimal glitches as compareci with Figure 5.32, The sine wave responses are not included since they do not show any noticeable degradation due to the small time mismatch. Time (s) ir) Time (s) Tirme (s) Figure 5.34. Rwponse to unit stqAmction input with 1 second system thne deiays and 2.002 second Iwp deiap (ii.0.1% emr) in the Smith amtroîias, including the hunun apei.ltor in the lwp: (8) apdnspoase, iad (b) steaine*enspoase- The nairal network predidar used in the previow simulations is obviously highly acauase over the total input9utput space judging by the dîsin 5.4.3. In order to see the effeet of a less acairatt predictor, another neural network was trained by wing the same input pattern and targd dara as for the original network The new network is a 5,5,2 newon ncmork using the same activation Mons(ie. tansig, tansig, Iinear) as the original. Using fèwer neutons should reâuce the accuracy of the network particularly when using fiwer training epochs. By testncting the training of this 5,5,2 network to 100 epochs, îhe resulting mean-squared-enor (MSE) wu5.7 x lû5 as comparai to MSE of 1.26 x 1LOd for the onginai 8,82 network This may not seem signifiant but th tesuhing speed vs. speed set response hwnin Figure 5.3 5 obviowly shows that the network is not well trained a! ai5 puticulariy compareci with the resuhs of the 8,8,2 networlt in Figun 5.18(b). The staring angle vs. steering set response tumed out to be eadaliy as perfect as for the 8,8,2 network in Figure 5.19@) and is not inciuded here. in rârospect, a pedictor with even worse results should have ban used for wmparison purposes. By including this ncw prcdidw in the basic system sbown in Appendix 4 the spad and staring angle te~po11scswat a@ plottcd Se-on responses KC shown in Figure 5.36. Sisoidrl mqmnm~arc rhown in Figure 5.37. Tbe spad response showing the depdncy on the atœdq mgie is included in Fi- 5.38- These plots were al1 deusing tbe srme sdes as used fa tbe aiginal 8,8,2 n#iron prediaor in Figures 5.20 through 5.24 fbr weof cornpuison. The only signifiant dinaea# between tbe unit-step mpows in Figure 5.36 and those in Figures 53qa) and 5.2l(a) is timt the stcady-sute unplitudes arc lower whm using the 5,5,2 nauon predictor. Tbe duciion in the speed mpmc is undcrswuidable hmthe poody-~ainedDctworYs spad respamc in Figure 5.35, but the nctwork's steaing angle rrspoase ldvir(uaüy @kt,so one wauld not cxpectthesteeringm@e'sunitstcptcspoll~~tohveareducedsttady-state~e~ Recheckùigthedn~rndthepedictornspo~hrs~edthatthese resuhsareaccumîe. ThepndictorwrsgumataiGromthetninednainlnetwork using the same "gensim" Monin Uitlrb as usai br the originrf predictor. The teason fOr the steering rnglefs teduced stady-sîatc Mhre is not how~At any 1.2 I I I 1 1 1 1 1 1 1 - I 0.8 - B a O O O z 0.6 - O 8 I O O O € O I O O O Q 8 2 0.4 - 1 8 O 8 8 O rn 1 08 0.2 - 8 - a d t - 42 I I I t 1 1 t I I O 0.1 0.2 0.3 0.4 Ob 0.6 0.7 0.8 OS 1 Spad Set O 2 4 6 8 10 12 14 16 18 20 rie(s) (4 Figure 5.38. Spccd r#po119t of adJystem ind $53 nnrron predictor (a) specd sctdsteeringsetinprtsat0.15~witbmpli~ges[011 for Jpeedsetaad[-1 l]hrstaamg~ia,ind@)spadscrinprtitO.l Kzd st#rinssetinprtit0,15~*unplinide~[ol]fOtspcsd*rad [-1 11 hs&caingret. rate, it shows the reduced acairacy ofthis predidar. Comparing the siisoidal responses in Figure 5.37 with th& counterparts in Figure 522(a) auci Figure 5.23(a), there is some degdation, most prominentiy at the positive maximum values on the speed response, although the stemhg angle response ais0 shows reduction at these points. Figure 5.37@) aiso reveals a degree of non-symmetry since the negative response is not the same as the positive. Figure 5.38(a) shows the reduction in amplitude for speed compareci to Figure 524(a). Figure 5.38@) also shows the amplitude reduction for speed, but is ais0 more noticeable distortion between the actual system and predicted responses, where there was virhially none in Figure 5.24@). These results are sutticient to demonstrate the inadequacy of the 5,5,2 neunin predictor testeci and to validate the 8,8,2 newon predictor d. The obstacle-avoidance subsystem design hm been dmui'bed in section 4.4.2. Proper @ormance of this subsystem cannat easily be verified without taking this pmject to the next logical stage, which wdbe animation in a dynamic environment. The robot vehiclt nads to be sirnulated in a mine environment on a video screen wherc various obstacles can be placed and whtre îhe human operator can drive the vehicle over a tirne-delayed iink A few ùmufuions wae nui to give somc exampla of the fiinctioning of this subsystcm in the ovall system. Since tùc obstacle hctor can dirdy influence the spad, this depeadency is show11 in Figure 5.39. Obstacle factors ranghg hmO to 1 wae ippüed dinctly to the RTC specd control systcm (sec Figure A6) while the spad bad a unit stcp iapuî (maximum spced) and the stccting angle was hed at an angle of zero. Fii5.39 shows how the obstrcIe fiictor reduces the speed as the obstacle factof deaerues (imaeasing likelibood of hittiq an obstacle), Automatic steering to avoid ohcies is rlso designcd into the system. For the case when the steering set input is zero ci. süai&t ahcad), r unit ssep obstacle ad input was appiied to the Jtaring coatrol sr~tem(set Figure A6). The staring angle responsc is shown in Figure 5.40(a). Tbe maximum valut at!ahable with the muhimiable sy~temis 0.785. This shauld be sufncient for most obstacle avoidancc scenarios. Ifnot, the operatcir W aiways ultimateiy amilable for intervention via teleoperation overthc timaddayed iink. Tbe nspaase ofthe stdng 8agie to a Smimnidrl obstrcle iMid input wu iIso ploatsd and inciuded m Fiepue S.qO(b)hrthecasewbentbes&ajngwtinpitiszrro. Nonlincrritiwinîbesystem are evident. NCvathC1ess, it should be daentiy lineir SPr rdsqwte thtade- a~oidance~Fiiy, two more nins were mide to test the dEect ofthe obeircIt mid inputonstaringwmmandinputsfiomtheopntor. U~~~steuiagret input, the ~rngIe~wuplottsdforpositiVefiXed~rvoidinputs ofO~and0.75showninFi~5.41.TbcgenenieffietWsantopovidcapositive o~(tht&ectissymmetricil~nqptivcmpAs). AIimitawuinsrrlledatbe ùiputtotf#vebicletopreventtbespadhmucœedingunity.Wbiletisnots~ ~obstack.void~of1.0~duemto~tbitthessariag~enmiiitsd positive througbuî the cycie, even wben the dgset input rciebsd -1.0. There is li#ledoubt~~&strelemidcoatrdpmulclii~ttievehidemdependentlyof an OpemK 0.8 1 1 1 1 I I I 1 O 0.1 02 0.3 0.4 0.5 0.6 0.7 I rime (s) (a) Figure 5.41. Steaing uiglc rcspanse to gzairnidaI steerMg set input of 0.15 Hz md ampliringe [-1 11 with ob#acle amid nxsd inputs of: (8) 0.25, rad (b) 0.75. Tests conducted on the obstacleavoidana subsystem alone do not provide a great deal of information. There are too many variables for a meaningful appraisol without getting swamped with data Two examples are included here. The obstacle factor vs. muge to obstacle bas bem plotteû as a fundon of the angle to obstacle for the case when the steering angle is zero (straight ahead) and the speed is 1 (fidl speed). This is shown in Figure 5.42. The inmments are 0.1 so the ames are wry angular. We can see that at very close range, the obstacle factor is roughly proportional to the angle to obstacle, which is as one would expect. Small angle to obstacle inputs gives a small obstacle fictor signifying high risk of collision. As the range to obstacle increases, the obstacle &or increases, sipiQing less risk of collision. Figure 5.43 includes results of another test showing the obstacle avoid steering adjustment vs. range to obstacle as a hdonof the angle to obstacle for the same case of zero steering angle and maximum speed. At very close range the steering adjustment is maximum for al1 angles to obstacle and decreases with increasing range as expected. At large angles, the adjustment is very small as expected. The specific case of zero angle to obstacle is not shown since it is strongly negative. Note that the adjustment angles are positive for negative @es to obstacle, as we would expect. Positive angles to obstacle will resuit in negative adjustmcnt angles by symmctq. The tunnel-tracking mode is rctivided by simply closing the teleoperationttracking mode switch in the îuqsteuing coatrol bld(sa Figure k 11) using a supervisory control wmrnand hmthe local terminal. The tunnel error signal is presumed to corne hman erdanally-supplied sensor signal-proce~sing systea This signal is the emrbetwan the robot vehicle's heading and the angle to the estimated centreline of the tunnel. A single simulasion was mu using a sinusoida1 input varying over the ftll range of 11 which npnscnted the location of the ûmel centrelime (ie. swinging baek and forth betwan hard ldt to hard right). Tho difference between bissignal and the steuing angle was used as the tunnel arot signal. The ste-g angle was used since the herding would only be availabIe when system animation is prwided Figure 5.44 is the staring angit response. Just as the response to the obstacle avoid signal was less than the dablecange [ml 11, so thW response also is Iesq spanning [4.6 0.61. Exkaskc trkl lad cnor in varying tbe available parameten ofthe multivanable fiizzy stecriq C011$01 systan couid not inaeise tbis range without advctscly affecting other penmeters. Once again, however, we can pfesume that this range in steaing angle should be suficient for tunnel trockin& evea in the presence of obstacles. Forûmatcly, in times of diffïcuity, the opaator is there. Figure 5 42. Obstacle tiictor M. ~gcto obrride as 8 Wonof angle to obsticle withaeeringangleofoaorndspeedof1. Figure 5 43. Obstacle avoid stdng adjutment vs ~geto obstacle as a Monof angle to obstacie with steaiag @e of zero and speed of 1. Figure 5.44. Staring augie nsponse to amaekncking am. 6. CONCLUSIONS 6.1 Summary and Conclusions This thesis presents the resuhs of a nsearch project to deveiop a novel design for remotely controlling a robot vehicle over a timdelayed communication link. A fundamental requinment was to incorporate methods of intelligent wntrol[28] wherever these concepts seemed proper to off' th& unique approaches and solutions. This has been accomplished. A multivariable fiuy control system is the basia for the system. A neural netwatk predictor is indudeci to provide the human operator with a non-delayed estimate of the system output. The fiiadamental technique usad m wmpensating for the time deiay is Smith comrol. White this technique was onginated in the 1950's for process control problcrns, the evolution of tectmology (primady digital cornputers and digitai proceaine) mhances its viabiIity in vsrious opplicasions of wntrol involving time delay in the loop. It bss bcen shown that Smith control can be simplified in rcmotc-control applications [Il]. Its capability ofrccomplishing th- delay compensation is vay obviow hmthe resuits obtabi and prescnted in this thesis. Simulations wing a linau plant with no dishirboa#s show that end-tand closed-loop control is asdalyindepeadent of the tirne delay. Since the primery mode of conttol m this system is teieoperatio~the idditid complication of incorporating a human operator in the systeni hrs beai a sipikant extra challenge. Modeling human operators U not a trivial task Since a human operator introduces additionai timcmsponsc dynamics m teleopaated systans, it bas been discussed hcrcin and shown how detrimentri this CM be for watrol, pactîdady if tirne delay is present. These additionai dynunics must be compensated f9r ifthe systan is to be abIe to firaction with time deiry in the 1oop It has ban shown îhat it is particuiarly convenient to accompüsh this by using the same simplifieci Santh control system as useci for the time dday. This does rerlugz a good widd ofthe operator's dynamics for Smith control compumtion. This U casy in mnurlt;lns. How well it wouldworkinrealityisaquestiontbatcrno~~be~by~~ subjectsincaniroiledtests. Itisknownthitmnlrtivdy~manud control tasks (with no delay or very mail deiay in the loop) tbot a luunan operator wül compenaate for the system dyMmics to a bigh depe. Al1 tmhed buman operoton such as automobile cirivers or airphne pilots do this autornaticaüy. The resuiting response is approximately modded by the crossaver modd [701. This is esscnWy the frequency response of the human's neuroddyarY5cs in the crossover region (ie.0 dB gain). This fiequency is approxhwefy 0.4 Hz on average for most people. Theref'ore, the ûansfer îunction modd of the human operator used in this systan should be valid. The actual theconstarit used may aeed some tuning depaiding on the individuai operator, but the ~stcapabiiity of the buman information-proceshg system can most likely provide the necessary d.mt.It has ahdybœn disaisseci that a human operator wili tend to adopt the nectssary qitalitation nquued to accomptish a particular wntrol task as long as it is within a Wscapability. This cannot be modeied easiiy. This suggests on ara for fkkwork invoiving human operator testing to detennine this capability and the degree of sensitivity to spe&iîic models. Another issue with hmmoperatcin is madon delay. This could be cornpcnsated for with the systern delays in the simplibed Smith controiler, but ït has ban shown that with preview abiiity?such as seeing the road (or tuiuiei) ahepd, an opaator can compensate for reaction delay by ushg precognitive eantrol. On this buis, don delay was not wnsidered to be partieularty sîgnifi~~lltand was not iaciudui in tests. Multivariablc6iizzy~0~lwiisustdinthespadd~conbolsystmis descrii The specd contrd system uses two pM114 error-ôased loops to provide the neaswy fadback control. The speed set primasr contrd loop uscs conventionaJ errordetection and operata on an d-& bais over the time de@? compaiseted by using Smith controt. The speed fietof contrd loop uscs fûzzy enor dctection siace the thne speed factor inputs are best desaibd m tams of 6iizy linguistic variables. The steering control system is Jso a ciod-loop nwhuiable system using fûzzy control. The steering set pximary consr01 loop uses comrentionai mordetcction and operata on an end-toad basis over the time *y, iIso cornpdby ushg Smith control. The other loops are used for obstreleawidrna and d-mîonomow modes inchiding tufinel-traclEmg ad atmnratic sacringinto iatmahg tunnels. Ia genenl, the spad and stccring controI systems pedbmiana W deat. predictor in real tirne. This configuration provides an alternate mahad of remotdy controlling a robot vehicle over a timbdelayd link. Wtthout feedback infdon fiom the runote teminal, this configuration is #isemially open-bop and would not rquire Smith control. ûfcaurse, d (dciayed) systan output would be kd back to the local terminal to keq the display updated. The operator would then provide commands based on the error betwccn the desired position, heading and motion of the predicted vehicle and the actual updated parameters shown on the display, based on the delayed feedback fiom the remote terminal. The success of this mahod of remote control obviowly depends on a number of Mors including the accuracy of the displayed emriroament, the length of the tirne delay and the spœd of the vehicle relative to the size of the display. This is another area for fiirther research, WMe a wmplete system has bem dcscrii in this thesis, it is basically theoreticai since its fwiction bas primarily beai to study the wntrol aspects of the system It is recognized that additionai work is necessary Mon aü of thest features wuld be implemented in an actual systun. Certainiy, the fiindamental concept of wing simplified Smith control and the gmeral stnictuec of the basic system should be sound. Simüariy, wing the alternate configuration, the operotor should be able to drive the robot vehicle over the tirne delay by driving the prtdictor in reai tirne on the console display. The "anciihy" féatures inchidhg obstaci~avoidmce,tunnd-üacking and automatic intersection-turning bave only ban tested in a dhmtarymanncr since thty really nquire taking this systcm to the nuct levd of simulation fOf fiaal twhg rad dedopmcnt. Thaî levd would be an mimaîed sinnilation environment. Withwt this level of simulation, system and subsystem tests can go onîy so fàr. 6.2 Unique Contributions The contriions ptcscllted by the reJeareh and ddopmcnt m this thesis arc outlined bdow. 1. This thesis hris demoastmtd îht auccessfiit use of Smith control m compaisating for time dday in temote-control applications. Sysîem siimilatiau have shown tbaî the &c length of the delsy is irrdcvaui (subject to dequate cornpiter memory) in maintainin8 stable, cioseci-loop conirol. A simplitied mcrhod of Smith wntrol bas bœn used which requins only tbe loop timc dday. Bued on a 2 second loop delay, this delay needs to be meawed to an accuracy m the order of - +M.1% to avoid any si@cnnt visual dcpdation in syJteni respoase. 2. Multivariable fiizsr consr01 bas been wed in the spad and steering control systems. The fhqspeed wntroUer uses two paralid closed loops: a conventional error-based loop op* over the time delay (wùig Smith contrai compensation) and a fbzq mr-bascd loop operathg at the remote terminal. The fuzzy steerhg coutr011cr mciudes a wnvartid error-based loop operating end-t0-4 across the tirne delay (wing Smith control wmpeasation) and semi-autonomous intersection-timing, tunnel-traekin~and obstacieavoidancc in a hiddstructure. 3. A fiizy logic obstacle-avoidance subsystem based on the use of obstacle factors has been introduced. These obstacle factors are based not on the pfesent heading ofthe vehicle but on the steeriag ande of the whtels, the angle to obstacle and the angles adjacent to the staring angle. This meuiod has been designeci to wntrol both the speed and stcuing angie of the vehicle to avoid collisions with obstacles. 4. Human operator dymmics in the timedelayed cantrol loop cm be catastrophic. Compaisating fbr thest pafhmcdegrading flects bas been successfi~llyaccomplished using the same simphiïed Smith control system as used for the tirne delay. 6.3 Future Work 1. A major disappointmeat is îkfiilme to develop a nadnetwork pndictor that can accommodate the -ait response ofthe systan, A tnie systcm pndictor would accept any initid conditions ofthe inputs and respond cxady the same as the actual system. This appears to be a simami cbdlcngc tkst must be imnstigsted. - 3. The ncxt logid stagt in devdopment of this systmi û&rc acRial Unplementation in hardware is to deveiop a SuitOble simulation mviroment where a wmplete animation of the systun can be tested. This project bas takm the design as fâr as reasanabiy pssiile usiag the levei of simulation desdeci herein. A dynamic simulation environment is necwsary to ve&jr and tunc the obstacleavoidance subsystem as weü as the semi-uut0110mous modes of~agclongand automatic intersection-tuming. 4, Witb a total system and emRronment available in animation, this would be an appropriate time to begin testing a human operetor in teleapetation trials over tune delay using a suitable cantrol device such as a joystick and a video display. This wouId aiso be the ideai time and phto observt bow tbc operator prefcfs to drive the systém. This level ofsystem rescerch and dcvclopmeut ddbe vay infi,rmative and udfor enbanchg remote-conttol systcm design. 5. Ehhg the systcm animation available wüi also dow testing sensitivity of human operator dynamics modds with doperotors in the system. This may lead to various unique improvements in the design of teieoperated systemsems [il J.L. Adams, "An Investigation of the Eftècts of the Lag Due to Long Transmission Distuices Upon Runote Control, Phase II-Vehicle Experiments & Phase III-Conclusionsw,NASA Technical Note D-1351, Nationai Aeronautics & Space Administration, April1962. [2] L. Conway, RA. Voh and MW. Walkcr, "Teieautommous Systans: Projecting and Co-ordinating Intelligent Action at a Distance", EEE Transactions on Robotics and Automation: 6(2), 146-158, April1990. [3] iEEE Standard Dictionary ofElectrical and Elecüonic Tcrms, The Instinite of Electricai and EIectronic Engheers, Inc., New Yok 1984. [4] J.G. TNxal, M.J. Shoomq WR Blessa and J.W. Clark, Remote Controi, in the Handbook of TelerneÉry and Remote Controi, ELGnicnba%, Editor-in-Chie McGraw-Hill, New Yotk, pp.15-1 to 15-169, 1967. [SI RA Hess, Hunan-in-thehop Controi, in The Control Handbook, W.S. Levine, ed., CRC Press & IEEE Press, Boca Raton, Fi,, pp. 1497-1505,1996. [q WM. Smith, "yd VIFeedback and Behaviot', Science, 132,1013-1014, October 1960. [7J O. JM Smith, "Closer Control of Loops Wth DdTii, Chenrical Engineering Progress, 53(5), 217-219, 1957. [8] KJ. Astrom, "Frequency-Domin hpertits of Otio Smith Rqdaîors", Intdonal Journal of Contrai, 26(2), 307-3 14,1977. [9] JEMarshan, Control of Ti~DelaySystems, SteveMge Umted Kiugdom: Peta Peregrinus Ltd, 1979, [IO] Z.J. Palmor, TheDelay Compensation - Smith Pndictor and its Modifications, in The Control Hanôbook, W.S. tevine, ed., CRC Press & EEE Pnss, Boca Raton, FL, 224-237, 19%. [i 11 XC. Wood and UR Oison, Tnnc-Ddayed Remotd=ontrol Systcms Ushg Smith Contmiid, Proceedings of the IASTED hanationai Confirrnce on Intclligent Systcms and Contrai, Hdi& N.S., 3639, Juae 1-3,1998. [12] A Kheddar, R Cbellrü and P. Coiffit, ViRcaiity Assisted Telcopemtiog in VEHadbo~iqKM,Stmq,ed,~Etibgmi~Iac., CbapterS6, Feb- 2002 (atpeeted)- [13] TB. Sheridan, Telerobotics, Automation and Humas Supemhry Controi, Tite MIT Pnss, Cambridge, Mass., 1992. [14] A Hemami, "Robotics and Automation in Mimng", IEEE Canadian Review, 11-15,Fdl1993. [15] P. Corke, J. Roberts and G. Wrnstaniey, "Viion-Bad Control for Mining Automation", IEEE Robotics and Automation Magazine: 5(4), 44-49, Decesuber 1998- Il61 S. Scheding, G. Dissana* E.M. Nebot and H.Durranî-Whyte, "An Experirnent in Autonomous Navigation of an Underground Mining Vehiclen, IEEE Transactions on Robotics & Automation: lyl), 85-95, Febniary 1999. [17J J. Yen, R Langari and L.A Zadeh, eds., Industrial Applications of Fuay Logic and Intelligent Systems, IEEE Press, New York, 1995. [I8] P.J. Antsakliq M. Lemmon and J.A Stivcr, LecuMng to be Autonomous, in Intelligent Control Systems, UM, Gupta and N.K. Sieds., IEEE Press, Picataway, NJ, 28-34,1996, [19] L.A Zadeh, Foreward, in Inteligent Control Systems, MM.Gupta and NK Sith, eds., IEEE Press, Picataw~y,NJ, Iàx-q 1996. [20] T.I&ss, Fuzy Logic With Engincehg Applications, McGraw-Hüi, New Yok 1995. [21] S. Haykin, Neural Networiy EEE Press, Macmillrin CoIIegc Publishing Company, New York, 1994. [22] N. Wakami, H NomS. qlaLi, "Fuay Logic for Home AppliMces", in Fuzy Logic and NdNetwork EIandbook CH. Cùa, ed., Mcûraw-Hill, New Y* pp. 21.1-21.23, 1996. [23] B. Kosko, NadNetworks ind Fuzy Systemq Preati~~Kngld N.J., 1992. [24] T.J. Nelson, UR Olson and KC. Wood, "Long-Delay Teleumîrol of Luuar MiMng Equipment", Space 98, Pmcc&gs of îhe SMInanMiod Confettllct d Exposition on Enginaring, Coustruciion ad Operations in Space, Albuquerque, NM, 477-484, Aprü 26-30, 1998. 1251 TB. Sheridan and WJL Fcdi, "Rmiote Msnipuiative Contrd Wrtb Transmission Delay", LEEE Trannaaim on Himm Factors m Enghahg 25-29, Sept& 1%3. [26] WR Ferrcll, "Remo& Manipulation With TransmisSionDelayw,EEE Transactions on Human Factors in ihgkmhg: FIFE4,24-32, September 1965. [ZTJ R FerreIl, "RemoteManipulation Wrth Transmission Delay",NASA Technical 'on, Febw1965. Note D-2665,Nationai Aero&cs & Space Admmüab. . [28] L. Adams, "An Investigation of the E&cts of the Tune Lag hie to Long Transmission Distances Upon Remote Contrai, Phase 1-Tracking. . Expaiments", NASA Technical Note D-1211, National Aerodcs & Space Adminisbati'on, Dec«nbet 196 1. [29] RH. Lewis, "Preliminary Lunar Tdeoperotions Imrestigations Usiag a Low-Cost Mobile Robot", noceedings of the Ntnth RincetonlAIANSpace Studies Institute Conférence, 138-143, May 10113,1989. [30] Robert J. Anderson and Marlr W. Spong, "Bilateral Control of Teleoperaton With TieDelay", EEE Tdonson Automatic Consrol: 34(5), 494-501, May 1989, [3 11 G. Niemeyer and J-JE. Slotine, "StableAdaptive Tdeaperation", IEEE Transactions of ûceanic Engimmhg: 16(1), 152-162, Jamiary 1991. [32] W.S. Kim, B. Hadord and A.K. Bejczy, "Force-Reflectioa and Shared Cornpliant Control in ûperator Teldpuiaton With Tme Delay", iEEE Tdonson Robotics & Automati011: 80,176185, April1992. [33] S. Lee aad H.S. Lee, "Modame, Des@ Bt Evaldon of Advand Teleoperator Control Systems Witb Short Tmie Delay", IEEE Tdnson Robotics and Automation: 9(5), 607423, ûcîober 1993. [34] D.A Lawrence, "Stabüity d Transprrency in Biiateml Tdeoperation", IEEE Tdonson Robotics and Automotion: 9(5), 624637, October 1993. [35] G.M.H. Lcung, BAFrsncW d J. ApApkariaiS %ilPtail ControUer for Teleopemors With TiDelay via Mu-Syntbesis", EEE Tmnsactiom on Robotics and Automation: 11(1), 105416, Febnury 1995. [36] US. TrbtdyiIou & MA GrosenbiPib "Robust Coutrol for Undemater Vehicle Systems With TiiDdays", IEEE JaurnJ of Occanic EnginCamg: 16(1), 146-151, Jaa~ary1991. . . [37] CRKelly, "Pndidor Instnimmu Lwk Imo The Futuren, Consr01 Enpœmg 8690, March 1962. [38] LE. Fargel and E.A Ulbrich, "nebictor Dispiays Extend Manual Operation", Control Engineering, 57-60. Augus&1963. [39] CKKelly, "Predictor Dispiays -BeControl for Cornplex Manual Systems", Control Engineering, 8690, Aupst 1%7. [40] JEArnold and P& W. Btaisted, Pesign and Evaluation of a Predictor For Remote Contml Systems Operatin8 With Signai Transmission Delays", NASA Technical Note D-2229,N&d Aeronauîics & Space Administration, Daunber 1963. [41] L. SeW.S. Kim, F. Tendick, B. Ha~ahrd,S. EUW, M. Denom~~,M. Da, T. Hayes, T. JO* M Lawtoq T. Milis, R Petason, K. Sandem, M. Tyler and S. Van Dyke, "Telerobotics:Display, Conid & Communication Problerns', lEEE Transactions ofRobotics & Automatim 3(1), 67-75, Febniary 1987. [42] JME. van de Vegte, P. Milgram and RH. Kwong, "Teleoperator Contml Models: Wécts of rie Delay Md Impafect Systcm Knowledge", ïEEE Tdons on Systems, Man and Cybernetics: 2q6) 1258-1272, NovembcT/Decrmber 1990. [43] TB. Sheridan, "Space Teleoperation hghTic Delay: Rcview & Prognosism. IEEE Tdonson Robotics & Automation: 9(5), 592606, October 1993. [44] G. Hirzinea, B. B~aner,J. Dierràeh ind J. Hclldl, "Sensor-Bad Space Robotic~~ROTExand its Tderobotic Feahires", IEEE Tmnsactiom on Robotics and Automation: 9(5), 649-663, October 1993 . [4S] W.S. Kim and AIL Bcjczy, "Daaonstrrtioil of a Higù-Fidelity nedictiw/neview Dispiay Technique fbr Tderobotic S&&g in Spa,JEEE Tdonson Robotics & Automation: 9(5), 698-702, Odober 1993. [46] S.G. Lantcigx, "contra1 of Tdananipuiptors With WCTii Delays", M.ScEE. Thesis, Uninrsity ofNew Bmmkk, Mny 1994. [47J W.R. Fdand TB. Sheridan, "Supervisory Cantrol of Runote IEEE Spectnim: 4(10), 81-88, October 1967. [48] D. Pivirott0,"Rovetsl Us@ Mobile Robots as Planetary Explorers", The Planetary Rcport, X(4), û-13, JdyiAugust 1991. [49] T.S. Lindsay, TeieproBnmming: Remotc SiRobot Task Execution, PhD Dissertation, Unmrsity ofpamsyfvmir 1992. [5 11 CMAnderson, "AdVanang Our Ambitions: The 1994 Mars &ver Test", The Planetary Report, XIV(5), 1647,1994. [52] B .S. Graves, A Genaalized Tdeau!onomous Arcbiteaure Using Sion-Baseâ Action Sdection, PhD Mon,Texas A & M University, 1995. [53] G. Meinsrna and H Zwart, "Oa ti, Control for Dead-T'me Systeinsn, IEEE Transactions on Automaîic Control: 45(2), 272-285, Febniary 2000. [54] RR $elmi6 and F.L. Lewis, r)eab;ane Compensation in MonControL Systems Using NdNetworksn, IEEE Tmnaactions on Automatic Control: 45(4), 602613, Apd 2000. [55] H-XLi and SKTso, "Higher-ûrder Fuay Coml Stnichire fw Hjgher Order or Tune-Dehy Systems", IEEE Transactions on Fuzty Systems, 7(5), 540-552, October 1999. [56f FL.Lewis, WX.Th, L.-Z. Wang and ZXLi, "Deadzone Compensaiion in Motion Control Systems Using Adaptm Fuay Logic ControP', IEEE Transactions on Control Sysîems Technology, 7(6), 73 1-742, Novanber 1999. [57] K. Taylor and B. Dalton, "Iaternet Robots: A New Robotics Niche", EEE Robotics and Automation Wgazk,7(1), 27-34, March 2000. [58j K. Goldberg, S. Gentmr, C. Sutter Md J, Wicgley, "Tbe Mereury Project: A Feasi'bility Study for Internet Robotsn, IEEE Robotics and Automation Mphe,7(1), 3540, Match 2000. [59] P. Saucy and F. Mondada, "KhcpûnnieWeb:Open Access to a Mobile Robot on the Internet", EEE Robotics and Automation Magahq 7(1), 4147, March 2000. [60] D. Schuiq W. Burgrvd, D. Fo& S. Thnm and A Cremas, Web Interfoces for Mobile Robots in Public Places", IEEE Robotics and Automation Mqazkc, 7(1), 48- 56, March 2000. [61] RC. hoand TM. Chen, "Devdopment of a Multi'bebwi-Based Mobile Robot for Remote Supcniisory Consrol Thou& the huemet", IEEE Tr(lZ1SICtim on MccbaEronics, S(4), 376-385, De#mkr 2000. [62] JM Mc,W&ts of Thne Deiay in tbe ViFeedback Loop of a Man- Machine Systemn, NASA Contnctor Report, NASA CR1560, National Acronsutics & Space Administration, Septeniba 1966. [63] D. McRuer and E. Krcndd, Dynamic Response of&msn Opartors, Teehnid Repn WADC-TR-56524, U.S. Ait FOKCC,~~~~. 1641 E. Gabay and S.J. Merhav, "Identification of a PdcModd ofthe Ebm Operator in Closed-hop Control Tasks", EEE Tmnsactions on Systems, Man and Cybemaic~:7(4), 284-292, Apd 1977. 1651 M. Tomiaika and M Fujii"Extended Signal Quickening for Md Controln, DEEE Transactions on Systems, Man and Cyknetics: 9(10), 668-676, October 1979, [66] F. Osafo-ChatIcs, G.C. Agamai, W.D. O'Neill and GL.Gottlieb, ''Application of Thne-Series Modehg to Human -or Dynamicsn, EEE TCBIlSaCtions on Systems, Man and Cybenietics: 10(12), 849-860, December 1980. [67J P.W.Koken, H.J.J. Jonker and C.J. Erkciens, "A Model of the Human Smooth Pursuit System Based on an UnsupeMsed Adaptive ControUer", lEEE Transactions on Systems, Man and Cybemetics, Put A: Systans and Humans: 26(2), 275-280, March 1996. 1681 G.0.A Zapata, R.K.H. Galvao and T. Yoneyama,"Extrading Fuay Control Rules From Experimental Huxnan Opaator Data", IEEE Td0119on Systems, Man and Cybemetics, Part B: Cybernetics: 29(3), 398-406, June 1999. [69] M Packand CH. Houpis, "Fli&ht Control of Piloteci AircraAn, in The Contml Handbook, W.S. Levine, cd., CRC Press & IEEE Pnss, Boca Raton, FL, 1301-13Q 1996. [71] D.T. McRuer, RW.Allm, DEWeir and RKKlein, "New Results in Driver Steering ControI Modelsn, Human Factors, 19(4), 381397, 1977. [72] C.C. hhddaq "Application of an Optimal Previcw Controt for Sionof Closed-Loop Automobile Dchhg", IEEE Tdomon Systems, Man ad Cybernetics: 1 1(6), 393399, June 1981. [73] D.L. Kleinmm, S. Baron and W.H. Levison, "An Optimal Control Modd of EIuman Respoasc, Part 1: Tbeory aaâ Validation*, Automotica, 6,357469,1970. [74] G.O. Burnhem and GA Bekey, "A Hairistic FmitSStitc Model of the Hman Driver in a Car-FoUo- Situationn,IEEE Tdonson Sysîans, Man and Cybanetics: 6(8), 554562, August 1976. [75] KD.Enstrom and WB. Rouse, "Rerl-li~~~Dctnmnrb'on of bow r rTuFui &Is Auocated His Attention Between Co-1 d Monitoring Tub", EEE Tdm on Systems, UPn aad Cybernetics: 7(3), 153-161, Much 1977. [76] DHLee and W.Hm, "Analysis of Human operator's Behavior in High-ûrder Dynamic Systems", IEEE Tdonson Systems, Man and Cybernetics: 10(4), 207- 213, April, 1980. \ [77l RA Hess and BD. McNaUy, "Automation Effeets in a Muitiloop Manual Control Systemn, IEEE Tmnsactions on Systems, Man and Cybenietics: 16(1), 111- 121, January/February i986. [78] A Abdel-Malek and VZ.Mannarelis, "Modeiing of Task-Dependent Cbaracteristics of Human Operator Dynamics During Pursuii Manuai Tracking", EEE Transactions on Systenis, Man and Cybdcs: 18(1), 163472, January/February 1988. [79] RA Hess, "nusuit Trockmg and Higber Leveis of Skiil Development in the Human Pilot", EEE Tramactions on Systems, Man and Cybenretics: 1I(4), 262-273, Apd 1981. [80] M. Yoshizawa and H. Talreda, "An Output Regdation Model for Human Input Adaptiiility in the Manuai Control Systan", EEE Transactions on Systems, Mau and Cybemetics: 18(2), 193-203, March/April1988. [81] DE. Greene, RE. Barr, C. Fulcber, L. Hwang and S.G.K. Rao, "A Stochastic Sequential Model for Man-Machine Tracking Systems", IEEE Transactions on Systems, Man and Cybcrnetics: 18(2), 3 16-326, UardApril1988. [82] RA Hcss,"Modeling the Eff'ofDisphy Quality upon Humui Pilot Dynarnics and Perceived VehicIe Hhdibq~Quriities", TEEE Tmnaactions on Systerns, Man and Cybemetics: 25(2), 338544, Febnury 1995. [83] EX. Boer and RV.Kcnyon, "Estimrsioa of Tme-Varying Delay Tiin Nonstationary Lincar Sy~tans:An Appdto Monitor Human Opaator Adaptation inManualTrackingTasks",~T~onSystrms,ManuidCybetnaics,Part A: System~~d Humans: 28(1), 89-99, JUIUUY 1998. [84] E. Donges, "A Two-Levd Uodtl of Driver Stecring Behivior", HIFactors, 20(6), 69 I-707,1978. [8s] AT. Bahill, "A SileAdoptivt Smith Pndictor for Controllhg nmeDeiay Systemsn, IEEE Control Systems M@m, 3(2), 1622, May 1983. [8q T. Furukawa ind E. S- "Predictive Corn1For Systems With Ti Delay", International Jdof Conirol, 37(2), 399412,1983. [87] KW~andUIto,"A~ModdCoatroIforLMearSystemsW& Delay", EEE Tdomon AutdcCmîml, 26(4),1261-1269, Decanber 1981. 1881 T. Hag$und, "A PrediCtm PI Controiier for Processes With Long Dead EEE Control Systms Mapine, 12(1), 5760, Febniary 1992. 1891 KJ. Astrom, C.C. hgand B.C. Lh,"ANew Smith Predictor for Contro~a Process With an Integraior and Long Dead-Ti, EEE Tdnson Automatic Coriboi, 39(2), 343-345,Feb- 1994. [9û] hd.R. Matausek and AD. Micic, "A Modifieci Smith Predictor for Controlluig a Process Wth an Integraior and Long Dead-Ti, lEEE Transactions on Automatic Controi, 41(8), 1199-1203,August 1996. [91] P. BIaha and P. Pivonka, "Intelligent Corrector for the Systems with Tm DeIay", 42d Intanational Scicnt%cCoUoquium, flrncnau, 637641,1997. [92] M.R, Maîausek and AD. Micic, "On the MOdiîied Smith Mctorfor ControUing a Rocess Witb an integrator and Long DeadDeadTlen,IEEE Tmnsactions on Automatic Control, 44(8), 1603-1606,August 1999. [93] JE. Normey-Rico and EF. Caaacho, "Robust Tuning ofDead-Time Compensators fbr Processes with an Imegrator and Lang Dead Tlen,EEE Tdonson Automatic Controi, 44(8), 1597-1603,Augwt 1999. [94] 2,-Q.Wang and S. Skogestad, "Robust Control of Tirne-Delay Systems Using the Smith Predicto?', Int. J. of Controi, 57(6), 1405-1420,1993. [95] E. Coq The Fuay Systems Handbook, AP Profession& Cambridge, Mq 97-98, 1994. [96j UK Liubaldrq D.S. Rhode, J.R WAdman, P.V. Kokotovic, "Adaptive Automotin Speed Contrai", m The Control Hiiadbook W.S. Lcvinc,cd., CRC Pnss & IEEE Pr- BO= Raton, FL, 1274-1285,1996. [971 T.D. Gillespie, Fun- of Vehiclc Dymmics, Society of Automotivc Engineer~,k., W~C, PA, 21042,1992. [98] J.Y. Woag, Thtory of Ground Vehicles, John Wiley rnd Sons, New York, pp. 149-153,1978. [99] J.Y. Won& Theory ofGround Vebicles, John Wiley and Sons, New York, pp. 21&239,1978. [lOO] JREllig Venicle Dynamics, Business Books, Londoq En&& pp. 60-92, 1%9. [loi] K. Breideatûai aad TAKomarek, "Radio Tracking Systemw, m Deep Spaœ Teiecommuaications Systems Eneiiieerin$ J.H. Yu- cd., Plenum Press, New York, p.144, 1983. The general test configuration ofthe basic remo~nüolsystem modeled and simulateci using MatWSimulink is presensed in this Appendix. Each subsystem diagram in the figures uses the same nomenclatm as used in the ncxt bigha level. For example, the fuey &riag wntml diagram in Figure AM is the expanded view of the fiizzy steering control block in the RTC dngcontrol diagram show in Figure A13, This latter dhpm, in hm, is the srpanded view ofthe RTC steering wntrol block in the mnote teleautonomous controiler diagram shown in Figure AS and so on up to the complete system block diagram show11in Figure Al. The system description has been covdin Chaptcr 4. The simulation diagrams here show dI the detailed components required or inciuded in the various simuIaîions. These components include scaling amplifiers, wastants, lisniters, switches and blocks hrpdorming dgebraic and logical opaafions. Scaling amplifiers are used to provide adequate ranges for the parcicular parameters by amplXjing or attmdng the signals as neasq. bmbnts rue included to provide the neccssary DC SMin level for the same nason, This will be described in each relevant section. A. Matlab/Simuiink Basic System Configuration A.1 OvmüTest Configuration The toplevel diagram uscd in perfi,rming systcm simulations is shown in Figure A 1. Stcp-Won or sinusoidd test si@ for spad set and stcaing set inputs are selected by the input swit&cs rad monitored using the input oscilloscopes. The adsystem speed and staring angle outprrts are multiplied by the sding ampliners of gain 2.6 and 1.01, nspectiVely, to provide ml mges of [O 11 ad [-1 il, respectivelyYThe praücted system outputs & not requin dmggains sincc they are included in the d network pndictor. These output signais are passed through offset test delays that an set to the 01~bwaysystcm delays in order to perform compatisons ofthe pndicted outputs with the actual outpuis. The various multiplexas, oscilloscopes and Sirnulii output ports albw Mewing and pl- the four autput sipais in appqxiüc cornbinitioaa The major system blocks are shown in Figun A2. This shows the gened layout of the system with the brwatd and reverse signal patbs betmm the various blocks and over the system (the) delays. The constant block inputs to the RTC represent the signals cxpcctd to be provided by an dly-provideci seasor- processing system located on-board. These signais are the angle to intasection (IntAng), range to intasection (IntRan), angle to obstacle (ObAng), range to obstacle (ObRan) and tunnel-tracking error (TrackErr). For simulations, these inputs bave been co~ectedto appropriate step-fiuiction or sine wave generators. The operabr/console shown in Figure A3 includes the buman operator dynamics for both the actual and pndicted paths. An input scaling gain of 0.4 is included in the spad path to provide the proper scaüng for the correct range (ie. 1-2 21) for the spad set ercor signal hmthe operator. Figure A4 shows the tnnsfer fiinction of the human operator used in tbis ptoject. The swiîches bave becn udto tesr the variousconfiigiirations, primarily human operator in and out of the loop. The rcadion delay is trivial to compensate for with the Smith controUa, but as discussed previously, it is expected that the oparitor can provide the necessary compmsati*ongivm a vidm view of the patb in fiont of the robot vehicle. The reaction delay has not been included in any of the simulrtions. The errorddecror is included since the opmtor is expccted to prode this fiinction. AS. Opmtor Speed/Steerhg Rcrpowe Figure A5 provides the same hum8n operator dymmics as the pmious figure for the predicted output without the eriorddeaot. INPUT OPERATOR GAIN SPEEO a RE SPONSE SpdSit I Figure A4. ûpmtot spadleemr mpome. REACTION NEUROMUSCUUR RESPONSE Figure AS, Operator spead/steering tesponst. The only component in the LTC as shown in Figure A6 that is relevant to this thesis is the predictor. This is the neural network-based subsystem that provides the (steady-state) mode1 of the subsequent system without the time delay so the operator can see the irtunadiate &a of the control cornmands wEthout waiting fbr the delayed feedback hmthe advchicle. The opaasor's codinputs, speed set enor (SpdSetErr) and Jtaring set aror (StrSctErr) pas Mythtough the LTC. a Pd- -mstr PREDICTûR A.7. System Ddrys The forward and rmrse system thedelays an shown in Figure A7. Figure A7. System delays. Figure A8 shows the major subsystem blocks in the RTC. These are îk obstacle avoidance, RTC speed control ind RTC control bloclts dong wiîh theu various input and ouîput signais, Tbe spad set enor (SpdSaEn) and stcubg set mot (StrSaEn) inputs arc tbe wntrd commrnds hmthe LTC. Speed and 8tœring angle (Strhg)inputs are the fadbeclr wtpits fiom the vehicle. Tbe osha inputs rn fiom the onhard si@-pnnxsàng sptem The outpub of tht obstrcle rwiibnce block, obstacle factor (ObFac) rnd obstick rvoid (ObstAv) arc shown comcctal to the othcr two bloclrs. Tiie two ou- ofthe RTC are the spad control (SpdCntl) lad steering conid (StrCnsI) Sig& which uc wed to drive the speed rad stœrhg actuaton, rtspcctivcly, on the robot VebicIe. Figure A9 shows the detailad obstacle-avoidance subsystem. The top row is used fiir calculatiog the primary obsîacle &or (which is simpiy refefced to as the obstade factor outside of this iiubsytem). The two firay controllers are shown dong with scaIing amplifiers and DC offset. The first &ng amplifier with gain 1.12 was used to ensure the fiill range [-1 11 ofthe hmrd Woninput. The constant 0.1056 was provideci to shift the obstacle factor signal level downward to ensure ttraS it would reach a minimum value of zero. The amplification by 1.27 imrnediatcly following was necessary to realize a fidl "dynamic" range &om O tol. A number of litniters were found to be necessary to ensure that signals beyond a certain range would not couse problems in the simulations. It is quite probable tbat many ofthem wdd be removeci in practice, but this is not lmwn with certainty at this time so îhey have been left U1. A number of other blocks, such as comparaton and multiplias are used for calculating parameters such as left and right steaing adjutmats and autosteering wntrol output in Figures A9 and A 15, rcspcctiveLy. The second and third rows are used to caldate the left and right obstacle factors, respectively, in an identical mariner to the first, except tbaî these vaiues are now based on views that are s6ifted by 2Ph on either side of the cunait steexing angle. The (arbitnrily chooea) 2û% Mtsin the steering aagle are sœn as DC offsets of +/- 0.2. The presence of the satudon bloclrs at the inputs is to -thot the Uiputstothe~~wn~roUersdonotarcadthe~e[-111 which, ofcourse, would ocarr when the steering mgle txacds +/- 0.8. The angle to obstacle (Ob-) and range to obstacle (ObRan) inputs npresart the obstacle having the lowest obstacle factor (ie. highest danger levei). The steering aagie (StrAng) and speed inputs ue the outputs hmthe robot vehicle. The tirne to collision input is caldatai by mdtiplying the r~~eto obstacle by the raiprocal of the sped Sithe spd varies betwegi O and 1, r nm#tO DC offset of0.001 has been ch- to allow the reciprocal to vary beâween 0.999 (-1) and 1000. This is then ûmca!ed to the range [l 101 before being muiîiplied by the mnge to obstacle. This eff6ainIy covers a speed range of [O. 1 11, which should be dcient, Al1 three ofthe obstacie faetors are used to dmeif- Jteering djustment is to be made. The M and right oWefhm arc oompued to sa ifom is larger. That value is then compared with the primiiry obstacle Eactor to see wbich of those two is larger, and an appropriate steerhg adjwtment for obstacle avoidance is generated as the output obstacle avoid (ObstAv) signai. A.10. RTC Spad Centrol The block diagcam in Figure A 10 shows the spdcontrol subsystem in the RTC. This includes the Smith control bloek and îhc fiizzy speed control block with their respective inputs and outputs. The spced set wrc~rnmand fiom the LTC is fust processed in the Smith wnüuller before going to the fiiay specd control block. The range to intersection (IntRan) is wcd to nstrict the speed of the vehicle to - 0.25 (ie. 25%) of maximum as iî approachts and passes through intersections without stopping. The range to intcrseetian is disabled as shown in Figure 4.9 during teleoperaiion modo by supavisory control command. The obstacle firctor is designesi to slow the vehicle in the presence of obstacles and to stop it if its numerical value reaches a minimum of - 0.25. The steering angle is the sngle of the wheels refcrcnced to zero (ie. "straight ahead") and measured My.The dependency on the speed by the steering angle is such that maximum Jpeed will be maintaineci for st-g angles within about +/- 0.25 of zero. Beyond this range, tbe spœd will be pmpssively reduced with increasing staring uigle down to about 0.33 (ie. 33%) of maximum. The final input, spced, is the dvehicle output spœd. A bop gain of 10 hs dso bccn included. This was detennined empiricilly during systcm development to to provide satisfactory systern responae. The fwzy Jpeed comlblock eontains the two kzyspeed controilas and the speed faetor selector as shown in figure A 11. Tblatter block simply tJres the minimum dueof the thne speed factor inputs, rngle to intasection (IntAng), obstacle &or (ObFac) and stecring angie (SbAag) to ensun that the output spœd of the vehicle is appropriatcly amûahd, ifheassaiy. The limiter bloclrs wae added to ensure that the inputs to the fiiay coiltlollen do not cxdthe mages of the respective universes of discoune. SPEEO FACTOR SPEED FACTOR ObF- 1 SELECTOR ERROR Figure A, 1 1. Fu- speed control. A.12. Smith Control The Smith control block shown ia Figure A 12 includes both the Ioop delay theand the human operator dynamics, This is essentially the operator's neuromuscular response since the operator's rdondelay bas not been included in the simuIations. The loop delay switch accommodates testing the system with and without the human operator in the loop. The same bIock is wed for the steering control subsystem. A.13. RTC StdgCoatrol This bb& contains the tkysteerjng control block dong with its Smith corrtrollw and loop gain as shown in Figure A13. The steering set enor (StrSdEn) command fiom the LTC is pfocessed by the Smith controller before going to the lÙzzy steering mntrol block. The dersteering wntrol inputs to the Latter block inchde angle to intersection (MAI@, tange to intersection (intRan), tunnel tracking error (TrackErr) and obstacle avoid (ObstAv), The st- augk (StrAng) W the output Eom the robot vehicle fd back. Tht loop gain of 15 was ddermined empirically to pmvide &&tory system re~ponse. kl4. F~uzyStdg Control This block containsthe fllzy sieaing controllers as well as the autosteering wntrol block as shown in Figure A 14. The switches, activated under supervisory control and used fbr scmi-autonomow opaition arc inciuded. in tdeoperation mode, oniy the steering set aror (StrSetErr) Gommrnd input and the obstrck avoid (ObstAv) input are in the Ioop. Ona again, ding unplificm and liare included to en- that the fiil1 ranges of inpuîs and outputs are provided. The fiizy controlier fbr intérsection control is included hm, AIS. Autosteering Control The autosteaing control bluck sbwn in Fi- A. 15 W tbe implcmentaiion of theauto~~lJchemestrowninFigure4.17.Thethneinputs,usbownin Figure A 14 are intcnedion con$ol (TntCd), fange to imaSeaion @nthn)and tracking cmPtPckErr). Depadhg on the mode scleded, these inputs are wmbined to provide the reqWed outosteering wn$~lsignal. This block includes the speed and steuing dyiiamics rnodels of the robot vehicle as shown in Figure A 16. Figure A 12. Smith wntrP1. Figwe A 13. RTC c0-L El- Et- Figure A 15. Auto- cbatrol. Figure Al6. R&t ddë. B. MitiabBimuiink Alternate System Configuration This Appendix pmvides the system block diagram for the aiternate configuration described in Cbaptcr 3 and ghown in Figure 3.15. Since this configuration is a modification ofthe basic remotscontrol sys- only those blocks thaî bave benchmgad' bom those show11 in Appendîx A arc includcd. Sincc tbu is an open-loop system on an end-tolend buis, the system delays an only in tbc forward direction and the simplified Smith controllas in the RTC have been bypassed. These changes are not aplicitiy shown. 1 Altemate Corn plete System The major system blocks, shown in Figure A2 ue again shown in Figure B. 1. This shows the gend layout of the modified sya~nwith the revisai signai paths between the various bloclcs and ova the o~lbwaysystcm deiays. The command signals are now the outputs noni the pndiaar. The altunative opaitor/ooiw)le black show11 in Figure B.2 includes only the human0~r'~neuromusailudynuniCa. This block includes dytk @cîœ u sbown in Figure B.3. Thedyc~tequisedintheRTC~rtfvWrhanrteconfiguartio11sbownin Figure B.4 is the inchision of îkinput suhg gain of 0.4 rnd the bypusing of the simplified Smith controU~.Tbis hm ban dwarssed in Section 5.4.1. -1 02s+1 1 NEUROMUSCULAR SPEEù RESPONSE NEUROMUSCUUR STEERING RESPONSE Figure B.2. Aiternate operatorlwmle.