HAL Id: tel-00480931
https://tel.archives-ouvertes.fr/tel-00480931
Submitted on 5 May 2010
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entific research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
systems - Application to mechanical systems
Antoine Chaillet
To cite this version:
Antoine Chaillet. On stability and robustness of nonlinear cascaded systems - Application to mechan-
ical systems. Automatic. Université Paris Sud - Paris XI, 2006. English. �tel-00480931�
THÈSE
Présentée pour obtenir
Le GRADE de DOCTEUR ÈS SCIENCE S
DE L'UNIVERSITÉ PARIS XI ORSAY
par
Antoine CHAILLET
Sujet:
On stability and robustness of nonlinear systems
Appli ations to as aded systems
Stabilité et robustesse des as ades non-linéaires
et appli ation aux systèmes mé aniques
Soutenue le 7 juillet 2006 devant la Commision d'Examen:
M.Jean-Mi helCoron Examinateur
M.Thor Fossen Rapporteur
M.Antonio Loría Examinateur
M.RodolpheSepul hre Examinateur
M.Eduardo Sontag Rapporteur
M.Andrew Teel Examinateur
A Thibaut.
Remer iements - A knowledgement
I sin erely thank Profs. Fossen and Sontag for having done me the honor of reviewing
thisdo ument,and theexaminersofmy ommittee, Profs. Coron,Sepul hre andTeel,for
their parti ipatio n to the evaluation of this work and their pre ious s ienti omments
thatsurely greatly improvedthe qualityof the thesis.
I warmly thank my supervisor, Dr. A. Loría who, with his s ienti ulture, his in-
tuition, his omprehension , his listening and his sympathy, made of these three years a
gilt-edged sour e of s ienti enri hment in a simple and friendly atmosphere. Few PhD
students,a ordingtome,areleftsu hafreedomintheirworkandgivensu hapromotion
of theirresults bytheir supervisor. Iamdeeplygratefulto him for this.
I also thank the personswith whom I had the privilege and pleasure to dire tlywork
with during the past three years: Alexei, David, Elena, Erik, Mario, Rafael and Ya ine.
These ollaborations have allbeen ofgreat interestto meandextendedmy ulture in the
eld. Animportant partofthe results presented in this do ument arefruitsof thesejoint
works.
Jetiensàexprimermaprofondere onnaissan eàmafamillesurl'amouretla onan e
de laquellej'aitoujours pu ompter.
J'adresse une pensée tout parti ulière à Hélène sans l'éternel soutien de qui je ne me
seraisprobablement paslan é dans e projet.
Enn, estroisannées n'auraient étéaussienri hissantes et animéessans les ollègues
etamisren ontrésauLSS,etplusparti ulière mentAlessandrodiRÔma,elSuperCabron,
Fernandinetta , Haïowen et Islem. Mer i enn à Vin e, mon ollo ' et ami, pour me sup-
porter même en période de réda tion!
Preliminary remark
Thisdo ument synthesizesthe resear h works ondu ted byA.Chaillet, under the super-
visionofDr. A.Loría,inordertoobtainthePhDdegreefromUniversitéParisSud. Please
notethata Fren hversionis alsoavailableon request.
For anyremark,question or omment, please feelfreeto onta tthe author at:
antoine hailletyahoo.fr.
Contents
Remer iements - A knowledgement 5
Preliminary remark 7
Preamble 11
Contribution of this thesis 15
Publi ations 17
Notation 19
1 Denitions 21
1.1 Stabilityofthe origin . . . 22
1.2 Stabilityofsets . . . 25
1.3 Semiglobal and pra ti al asymptoti stability . . . 28
1.4 Input toState Stability . . . 31
2 Semiglobal and pra ti al asymptoti stability 33 2.1 Su ient onditions . . . 36
2.1.1 Globalpra ti al stability. . . 38
2.1.2 Semiglobal pra ti al stability . . . 45
2.1.3 Semiglobal asymptoti stability . . . 51
2.2 Converse results. . . 55
2.2.1 Semiglobal pra ti al stability . . . 57
2.2.2 Semiglobal asymptoti stability . . . 63
3 Stability of nonlinear time-varying as aded systems 65 3.1 Semiglobal pra ti al asymptoti stabilityof as aded systems . . . 68
3.1.1 WithaLyapunovfun tionfor the driven subsystem . . . 68
3.1.2 Withouta Lyapunovfun tion for the driven subsystem. . . 73
3.2 Semiglobal asymptoti stabilityof as aded systems . . . 77
3.2.1 WithaLyapunovfun tionfor the driven subsystem . . . 77
3.2.2 Withouta Lyapunovfun tion for the driven subsystem. . . 78
3.3 Globalpra ti al asymptoti stabilityof as aded systems. . . 79
4 Set-stability 95
4.1 Prelimina r ydenitions andtools . . . 98
4.2 Onset-stabilityof as aded systems . . . 101
4.3 Example: ross-tra kformation ontrol ofundera tuated surfa evessels . . 106
5 Integral inputto state stability for as aded systems 113 5.1 Globalasymptoti stabilityfor as ades, Lyapunov-based . . . 116
5.2 Integral input to state stabilityfor as ades, Lyapunov-based . . . 120
5.3 Integral input to state stabilityfor as ades, traje tory-based . . . 126
6 Appli ation to me hani al systems 135 6.1 PID ontrol of robot manipulator s . . . 135
6.1.1 Robustnesswith respe tto external disturban es . . . 137
6.1.2 PID ontrol onsidering a tuators'dynami s with disturban es . . . . 141
6.1.3 Experimental results . . . 144
6.2 Spa e raft formation . . . 146
6.2.1 Problemformulation . . . 147
6.2.2 Measurements available . . . 149
6.2.3 Whenonly bounds areknown . . . 151
6.2.4 Simulation results . . . 153
6.3 Underway shipreplenishment . . . 155
6.3.1 Preliminaries . . . 156
6.3.2 Virtualvehi ledesign . . . 160
6.3.3 Follower vehi ledesign . . . 161
6.3.4 Stabilityanalysisof the overallsystem . . . 162
6.3.5 Simulation study . . . 163
Con lusion and further resear h 169 A Proof of auxiliary results 171 A.1 Proofof ofLemma 2.7 . . . 171
A.2 Proofof Proposition1.16. . . 172
A.3 Proofof Theorem 3.38 . . . 174
A.4 Proofof Corollary 3.40 . . . 176
A.5 Proofof Theorem 3.42 . . . 177
A.6 Proofof Claim 6.4 . . . 178
A.7 Proofof Claim 6.5 . . . 178
Referen es 194
Preamble
Before introdu ing, in amore detailed manner,the subje tof study ofthe present thesis,
we give, in informal terms, a on ise overview of the motivations for this work and its
ontribution s.
Stability. Roughlyspeaking,stabilityisthepropertyofadynami alsystemthatanyerror
signals an be madearbitrarilysmall providedthatthe initialerrorsaresu iently small.
It is a ru ial notion froma ontrol point of view as itensures an a eptable behavior of
the plant if its initial onguratio n is not too far from the nominal one. If, in addition,
theerrorsignalseventuallytendtozero, wesaythatthisoperatingpointisasymptoti ally
stable. The domain ofattra tion onsiststhe set ofall initial states fromwhi h solutions
gotozero. Wetalkabout globalasymptoti stabilitywhenthedomainofattra tionisthe
whole state-spa e.
Obsta les. Through intuitive examples,weexposesome ofthe reasonsthatmayprevent
the error signals from onverging to zero, as, for instan e, the presen e of an external
perturbation, measurement impre ision, fri tion, et . In the same way, we showthat the
domain of attra tion may be restri ted to a ompa t neighborhood of the origin, notably
in the ase of negle ted high order nonlineariti es. In these situations, most of existing
toolsfail at ensuringbetter than ultimate boundedness ( onvergen e ofsolutions to some
neighborhood of the origin)or to lo alstability(restri tion ofthe domain of attra tion).
Semiglobal and pra ti al stability. Su h adegradation of performan e is not a ept-
ableinmany on reteappli ations,asthismayresultinatoolittleoperatingbandwidthor
atoolarge impre ision. Nevertheless, for ontrolledsystems,thedomain ofattra tion an
often be arbitrarily enlarged provided su iently large gains. We refer to this property
as semiglobal asymptoti stability. For a given system, semiglobal asymptoti stability
ensuresmu hmore interesting propertiesthan simplylo al properties, sin eitestablishes
thatnotheoreti alobsta lepreventsfromin ludinganygivennitesetofinitial onditions
to the domain ofattra tion.
Inthesameway, thesteady-stateerrors anoftenbediminishedatwillunderasimilar
tuningofthegains: we allthispropertypra ti alasymptoti stability. Again,this on ept
shouldbeseenasafarstrongerpropertythanthesimpleultimateboundednessofsolutions.
Indeed,pra ti al asymptoti stabilityimposesthatthe pre ision, afterthe transients, an
be made as ne as desired. In addition, as we will see in more details in the sequel, it
suggestsareasonablebehaviorofthetransientdynami swhi hisnotthe ase,ingeneral,
of ultimate boundedness.
Whenitfollowsfromalimitationoftheperforman esofthesystem,semiglobalpra ti-
external perturbations and model un ertainty. But, aswe will see, this stability property
doesnotariseex lusivelyfromadegradationofglobalasymptoti properties,butmayalso
be established byexistingresults in the literature, su h asaveragingte hniques.
A rst goal of the present thesis is to provide a rigorous framework for the study
of semiglobal and/or pra ti al asymptoti stability properties. To this end, we provide
su ient onditions,expressedintermsofLyapunovfun tions,thatguaranteethisstability
properties. Aswewill see, these onditions allowthatthe Lyapunov fun tion depends on
thetuningparameter. Indeed,su ha hoi eallowsto in ludeinouranalysisamu hwider
s ope of appli ations than if the Lyapunov fun tion was assumed uniform in the tuning
parameter. Forinstan e, forme hani alsystems,itis ommon to hoosethe energyofthe
system asa Lyapunov fun tion, in whi h ase the ontrol gains, then playing the role of
the tuning parameter, naturally appear in the Lyapunovfun tion.
Due to this non-uniformity, ompared to lassi al results, an additional assumption is
required that links the upper and lower bounds on the Lyapunov fun tion. We infer the
ne essityof su han additional onditionthrough anexample.
Westress thatthe stability on ept thatwe will usealongthe do ument makesuseof
two measures: the distan e to the ball for whi h we want attra tivity and the Eu lidean
norm. Here also,this hoi e is motivated by simpli ityand generality reasons. Indeed, if
wehad hosentouseonesinglemeasure,the orrespondingLyapunovfun tionwouldthen
have had to vanish on a whole neighborhood of the origin, whi h would have prevented
the use of the Lyapunov fun tion asso iated to the nominal system. On the opposite,
with this hoi e, most of the semiglobal or pra ti al stability properties that result from
a degradation of an ideal system due to perturbations an be inferred by using the
Lyapunovfun tion of the unperturbed system.
Hen e, we propose tools that allow to establish powerful stability properties, i.e.
semiglobaland/orpra ti alstability,whi husuallydonotrequiremu hmore onservative
assumptions than thoseneeded for the (weaker but ertainlymore lassi al) properties of
ultimate boundedness andlo al stability.
We also present a so- alled onverse theorem for semiglobal pra ti al asymptoti
stability, i.e. a result that guarantees the existen e of a Lyapunov fun tion under the
assumptionofsu hastabilityproperty. Thegeneralityofthe on eptthatweuserequires
spe i pre autions ompared to the properties that would be uniform in the tuning pa-
rameter. We will see that, asthis result generatesan autonomous bound on the gradient
ofthe generated Lyapunovfun tion,itwill be of greathelpin lightening the assumptions
in our results on as ades.
Cas ades. In orderto simplify the study ofa omplex system,it is ommon, in stability
analysis, to divide it into smaller inter onne t ed subsystems. In this way, the di ulty
of the analysisis often redu ed. A parti ular type of su h inter onne tion is the as ade
stru ture. In this situation, the subsystems are inter onne ted in a unilateral way, i.e.
the output of a driving subsystem is the input of a driven subsystem. The modularity
oered by this so- alled as ade approa h gave rise to powerful results, both in analysis
and ontrol design.
However,most ofthe existingresultsin thisdomain onlytreat lo alorglobal stability
of the origin. Hen e, they do not apply to the on epts, although ommon and powerful,
of semiglobaland/orpra ti al stability.
Asa se ondobje tive,we provide su ient onditions under whi hsemiglobal and/or
ing, we show that this is the ase provided that we expli itly knowa Lyapunovfun tion
for the driven subsystem and that the solutions are uniformly bounded. In the ase of
global pra ti al asymptoti stability, we provide a stru tural riterion to ensure this uni-
form boundedness of solutions,therefore yielding an easy-to- he k ondition to guarantee
global pra ti al asymptoti stability of the as ade. An illustration of these results on-
sists in stabilizing, by a bounded output feedba k, the double integrator ae ted by a
persistently ex iting signal. Asanother appli ation, we rigorouslyshowthatsmoothing a
feedba k ontrol lawmayresult in pra ti alstability. Furthermore,weprovide a onverse
Lyapunovresult for semiglobalpra ti al asymptoti stabilitythat permits us to relaxthe
requirement ofexpli itlyknowingaLyapunovfun tion forthe drivensubsystem. Asillus-
tratedbyanexample,this latterfeature isparti ularly usefulwhenstabilityisestablished
basedon averaging te hniques.
Set-stability. Thegeneralityoeredbyset-stabilitymakesit,aswefurtherdevelopupon,
another interesting tool for the stability and robustness analysis of perturbed systems.
Indeed, this notion in ludes, as parti ular ases, the stability of a single operating point,
of a traje tory or even a more omplex domain a ording to the set that is onsidered.
Moreover, asthe latter isnot assumed to be ompa t, itis also possible to in lude to the
study the partialstability, whi h refersto the situation when the behaviorof only a part
of the state is onstrained. We will see that the latter appears very useful when dealing
with adaptive ontrol.
The thirdobje tive ofthis workis to providesu ient onditions forthe preservation
ofthe set-stabilityfor as adedsystems. Therequirement isrstgivenasaglobal bound-
edness of the solutions of the overall as ade. We establish that, in some situations, this
anberelaxedtojustforward ompletene ssprovided agrowthrestri tiononthe inter on-
ne tion term. Asan illustrative appli ation,wepropose aproof for are ently established
resultin marine ontrol.
ISS and iISS. So far, we have dis ussed Lyapunov stability for systems without inputs.
A eld of stability analysis, regrouped under the paradigm of input to state stability
(ISS), isespe ially on erned bythe impa tof external signalson the performan e ofthe
system. Without going into details, this property imposes that the norm of the urrent
statebeboundedbyafun tionoftheamplitudeoftheperturbingsignalplusafadingterm
dependingoninitial onditions. Arelaxedextensionofthispropertyis alledintegralinput
tostatestability(iISS).Insteadoftheamplitudeoftheexternalsignal,thispropertytakes
into a ount the energy that the latter feeds to the system. The iISS property is very
general in stability analysis and provides interesting information about the system. For
instan e,iftheinputenergyisnite,thenthestate onvergestozero. Inthissense,iISS(as
well asISS) therefore onstitutesanother powerfulmeasure of the robustness of asystem
to external perturbations.
A fourth part of this text is devoted to the behavior of iISS systems when pla ed in
as ades. Weprovide elementary onditions underwhi h the as ade omposedofaniISS
systemdrivenbyaglobally asymptoti allystableoneremainsglobally asymptoti allysta-
ble. These onditions areexpressed in termsof theLyapunovfun tionsasso iatedto ea h
subsystem,thus generalizing existingtraje tory-based results. Under mildly onservative
additionalassumptions, weestablishthatthe as ade oftwoiISS subsystemsisitselfiISS.
Thislatterresultisrstlyexpressed in termsofLyapunovfun tions, andthen in termsof
Appli ations. Many of the results presented in this do ument have been applied in
pra ti e, and we exposesome ofthese results in afth step.
We study the robustness of PID- ontrolled robot manipulator s to fri tions, external
disturban es, model un ertainty and taking into a ount the dynami s of a tuators. We
prove that, under these environmental onstraints, the system is semiglobally pra ti ally
asymptoti allystable. Thisis onrmed byexperimentalresults.
In an other domain, we show that the leader-follower strategy adopted for the on-
trol ofspa e raftformations yields globalasymptoti stabilitywhenall measurementsare
available. However,inpra ti e,someinformationontheleader'spositionmaynotbeavail-
able. We showthat, provided that these signalsarebounded, global pra ti al asymptoti
stability an be on luded.
Finally,in the ontextofunderwayshipreplenishment,wherethe ontrolofthesupply
vesselaimsat preservinga onstant distan efromthe mainshipduringtheoperation,the
only measurements available for the main ship areposition and heading. No information
onitsmodelisatdisposal. Underthis onstraints,weshowthatavirtualvehi leapproa h
ensuresglobal pra ti al asymptoti stabilityof the system.
We eventually stress that, although some of the results presented here impose rela-
tively heavy notations for the sake of rigor, the do ument also aims at giving intuitive
explanations of the utilized on epts. In this dire tion, we give several simple examples
to illustrate the purpose and, when possible, provide simplied orollaries that are less
general but easierto usein pra ti e.
Also, even though the results presented along this do ument on ern more stability
analysisthan stabilization,in the sense thatnoexpli itdesign of ontrol lawispresented,
theystill onstituteapres riptiveframeworkonwhi hone ouldbase ontroldesignstrate-
gies,asillustrated bythe on rete appli ationsof Chapter6.
Contribution of this thesis
We briey summarizethe main results of this thesis, hapterby hapter, and ite related
publi ations. Labels orrespond to the listof publi ationspresented in p. 17.
- Chapter2: We present new toolsfor the study of semiglobal and pra ti al stability
ofnonlinear time-varyingsystems. Somesu ient onditions, interms ofLyapunov
fun tions, are proposed. Compared to lassi al Lyapunov onditions, an additional
onditionappears,thattakesintoa ount thenon-uniformityoftheLyapunovfun -
tion in the tuning parameter. We underline the ne essity of su h a requirement
throughanexample. Conversely,weprovethatsu haLyapunovfun tion anbede-
rived providedsu ient regularityofthe right-handside ofthe ordinary dierential
equation.
This hapterformed the subje t ofthe following publi ationswith A. Loría:
[(i), (ii), (iv), (viii), (x),(xiii), (xvii)℄.
- Chapter 3: We extend the results of Chapter 2 to nonlinear time-varying systems
presenting a as ade stru ture. We prove that, under a boundedness ondition on
thesolutions ofthe overallsystem,bothsemiglobalandpra ti alstabilityproperties
arepreserved bythe as ade inter onne t ion of two subsystems. We also givesome
su ient onditionstoensurethe boundedness onditiononthesolutions,whi hare
parti ularly easytouseinthe aseofglobalpra ti al stability. Illustrativeexamples
areprovided in ea h ontext.
Theseresults wereoriginally presented in the following publi ationswith A.Loría:
[(i), (ii), (iv), (viii), (x),(xiii), (xv), (xvii)℄.
- Chapter4: Weanalyzethebehaviorofnonlinearsystemsthataregloballyasymptot-
i allystablewith respe ttoa(nonne essarily ompa t)set,whenpla edin as ade.
We provide su ient onditions under whi h set-stability, dened with respe t to
two measures,ispreserved bythe as ade inter onne tion.
Theseworks orrespond to the ollaboration [(xiv)℄ with E. Panteley. An extension
wasproposedwith the same oauthor, J.Tsønnas and T.A.Johansen in [(xix)℄.
- Chapter5: Westudythe preservationoftheintegralinputtostatestabilityproperty
of nonlinear time-invariant systems in as ade. We give some su ient onditions
for the as ade omposed of an iISS subsystem driven bya globally asymptoti ally
stable(GAS) subsystemto be GAS.These onditions are expressed in termsof the
traje tory-based results. We also provide onditions under whi h two iISS systems
pla ed in as ade remain iISS. These su ient onditions are rst expressed based
on the Lyapunovfun tion forea h of the twosubsystems, andthen on the estimate
oftheir solutions.
Theseresults wereprepared with D.Angeli in [(iii), (xii)℄.
- Chapter6: We present on rete appli ationsof our maintheoreti al ndings insta-
bilizationproblems ofme hani al systems. Weshowthat, whentakinginto a ount
external perturbation s (su h as fri tion, torque ripping, et . ) and the dynami s of
the a tuator, PID- ontrolled robot manipulator s aresemiglobally pra ti ally stable,
ba ked upwith experimentalresults.
Onthe otherhand, a ontrol for underwayfuel replenishment ofvesselsis designed,
using a virtual ship approa h, whi h requires neither a priori model knowledge nor
velo ity measurement for the ship to be replenished. Global pra ti al asymptoti
stabilityis obtained.
Athird appli ation on erns the ontrol ofa spa e raftformation, whentakinginto
a ountboundedexternaldisturban es. A ordingtotheassumedlevelofknowledge
we have on the orbital parameters of the leader, various stability properties are
derived.
Theseappli ations werethe obje tof the following joint publi ationswith R.Kelly,
E.Kyrkjebø,R.Kristiansen,A.Loría,E.Panteley,K.PettersenandP.J.Ni klasson:
[(iv), (vi) , (vii), (xvi), (xviii)℄.
Although not presented in this do ument, these three years of PhD gave rise to other
fruitful ollaborations:
- The publi ation [(ix)℄ is a joint work with J. de León Morales, A. Loría and G.
Besançon where we proposed an adaptive observer for systems that an be put in
the so- alled output feedba k form, based on a onvenient persisten y of ex itation
property.
- With A. Loría, G. Besançon and Y. Chitour, we have posed open problems for
stabilization of persistently ex ited systems, and partially solve them in the ase of
the doubleintegrator, f. [(xi)℄.
- In[(xx)℄,with M. Sigalotti,P. Mason, Y.Chitour and A.Loría, thislatter problem
was further extended and solved with a linear time invariant feedba k, with gains
uniform in the persistently ex iting signal.
Publi ations
Thefollowingisanexhaustivelistofpubli ationswrittenduringthepastthreeyears,that
areeither published, a epted for publi ation,or stillunder review. It ontains but isnot
restri tedto the ontents ofthis do ument.
Journal Papers
i/ A.ChailletandA.Loría,A onversetheoremforuniformsemiglobalpra ti al asymp-
toti stability: appli ation to as aded systems,A epted for Automati a,2006.
ii/ A.Chaillet and A. Loría,Uniform semiglobal pra ti al asymptoti stability for non-
autonomous as aded systems and appli ations, A eptedfor Autmati a, 2006.
iii/ A.ChailletandD.Angeli,Integralinputtostatestablesystemsin as ade,Submitted
to SystemsandControl Letters, 2005.
iv/ A. Chaillet and A. Loría, Uniform global pra ti al asymptoti stability for non-
autonomous as aded systems, Submitted to European Journal of Control, 2006.
A.Chaillet, A. Loríaand R.Kelly, Robustness of PID- ontrolled manipulators vis a
vis a tuator dynami sand external disturban es,Submitted to European Journal of
Control, 2006.
Book hapters
vi/ R.Kristiansen, A.Loría, A. Chaillet, and P. J. Ni klasson, Output feedba k ontrol
of relative translation in a leader-follower spa e raft formation, In Pro eedin gs of
the Workshop on Group Coordinatio n and Cooperative Control, Le ture Notes in
Control and InformationS ien es, Springer Verlag,Tromsø,Norway, 2006.
vii/ E.Kyrkjebø,E.Panteley,A.Chaillet,andK.Pettersen,Avirtualvehi leapproa hto
underwayreplenishment ,InPro eedin gsoftheWorkshoponGroupCoordinatio nand
Cooperative Control, Le ture Notes in Control and Information S ien es, Springer
Verlag,Tromsø, Norway, 2006.
viii/ A.Chaillet, Tools for semiglobal pra ti al stability analysis of as aded systems and
appli ations, InPro eedin gs oftheCTSHYCONWorkshop,International S ienti
and Te hni al En y lopedia, Paris,Fran e, July 2006.
Conferen e papers
ix/ J.de LeónMorales, A.Chaillet, A.Loría,andG. Besançon,Output feedba k ontrol
viaadaptiveobserverswithpersisten yofex itation,InIFACWorldCongress,Praha,
T he kRepubli , July 2005.
x/ A.Chaillet and A. Loría,Uniform semiglobal pra ti al asymptoti stability for non-
linear time-varying systems in as ade, In IFAC World Congress, Praha, T he k
Republi ,July 2005.
xi/ A. Loría, A. Chaillet, G. Besançon, and Y. Chitour, On the PE stabilization of
time-varying systems: openquestions andpreliminary answers, InCDC2005, pages
68476852, Sevilla, Spain,De ember 2005.
xii/ A. Chaillet and D. Angeli, Integral input to state stability for as aded systems, In
MTNS2006, Kyoto,Japan, July 2006.
xiii/ A.Chailletand A.Loría,Uniformglobal pra ti al stability for non-autonomous as-
aded systems, InMTNS2006, Kyoto,Japan, July 2006.
xiv/ A. Chaillet and E. Panteley, Stability of sets for nonlinear systems in as ade, In
MTNS2006, Kyoto,Japan, July 2006.
xv/ A.Chaillet, Estimée du domained'attra tionpourla as ade desystèmes uniformé-
ment asymptotiquement stables, In CIFA 2006,Bordeaux, Fran e, June 2006.
xvi/ A.Chaillet, A.Loría, andR.Kelly, Robustness of PID ontrolled manipulators with
respe t toexternal disturban es,InCDC 2006,San Diego,USA, De ember 2006.
xvii/ A. Chaillet and A. Loría, A onverse Lyapunov theorem for semiglobal pra ti al
asymptoti stability and appli ation to as ades-based ontrol, In CDC 2006, San
Diego,USA, De ember 2006.
xviii/ R.Kristiansen,A.Loría,A.Chaillet, and P. J.Ni klasson,Adaptive outputfeedba k
ontrol of spa e raft relative translation, In CDC 2006, San Diego, USA, De ember
2006.
xix/ J. Tsønnas, A. Chaillet, E. Panteley, and T. A.Johansen, Cas aded lemma for set-
stable systems, Submitted toCDC 2006,San Diego,USA, De ember 2006.
Other reports
xx/ M.Sigalotti, A.Chaillet,P. Mason,Y.Chitour, andA.Loría,Lineartime-invariant
stabilizationwith persisten y of ex itation, Internal report, 2006.
xxi/ A.Chaillet and A. Loría,Uniform semiglobal pra ti al asymptoti stability for non-
autonomous as aded systems and appli ations, 2006,
http://arxiv.org/PS_ a he/math/pdf/0 503/0503039.pdf.
Notation
All properties su h as positive, greater, in reasing, et . are to be understood in the
stri t sense.
N
andR
denote the sets ofall nonnegative integers and allreal numbers respe tively.N ≤N
ontains all the nonnegative integers lessthan or equal toN ∈ N
. In the same way,R ≥0
is omposedof allnonnegative realnumbers.I
denotes the identitymatrix of appropriatedimension.A ontinuous fun tion
α : R ≥0 → R ≥0
is of lassK
(α ∈ K
), if it is in reasing andα(0) = 0
. It is said to belong to lassK ∞
if, in addition,α(s) → ∞
ass → ∞
. Aontinuous fun tion
σ : R ≥0 → R ≥0
is of lassL
(σ ∈ L
) if it is de reasing and tends tozero as its argument tends to innity. A fun tion
β : R ≥0 × R ≥0 → R ≥0
is said to be alass
KL
fun tion ifβ( ·, t) ∈ K
for anyt ∈ R ≥0
, andβ(s, ·) ∈ L
for anys ∈ R ≥0
.We denote by
φ( ·, t 0 , x 0 )
the solutions of the dierential equation˙x = f (t, x)
withinitial ondition
φ(t 0 , t 0 , x 0 ) = x 0
.We use
|·|
for the Eu lidean norm of ve tors and the indu edL 2
norm of matri es.We use
k · k
for the essential supremum norm, i.e., for a signalu : R ≥0 → R p
,kuk :=
ess
sup t≥0 |u(t)|
.We denote by
B δ
the losed ball inR n
of radiusδ
entered at the origin, i.e.B δ :=
{x ∈ R n : |x| ≤ δ}
. We use the notationH(δ, ∆) := {x ∈ R n : δ ≤ |x| ≤ ∆}
. By anabuseof notation,
B 0 = H(0, 0) = {0}
andB ∞ = H(0, ∞) = R n
.δ
being anonnegative onstant,we dene|x| δ :=
infz∈B δ |x − z|
. Moregenerally, for alosed set
A
,|·| A
representsthe distan eto this set:|x| A :=
infz∈A |x − z|
.For agiven set
E
ofR n
,E ◦
denotes itsinterior.Let
a ∈ {0, +∞}
andq 1
andq 2
be lassK
fun tions. Wesaythatq 2 (s) = O(q 1 (s))
ass
tendstoa
ifthereexistsanonnegative onstantk
su h thatlim sup s→a q 2 (s)/q 1 (s) ≤ k
.We say that
q 2 (s) = o(q 1 (s))
) ifk
an be taken to be zero, and thatq 1 (s) ∼ q 2 (s)
iflim s→a q 2 (s)/q 1 (s) = 1
.We say that
f : R ≥0 × R n → R n
satises the Carathéodory onditions iff ( ·, x)
ismeasurable forea h xed
x ∈ R n
,f (t, ·)
is ontinuousfor ea hxedt ∈ R ≥0
and,for ea hompa t
U
ofR ≥0 × R n
, there exists a integrable fun tionm U : R ≥0 → R ≥0
su h that|f(t, x)| ≤ m U (t)
for allall(t, x) ∈ U
.Afun tion
f : R n → R n
issaidtobelo allyLips hitz if,forany ompa tU
ofR n
,there existsanonnegative onstantk U
su h that|f(x) − f(y)| ≤ k U |x − y|
for all(x, y) ∈ U 2
.When the ontext issu iently expli it,wemayomit the arguments ofa fun tion.
Chapter 1
Denitions
Lyapunov stability. The works presented in this do ument appeal to many dierent
typesof stabilityproperties. Stabilityshould be understood in the Lyapunovsense. Gen-
erally speaking,itrefersto thepropertyofa point,a setor atraje torythatanysolution
starting su iently nearremains arbitrarily lose at all time. It onstitutes a ru ial fea-
ture in ontrol of dynami al systems, as it ensures an a eptable behavior of the plant
provided thatits initial onditions aresu iently lose to the nominal ones.
The notion of stability may easily be grasped in the ontext of me hani al systems.
Considering a ball on a non at surfa e, an equilibrium position is stable if, after any
su iently small perturbations on the position of the ball, itremains for ever arbitrarily
near to it. The equilibrium is said to be asymptoti ally stable if, in addition, the ball
approa hes itasymptoti ally. This isillustrated bythe drawings ofFigure 1.1.
Instability Asymptotic
stability
Global
asymptotic
stability
Figure1.1: Illustrationof dierent typesof stability.
In some situations, it is interesting to know how far from the asymptoti ally stable
equilibrium the ball an start and nally return to it. The region of the state spa e
that leadasymptoti onvergen e isreferred to asdomain of attra tion. Ifthe domain of
attra tion is the whole spa e, then the equilibrium under onsideration is alled globally
asymptoti allystable.
While very intuitive in the ontext of me hani al systems, Lyapunov stability is far
from being onned to this area. Generally speaking, the systems onsidered throughout
the do ument arerepresented asanite dimensionaldierential equationof the form
˙x = f (t, x) ,
(1.1)where
t ∈ R ≥0
representsthetime,x ∈ R n
isthe stateandf : R ≥0 × R n → R n
isassumedto satisfy Carathéodor y onditions ( f. p. 19) and to be lo ally Lips hitz in
x
. Morepre isely, for ea h ompa t
U
ofR ≥0 × R n
, we assume that there exists an integrablefun tion
k U : R ≥0 → R ≥0
su h that,for all(t, x) ∈ U
andall(t, y) ∈ U
,|f(t, x) − f(t, y)| ≤ k U (t) |x − y| .
By virtue of [Hal69, Theorem 5.3℄, these ombined onditions ensure both existen e and
uniqueness of the solutions of(1.1).
It is worth pointing out the wide varietyof systemsthat an be des ribed by su h an
equation. To ite a few, it overs a very large number of ontrol problems in me hani s,
ele tri alsystems,biology, ele troni s,ele tro-mag net i s,et . , f. e.g. [OLNSR98,KSV91,
Son05b℄.
Why time-varyingsystems ? Thefa tthatthe right-hand sideterm ofthe onsidered
dierential equation is time-dependent allows to in lude in the study many problems of
traje torytra king. Thisaims at designinga ontrol
u
in su h awaythat the solution ofthe the dynami al system
˙x = f (x, u)
follows asymptoti ally apres ribed referen ex d (t)
.As the adopted ontrol law depends on the time-varying referen e traje tory
x d (t)
, thesystemin losed-loop,although originally time-invariant,is ofthe form
˙˜x = g(t, ˜x)
, where˜
x := x − x d
. Thesostated tra king ontrol problemapplies to manyphysi al systems,asfor instan e in the area of ontrol of me hani al and ele trome hani al systems ( f. e.g.
[OLNSR98℄ andreferen es therein).
Another typi al situation in whi h expli it time-dependen e of the dynami al system
o ursis thatof regulationsproblems (thatis, stabilizationof xedoperating point)that
donotsatisfyBro kett's ondition[Bro83℄orthemore onservative onditionpresentedby
Coron in[Cor90℄. Inthis ase, the open-loop plant isnot stabilizable byany ontinuously
dierentiable time-invariant feedba k. A time-varying ontroller is then on eivable. For
instan e,itwasshownbyCoronin[Cor92℄thatany ompletely ontrollablesmoothsystem
without drift(in luding nonholonom i me hani al systems) an be stabilizedbymeans of
a smooth periodi time-varying state feedba k. Also, it was shown in [KT03℄ that, if a
system an be stabilized by a ontinuous state-feedba k, then itis stabilizable (although
possibly in anon-uniform way)byasmooth time-varyingfeedba k,whi h may onstitute
an interesting featurefor some appli ations.
However, this expli it dependen e in time of (1.1) an also be of interest from an
analysispoint ofview. Some te hniquesin the literature, seefor instan e[Kha96, Lor04℄,
onsistin simplifyinga omplexnonlinear systeminto amore simpletime-varying one by
onsidering partof the state asasimple fun tion oftime.
Although the results presented along this do ument on ern more stability analysis
thanstabilization, inthesensethatnoexpli itdesign of ontrollawispresented, theystill
onstitute a pres riptiveframeworkon whi h one ould base ontrol design strategies.
1.1 Stability of the origin
Many ontrol appli ations an be formulated asa stabilization problemof the origin of a
dynami al system. Typi ally, one requires thatthe error between the desired behavior of
the system and the a tual one onverges to zero, leading to the notion of attra tivity of
the dieren e between the desired behavior and the a tual one remains arbitrarily small
at alltime: in otherwords, stability ofthe origin is desired.
We briey re all these notions in a nonlinear time-varying ontext. In this se tion,
I
denotea losed(butnotne essarilybounded)subsetof
R n
that ontainstheorigin. Please referto the Notationpart(p. 19) fora denitionofthe mathemati al on epts usedhere.Westart with the notion of uniform boundedness of solutions.
Denition 1.1 (UB/UGB) Let
I
be a losed subset ofR n
. The solutions of (1.1) are said to be uniformlybounded onI
if, for any nonnegative onstantr
, there existsa non-negative
c(r)
su h that, for allt 0 ∈ R ≥0
, they satisfyx 0 ∈ I ∩ B r ⇒ |φ(t, t 0 , x 0 ) | ≤ c , ∀t ≥ t 0 .
If
I = R n
, then the solutions are uniformlyglobally bounded .Based on this, we an introdu e a pre ise denition of the stability on ept thatwill
be usedthroughout the do ument.
Denition 1.2 (US/UGS) Let
I
be a losed subset ofR n
. The origin of (1.1) is said to be uniformly stable onI
if its solutions are uniformly bounded onI
and, given anypositive onstant
ε
, there exists a positiveδ(ε)
su h that, for allt 0 ∈ R ≥0
, the solution of(1.1) satises
|x 0 | ≤ δ ⇒ |φ(t, t 0 , x 0 ) | ≤ ε , ∀t ≥ t 0 .
(1.2)If
I = R n
, then the origin is uniformly globally stable.Stri tly speaking,stabilityofthe origin is apurely lo al on ept whi his summarized
by(1.2). Inmanyappli ations,itisalsointerestingtoknowadomainfromwhi hsolutions
remainbounded,whi h explains why aboundedness requirement is imposed in the above
denition.
Next, we re allthe notion ofattra tivity ofthe origin.
Denition 1.3 (UA/UGA) Let
I
be a losed subset ofR n
. The origin of (1.1) is said to be uniformlyattra tive onI
if, for all positive numbersr
andε
, there exists a positivetime
T (r, ε)
su h that, for allx 0 ∈ B r ∩ I
andallt 0 ∈ R ≥0
, the solutionof (1.1)satises|φ(t, t 0 , x 0 ) | < ε , ∀t ≥ t 0 + T .
If
I = R n
, then the origin is uniformly globally attra tive.When the two latterproperties are ombined, the resulting propertyis alled uniform
asymptoti stability.
Denition 1.4 (UAS/UGAS) Theoriginof(1.1)issaidtobe uniformlygloballyasymp-
toti allystableon
I
ifitisbothuniformlystableanduniformlyattra tive onI
. IfI = R n
,then the origin is uniformlyglobally asymptoti allystable.
The uniformity requirement in the above denitions refers to the initial time. It
orrespondstothe independen e of
δ
andT
int 0
. Inotherwords: nomatteratwhattimethe system'straje tories start, onvergen e-ra te to zeroandovershoot remainun hanged.
The importan e of uniformity. Uniformity is a ru ial property of time-varying sys-
tems, as it provides a ertain robustness with respe t to external disturban es. More
pre isely, as more detailed in [LLLP 05, Se tion 2.1℄, it an be shown that the uniform
asymptoti stability of the origin of a zero-input system
˙x = f (t, x, 0)
ensures stabilitywith respe t to onstantly a ting disturban es of
˙x = f (t, x, u)
, whereu
denotes an ex-ternal signal 1
, provided that the
f (t, x, u)
is lo ally Lips hitz inx
uniformly int
. Thison ept, alsoknownastotalstability,wasintrodu edbyMalkin, f. e.g. [Mal58℄. Itstates
that the traje tories remain arbitrarily small at all time ifthe initial state and the input
signalaresu iently small. Morepre isely, it isdened asfollows.
Denition 1.5 (Total stability) The origin of
˙x = f (t, x, 0)
is said tobe totally stableif, for ea h
ε > 0
, there existsδ(ε) > 0
su h that, for allt 0 ∈ R ≥0
, the solution of˙x = f (t, x, u)
satisesmax {|x 0 | , kuk} ≤ δ ⇒ |φ(t, t 0 , x 0 , u) | ≤ ε , ∀t ≥ t 0 .
In a nutshell, byestablishing uniform asymptoti stability, we guarantee thatthe be-
haviorofthe systemis not toomu h alteredbythe presen eofsu iently smallexternal
disturban es. This robustness property does not hold for non-uniform properties, as il-
lustrated by [LLLP 05, Example 2.1, p. 28℄ in whi h a simple s alar time-varying system
is exhibited with the following properties:
˙x = f (t, x, 0)
isglobally asymptoti ally stable (but not uniformly), nevertheless one an design an arbitrarily small perturbationu
insu h awaythat
˙x = f (t, x, u)
generatesunbounded solutions.Larger perturbations. While uniform asymptoti stability thus ensures a natural ro-
bustnessto small externaldisturban es, itprovidesno information onthe behaviorofthe
systemsubje t to larger perturbations. Instabilityanalysis, itis lassi al to observe that
the presen e ofa bounded non-vanishing disturban e impedes asymptoti stability, yield-
inginsteadthe onvergen etoa(possiblylarge)neighborhoodoftheoperatingpoint. This
propertyis referredto asultimateboundedness, f. e.g. [Kha01, Yos66℄.
Denition 1.6 (Ultimate boundedness) The solutions of
˙x = f (t, x)
are said to beuniformly ultimately bounded if there exist positive onstants
∆ 0
andc
su h that, forevery
∆ ∈ (0; ∆ 0 )
, there exists apositive onstantT (∆)
su h that, for allx 0 ∈ B ∆
andallt 0 ∈ R ≥0
, theysatisfy|φ(t, t 0 , x 0 ) | ≤ c , ∀t ≥ t 0 + T .
If this holds for arbitrarily large
∆
, then the solutions are globally uniformly ultimatelybounded .
Inmanysituations,thispropertyisnotenoughtoensure orre tperforman es. Indeed,
weseethatuniformultimateboundednessisonly on ernedwiththebehaviorofthesystem
afterasu ientlylongtimeand,hen e,doesnottakeintoa ountthetransientdynami s.
Inaddition,the domain towhi hsolutions onvergemaybe large,then preventingagood
pre ision.
The aim of the following se tions is to introdu e stability properties that may help
guaranteeing stronger featuresto perturbed systems.
1
u : R ≥0 → R m
may onsistinanymeasurablelo allyessentiallyboundedfun tion.1.2 Stability of sets
In the above denition of ultimate boundedness, solutions are required to onverge to
someball,ofradius
c
, enteredat theorigin. Itisnaturalto extend thispropertyto moregeneralsets. Inaddition,itisinterestingto onstrainthebehaviorofthesystemduringthe
transients, in orderto avoiddisproportioned overshoots. Thismotivatedthe introdu tion
of set-stability[Zub57, HP73℄.
A general on ept. The analysis of set-stability is very general and onsequently very
ommon in ontrol pra ti e. This ensues from the fa t that the set under onsideration
may onsistinasingleoperatingpoint(then orrespondingtoDenitions1.2,1.3and1.4),
a path,or a more omplex,possiblyunbounded, regionof the state-spa e.
As it will appear more learly in the following denitions, stability (and, similarly,
attra tivity)of asingleoperating point
x ∗
isobtainedby onsideringthe set{x ∗ }
. Inthisrespe t, we always onsiderthatthereferen epoint
x ∗
isthe origin. This an beassumedwithout lossof generality, sin e, if
x ∗
is an equilibriumfor (1.1),then0
is an equilibrium for˙z = g(t, z) := f (t, z + x ∗ )
with the oordinate hangez := x − x ∗
.In the same way, stabilityof a path maybe onsidered by hoosing the set ontaining
all the points ofthis path.
Inthe asewhenthe appli ationdoesnotrequire onvergen eto theorigin but justto
a small neighborhood of it, it is appropriate to onsider the set as a ball of small radius
entered at zero. This allows to dene a rigorousformulation ofthe problems for whi h a
steady-state error is tolerated, and isalso at the basisof pra ti al stabilityas we will see
in the next se tion.
The set may also be de omposed as
R n ′ × {0}
, withn ′ ∈ N <n
, whenonly partof thestate is required to be stable. We refer to this property as partial stability, f. [Vor98℄.
Many appli ations indeed require the onvergen e of a redu ed number of variables to
operate orre tly. This on epthasalsoproved usefulin presen eof superuous states,or
whenthe plantisinherently unstablewith respe ttopartofthe states. SeeChapter4for
details.
Thefollowing stabilitydenitionsshouldthereforebe seenasgeneral statements, from
whi h allthese parti ular ases maybederived.
When dealing with set stability, spe ial attention has to be paid to the existen e of
solutions forallpositivetime.
A
andI
denoting two losed (butnot ne essarilybounded) sets ofR n
that ontain the origin2
, we therefore start by re alling the notion of forward
ompleteness. Please see[AS99℄for a Lyapunov hara terizat ion of thisproperty.
Denition 1.7 (Forward ompleteness) The system (1.1) is said to be forward om-
plete on
I
if, for allx 0 ∈ I
and allt 0 ∈ R ≥0
, its solutionφ(t, t 0 , x 0 )
est dénie pour toutt ≥ t 0
.Based onthis,we anextendDenition 1.1to the asewhenwearenotinterestedin a
boundedness of the distan eof thesolutions from the origin,but froma given losed (not
ne essarily ompa t) set
A
.2
This assumption, whi h an be made withoutloss of generality, is imposed inorder to ensurethat
|·| A ≤ |·|
.Denition 1.8 (UB/UGB with respe t toa set) The solutions of (1.1) are said to
be uniformly bounded on
I
with respe t toA
if (1.1) is forward omplete onI
and, forany nonnegative onstant
r
, there exists a nonnegativec(r)
su h that, for allt 0 ∈ R ≥0
,they satisfy
x 0 ∈ I ∩ B r ⇒ |φ(t, t 0 , x 0 ) | A ≤ c , ∀t ≥ t 0 .
If
I = R n
, then the solutions are uniformly globally bounded with respe t toA
. Fur-thermore, for the ase that
A = {0}
andI = R n
we simply say, with a slight abuse ofterminology, that the solutions of (1.1)are uniformlyglobally bounded .
Inthe asewhen
A = {0}
,were overuniform boundednessasintrodu edin Denition 1.1. We see that an additional requirement, namely forward ompletene ss, is imposed inthe above denition. As the set
A
may be unbounded, traje tories mayexplode in nite time while the quantity|φ(t, t 0 , x 0 ) | A
remains bounded at all time. Assuming forwardompleteness ex ludes this possibility. It should be stressed that, in the ase when
A
is aompa t set, this additional requirement is not needed anymore. These remarks hold as
well for the nextthree denitions.
Denition 1.9 (US/UGS of a set) Assume that (1.1) is forward omplete on
I
. Theset
A
issaidtobe uniformlystableonI
for(1.1)ifthesolutionsofthelatterare uniformlybounded on
I
with respe t toA
and, given any positive onstantε
, there exists a positiveδ(ε)
su h that, for allt 0 ∈ R ≥0
, the solutionof (1.1)satises|x 0 | ≤ δ ⇒ |φ(t, t 0 , x 0 ) | A ≤ ε , ∀t ≥ t 0 .
If
I = R n
, then the setA
is uniformlyglobally stable.Denition 1.10 (UA/UGA of a set) Assumethat(1.1)isforward ompleteon
I
. Theset
A
is said to be uniformly attra tive onI
for (1.1) if, for all positive numbersr
andε
, there exists a positive timeT (r, ε)
su h that, for allx 0 ∈ B r ∩ I
and allt 0 ∈ R ≥0
, thesolutionof (1.1)satises
|φ(t, t 0 , x 0 ) | A < ε , ∀t ≥ t 0 + T .
If
I = R n
, then the setA
is uniformlyglobally attra tive.Denition 1.11 (UAS/UGAS of a set) Assume that (1.1) is forward omplete on
I
.The set
A
issaid tobe uniformlygloballyasymptoti allystableonI
for (1.1)ifit isbothuniformly stable and uniformly attra tive on
I
. IfI = R n
, then the setA
is UniformlyGlobally Asymptoti allyStable .
Twomeasures. Itisworthpointingoutthatthesedenitions arespe ial asesofstability
withrespe ttotwo measures, f. [Mov60,LL93℄. This on eptisverygeneral andin ludes,
aswe have seen,stabilityofa singlepoint,of a ompa t set, ofa pres ribed traje toryas
well as partial stability [Vor98, Vor02℄. It was used in e.g. [LS76, TP00, Lee04℄. Here,
the rst measure is the distan e to the set under onsideration
|·| A
, while the se ond isthe Eu lidean norm
|·|
. As we will see later (see Se tion 2.1), for perturbed systems orwhendealingwithadaptive ontrol,this hoi eallows,in manysituations,tousethesame
Lyapunovfun tionasthenominalsystem,whi hmakesthisstabilitypropertymu heasier
In thisrespe t, we stressthatthe term uniform usedin the above denitions on erns
only the dependen e in the initial time. More pre isely, the onstants
c
,δ
andT
inDenitions1.8,1.9and1.10areallrequiredtobeindependentof
t 0
. Otherexistingresultsin the literature( e.g. [Yos66,LSW96,TPL02℄)usethisterminology to underlinethatthe
set-stabilityisdenedwiththesamemeasure,notablyimplyingthattheset
A
ispositivelyinvariant, whi his not the asehere.
As in the spirit of Hahn's formulations [Hah63℄ of stability in terms of
K
andKL
estimates (seealso [Son98a℄), the properties dened above an be written in the following
pre iseway.
Proposition 1.12 (
K
hara terization of UB/UGB) Assumethat(1.1)isforward om- plete onI
. The solutions of (1.1) are uniformly bounded onI
(resp. uniformly globallybounded)withrespe tto
A
ifandonlyifthereexistsa lassK
fun tionη
andanonnegative onstantµ
su h that, for anyx 0 ∈ I
(resp.x 0 ∈ R n
) and anyt 0 ∈ R ≥0
, the solution of(1.1) satises
|φ(t, t 0 , x 0 ) | A ≤ η(|x 0 |) + µ , ∀t ≥ t 0 .
Proposition 1.13 (
K
hara terization of US/UGS) Assumethat(1.1)isforward om- plete onI
. A losed setA
is uniformly stable onI
(resp. uniformly globally stable) for(1.1) if and only if there exists a lass
K
fun tionγ
su h that, for anyx 0 ∈ I
(resp.x 0 ∈ R n
)and anyt 0 ∈ R ≥0
,the solution of (1.1) satises|φ(t, t 0 , x 0 ) | A ≤ γ(|x 0 |) , ∀t ≥ t 0 .
Proposition 1.14 (
KL
hara terization of UAS/UGAS) Assume that (1.1) is for- ward omplete onI
. A losed setA
is uniformly asymptoti ally stable onI
(resp. uni-formlyglobally asymptoti ally stable)if andonlyifthere existsa lass
KL
fun tionβ
su hthat, for all
x 0 ∈ I
(resp.x 0 ∈ R n
)and allt 0 ∈ R ≥0
, the solutionof (1.1) satises|φ(t, t 0 , x 0 ) | A ≤ β(|x 0 | , t − t 0 ) , ∀t ≥ t 0 .
The proof ofthese hara terizations follows along the same linesas[Vid93, Theorems
53 and61℄, we thereforedo not re allthem here.
Whenthe onvergen eratetotheset
A
isexponentialandthedependen eintheinitial state islinear, the stabilityis saidto beexponential.Denition 1.15 (UES/UGES of a set) If, in Proposition1.14 , the lass
KL
fun tionan be pi ked as
β(s, t) = k 1 se −k 2 t , ∀s, t ∈ R ≥0
forsome positive onstants
k 1
andk 2
, thenthe setA
issaid tobe uniformlyexponentially stableonI
(resp. uniformlyglobally exponentially stable) withparameters(k 1 , k 2 )
.Forthestudy ofthealteration ofastabilitypropertyunder theinuen edisturban es,
anoteworthyparti ular aseof the above denitionsis whenthe sets under onsideration
are losed balls. It is indeed at the basisof all the denitions ofsemiglobal and pra ti al
stability properties introdu ed next. The following proposition, thatfollows from Propo-
sitions 1.12 and 1.14, establishes a strong linkof this on ept with the
(σ → ρ)
-stabilityProposition 1.16 (UAS and
σ → ρ
stability) Let∆ > δ > 0
. Then the followingimpli ations hold:
- If
B δ
is UAS onB ∆
, then (1.1) is(∆ → δ)
-stable;- If (1.1) is
(∆ → δ)
-stable, thenB δ
is UAS onB ∆ ′
, for all∆ ′ ∈ (δ, ∆)
.Theproofofthispropositionisdetailedin Se tionA.2 . We annoti e thatno forward
ompleteness assumption is needed anymore as the set under onsideration, namely
B δ
,is ompa t. In this ase, uniform asymptoti stability naturally ensures the existen e of
solutions for allforward time.
1.3 Semiglobal and pra ti al asymptoti stability
The need of a ner analysis. As already pointed out by Hahn in [Hah63℄ and byLa
Salle and Lefs hetz in [SL61℄, pra ti al onsiderations should be taken into a ount when
studyingtheasymptoti stabilityoftheequilibriumofagivenplant. Toquoteanexample
of thelatter referen e, the asymptoti stability ofan ele tri alsystemoperating at 110
V
ensuresthatsmallvariationswill be an elledout. However,iftheamplitude ofthesetol-
eratedvariationsistoolsmall,sayofsomemillivolts,thesystemmaynotoperate orre tly.
Onthe opposite, the operating point of a given system maybe mathemati ally unstable,
thusgeneratingsmallos illationsaround it,butstillguaranteeasu ientpre isionfor an
a eptable behavior. Using the intuitive illustration, already used in Figure 1.1, of a ball
on anon-at surfa e,these would orrespond to the following situations:
Asymptotic stability with a
small domain of attraction
Instability with a small
steady-state error
Figure1.2: Pra ti al onsiderations about stability.
A tighter analysisisthen apital.
Steady-state errors and restri ted domain of attra tion. As already noted, non-
vanishingperturbations a ting onthe plant ormeasurement impre isionsmayimpede the
onvergen e to theorigin byyielding asteady-state error. Inthe same way, itisoftenthe
ase thatsome negle ted high-order nonlinearityin the dynami s prevent global stability,
generating instead an unbounded basin of attra tion. In ea h of these situations, an we
In the stability analysis of losed-loop systems, but also in some ontexts that are
developed later (su h as averaging te hniques or output feedba k ontrol; see Chapter
2), the tuning of some free parameters (typi ally ontrol gains) often allow to arbitrarily
enlarge the domain ofattra tion, or to diminish atwill the magnitude ofthe steady-state
errors. Theseproperties arerespe tively referredto assemiglobal and pra ti al stability.
In moreformalterms, semiglobaland pra ti alstabilityproperties pertain to parame-
terizednonlinear time-varying systemsof the form
˙x = f (t, x, θ) ,
(1.3)where
x ∈ R n
,t ∈ R ≥0
,θ ∈ R m
is a onstant parameter andf : R ≥0 × R n × R m → R n
is lo ally Lips hitz in
x
and satisesCarathéodory onditions for any parameterθ
underonsideration.
Denition 1.17 (USAS) Let
Θ ⊂ R m
be a set of parameters. The system (1.3) is said to be uniformly semiglobally asymptoti allystable onΘ
if, given any∆ > 0
, there existsθ ⋆ (∆) ∈ Θ
su h that the origin is uniformly asymptoti ally stable onB ∆
for the system˙x = f (t, x, θ ⋆ )
.Denition 1.18 (UGPAS) Let
Θ ⊂ R m
be asetofparameters. Thesystem(1.3)issaid to be uniformly globally pra ti ally asymptoti ally stableonΘ
if, given anyδ > 0
, thereexists
θ ⋆ (δ) ∈ Θ
su h that the ballB δ
is uniformly globally asymptoti ally stable for the system˙x = f (t, x, θ ⋆ )
.Denition 1.19 (USPAS) Let
Θ ⊂ R m
be aset of parameters. The system(1.3)issaid tobe uniformlysemigloballypra ti allyasymptoti allystableonΘ
if,givenany∆ > δ > 0
,there exists
θ ⋆ (δ, ∆) ∈ Θ
su hthat the ballB δ
isuniformly asymptoti ally stableonB ∆
forthe system
˙x = f (t, x, θ ⋆ )
.In the above denitions,
θ
represents the tuning parameter, e.g. ontrol gains or anyfreedesign parameter.
Θ
is the set of allowed tuningparameters, whi h maybe bounded due to physi al onstraints su h as limitation of the output of a tuators.∆
an be seenas the radius of the estimate of the domain of attra tion; in most appli ations, a larger
∆
indu es better performan e sin e the operating bandwidth is enlarged. In ontrast,δ
represents the radius of the ball to whi h solutions ultimately onverge; therefore it istypi allyrequiredtobesmall,inordertoredu ethesteady-stateerrorasmu haspossible.
Pra ti al stability and ultimate boundedness. As it is further dis ussed in the
sequel (see Chapter 2), pra ti al stability shares similarities with the lassi al ultimate
boundedness property ( f. Denition 1.6), in the sense that solutions eventually rea h a
neighborhood of the operating point. It should however be lear to the reader that the
above Denitions 1.18 and 1.19 are usually more interesting in pra ti e, as they require
the size of this neighborhood to be redu ible at will by an adequate tuning and as they
requirethe ball
B δ
notonlyto beattra tivebutalso stable(inthe senseofDenition??).We also stress that Denitions 1.18 and 1.19 do not require the origin to be an equi-
librium for the system(1.3). This indeed failsfor many pra ti ally stable systemsas, for
instan e, Examples 2.2and 2.8givenbelow.
In viewof Proposition1.14, USPAS an be expressed in termsof