Università degli Studi di Roma Tor Vergata. Facoltà di Ingegneria


 Candice Jennings
 1 years ago
 Views:
Transcription
1 Università degli Studi di Roma Tor Vergata Facoltà di Ingegneria Corso di laurea in ingegneria dell Automazione Tesi di Laurea Magistrale On Quantization in Control Systems: Stabilization of Saturated Systems Subject to Quantization. Relatore Prof. Sergio Galeani Candidato Francesco Ferrante Corelatori Frédéric Gouaisbaut Sophie Tarbouriech Anno Accademico
2 On Quantization in Control Systems: Stabilization of Saturated Systems Subject to Quantization. Francesco Ferrante dell autore:
3 Ad Aldo e Luana, i miei genitori A Marta, mia sorella A Viola, il mio amore Coloro cui devo tutto.
4 ACKNOWLEDGEMENTS My gratitude goes to Luca Zaccarian, who allowed me to perform the activity which led to the present thesis. During this time, he constantly provided me valuable advices, being for me a mentor but mostly a friend. I am thankful to him since he supported my future career as PhD student, giving me the confidence. I would like to thank Sophie Tarbouriech and Frédéric Gouaisbaut, who provided me their experiences and their time during my activity at the LAASCNRS. I am very grateful to them, since they have introduced me to the research activity in automatic control and gave me the chance to go on, fulfilling my desire. I have learned a lot from their and I hope to learn much more in the future. I also thank all the members of the MAC group, who welcomed me in the best way, making me feel at home.
5 RINGRAZIAMENTI Arrivati alla fine di un percorso così lungo e così impegnativo, il capitolo dei ringraziamenti potrebbe superare di gran lunga la lunghezza di tutta la tesi! Per questo cercherò di riassumere, senza recar dispiacere a nessuno, o almeno lo spero. Voglio ringraziare per cominciare la mia famiglia, la quale mi ha permesso di giungere fino a questo punto, sopportandomi e supportandomi in ogni modo sempre ed incondizionatamente. Tra loro, ringrazio mio padre Aldo e mia madre Luana, i quali hanno permesso tutti ciò ed hanno sopportato il mio incessante brontolare. Un sentito grazie va a mia sorella Marta, del resto come non ringraziare espressamente colei che innumerevoli volte ha sentito parlare di esami durante le sue cene, ma che ha sopratutto sostenuto notti insonni permettendomi di scrivere tesi e tesine varie. Grazie anche a Viola, il mio amore, che con amore e pazienza mi è sempre stata accanto. Senza di lei sarebbe stato impossibile per me completare questo percorso, che rappresenta una delle tappe più importanti della mia vita. Spero che tutti i sacrifici da lei fatti in questo periodo, si trasformino in dei frutti che potremmo raccogliere insieme giorno dopo giorno. Voglio ringraziare anche mia suocera Miriam per l asilo ed i buoni pasti offertimi numerose volte durante questo periodo. Ringrazio anche la famiglia Baldini, la quale mi è stata accanto in ogni momento durante il mio primo periodo a Tolosa e continua
6 a farlo tutt oggi. Senza di loro tutto ciò sarebbe stato quasi impossibile. Un grazie va anche ad i miei amici: Emanuele Pellegrini il quale che ha sempre creduto in me. Giovanni Tancredi l amico di sempre, il quale non ha esitato a dispensare numerosi consigli di natura grafica durante la stesura di questa tesi. Raffaello Bonghi, l amico con cui ho condiviso gran parte di questa esperienza e con il quale ho superato numerosi esami, non dimenticherò mai il tempo trascorso insieme in una delle fasi più importanti della mia vita, l approccio con l automatica. Antonio De Palo, come dimenticare le giornate passate insieme a studiare con un entusiasmo senza fine che non dimenticherò mai. Federico Celletti compagno di studi ma sopratutto amico, ore e ore passate a studiare con una voglia insaziabile di capire, condividendo la passione per la teoria del controllo. Le numerose giornate trascorse a discutere hanno dato spunto a numerose ed interessanti riflessioni. Oltre gli amici, vorrei ringraziare anche i docenti che ho incontrato durante il mio percorso. Un grazie al prof. Lavino Ricciardi che ha dato inizio al meccanismo che mi ha condotto fino qui. Ringrazio anche il prof. Salvatore Monaco, il quale ha dato il via alla mia passione per l automatica, introducendomi alla teoria dei sistemi con curiosità e donandomi le solide basi che mi hanno permesso di giungere fin qui. Grazie anche al prof. Alessandro Savo, che ha saputo trasmettermi un metodo che tutt oggi caratterizza il mio modo di affrontare i problemi. Un ricordo anche alla memoria del prof. Alessandro De Carli. Ricorderò per sempre i suoi insegnamenti ed il suo saper essere ingegnere. Ringrazio infine, il prof. Sergio Galeani per la sua disponibilità sia come relatore, sia come professore sempre pronto a suggerire interessantissimi spunti ed esempi, ed il prof. Osvaldo Maria Grasselli per la sua cordialità e per il rigore e la passione con cui conduce il suo corso intriso di insegnamenti fondamentali.
7 Don t just read it; fight it! Ask your own questions, look for your own examples, discover your own proofs. Is the hypothesis necessary? Is the converse true? What happens in the classical special case? What about the degenerate cases? Where does the proof use the hypothesis? Paul Halmos, I want to be a Mathematician, Washington 1985
8 CONTENTS Introduction VII 1 Quantization in control systems Preliminary definitions Quantized systems Issues on discontinuity Stability and stabilization The works of R. Brockett and D. Liberzon Control by static quantizedstate feedback Stabilization via dynamic quantizer The work of S.Tarbouriech and F.Gouaisbaut Quantization in Linear Saturated Systems Preliminary Definitions The OpenLoop System The ClosedLoop System Boundedness/Ultimate Boundedness via Lyapunov Like Analysis Existences of Limit Cycles: The Planar case I
9 2.4 Static State Feedback Control Design Problem statement and first steps toward the solution Models for the Saturation Nonlinearity Sector Conditions for the Uniform Quantizer Main results State Feedback with Quantized Input State Feedback with Quantized Measured State Optimization issues and numerical results Optimization Based Controller Synthesis Size criteria for ellipsoidal sets LMI Formulation Numerical Examples Quantized Input Case Quantized Measured State Case Conclusions and Outlooks Summary Critical Aspects Outlooks A Further Clarifications on the LMI Formulation 85 A.1 Quantized Input Case A.2 Quantized Measured State Case B Mathematical Review 89 B.1 Convex sets and Functions B.2 Convex Combination and Convex Hull B.3 Linear Matrix Inequalities B.4 S Procedure B.5 Ellipsoidal set contained in a symmetric polyhedron Bibliography 97 II
10 List of Figures 99 III
11 LIST OF SYMBOLS = equal to not equal to < (>) less (greater) than ( ) less (greater) than or equal to defined as : such that A\B max Subset of Set difference x A\B if x A and x / B exists for every belonging to summation maximum IV
12 min N Z Z Z l R n R n + R n m x x minimum strictly positive integer numbers integer numbers integer numbers multiple of lcomponent vector of integer numbers multiple of nth dimensional Euclidean space subset of the nth dimensional Euclidean space composed by vectors with strictly positive components Real n m matrices space the norm of a vector x the absolute value if x R, the vector componentwise absolute value if x R n B m the ball centered at the origin of radius m {x R n : x m} ( ) vector componentwise less (greater) than ( ) vector componentwise less (greater) than or equal to Co{x 1,..., x n } The convex hull of {x 1,..., x n } 0 the null scalar or the null matrix of appropriate dimension 1 m unitary vector belonging to R m, i.e. 1 m = [1,..., 1] } {{ } m I n n n identity matrix i.e } 0 0 {{ 1 } n V
13 diag{a 1 ;... ; A n } Denotes the blockdiagonal matrix whose diagonal elements are A 1,..., A n λ min (P )(λ max (P )) the minimum (maximum) eigenvalue of a symmetric matrix P x, (A ) the transpose of a vector x (a matrix A) He(A) means A + A trace(a) trace of the matrix A P > 0 a symmetric positive definite matrix P 0 a symmetric positive semidefinite matrix A > B means that the matrix A B > 0 A B means that the matrix A B 0 f 2 f 1 it stands for symmetric blocks in the expression of a matrix the composition of two function LM I Linear Matrix Inequality ẋ f x the derivative of x respect to time the Jacobian matrix f : S 1 S 2 a function f mapping a set S 1 into a set S 2 sat( ) the saturation function VI
14 INTRODUCTION In the real realization of control systems, most of the time, some constraints are present on the variables that determine the evolution of the controlled system, these constraints arise from limited resources or technical limitations present in the real situations. One of the major limitations in real cases consists in a limit on the number of possible values for certain variables. This phenomena is commonly called quantization and the systems, whose variables are subjected to the quantization constraints, are commonly called quantized systems. For example, quantization arises when a plant is controlled by a digital system, which uses a finiteprecision arithmetic to compute the control action, or when a controlled system is only a subsystem which exchanges information with a main system via a communication channel, with a limited set of symbols. Figure 1 depicts a typical situation wherein quantization is involved. The previous list of examples is only a small part of a wide class of real cases, since the presence of the quantization is massive in practice. It is straightforward find more and more examples that clarify the importance of a systematical study of the quantization phenomena. In the literature many works that deal with the quantization aspects in control system have been presented. For instance in [11] the effects induced by the quantization in digital control loop had been studied, VII
15 Figure 1: Quantization in control system showing the presence of chaotic behavior or limitcycle in quantized system. Nevertheless, until the late of 80 s the quantization effect was considered as an undesirable effect and hence the presence of this phenomena was neglected in the control synthesis. Then, to estimate the effect induced by the quantization, the difference between the real signal and the quantized one was modeled as a noise (stochastic signal), with certain characteristics, and the behavior of the controlled system was studied in presence of that noise see [17]. In other words the quantization effect was been considered for a long time as an approximation and it was studied by the means of the tools provided by the information theory. However, since digital devices were becoming pervasive in control systems, new systematical analysis methods were necessary. D.F.Delchamps in 1998 [6] marked a watershed, proposing an alternative approach to deal with stability and stabilization in quantized control systems. This approach entails modeling the quantization phenomenon via a static nonlinear function, quantizer, which maps a real variable in a variable belonging to a given discrete set Q, that is: q: R Q The methodology proposed by D.F.Delchamps in [6] allows one to use the modern control theory tools in order to deal with quantized control systems. VIII
16 Such tools are based on the statespace description of the controlled plant, that is ẋ = f(x, u) where x: R R n, u: R R m, are respectively called state and control and f : R n+m R n. For example, describing the quantizer as a nonlinear map q( ), the effect induced by a quantizer on a given state feedback law v = k(x), can be considered, setting: u = q(k(x)) Accordingly to this approach, recently D. Liberzon and R. Brockett have been proposed a new framework, for studying quantized systems, based on the input to state stability concept proposed by E. Sontag in [19]. Moreover, in [16] D. Liberzon has highlighted that in many case, due the presence of the quantization, the Lyapunov asymptotic stability property may not be achieved. On the other hand, besides the stabilization problems, several aspects related to quantization effect in control systems has been studied. For instance, in control schemes whose control variables are subject to quantization, it is clear that the control properties could be influenced by the quantization phenomena. Regarding this aspect, for example, in [1] issues on reachability, for quantized linear discrete time, has been pointed. Let us take a step back to stabilization problem, in [3, 15, 16] D. Liberzon and R. Brockett have proposed new techniques aimed mainly at removing the limitation introduced by a quantizer in achieving the asymptotic stability. Such techniques deal with quantizers whose parameters can be varied online. Especially, in [3, 15, 16] a hybrid control scheme is proposed in order to obtain the global asymptotic stability for nonlinear systems, under precise conditions. However, these strategies can be adopted only in particular cases. Indeed, usually varying quantizer parameters is not allowed. In such situations, one have recourse to more classical control structure, hence disposing of synthesis conditions in these cases can be suitable. Concerning this aspect, in [22] some synthesis conditions have been provided in a quantized static state feedback control context and with respect to a restricted IX
17 class of system, that is linear saturated systems. Such class of systems can be represented by the following model: ẋ = Ax + B sat(u) This class of systems holds the interest of engineers and researchers as many real plants can be approximately modeled through a linear model, whereas saturation aries, since in real cases input variables cannot assume arbitrarily large values. Especially, in [22], via a slightly modification of the approach proposed by D. Liberzon, the authors have provided a synthesis procedure, which can be carried out through a solution of an associate LMI optimization problem, that is an optimization problem like: min c 1 x c n x n subject to A x 1 A x n A 1 n 0. A m 0 + x 1 A m x n A m n 0 where, x 1,..., x n R are the decision variables and A i 0, A i 1,..., A i n, i = 1,..., m are some given matrices. This latter feature represents the strong point of the exhibited methodology, since LMI optimizations problems can be efficiently solved by numerical procedures [2, 4]. On the other hand, as highlighted by the authors, putting an effort to simplify the problem, some conservativeness has been introduces. Such a conservativeness sometimes reveals some limitations in the applicability of the exposed method. In this thesis, we extend the work proposed in [22] reducing the above mentioned conservatism. To this end, a different approach to modeling the saturation nonlinearity is adopted. The research activity, which has led to this thesis, was conducted at: Laboratorie d Analyse et d Architecture des Systèmes (LAAS CNRS) of Toulouse (France), under the supervision of Sophie Tarbouriech and Frédéric X
18 Gouaisbaut. For sake of clarity, we will expose as follow, which is the structure of the present thesis. Thesis Structure The thesis is composed of four chapters. Chapter 1 In this chapter, we will present an overview on the literature related to quantization in control systems. Firstly, we will provide some basic definitions and afterwards we will show how a quantizer can be represented and which are the related technical problems. In the end we will focus on the concept of stability for quantized system and we will show some stabilization techniques presented in the literature. In the end we will drops an hint on the works proposed in [22]. Chapter 2 In this chapter we will focus our attention on the effect induced by a quantizer in a control scheme involving a restricted class of controlled systems, that is the class of continuoustime LTI systems with saturated input. Especially in this chapter, we will firstly explain which are the models used in order to describe the controlled plant, afterwards we will illustrate which are the closedloop systems analyzed in this work. Concerning this aspect, we will show how to cope with the difficulties due the presence of the saturation and the quantization. With respect to the saturation, a polytopic representation of a linear saturated system will be provided, instead with regards to quantization, sector conditions proposed in [22] will be shown. Through such tools some simplified models, of the closedloop systems analyzed, will be proposed. In the end, we will expose the formal problem we want to solve. XI
19 Chapter 3 In this chapter, we will show some theoretical results aimed at solve the Problem 1 exposed in the Chapter 3, for both the case presented. These results are based on the simplified models shown in the previous chapter. Especially, these results represent sufficient conditions. Chapter 4 In this chapter, we will show how the conditions stated in the Chapter 3 can be effectively used in order to solve numerically the problem stated in Chapter 2. To this end, an optimization procedure will be presented in order to develop a suitable synthesis procedure, however this procedure is characterized by nonlinear constraint. Therefore, some additional constraints will be considered in order to obtain, in both cases, a LMI optimization problem. Finally, some numerical results will be shown. XII
20 CHAPTER 1 QUANTIZATION IN CONTROL SYSTEMS In this chapter we will present an overview on the literature related to quantization in control systems. Firstly, we will provide some basic definitions and afterwards we will show how a quantizer can be represented and which are the related technical problems. In the end we will focus on the concept of stability for quantized system and we will show some stabilization techniques presented in the literature. 1.1 Preliminary definitions In the sequel, we will deal with dynamical systems described by an ordinary differential equation with a specified initial condition, that is: ẋ = f(x, u) (1.1) x(t 0 ) = x 0 1
21 1.2 Quantized systems 2 where, considering only solutions that run forward in time, x: [t 0, + ) R n, u: [t 0, + ) R m, x 0 R n are said to be respectively, "state", "control" and "initial state" and ẋ denotes the timederivative of x, i.e: ẋ(t) = dx(t) dt Moreover, without loss of generality, we will usually assume t 0 = Quantized systems By a quantizer we mean a function q that maps the Euclidean space R l to a discrete set Q, that is: R l Q q: x q(x) Depending on the model adopted for describing a particular situation, the set Q can be bounded or unbounded, but anyway we can state that a quantizer is a discontinuous map since it maps a connected space, R l, in an unconnected space Q [20]. The fact that the quantizer map is discontinuous can represent a non trivial problem in the field of quantized control systems. As stated previously, we are dealing with dynamical systems described by ordinary differential equations, hence the presence of a discontinuous element in the controller implies that the closedloop system will be described by a discontinuous righthand side ordinary differential equation. For instance, let us consider a static statefeedback control law for the system (1.1), obtained using a quantized state measure that is u = γ(q(x)), this controller provides the following closedloop system ẋ = f(x, γ(q(x)))
22 1.2 Quantized systems 3 which is obviously described by a discontinuous righthand side ordinary differential equation. Therefore in order to handle quantized systems we must face with the problem of discontinuous righthand side differential equations Issues on discontinuity In the literature many examples are shown, wherein a discontinuous righthand side differential equation does not admit any solution in usual sense. For a recent survey we refer to [5]. Therefore in order to deal with systems described by a discontinuous righthand side differential equation, we must extend the classical concept of solution for differential equations. Unfortunately there is not an unique way for doing this. Depending on the problem and on the objectives at hand, different notions have been proposed and each of them arises from a different point of view adopted in facing the righthand side discontinuity. For example, in [8], a new concept of solution is introduced, looking at the values assumed by the vector field f in a neighborhood of each point rather than the single point. Formally a set valued map is associated to the vector field f. Specifically, for every x R n the vector field f is evaluated in the open ball B(x, δ) centered at x with radius δ and the analysis is carried out for smaller and smaller δ, that is when δ approaches zero. In [13] a similar approach is proposed. Another notion of solution is provides by the concept of Carathéodory solutions, shown in [8]. Especially, regarding at the differential equation (1.1), a Carathéodory solutions essentially is a classical solution, that satisfies the relation (1.1) for almost all t. In order to define formally what we mean with Carathéodory solutions it is suitable to introduce the concept of absolutely continuity Definition 1. [5] A function γ : [a, b] R is absolutely continuous if, for all ε (0, + ), there exists δ (0, + ), such that, for each finite collection {(a 1, b 1 ),..., (a n, b n )} (a, b)
23 1.2 Quantized systems 4 with it follows that n (b i a i ) < δ i=1 n γ(b i ) γ(a i ) < ε i=1 Clearly, every absolutely continuous function is continuous, but it can be shown by examples that the converse is not true [5]. Now we can provide the definition of Carathéodory solutions. Definition 2. A function x: I R n is a Carathéodory solution of the equation (1.1) on the interval I if is an absolutely continuous function that satisfies the following integral equation x(t) = x 0 + t t 0 f(x(τ))dτ t, t 0 I, t t 0 (1.2) The foregoing overview represents only an outline of the wide area concerning discontinuous righthand side differential equations (see for example [5, 8, 13]). Looking at the above definitions, one may argue that every notion faces with the discontinuity in the righthand side in different manners. Hence, choosing different notions of solution can provide different outcomes. Obviously, in choosing the notion of solution, one should be careful in obtaining a faithful model of the real problem, without adding any undesired phenomena. In fact, depending on the considered solution notion, different behaviors can be considered or excluded. For instance to consider Filippov s solutions allows us to analyze sliding mode motions, which are not solution in Carathéodory sense. However in this work, we will focus only on Carathéodory solutions. Therefore we will accordingly excluded particular solution like sliding mode motions between the quantization regions.
24 1.2 Quantized systems Stability and stabilization The most interesting topics in control area are problems related to stability and stabilization of an equilibrium point. In quantized system, the problem of stability plays a relevant role. Indeed, since almost all the controllers are feedback controller, they use the measured variables in order to compute the control action. Hence, when these variables are subject to quantization, it is clear that the closedloop stability could be influenced by the quantization effect. Furthermore, without loss of generality, in the sequel we will consider the origin as equilibrium point. To deal with the problem of stability for quantized systems, first of all it should be noticed that in controlled systems involving a quantizer, sometimes the classical concept of Lyapunov asymptotic stability can be meaningless. Indeed in order to steer the state from a given initial condition to the origin, it is necessary to get an infinitely arbitrarily high precision close to the origin, but this requirement do not fit with the quantization constraint. For instance in a system controlled by a quantized input it is possible to drive the system only with a discrete set of values. However the impossibility in achieving the asymptotic stability objective it is not always so serious, indeed the real important task is often to keep the system trajectories bounded and sufficiently close to the origin. In order to handle this aspect it is suitable to introduce a weaker property than the asymptotic stability. At this end, in the literature several properties have been introduced. For example in [25] a notion of containability is provided in a particular case, whereas in [14] the concept of practical stability is introduced and studied. Furthermore, in [12] a more general property is introduced for nonlinear continuoustime systems. Such a property is called boundedness/ultimate boundedness property [12] and it is defined as follow. Boundedness and ultimate boundedness With respect to the system (1.1), the following definition holds.
25 1.3 The works of R. Brockett and D. Liberzon 6 Definition 3. [12] The solutions of (1.1) are: uniformly bounded if there exists a positive constant c and a (0, c) there is a β = β(a) > 0 such that: x 0 R n : x(t 0 ) a x(t) β t t 0 (1.3) globally uniformly bounded if (1.3) holds for arbitrary large a uniformly ultimately bounded with bound b, if there exist two positive constants, b, c and a (0, c) there exists a time T = T (a, b) 0 such that: x 0 R n : x(t 0 ) a x(t) b t t 0 + T (1.4) globally uniformly ultimately bounded if (1.4) holds for arbitrary large a Remark If the system (1.1) is autonomous, then the term uniformly can be omitted without ambiguity. 1.3 The works of R. Brockett and D. Liberzon The quantization phenomena and related problems discussed above can be englobed in a wider field called "Control with limited information" of which R. Brockett is one of the pioneers. Some representative references, concerning this research area, include [3, 6, 7, 23 25]. Many approaches can be adopted in order to cope the limited information control problem, for
26 1.3 The works of R. Brockett and D. Liberzon 7 instance in [15] a general methodology for handle this class of problems is proposed for state feedback case. Especially this approach can be summarized as follows. 1. Model the quantification effects via deterministic additive error signal e; 2. Design a nominal control law ignoring these errors, i.e. a control law acting on perfect information, which we assume by simplicity to be a static state feedback u = k(x); 3. ( certainty equivalence ) Apply the above control law to the imperfect/corrupted signals, resulting in u = k(x + e) and combine it with an estimation procedure aimed at reducing e to 0. Using this approach requires that the closedloop system is robust, in some sense, with respect to e. Indeed one can show, by example, that even if the error converges to zero and the synthesized controller stabilizes the closedloop system when e 0, the state can blow up in finite time making the control strategy unable to stabilize the plant [15]. Hence the main problem is how to characterize the necessary closedloop robustness with respect to e. In the works of D. Liberzon and R. Brockett this characterization is founding on the "Input to state stability"(iss) concept introduced by Eduardo Sontag [19]. Keeping in mind the approach provided by the control with limited information, we can cope with the problems of stability and stabilization for quantized systems Control by static quantizedstate feedback In [16] D. Liberzon has shown that it is possible to obtain the boundedness and the ultimate boundedness properties for a general nonlinear system controlled by a static quantizedstate feedback, under precise assumptions. Let us start to specify which the author means as quantizer in this work.
27 1.3 The works of R. Brockett and D. Liberzon 8 If the variable subject to quantization belongs to R l then the quantizer is a mapping q: R l Q where Q is a finite subset of R l. Therefore the quantization regions are the sets {z R l : q(z) = i}, i Q Furthermore the following assumption on the quantizer holds. Assumption 1. [16] There exist two positive real numbers M, such that: 1. If then 2. If then z M q(z) z z > M q(z) > M The above conditions define the socalled saturated quantizer. Condition 1 gives a bound on the quantization error when the quantizer does not saturate whereas condition 2 provides a way to detect the quantizer saturation. The numbers and M represent the quantizer sensitivity and the quantizer range. In Figure 1.1 the described quantizer is depicted. It should be noticed that the author does not provide a real description of a quantizer, but rather a set of properties which are commonly satisfied by a general quantizer. In this way, the results provided are not relate to a particular definition of the quantizer and they can be extended in every situation as long as Assumption 1 is verified. After the previous clarifications, assuming
28 1.3 The works of R. Brockett and D. Liberzon 9 Figure 1.1: Graphical description of quantizer defined by D. Liberzon that a general nonlinear system ẋ = f(x, u), x R n, u R m (1.5) and a state feedback law u = k(x) (1.6) that renders the origin globally asymptotically stable for the system (1.5) are given. Now suppose that the state x is quantized, hence the control law (1.6) becomes u = k(q(x)) = k(x + e) (1.7) where e q(x) x is the quantization error. One can straightforwardly argue that in general the origin of the obtained closedloop system is not still globally asymptotically stable. Nevertheless one may expect in place of the global asymptotic stability property a weaker property like trajectories
Controllability and Observability of Partial Differential Equations: Some results and open problems
Controllability and Observability of Partial Differential Equations: Some results and open problems Enrique ZUAZUA Departamento de Matemáticas Universidad Autónoma 2849 Madrid. Spain. enrique.zuazua@uam.es
More informationWMR Control Via Dynamic Feedback Linearization: Design, Implementation, and Experimental Validation
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 6, NOVEMBER 2002 835 WMR Control Via Dynamic Feedback Linearization: Design, Implementation, and Experimental Validation Giuseppe Oriolo, Member,
More informationTHE PROBLEM OF finding localized energy solutions
600 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 45, NO. 3, MARCH 1997 Sparse Signal Reconstruction from Limited Data Using FOCUSS: A Reweighted Minimum Norm Algorithm Irina F. Gorodnitsky, Member, IEEE,
More informationFeedback Control of a Nonholonomic Carlike Robot
Feedback Control of a Nonholonomic Carlike Robot A. De Luca G. Oriolo C. Samson This is the fourth chapter of the book: Robot Motion Planning and Control JeanPaul Laumond (Editor) Laboratoire d Analye
More informationSubspace Pursuit for Compressive Sensing: Closing the Gap Between Performance and Complexity
Subspace Pursuit for Compressive Sensing: Closing the Gap Between Performance and Complexity Wei Dai and Olgica Milenkovic Department of Electrical and Computer Engineering University of Illinois at UrbanaChampaign
More informationFEEDBACK CONTROL OF A NONHOLONOMIC CARLIKE ROBOT
FEEDBACK CONTROL OF A NONHOLONOMIC CARLIKE ROBOT Alessandro De Luca Giuseppe Oriolo Dipartimento di Informatica e Sistemistica Università di Roma La Sapienza Via Eudossiana 8, 84 Rome, Italy {deluca,oriolo}@labrob.ing.uniroma.it
More informationOPRE 6201 : 2. Simplex Method
OPRE 6201 : 2. Simplex Method 1 The Graphical Method: An Example Consider the following linear program: Max 4x 1 +3x 2 Subject to: 2x 1 +3x 2 6 (1) 3x 1 +2x 2 3 (2) 2x 2 5 (3) 2x 1 +x 2 4 (4) x 1, x 2
More informationINTERIOR POINT POLYNOMIAL TIME METHODS IN CONVEX PROGRAMMING
GEORGIA INSTITUTE OF TECHNOLOGY SCHOOL OF INDUSTRIAL AND SYSTEMS ENGINEERING LECTURE NOTES INTERIOR POINT POLYNOMIAL TIME METHODS IN CONVEX PROGRAMMING ISYE 8813 Arkadi Nemirovski On sabbatical leave from
More informationFeedback Control Theory
Feedback Control Theory John Doyle, Bruce Francis, Allen Tannenbaum c Macmillan Publishing Co., 1990 Contents Preface iii 1 Introduction 1 1.1 Issues in Control System Design.............................
More informationSampling 50 Years After Shannon
Sampling 50 Years After Shannon MICHAEL UNSER, FELLOW, IEEE This paper presents an account of the current state of sampling, 50 years after Shannon s formulation of the sampling theorem. The emphasis is
More informationAircraft Pitch Control Via Second Order Sliding Technique
Aircraft Pitch Control Via Second Order Sliding Technique A. Levant and A. Pridor Institute for Industrial Mathematics, BeerSheva, Israel R. Gitizadeh and I. Yaesh Israel Military Industries, RamatHasharon,
More informationAn Introduction to the NavierStokes InitialBoundary Value Problem
An Introduction to the NavierStokes InitialBoundary Value Problem Giovanni P. Galdi Department of Mechanical Engineering University of Pittsburgh, USA Rechts auf zwei hohen Felsen befinden sich Schlösser,
More informationA Googlelike Model of Road Network Dynamics and its Application to Regulation and Control
A Googlelike Model of Road Network Dynamics and its Application to Regulation and Control Emanuele Crisostomi, Steve Kirkland, Robert Shorten August, 2010 Abstract Inspired by the ability of Markov chains
More information4.1 Learning algorithms for neural networks
4 Perceptron Learning 4.1 Learning algorithms for neural networks In the two preceding chapters we discussed two closely related models, McCulloch Pitts units and perceptrons, but the question of how to
More informationWHEN ARE TWO ALGORITHMS THE SAME?
WHEN ARE TWO ALGORITHMS THE SAME? ANDREAS BLASS, NACHUM DERSHOWITZ, AND YURI GUREVICH Abstract. People usually regard algorithms as more abstract than the programs that implement them. The natural way
More informationThe Set Covering Machine
Journal of Machine Learning Research 3 (2002) 723746 Submitted 12/01; Published 12/02 The Set Covering Machine Mario Marchand School of Information Technology and Engineering University of Ottawa Ottawa,
More informationIEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 4, APRIL 2006 1289. Compressed Sensing. David L. Donoho, Member, IEEE
IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 4, APRIL 2006 1289 Compressed Sensing David L. Donoho, Member, IEEE Abstract Suppose is an unknown vector in (a digital image or signal); we plan to
More informationAnalysis of dynamic sensor networks: power law then what?
Analysis of dynamic sensor networks: power law then what? (Invited Paper) Éric Fleury, JeanLoup Guillaume CITI / ARES INRIA INSA de Lyon F9 Villeurbanne FRANCE Céline Robardet LIRIS / CNRS UMR INSA de
More informationElementary Topology Problem Textbook. O. Ya. Viro, O. A. Ivanov, N. Yu. Netsvetaev, V. M. Kharlamov
Elementary Topology Problem Textbook O. Ya. Viro, O. A. Ivanov, N. Yu. Netsvetaev, V. M. Kharlamov Dedicated to the memory of Vladimir Abramovich Rokhlin (1919 1984) our teacher Contents Introduction xi
More informationThe Backpropagation Algorithm
7 The Backpropagation Algorithm 7. Learning as gradient descent We saw in the last chapter that multilayered networks are capable of computing a wider range of Boolean functions than networks with a single
More informationHow to Use Expert Advice
NICOLÒ CESABIANCHI Università di Milano, Milan, Italy YOAV FREUND AT&T Labs, Florham Park, New Jersey DAVID HAUSSLER AND DAVID P. HELMBOLD University of California, Santa Cruz, Santa Cruz, California
More informationData Quality Assessment: A Reviewer s Guide EPA QA/G9R
United States Office of Environmental EPA/240/B06/002 Environmental Protection Information Agency Washington, DC 20460 Data Quality Assessment: A Reviewer s Guide EPA QA/G9R FOREWORD This document is
More informationHYBRID systems involve a combination of discrete and continuous
UNPUBLISHED REPORT Formal Semantics and Analysis Methods for Simulink Stateflow Models A. Tiwari Abstract Embedded control systems typically comprise continuous control laws combined with discrete mode
More informationFast Solution of l 1 norm Minimization Problems When the Solution May be Sparse
Fast Solution of l 1 norm Minimization Problems When the Solution May be Sparse David L. Donoho and Yaakov Tsaig October 6 Abstract The minimum l 1 norm solution to an underdetermined system of linear
More informationLevel sets and extrema of random processes and fields. JeanMarc Azaïs Mario Wschebor
Level sets and extrema of random processes and fields. JeanMarc Azaïs Mario Wschebor Université de Toulouse UPS CNRS : Institut de Mathématiques Laboratoire de Statistiques et Probabilités 118 Route de
More informationRobust Set Reconciliation
Robust Set Reconciliation Di Chen 1 Christian Konrad 2 Ke Yi 1 Wei Yu 3 Qin Zhang 4 1 Hong Kong University of Science and Technology, Hong Kong, China 2 Reykjavik University, Reykjavik, Iceland 3 Aarhus
More informationMaximizing the Spread of Influence through a Social Network
Maximizing the Spread of Influence through a Social Network David Kempe Dept. of Computer Science Cornell University, Ithaca NY kempe@cs.cornell.edu Jon Kleinberg Dept. of Computer Science Cornell University,
More informationA Case Study in Approximate Linearization: The Acrobot Example
A Case Study in Approximate Linearization: The Acrobot Example Richard M. Murray Electronics Research Laboratory University of California Berkeley, CA 94720 John Hauser Department of EESystems University
More informationOn SetBased Multiobjective Optimization
1 On SetBased Multiobjective Optimization Eckart Zitzler, Lothar Thiele, and Johannes Bader Abstract Assuming that evolutionary multiobjective optimization (EMO) mainly deals with set problems, one can
More informationDistributed Control of Robotic Networks
share March 22, 2009 Distributed Control of Robotic Networks A Mathematical Approach to Motion Coordination Algorithms Francesco Bullo Jorge Cortés Sonia Martínez share March 22, 2009 Copyright c 20062009
More information