This document summarizes a research paper that presents a novel observer-controller design to solve the global output-feedback tracking problem for a benchmark nonlinear system called RTAC. The design transforms the system equations such that the unmeasured state depends linearly on the system, allowing an observer-controller approach. This is the first solution capable of accommodating all initial conditions with the same controller and guarantees asymptotic tracking for a larger class of reference signals than prior methods.
This document summarizes a research paper that presents a novel observer-controller design to solve the global output-feedback tracking problem for a benchmark nonlinear system called RTAC. The design transforms the system equations such that the unmeasured state depends linearly on the system, allowing an observer-controller approach. This is the first solution capable of accommodating all initial conditions with the same controller and guarantees asymptotic tracking for a larger class of reference signals than prior methods.
Original Description:
Global Output-Feedback Tracking for nonlinear systems.
This document summarizes a research paper that presents a novel observer-controller design to solve the global output-feedback tracking problem for a benchmark nonlinear system called RTAC. The design transforms the system equations such that the unmeasured state depends linearly on the system, allowing an observer-controller approach. This is the first solution capable of accommodating all initial conditions with the same controller and guarantees asymptotic tracking for a larger class of reference signals than prior methods.
This document summarizes a research paper that presents a novel observer-controller design to solve the global output-feedback tracking problem for a benchmark nonlinear system called RTAC. The design transforms the system equations such that the unmeasured state depends linearly on the system, allowing an observer-controller approach. This is the first solution capable of accommodating all initial conditions with the same controller and guarantees asymptotic tracking for a larger class of reference signals than prior methods.
Phoenix, Arizona USA December 1999 Fr Ml l 14: OO Global Output-Feedback Tracking for a Benchmark Nonlinear System * Zhong-Ping J iang Ioannis Kanellakopoulos Department of Electrical Engineering Polytechnic University University of California Brooklyn, NY 11201. zjiang@control.poly.edu ioannis@ee.ucla.edu Department of Electrical Engineering Los Angeles, CA 90095-1594 Abstract: In this paper, the output-feedback global tracking problem is solved for the well-known nonlinear benchmark RTAC system, where one of the unmea- sured states appears quadratically in the state equa- tions. Our novel observer-controller backstepping de- sign yields a nonlinear output-feedback controller that forces the translational displacement to globally asymp- totically track an appropriate time-varying signal. The proposed solution is new even for the case of global output-feedback stabilization, namely when the refer- ence signal is zero. 1 Introduction The problem of controlling the nonlinear benchmark mechanical system usually called RTAC (for Rotational- Translational Actuator) or TORA (Translational Os- cillator with a Rotational Actuator) was introduced in [lo]. This problem has recently received a consid- erable amount of attention from several researchers [l, 2, 3, 4, 10, 111. Due to the weak, sinusoid-type non- linear interaction between the translational oscillations and the rotational motion, the dynamic model of this RTAC is not globally feedback linearizable. There- fore, the direct application of well-known nonlinear control schemes such as feedback linearization does not solve the global stabilization problem. In the afore- mentioned papers, several novel nonlinear approaches have been developed, based on integrator backstepping and passivity techniques, for the state- and output- feedback stabilization and tracking problems. More- over, some of these approaches have been validated through experimental results. The objective of this paper is to address the problem of output-feedback global tracking for this benchmark system, following our previous results for the state- feedback [ll] and output-feedback [4] problems. We 'The work of the first author was supported in part by a start-up grant from Polytechnic University. The work of the second author was supported in part by NSF under Grant ECS- 9502945. consider the translational position and the rotational angle as the two outputs and assume that the linear and angular velocities are not available for feedback. The main difficulty here is that the system equations depend nonlinearly on an unmeasured state; this ren- ders existing global output-feedback stabilization and tracking methods [6, 7, 81 not applicable to the RTAC system. The semiglobal approach proposed in [4] forces the translational displacement to track a suitably de- fined reference signal for any given bounded region of initial conditions. In this paper, we give a global so- lution to the output-feedback tracking problem. By exploiting the physical structure of the RTAC system, the proposed tracking methodology not only accom- modates all initial conditions with the same controller, but also guarantees asymptotic tracking for a larger class of reference signals than the semiglobal scheme of [4]. Notation: For a vector 2 E Rn, zT denotes its trans- pose and 1%) its Euclidean norm. For a time-varying system E =g ( t , J ) +~ ( t ) , W+(.,t ) denotes the tra- jectory starting from $ at t =0 .and W(.,t ) means @99(.Lt)I i.e., the trajectory of =g ( t , t ) starting from 5 at t =0. I nxn is the identity matrix of or- der n while On, , is the null matrix of order n. GAS means global asymptotic stability while UGAS stands for uniform global asymptotic stability. 2 A benchmark problem The nonlinear benchmark system considered in this section was introduced by Wan, Bernstein and Cop- pola in [lo] and is known as RTAC or TORA. It is a mechanical system in which the translational oscilla- tions of the platform are controlled via the rotational motion of an eccentric mass. Assuming that the plat- form moves in the horizontal plane, the dynamics of the system are described by 0-7803-5250-5/99/$ IO.00 O I999 IEEE 4802 Authorized licensed use limited to: Sharif University of Technology. Downloaded on December 28, 2009 at 01:44 from IEEE Xplore. Restrictions apply. w1ierc.I; is the disturbance force actirig on the cart, N is the control torque applied to the proof mass, x, is the translational position and 8 is the rotational angle. In (l), 11.1represents the total mass of the disk and the cart, m and I represent the mass and the moment of inertia of the eccentric mass, respectively, and T is the radius of rotation. Asin [lo], after some appropriate normalized transfor- mations, equations (1) and (1) are simplified to $ , I + Xd = &(e2sino - ecosq f Fd (2) 8 = U-&XdCOS8, where 0 <E <1 and U is the control input. In the new coordinates defined by (3) the system (2) is described by the following state-space representation x 1 = 22 x2 = - 2 1 +tsi nx3 +Fd X B = x4 (4) x 4 = U + X E cos 2 3 1 - E2 cos2 x3 1 - E2 cos2 5 3 e2 cos 2 3 sin 2 3 xi. 1 - E2 cos2 2 3 (21 - t sin 23 - Fd) - In the rest of the paper, weassume that there is no dis- turbance, i.e. Fd =0 in (4). We further assume that y =(21, ~ 3 ) ~ is the measured output. In other words, weonly require that the translational displacement x, and the rotational angle 8 of the original system (1) and (1) be measured; the corresponding linear and an- gular velocities i, and d are assumed to be unavailable for feedback. This assumption is meaningful from a practical point of view, since speed sensors are more expensive and noisier than position sensors. The control task weaddress in this paper is the prob- lem of output-feedback global tracking for the RTAC system described by (2). In particular, given an ap- propriately defined reference signal x, ( t ) , we want to find a dynamic output-feedback law of the form 6 =.(t,Y,O) , U =P(t,Y,C) (5) such that. the states of the closed-loop system (4) and (5) are bounded on [0, fw). Furthermore, for any ini- tial value x d ( O ) , the tracking error xd( t ) - x r ( t ) con- verges to zero as t goes to fw. In a previous paper [4], weproposed a semiglobal solu- tion to this problem, which forces xd( t ) - zcr(t) to zero, provided that a fixed but arbitrary bound on the size of the initial conditions set (xd(o), kd( o) , 8(0), 8(0)) is known. A family of reference signals x,.(t) which can be tracked was also identified in [4]. The semiglobal output-feedback tracking design of [4] relies on the knowledge of a global state-feedback tracking controller and the incorporation of a high-gain observer. It was shown in [4] that semiglobal tracking was achieved when some observer parameters were chosen to be suffi- ciently large. In this paper, weproceed with a new idea which for the first time allows the global stabilization through output-feedback of a system whose nonlinear- ities depend on unmeasured states. It is of interest to note that we are able to track an even larger class of reference signals by means of the global solution pro- posed in this paper. For a subclass of reference signals, our global solution yields the UGAS (Uniform Global Asymptotic Stability) property for the null solution of the closed-loop system written in appropriate coordi- nates (see Section 4 for details). To identify these reference signals, we first note that the state Xd satisfies which leads us to introduce an additional differential equation (7) 1 ij =- - ( &+ x r ) . & Now, our class of reference signals can be characterized by the following assumption. Assumption 1 The reference time-varying signal x r ( t ) and its derivatives &(t ), &( t ) are bounded on [0, m). In addition, there is at least one solution ( ~ ( t ) , i ( t ) ) of (7), with appropriate initial values, which satisfies the inequalities where 0 <qmax <1 is a constant. 3 Global tracking via output feedback The purpose of this section is to introduce an observer- based output-feedback design method to solve the global tracking problem formulated in Section 2. As stated already, the major difficulty lies in the fact that the vector field in system (4) depends nonlinearly on the unmeasured state 24. The novelty of our controller de- sign is the introduction of a nonlinear state transfor- mation that eliminates this quadratic dependence on 4803 Authorized licensed use limited to: Sharif University of Technology. Downloaded on December 28, 2009 at 01:44 from IEEE Xplore. Restrictions apply. 2 4 in (4). That is, the obtained transformed system is made to depend only linearly on the unmeasured states. From there, a new observer/controller inter- laced backstepping procedure is introduced to achieve our design objective. 3.1 Observer design In order to design an observer for the unmeasured 2 4 - state, weintroduce the nonlinear transformation x =2 4 - 11,(X3)X4 (9) where 11, is a smooth function to be determined later. This type of transformation idea was used in [12] to deal with a class of nonlinear systems that are linear in unmeasured states and can be globally stabilized using output feedback. Equation (4) implies that If wechoose 11, to satisfy the equality then, weobtain I t is interesting to note that the quadratic term xz dis- appeared from the X-system (11). As a consequence, in the new coordinates (x1,22,23, x), the vector field of (4), with Fd =0, now depends linearly on the un- measured states (22, x): XI = 22 x2 = -21 +si n23 Clearly, wemust select a function 11, so that (10) holds and 1 - 11,(23) is nonzero for any 23 E R. Indeed, this is possible with the help of the following lemma whose proof is straightforward. Notice that the choice of such a function 11, is not unique. Lemma 1 The function 11,(23) =1 + (13) satisfies the equality (10) and the following property V X ~ E R (14) 1 2 5 d J ( X 3 ) I 1 +r> 7 We are now ready to design two observers to recon- struct the unmeasured states 22 and x of the system (12). Towards this end, weintroduce a new variable with L1 a positive design parameter. Noting that (1 =-L1G - (L; +1)xl +si n23 , (16) wenaturally introduce an observer of the form t1 h =-L1& - (Lf +1)xl +si n23 (17) h Denoting the observation error by el =51-cl, wehave e1 =-Llel (18) h Consequently, +Ll xl is an asymptotically exact es- timate of 22 =51 +Ll xl - el. Similarly, weintroduce a new variable 52 =x+ L223 with L2a positive design parameter. With (12), the &-dynamics satisfies (52 - L223) (19) L2 11,(x3) - 1 52 = - Therefore, weintroduce the following observer Letting the observation error e2 =52 - 52, weobtain As a direct application of Lemma 1 and Gronwall- Bellman inequality, wehave Lemma 2 For any initial condition e2(0) and any t 2 0, the solution e2(t) of (21) satisfies 3.2 A new backstepping design Before applying backstepping to design a *sired tracker based on the new coordinates (21~x2, x3,52), weintro- duce some suitable reference signals for these states. Using the time-varying signals v(t) and e( t ) which sat- isfy Assumption 1, we define (23) 1 cos 0, =arcsin(q) , 8, =- li 4804 Authorized licensed use limited to: Sharif University of Technology. Downloaded on December 28, 2009 at 01:44 from IEEE Xplore. Restrictions apply. Then, we introdlice the following variables Z = IC,, +E ~C OS O - I C . ~ - E&COSe, (24) z3 = e - er It is apparent that zd(t) converges to z,(t) as t -+00 if z =(21,z2,23,24)* goes to zero a5 t + CO. Thus, we will translate our problem of global tracking into a problem of global asymptotic stabilization (GAS) for the new z-system. In view of (3), (12), (7), (13) and (20), wehave 21 = z2 22 = -zl +E (sin(z3 +e, ) - sine,) 1 - E2 1 - E2 cos2 x3 (-L223 +z4 - e2) 23 = -\i 24 + Y 1 q / ( l - &2)(1-&2coS2x~) where Y is a function of ( e, , dr , &, z 3, z ~) . Note that 22 is not measured. However, wehave A 22 = 22 - el h 22 = +Ll(z1 +zr +EV) - j.r - ~ l j Next, wewill apply the backstepping approach to the system (25) in order to design a GAS feedback law U = u(t, z1,22, ~3 ~~4 ) . As explained earlier, this, in turn, solves the problem of output-feedback global tracking for the original model (2). Our constructive proce- dure has been inspired by the global state-feedback backstepping scheme proposed in [4, Sec. 4.11 and the cascade stabilization algorithm presented in [3, Sec. 31. As in [3,4], due to the weak effect of the virtual control input 23 via the bounded sinus function, wewill start the controller design procedure from the critically sta- ble (zl, z2)-subsystem. This explains to some extent why-we do-not work with the measured coordinates (21 9 t1, 23 t2). Step 1 : Consider the (zl,z2)-subsystem of (25) with z3 viewed as the virtual control input. Differentiating the quadratic function VI =$2; +i.2with respect to time gives VI = E Z ~ (sin(z3 +e,) - sine,) (26) Guided by the state-feedback design in [4, Sec. 4.11, we introduce the following Certainty Equivalence (CE) based virtual controller Z3 =z3 +I1arctan22 (27) where 0 <11 <2(1 - 20max/7r) is a design parameter and Om,, is a constant in (0,7r/2) such that le,(t)l 5 Omax. Why CE? Because z3 =-11 arctanzz makes VI in (26) negative semi-definite. However, this vir- tual stabilizer depends on the unmeasured state z2. With (26) and (27), letting 2 =(z1, ZZ) ~, the (zl, z2)- subsystem of (25) can be rewritten as Z = f ( t, z, el , i ) (28) In the next steps, we will apply backstepping to the (Z3, zr)-system (not the entire system (25)!) and design a nonlinear controller U of the form (5) to drive &(t) to the origin at an exponential rate as t + CO. Step 2 : Consider the quadratic function (29) V, =-g 1 +-e: 1 +-e2 1 2 2 2 2 and introduce two new variables 1 - E2 cos2 23 11 (-21 +E sin(z3 +e,) - E sin e,) = z4 - ai ( e r , er , z2+el , z3 +e,> where c1 >lTL1/2 +l / ( 2L 2dm) is - 24 a design pa- rameter. Notice that 22 +el =8 and z3 +0,. = z3 are .available for feedback design. Also note that al (~,,B,,O,~,) =0. Byvirtueof (18), (21) andLemma 1, completing the squares weobtain (32) - L 2 m e ; - Z3z4J 1 - E2 2 1 - E2 cos2 23 Step 3 : Consider the function (33) 1, V3 =V2(%,el1e2) +2z4 Differentiate V3 along the solutions of (18), (21) and (26) and choose the following tracking control law to make V3 negative definite: 4805 Authorized licensed use limited to: Sharif University of Technology. Downloaded on December 28, 2009 at 01:44 from IEEE Xplore. Restrictions apply. aal . aal .. aal . -- eT - -eT - ---eT aeT ao, ax3 r +e, (J--"-"^. 1 - E' COS' (z3 +0,) - 1)]} (34) where c2 >0 is a design parameter, and 2, 2, 2 and % stand for the partial derivatives of a1 with respect to its first to fourth argument, respectively. Indeed, when (34) is substituted into V 3 , from (32) and the completion of the squares, it follows that (35) Letting a =min {(c1- + - A) 2 L 2 m , c 2 , + , L z F } (36) V3 5 - 2a&( ~~, ~* , el , e~) (37) (35) implies The statement and the proof of our main stability re- sults are given in the next section. 4 Main results Weare now ready to state our main theorem. Theorem 1 For any reference signal xT( t ) satisfying Assumption 1, the problem of global output-feedback tracking is solvable for the benchmark system [2) with Fd =0. Before proving Theorem 1, wegive a technical stability result for a nonautonomous cascade system of the type [ = f ( t , S, Y) (38) t = f d( t , O (39) Y = HC (40) where [ E EtRn(, C E Etn<, y E EtRn. and H is a con- stant nC x ny matrix. We assume that the continuous function f is C1 in [, for each fixed ( t , y ) . We further make the following assumptions on the cascade system (38), (39) and (40). (Al ) The zero-input [-system (38), i.e. [ =f ( t , [ , O) := g ( t , [ ) , is globally asymptotically stable (GAS) at <= 0. ( A2) The driving <-system (39) is globally exponen- tially stable (GES) at =0. That is, there exist two positive constants lc'and a such that IC(t)l L klC(O)IexP(-at) , v t 2 0 (41) (A3) f is globally Lipschitz in (<, y ) on [0, +CO) x Rn ( x for some constant L >0, all t 2 0 and any pair of EtRn. 7 i.e. I f ( t , t, Y ) - f ( t , <I , Y' )l I LI(t - 5'7 Y - Y Y I (t, Y ) , ( <I , Y?. Lemma 3 Under Assumptions (Ai ), (A2) and (A3), if a >L, the cascade system is GAS at the equilibrium (<, c) =(0,O). Moreover, if the zero-input <-system is uniformly globally asymptotically stable [UGAS) at <=0, then the cascade system is UGAS at ( [ , C) = (0,O). Proof. We first note that, thanks to the global Lip- schitzness condition and the GES property, every so- lution (<(t), C(t)) of the cascade system is defined for all t 2 0. We will invoke the approximation Theorem 5.1 of [9] to establish Lemma 3. Let IC =f ( t , <, y) - f ( t , <, 0). By (A2), there is a constant a >0 such that Itc(t)l 5 Kexp(-at) , V t 2 0 (42) where K 2 0 is proportional to the magnitude of the initial condition <( O) . It is clear that the conditions C1 and C2 of [9, Theorem 5.11 hold for the system i =g( t , [) +IC. Choose an r >0 in such a way that a >L +K/ r . A direct application of 19, Theorem 5.11 yields that, for every 8 E Etn, there exists a r in the ball B( 8, r ) c l Rnc such that p9+(t,E) - ' ~ 9 ( t , r ) l I rexp(-at) , v t 2 o (43) Since K in (42) can be selected arbitrarily large, from (43), both statements of Lemma 3 follow readily. AAA Now, wereturn to prove Theorem 1. Proof of Theorem 1: Using the notations in Section 3, the closed-loop system (25), (18), (21) and (34) can be put into the cascade form (38), (39) and (40). In the present case, (38) is ( 28) and, letting =(el, Z ~ , & , e 2 ) ~ and H =( 1 2 x 2 , 0 2 x 2 ) , wehave c =fd(t,C) (44) As it can be directly checked from (37) with V 3 = + S T [ , the condition (A2) is satisfied. On the other 4806 Authorized licensed use limited to: Sharif University of Technology. Downloaded on December 28, 2009 at 01:44 from IEEE Xplore. Restrictions apply. hand, according to the definition of f in (25) and (28), it is directly verified that f is globally Lipschitz with L =niax(1, E . dl}, i.e. (A3) holds. In addition, by the definition of a in (36), the design parameters c1, c2, L1 and L2 can be chosen appropriately so that that a >L. I t remains to verify (Al ). When setting el =Z3 =0 in (28), the time derivative of Vi in (26) becomes 11arctan 2 2 2 11 arctan z2 - 28, 2 cos VI =- 2~~2si n which, together with the choice of 11 in (0, 2-4OmaX/r), implies that VI is nonpositive and is equal to zero if and only if 22 =0. Lyapunov stability theory tells us that the zero-input (-system is globally stable. Fur- thermore, from Barbalats lemma (see, e.g., [6, p. 491]), zz(t) goes to zero as t + +CO. As in [4, Sec.4.11, by an application of [5, Lemma 21 to the equation 22 =-z1 +E(sin(z3 +e,) -sin@,) , (45) we conclude that zl ( t ) also goes to zero as t -++CO. Therefore, the property (Al) has been proved. Finally, the proof of Theorem 1 is completed with the help of Lemma 3. AAA For a subclass of reference signals identified in As- sumption 1, that is, a set of periodic reference signals x, ( t ) , we can even conclude the uniform convergence of 2 d ( t ) - x,(t) to zero. Corollary 1 Under the conditions of Theorem 1, i f x,(t) and the deduced signal q(t) in Assumption 1 are periodic, all the states of the closed-loop system are uniformly bounded. In particular, Xd(t) - x,.(t) uni- formly converges t o zero as t -++ca. 5 Concluding remarks The problem of output-feedback tracking for the well- known nonlinear benchmark RTAC system has been solved globally for the first time. The novelty of this paper is the introduction of a nonlinear state transfor- mation which eliminates the quadratic dependence of the system equations on the unmeasured states. This transformation then allows the design of a novel ob- server/controller backstepping scheme, which leads to the desired global output-feedback tracking. Lyapunov-based backstepping designs are inherently robust to some types of disturbances, and can be robus- tified with respect to many other types. Our controller can also be robustified against several types of distur- bances using techniques very similar to those in [ll]. A complete study of the robustness issue is beyond the scope of this paper, but it undoubtedly is a topic that deserves further attention. References [l] G. Escobar, R. Ortega, and H. Sira-Ramirez, Output-feedback global stabilization of a nonlin- ear benchmark system using a saturated passivity- based controller, IEEE TCST, 7 (289-293) 1999. [2] R. A. Freeman and P. V. KokotoviC, Tracking controllers for systems linear in the unmeasured states, Autornatica, vol. 32, pp. 735-746, 1996. [3] M. J ankovic, D. Fontaine and P. V. KokotoviC, TORA example: cascade- and passivity-based control designs, IEEE TCST, 4 (292-297) 1996. [4] Z. P. J iang, D. J . Hill and Y. Guo, Stabilization and tracking via output feedback for the nonlinear benchmark system, Automatica, vol. 34, no. 7, pp. 907-915, 1998. [5] Z. P. J iang and H. Nijmeijer, Tracking control of mobile robots: a case study in backstepping, Autornatica, 33 (1393-1399) 1997. [6] M. KrstiC, I. Kanellakopoulos and P. V. Koko- toviC, Nonlinear and Adaptive Control Design. New York: J ohn Wiley & Sons, 1995. [7] R. Marino and P. Tomei, Nonlinear Control De- sign. London: Prentice-Hall, 1995. [8] L. Praly and Z. P. J iang, Stabilization by out- put feedback for systems with ISS inverse dynam- ics, Systems & Control Letters, vol. 21, pp. 19-33, 1993. [9] H. J . Sussmann and P. V. KokotoviC, The peak- ing phenomenon and the global stabilization of nonlinear systems , IEEE Trans. Automat. Con- trol, vol. 36, pp. 424-440, 1991. [lo] C.-J . Wan, D. S. Bernstein and V. T. Coppola, Global stabilization of the oscillating eccentric rotor, Nonlinear Dyn., vol. 10, pp. 49-62, 1996. [ll] J . Zhao and I. Kanellakopoulos, Flexible back- stepping design for tracking and disturbance at- tenuation, Int. J. Robust Nonlinear Control, vol. 8, pp. 331-348, 1998. [12] Y. Tan, I. Kanellakopoulos, and Z.-P. J iang, Nonlinear observer/controller design for a class of nonlinear systems, Proc. 37th IEEE Conf. Dec. Contr., pp. 2503-2508, Tampa, FL, 1998. 4807 Authorized licensed use limited to: Sharif University of Technology. Downloaded on December 28, 2009 at 01:44 from IEEE Xplore. Restrictions apply.