Chapter 5: Feedback Systems

Consider x = f (x, u), f (0, 0) = 0, (1) assume that u is obtained using state feedback: u = (x). The stability of the origin can be studied substituting (2) into (1): x = f (x, (x)). (3) (2)

Basic Feedback Stabilization : Consider the system x = ax2 + u where a is a non-zero constant (4)

Example 1

we look for a state feedback of the form u = (x) that makes x = 0 asymptotically stable. Consider rst setting u = ax2 x Substituting (5) into (4) we obtain x = x which is linear and globally asymptotically stable, as desired. Issues: this rst solution has 2 problems
(i)

(5)

It is based on the exact cancelation of the nonlinear term ax2 , thus requiring the exact knowledge of the system parameter(s). Canceling all nonlinear terms simplies the analysis but may not be a good idea.

(ii)

Example 2

: Consider the system given by x = ax2 x3 + u

following the approach in Example 1 we can set u which leads to x = x. 2 u1 cancels the terms ax2 and x3 , which are quite dierent:

u1 = ax2 + x3 x

The term in x2 is never desirable. It has a destabilizing eect. The term in x3 provides damping for x and can be benecial. Cancellation of the term x3 was achieved by incorporating the term x3 in the feedback law. Leads to very large input values.

Alternate solution: Given the system x = f (x, u) x Rn , u R, f (0, 0) = 0

we proceed to nd a feedback law u = (x) such that x = f (x, (x)) and x = 0 is asymptotically stable. We look for V1 = V1 (x) : D R satisfying
(i) (ii)

(6)

V1 (0) = 0, and V1 (x) is positive denite in D {0}. There exist L(x) : D R+ (positive denite) such that 1 (x) = V1 f (x, (x)) L(x) V x x D.

Example 3

: Consider again the system of example 2. x = ax2 x3 + u

2 dening V1 (x) = 1 2 x and computing V1 , we obtain

1 = ax3 x4 + xu. V In Example 2 we chose u = u1 = ax2 + x3 x. Thus 1 = ax3 x4 + x(ax2 + x3 x) = x2 V

L(x).

We now modify L(x) as follows: 1 = ax3 x4 + xu L(x) V choose u as, u = x ax2 . With this u, we obtain the following feedback system x = ax2 x3 + u = x x3 which is globally asymptotically stable. 2 Integrator Backstepping (x4 + x2 ).

We consider a system of the form x = f (x) + g (x), = u. f (0) = 0. Viewing the state variable as an independent input, we assume that there exists a state feedback control law that stabilizes the origin of the subsystem (7) = (x), (0) = 0 and a Lyapunov function V1 : D R+ such that 1 (x) = V1 [f (x) + g (x) (x)] L(x) 0 x D V x where L() : D R+ is a positive denite function in D. To stabilize now the system (7)(8) we proceed as follows:

x Rn , R

(7) (8)

We will make the following assumptions (see Figure 5.1(a)):

Adding and subtracting g (x)(x) to (7) (Figure 1(b)) we obtain the equivalent system x = f (x) + g (x)(x) + g (x)[ (x)] = u. (9) (10)

Dene z = (x) (x) = u (x) z = where (11) (12)

= x = [f (x) + g (x) ] (13) x x This change of variables can be seen as backstepping (x) through the integrator (Figure 1(c)). Dening v=z the resulting system is (Figure 1(d)) x = f (x) + g (x)(x) + g (x)z z = v (x) and (x) calulated as (13) V is dened as V = u Notice that (15)(16) is equivalent to (7)(8). To stabilize the system (15)(16) consider: 1 V = V (x, ) = V1 (x) + z 2 . 2 = V1 [f (x) + g (x)(x) + g (x)z ] + z z V x V1 V1 V1 f (x) + g (x)(x) + g (x)z + zv. = x x x We can choose V1 g (x) + kz , k > 0 v= x Thus = V1 f (x) + V1 g (x)(x) kz 2 V x x V1 [f (x) + g (x)(x)] kz 2 = x L(x) kz 2 .
4

(14)

(15) (16)

(17)

(18)

(19)

(19) implies that the origin x = 0, z = 0 is asymptotically stable. Since z = (x) and (0) = 0, the origin of the original system x = 0, = 0 is also asymptotically stable. The stabilizing state feedback law is given by u=z + (20)

u=

V1 [f (x) + g (x) ] g (x) k [ (x)]. x x

(21)

Example 4

: Consider the following system:

(22) (23)

Clearly this system is of the form (15)(16) with x f (x) g (x) = = = = x1 x2 3 f (x1 ) = ax2 1 x1 1

Step 1: Find = (x) to stabilize the origin x = 0. Dening V1 (x1 ) = 1 2 x 2 1 4 1 (x1 ) = ax3 V 1 x1 + x1 x2 Va (x1 ) x2 = (x1 ) = x1 ax2 1 we obtain x 1 = x1 x3 1. Step 2: To stabilize (22) (23), we use of the control law (21): u = V1 [f (x) + g (x) ] g (x) k [ (x)] x x 3 2 = (1 + 2ax1 )[ax2 1 x1 + x2 ] x1 k [x2 + x1 + ax1 ].

2 (x4 1 + x1 )

choosing

With this control law the origin is globally asymptotically stable. The composite Lyapunov function is 1 1 1 2 V = V1 + z 2 = x2 1 + [x2 (x1 )] 2 2 2 1 2 1 2 = x1 + [x2 x1 + ax2 1] . 2 2

Chain of Integrators

Consider now a system of the form x = f (x) + g (x)1 1 = 2 . . . k1 = k k = u For simplicity we focus on the third order system x = f (x) + g (x)1 1 = 2 2 = u (24) (25) (26)

To stabilize the origin we proceed as follows: we consider the rst subsystem (24). Assume that 1 = (x1 ) is a stabilizing control law for the system with Lyapunov function V1 . Consider now the rst 2 subsystems: x = f (x) + g (x)1 1 = 2 (27) (28)

We can stabilize this second order system using backstepping. Using the control law (21) and associated Lyapunov function V2 : 2 = (x, 1 ) = V2 =
(x) x [f (x) + g (x)1 ] 1 V x g (x) k [1 (x)] 2 V1 + 1 2 [1 (x)]

k>0

We now iterate view the third-order system as a more general version of (7)(8) with x= x 1 , = 2 , f= f (x) + g (x)1 0 , g= 0 1

Applying the backstepping algorithm once again, we obtain: (x) V2 u = x g (x) k [2 (x)], k > 0 x x (x, 1 ) (x, 1 ) 1 ]T V2 , V2 [0 , 1]T + [x, = , x 1 x 1 k [2 (x, 1 )], k > 0
7

or u = V2 (x, 1 ) (x, 1 ) 2 + [f (x) + g (x)1 ] + x 1 1 k [2 (x, 1 )], k > 0. 1 V = V2 + [2 (x, 1 )]2 2 1 1 = V1 + [1 (x)]2 + [2 (x, 1 )]2 . 2 2

The composite Lyapunov function is

Example 5

:
3 x 1 = ax2 1 x1 + x2 x 2 = u

Substitute x2 = x2d + 2 to the system equation, get

Design x2d as x2d = ax2 1 x1 , thus the rst system equation can be written as: x 1 = x1 x3 1 + 2 Since x2d is designed as x2d = ax2 1 x1 , then x 2d = d (ax2 1 x1 ) dt = (2x1 a + x1 )x 1 3 = (2x1 a + x1 )(ax2 1 x1 + x2 )

use (x1 , x2 ) to represent x 2d , then the second system equation can be written as: 2 = (x1 , x2 ) + u Design u as u = (x1 , x2 ) 2 + uaux The second equation of the system becomes: 2 = 2 + uaux Now for the following system: x 1 = x1 x3 1 2 = 2 + uaux
9

1 2 2 Using the following Lyapunov function candidate: V = 1 2 x1 + 2 2 , we have:

= x1 (x1 x3 V 1 + 2 ) + 2 (2 + uaux ) 2 x2 V 1 + x1 2 2 + 2 uaux if we let uaux = x1 , we can have

Therefore the stabalizing control law can be: u = (x1 , x2 ) 2 + uaux 3 = (2x1 a + x1 )(ax2 1 + x1 + x3 1 x1 + x2 ) (x 1 ) x1 2 3 4 = (2a + a)x1 (2a + 1)x1 + (2a + 1)x1 x2
Example 6

: Consider the following system, : x 1 = ax2 1 + x2 x 2 = x3 x 3 = u.

1 2 1 = ax2 Step 1: Consider the rst equation x 1 + (x1 ). Using V1 = 2 x1 , it is immediate that (x1 ) = x1 ax2 1 stabilizes the origin. Step 2: Consider the rst two subsystems. We propose the stabilizing law (with k > 0) and associated Lyapunov function:

(x1 , x2 )(= x3 ) =

(x1 ) V1 [f (x1 ) + g (x1 )x2 ] g (x1 ) k [x2 (x1 )] x1 x1

1 1 V2 = V1 + z 2 = V1 + [x2 (x1 )] 2 2 1 2 = V1 + [x2 + x1 + ax2 1] . 2 In our case, setting k = 1, (x1 ) = (1 + 2ax1 ) x1 V1 = x1 x1

2 (x1 , x2 ) = (1 + 2ax1 )[ax2 1 + x2 ] x1 [x2 + x1 + ax1 ]

Step 3: Consider the third order system with x= x1 x2 , = x3 , f= f (x1 ) + g (x1 )x2 0 , g= 0 1

From the results in the previous section we have that u = + g (x1 )x2 ]+ k [x3 (x1 , x2 )]
(x1 ,x2 ) [f (x1 ) x1 (x1 ,x2 ) x3 x2

V2 x2 +

k>0

is a stabilizing control law with associated Lyapunov function 1 V = V2 + [x3 (x1 , x2 ]2 2 2

11

Strict Feedback Systems

Consider now strict feedback systems of the form x = f (x) + g (x)1 1 = f1 (x, 1 ) + g1 (x, 1 )2 2 = f2 (x, 1 , 2 ) + g2 (x, 1 , 2 )3 . . . k1 = fk1 (x, 1 , 2 , , k1 ) + gk1 (x, 1 , 2 , , k1 )k k = fk (x, 1 , 2 , , k ) + gk (x, 1 , 2 , , k )u also called triangular systems. Considering rst the special case: x = f (x) + g (x) = fa (x, ) + ga (x, )u. If ga (x, ) =0 over the domain of interest, then we can dene 1 u = (x, ) [u1 fa (x, )]. ga (x, ) Substituting (31) into (30) we obtain the modied system x = f (x) + g (x) = u1 (32) (33) (29) (30)

(31)

which is of the form (7)(8). The stabilizing control law and associated Lyapunov function are thus: u = 1 (x, ) = 1 ga (x, ) [f (x) + g (x) ] x k1 > 0 (34)

V1 g (x) k1 [ (x)] fa (x, ) , x

1 V2 = V2 (x, ) = V1 (x) + [ (x)]2 . 2 Considering now the system x = f (x) + g (x)1 1 = f1 (x, 1 ) + g1 (x, 1 )2 2 = f2 (x, 1 , 2 ) + g2 (x, 1 , 2 )3
12

(35)

which can be seen as a special case of (29)(30) with x = fa x , = 2 , u = 3 , f = 1 = f2 , ga = g2 . f + g 1 f1 ,g = 0 g1 ,

The stabilizing control law and associated Lyapunov function for this systems are as follows: 2 (x, 1 , 2 ) = 1 g2 1 1 (f1 (x) + g1 (x)2 )+ (f + g1 ) + x 1 V2 g1 k2 [2 1 ] f2 , k2 > 0 1

(36)

1 V3 (x, 1 , 2 ) = V2 (x) + [2 1 (x, 1 )]2 . 2

(37)

13

Example 7

: Consider the following systems:

We begin by stabilizing the x subsystem. Using V1 = 1/2x2 1 we have that

Thus, x2 = (x1 ) = (x1 + a) results in

which shows that the x1 system is asymptotically stable. It then follows by (34)(35) that a stabilizing control law for the second-order system and the corresponding Lyapunov function are given by u = 1 (x1 , x2 ) = 1 3 2 3 {(1 + a)[ax2 1 x1 + x1 x2 ] x1 + 2 (1 + x2 ) k1 [x2 + x1 + a] (x1 + x2 )} , k1 > 0 1 1 2 = x2 1 + [x1 + x2 + a] . 2 2 2

V2

14