Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Nonlinear Control
Lecture 10: Back Stepping
Farzaneh Abdollahi

Department of Electrical Engineering

Amirkabir University of Technology

Fall 2011
Farzaneh Abdollahi

Nonlinear Control

Lecture 10

1/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Integrator Back Stepping

More General Form
Back Stepping for Strict-Feedback Systems
Uncertain Systems
Trajectory Tracking
Stabilizing
Stabilizing 1
Stabilizing 2

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

2/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Integrator Back Stepping

I

Let us start with integrator back stepping:

I
I
I

= f () + g ()

(1)

= u

(2)

[ T ]T R n+1 : is the state

u R: control input
f : D R n and g : D R n : smooth in a domain D R n ; = 0, f (0) = 0

Objective: Design a state FB controller to stabilize the origin

( = 0, = 0)

We assume both f and g are known

It is a cascade connection:

I
I

(1) with input

Second is the integrator (2)

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

3/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Suppose (1) can be asym. stabilized

by = () with (0) = 0:
= f () + g ()()

and V () is a smooth p.d. Lyap fcn:

V
[f () + g ()()] W ()
D, W () is p.d.

Now add g ()() to (1):

= [f () + g ()()]
+ g ()[ ()]
= u

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

4/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Suppose (1) can be asym. stabilized

by = () with (0) = 0:
= f () + g ()()

and V () is a smooth p.d. Lyap fcn:

V
[f () + g ()()] W ()
D, W () is p.d.

Now add g ()() to (1):

= [f () + g ()()]
+ g ()[ ()]
| {z }
z

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

4/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Suppose (1) can be asym. stabilized

by = () with (0) = 0:
= f () + g ()()

and V () is a smooth p.d. Lyap fcn:

V
[f () + g ()()] W ()
D, W () is p.d.

Now add g ()() to (1):

= [f () + g ()()]
+ g ()[ ()]
| {z }
z

= [f () + g ()()]
z

+ g ()z
= u

=
Farzaneh Abdollahi

[f () + g ()]

Nonlinear Control

Lecture 10

4/26

Outline Integrator Back Stepping

I
I

More General Form Uncertain Systems Trajectory Tracking

Fig b to Fig c is back stepping though the integrator

v = u
= [f () + g ()()] + g ()z
z

= v

It is similar to (1) But input zero

Now let us design v to stabilize the over all system:

Vc (, ) = V () + 12 z 2
V c W () + V g ()z + zv

origin is a.s.

g () kz, k > 0

Choose v =

So V c W () kz 2

origin is a.s. ( = 0, z = 0)

(0) = 0
=0

V
u=
[f () + g ()]
g () k( ())

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

(3)
5/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Lemma: Consider the system (1)-(2). Let () be a stabilizing state

fb control law for (1) with (0) = 0, and V () be a Lyap fcn that
V W () for some p.d fcn W (). Then, the state feedback control
law (3) stabilizes the origin of (1)-(2), with V () + [ ()]2 /2 as a
Lyap fcn. Moreover, if all the assumptions hold globally and V () is
radially unbounded, the origin will be g.a.s.

Example: Consider

x 1 = x12 x13 + x2
x 2 = u

Therefore = x1 , = x2

To stabilize x1 = 0: x2 = (x1 ) = x12 x1

the nonlinear term x12 is canceled: x 1 = x1 x13

Why x13 is not canceled?

V = x12 x14 x12 , x1 R

V (x1 ) = x12 /2

The origin of x 1 is g.e.s.

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

6/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

To backstep: z2 = x2 (x1 ) = x2 + x1 + x12

Hence
x 1 = x1 x13 + z2
z 2 = u + (1 + 2x1 )(x1 x13 + z2 )

I
I

Now take Vc = 21 x12 + 21 z22

V c = x 2 x 4 + z2 [x1 + (1 + 2x1 )(x1 x 3 + z2 ) + u]
1

u = x1 (1 + 2x1 )(x1

The origin is g.a.s

Farzaneh Abdollahi

x13

Nonlinear Control

+ z2 ) z2

Lecture 10

V c = x12 x14 z22

7/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

For higher order systems we can apply the recursive application of

integrator back stepping

Example: Consider
x 1 = x12 x13 + x2
x 2 = x3
x 3 = u

After 1 back stepping:

x 1 = x12 x13 + x2
x 2 = x3

that x3 is input is g.s. by:

x3 = x1 (1 + 2x1 )(x1 x13 + z2 ) (x2 + x1 + x12 )=(x1 , x2 )

and V (x1 , x2 ) = 21 x12 + 12 (x2 + x1 + x12 )2

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

8/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Backstep again: z3 = x3 (x1 , x2 )

x 1 = x12 x13 + x2
x 2 = (x1 , x2 ) + z3

2
(x1 x13 + x2 )
( + z3 )
z 3 = u
x1
x2

Define Vc = V + z32 /2 V c =

V
x12 x14 (x2 +x1 +x12 )2 +z3 [ x
x
(x12 +xx3 +x2 ) x
(z3 +)+u]
2
1
2

V
u = x
+
2

The origin is g.a.s

Farzaneh Abdollahi

2
x1 (x1

+ xx3 + x2 ) +

Nonlinear Control

x2 (z3

+ ) z3

Lecture 10

9/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Back Stepping for More General Form

I

Consider
= f () + g ()

(4)

= fa (, ) + ga (, )u

I
I

fa and ga are smooth

If ga (, ) 6= 0 over the domain of interest: define
u

1
[ua fa (, )]
ga (, )

if a stabilizing state feedback control law () and a Lyap fcn. V ()

exists s.t. satisfy the conditions of Lemma:
1
V
u = ga (,)
[
[f () + g ()] g () k[ ()] fa (, )] k > 0

and Vc (, ) = V () + 21 [ ()]2

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

10/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Back Stepping for Strict-Feedback Systems

I

By recursive backstepping strict-FB systems can be stabilized:

x

= f0 (x) + g0 (x)z1

z 1 = f1 (x, z1 ) + g1 (x, z1 )z2

z 2 = f2 (x, z1 , z2 ) + g2 (x, z1 , z2 )z3
..
.
z k1 = fk1 (x, z1 , . . . , zk1 ) + gk1 (x, z1 , . . . , zk1 )zk
z k
I
I
I
I

= fk (x, z1 , . . . , zk ) + gk (x, z1 , . . . , zk )u

x Rn
z1 to zk are scalar
f0 (0) to fk (0) are zero
gi (x, z1 , ..., zi ) 6= 0 for 1 i k over the domain of interest

strict FB fi and gi in z i only depends on x, z1 , ..., zi

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

11/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Start the recursive procedure with x = f0 (x) + g0 (x)z1

Determine a stabilizing state fb z1 = 0 (x), 0 (0) = 0 and

V0
x [f0 (x) + g0 (x)0 (x)] W (x), W (x) is p.d.

Apply backstepping, consider

x

= f0 (x) + g0 (x)z1

z 1 = f1 (x, z1 ) + g1 (x, z1 )z2

I

The parameters can be defined as

= x, = z1 , u = z2 , f = f0 , g = g0 , fa = f1 , ga = g1

The stabilizing state fb:

0
1 (x, z1 ) = g11 [
x (f0 + g0 z1 )

The Lyap fcn: V1 (x, z1 ) =

Farzaneh Abdollahi

V0
x g0 k1 (z1 )
V0 (x) + 12 [z1 1 (x)]2

Nonlinear Control

Lecture 10

f1 ], k1 > 0

12/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Now consider:
x

= f0 (x) + g0 (x)z1

z 1 = f1 (x, z1 ) + g1 (x, z1 )z2

z 2 = f2 (x, z1 , z2 ) + g2 (x, z1 , z2 )z3
I

The parameters can be defined as = [x z1 ]T , = z2 , u = z3 , f =

[f0 + g0 z1 f1 ]T , g = [0 g1 ]T , fa = f2 , ga = g2

The stabilizing state fb: 2 (x, z1 , z2 ) =

1
V1
1 1
g2 [ x (f0 + g0 z1 ) + z1 (f1 + g1 z2 ) z1 g1 k2 (z2 ) f2 ], k2 > 0

The Lyap fcn: V2 (x, z1 , z2 ) = V1 (, z1 x) + 12 [z2 2 (x, z1 )]2

This process should be repeated k times to obtain u = k (x, z1 , ..., zk )

and Layp fcn Vk (x, z1 , ..., zk )

If a system has not defined in strict FB system, one can

transform the states by normal transformation

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

13/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Back Stepping for Uncertain Systems

I

Consider the system:

= f () + g () + (, )

(5)

= fa (, ) + ga (, )u + (, )

I
I
I
I
I
I

in domain D R n+1
contains ( = 0, = 0); R n , R
all fcns are smooth
If ga (, ) 6= 0 over the domain of interest
f , g , fa , ga are known; , are uncertain terms
f (0) = 0 and fa (0, 0) = 0
k (, )k2
| (, )|

I
Farzaneh Abdollahi

a1 kk2

(6)

a2 kk2 + a3 ||

Note: The upper bound on (, ) only depends on .

Nonlinear Control

Lecture 10

14/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

V () for the first equation is

V
2
[f () + g ()() + (, )] bkk2 ; b is pos. const.

= 0 is a.s. Equ. point of = f () + g ()() + (, )

Suppose
(7)
|()| a4 kk2 , k k2 a5

Consider the Layp fcn for whole system:

Vc (, ) = V () + 21 [ ()]2
V c = V [f () + g ()() + (, )] + V g ( ) + ( )[fa +

I
I

ga u +

Farzaneh Abdollahi

(f

+ g + )]

Nonlinear Control

Lecture 10

15/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

V () for the first equation is

V
2
[f () + g ()() + (, )] bkk2 ; b is pos. const.

= 0 is a.s. Equ. point of = f () + g ()() + (, )

Suppose
|()| a4 kk2 , k k2 a5
(7)

Consider the Layp fcn for whole system:

Vc (, ) = V () + 21 [ ()]2
V c = V [f () + g ()() + (, )] + V g ( ) + ( )[fa +

I
I

(f
+
g

)]

1
V
I Choose u =
ga [fa + (f + g ) g k( )], k > 0
I Vc bkk2 + 2a6 kk2 | | (k a3 )( )2 =


2 T 

kk2
kk2
b
a6
; a6 = a3 a4 +a22 +a5 a1

| |
| |
a6 (k a3 )

ga u +

|
Farzaneh Abdollahi

{z
P

Nonlinear Control

}
Lecture 10

15/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

V () for the first equation is

V
2
[f () + g ()() + (, )] bkk2 ; b is pos. const.

= 0 is a.s. Equ. point of = f () + g ()() + (, )

Suppose
(7)
|()| a4 kk2 , k k2 a5

Consider the Layp fcn for whole system:

Vc (, ) = V () + 21 [ ()]2
V c = V [f () + g ()() + (, )] + V g ( ) + ( )[fa +

I
I

(f
+
g

)]

1
V
I Choose u =
ga [fa + (f + g ) g k( )], k > 0
I Vc bkk2 + 2a6 kk2 | | (k a3 )( )2 =

2 T 


b
a6
kk2
kk2

; a6 = a3 a4 +a22 +a5 a1
| |
a6 (k a3 )
| |

ga u +

|
I

Choose k > a3 +

Farzaneh Abdollahi

{z
P

a62
b

V c min (P)[kk22 + | |2 ]
Nonlinear Control

Lecture 10

15/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Lemma: Consider the system (5), where the uncertainty satisfies

inequalities (6). Let () be a stabilizing state fb control law that
satisfies (7), and V () be a Lyap. fcn that guarantee a.s. of the first
Equatin of (5). Then the given state feedback control law in previous
slide, with k sufficiently large, stabilizes the origin of (5). Moreover, if all
the assumptions hold globally and V () is radially unbounded, the origin
will be g.a.s.

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

16/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Trajectory Tracking for A Second Order System [1]

I

A 3 link underactuated manipulator

I
I
I

The first two translational joints are actuated

The third revolute joint is not actuated
The linear approximation of this system is not
controllable since it is not influenced by gravity

The dynamics:
mx rx m3 lsin() m3 lcos()2 = 1
my ry + m3 lcos() m3 lsin()2 = 2
I m3 lsin()rx + m3 lcos()r )y = 0
+ rx sin() + ry cos() = 0
where [rx , ry ]: displacement of third joint;
orientation of third link respect to x axis; 1 2 :
input of actuated joints; mi : mass, Ii : inertia;
= (I3 + m3 l 2 )/(m3 l)

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

17/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Transform
the dynamics by:



1
rx + (cos() 1)
1
2 =

tan()
;
=
2
3
ry + sin()


m3 lcos()2 + (mx I2 sin2 ())vx + ( I2 sin()cos())vy
m3 lsin()2 + (fracI 2 sin()cos())vx + (my I2 cos 2 ())vy


vx
where
=
vy
#

"
u1
2
cos() sin()
cos() +
; , I = I3 + m3l 2
2
2

sin() cos()
(u2 cos () 2theta tan()
Therefore

Farzaneh Abdollahi

1 = u1
2 = u2
3 = 2 u1
Nonlinear Control

Lecture 10

18/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Objective: The states track the prescribed path by dij , dij

The reference trajectory will be stated by following dynamics:

d11 = u1d
d21 = u2d
d31 = d21 u1d

Define the tracking error x = d

x 11 = x12 x 12 = u1 u1d

(8)

x 21 = x22 x 22 = u2 u2d
x 31 = x32 x 32 = x21 u1d + 21 (u1 u1d )
I

Now the problem is finding u1 and u2 to make system (8) g.a.s.

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

19/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Let us redefine the system into three subsystems:


x 31 = x32
1
x 32 = x21 u1d + 21 (u1 u1d )

x 21 = x22
2
x 22 = u2 u2d

x 11 = x12

x 12 = u1 u1d

First we find u1 to stabilize

Then find u2 to stabilize 1 and 2

Farzaneh Abdollahi

Nonlinear Control

u1 = u1d

Lecture 10

20/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Stabilzing

I
I

x 11 = x12
x 12 = u1 u1d
This system can be stabilized by defining

u1 = u1d k1 x11 k2 x12 ,

k1 > 0, k2 > 0

(9)

where P() = 2 + k1 + k2 is Hurwitz

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

21/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Stabilizing 1

Assuming
that u1 u1d 0, 1 can be written as

x 31 = x32
1
x 32 = x21 u1d
Objective looking to design a stabilizing feedback x21

Assume u1d is uniformly bounded in t and smooth

Considering x32 as virtual input

2 x , c > 0 can stabilize the

It can be easily shown that 1 = c1 u1d
31
1
first eq.

Following the back stepping procedure

stabilizing x21 is
2 x c (c u 4 + u 2 x )] =
x21 = 2 = u1d1 [(2c1 u 1d u1d + 1)x13 c1 u1d
32
2 1 1d
1d 32
1
3
(c1 c2 u1d + 2c1 u 1d + u1d )x13 (c1 u1d + c2 u1d )x32

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

22/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Stabilizing
2

I

x 21 = x22
x 22 = u2 u2d
With u1 = u1d , 1 is e.sabilized by x21 = 2

Apply backstepping to find u2 :

I
I

Define x21 = x21 2

x 21 = x22 d [2 ]

Now define x22 = x22 3 , 3 = c3 x21 +

It can be easily find that the following u2 can stabilize the system

dt

u2 u2d = c4 x22 +

d
dt [2 ]

d
[3 ]
dt

(10)

= c3 c4 x21 (c3 + c4 )x22 + c3 c4 2 + (c3 + c4 )

I

d
d2
[2 ] + 2 [2 ]
dt
dt

It has been shown that (9) and (10) can e.stabilized (8). [1]

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

23/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

Simulation Results

For the 3 link manipulator consider the following desired traj:

rxd = r1 sin(at) (cos(arctan(r2 cos(at))) 1)
ryd = r18r2 sin(2at) sin(arctan(r2 cos(at)))
d (t) = arctan(r2 cos(at))

Define: r1 = r2 = a = 1, k1 = 4, k2 = 2 2, c1 = 2, c2 = 2, c3 = 4, c4 = 4

The results of tracking is shown in Figs.

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

24/26

Outline Integrator Back Stepping

Farzaneh Abdollahi

More General Form Uncertain Systems Trajectory Tracking

Nonlinear Control

Lecture 10

25/26

Outline Integrator Back Stepping

Farzaneh Abdollahi

More General Form Uncertain Systems Trajectory Tracking

Nonlinear Control

Lecture 10

26/26

Outline Integrator Back Stepping

More General Form Uncertain Systems Trajectory Tracking

N. P. I. Aneke, H. Nijmeijerz, and A. G. de Jager, Tracking

control of second-order chained form systems by cascaded
backstepping, Internaitonl Journal Of Robust And Nonlinear
Control, vol. 13, pp. 95115, 2003.

Farzaneh Abdollahi

Nonlinear Control

Lecture 10

26/26