Adaptive Nonlinear Control—A Tutorial

Miroslav Krstić

University of California, San Diego

• Backstepping
• Tuning Functions Design
• Modular Design
• Output Feedback
• Extensions
• A Stochastic Example
• Applications and Additional Refer-
ences

main source:
Nonlinear and Adaptive Control Design (Wiley, 1995)
M. Krstić, I. Kanellakopoulos and P. V. Kokotović
 Backstepping (nonadaptive)

ẋ1 = x2 + ϕ(x1)Tθ , ϕ(0) = 0

ẋ2 = u
where θ is known parameter vector and ϕ(x1) is smooth nonlinear
function.

Goal: stabilize the equilibrium x1 = 0, x2 = −ϕ(0)Tθ = 0.

virtual control for the x1-equation:

α1(x1) = −c1x1 − ϕ(x1)Tθ , c1 > 0

error variables:

z1 = x1
z2 = x2 − α1(x1) ,

2
System in error coordinates:

ż1 = ẋ1 = x2 + ϕTθ = z2 + α1 + ϕTθ = −c1z1 + z2

∂α1 ∂α1 ! T
"
ż2 = ẋ2 − α̇1 = u − ẋ1 = u − x2 + ϕ θ .
∂x1 ∂x1
Need to design u = α2(x1, x2) to stabilize z1 = z2 = 0.

Choose Lyapunov function

1 2 1 2
V (x1, x2) = z1 + z2
2 2
we have
# $
∂α1 ! "
V̇ = z1 (−c1z1 + z2) + z2 u − x2 + ϕ T θ
∂x1
# $
∂α1 ! "
= −c1z12 + z2 u + z1 − x2 + ϕ T θ
∂x1
% &' (
=−c2 z2

⇒ V̇ = −c1z12 − c2z22

3
 z = 0 is globally asymptotically stable

invertible change of coordinates

x = 0 is globally asymptotically stable

The closed-loop system in z-coordinates is linear:

# $ # $# $
ż1 −c1 1 z1
= .
ż2 −1 −c2 z2

4
 Tuning Functions Design

Introductory examples:
A B C

ẋ1 = u + ϕ(x1)Tθ ẋ1 = x2 + ϕ(x1)Tθ ẋ1 = x2 + ϕ(x1)Tθ

ẋ2 = u ẋ2 = x3
ẋ3 = u
where θ is unknown parameter vector and ϕ(0) = 0.

Degin A. Let θ̂ be the estimate of θ and θ̃ = θ − θ̂,

Using
u = −c1x1 − ϕ(x1)Tθ̂
gives
ẋ1 = −c1x1 + ϕ(x1)Tθ̃

5
To find update law for θ̂(t), choose
1 2 1
V1(x, θ̂) = x1 + θ̃TΓ−1θ̃
2 2
then

V̇1 = −c1x2 + x ϕ(x ) Tθ̃ − θ̃ T Γ−1 θ̂˙

1 1 1! "
2 T
= −c1x1 + θ̃ Γ −1 Γϕ(x1)x1 − θ̂ ˙
% &' (
=0
Update law:

θ̂˙ = Γϕ(x1)x1, ϕ(x1)—regressor

gives
V̇1 = −c1x2
1 ≤ 0.
By Lasalle’s invariance theorem, x1 = 0, θ̂ = θ is stable and

lim x1(t) = 0
t→∞

6
Design B. replace θ by θ̂ in the nonadaptive design:

z2 = x2 − α1(x1, θ̂) , α1(x1, θ̂) = −c1z1 − ϕTθ̂

and strengthen the control law by ν2(x1, x2, θ̂)(to be designed)
∂α1 ! T
"
u = α2(x1, x2, θ̂) = −c2z2 − z1 + x2 + ϕ θ̂ + ν2(x1, x2, θ̂)
∂x1
error system

ż1 = z2 + α1 + ϕTθ = −c1z1 + z2 + ϕTθ̃

∂α1 ! T
" ∂α1 ˙
ż2 = ẋ2 − α̇1 = u − x2 + ϕ θ − θ̂
∂x1 ∂ θ̂
∂α1 T ∂α1 ˙
= −z1 − c2z2 − ϕ θ̃ − θ̂ + ν2(x1, x2, θ̂) ,
∂x1 ∂ θ̂
or
# $ # $# $   # $
ż1 −c1 1 z1 ϕT 0
= + ∂α1 T  θ̃ +
ż2 −1 −c2 z2 − ∂x ϕ − ∂α1 θ̂˙ + ν2(x1, x2, θ̂)
1 % ∂ θ̂ &' (
=0
remaining: design adaptive law.

7
Choose
1 1 1 1
V2(x1, x2, θ̂) = V1 + z22 = z12 + z22 + θ̃TΓ−1θ̃
2 2 2 2
we have
 
ϕT
V̇2 = −c1z12 − c2z22 + [z1, z2]  ∂α1 T  θ̃ − θ̃TΓ−1θ̂˙
− ∂x ϕ
1
- # $ 0
. / z
= −c1z12 − c2z22 + θ̃TΓ−1 Γ ϕ, − ∂α 1ϕ 1 − θ̂˙ .
∂x1 z2
The choice
# $ - τ 0
. / z1 1
'(%&
θ̂˙ = Γτ2(x, θ̂) = Γ ϕ, − ∂α1ϕ
∂x1 = Γ ϕz1 − ∂α
∂x1
1 ϕz
2
z2
% &' (
τ2

(τ1, τ2 are called tuning functions)

makes
V̇2 = −c1z12 − c2z22,
thus z = 0, θ̃ = 0 is GS and x(t) → 0 as t → ∞.
8
 2 3
z1
2 3 1 z2 2 3T τ2
θ #$ θ̃ ϕT #$ ϕT
+
!
!"
!
− ∂α1
ϕT +
!
!"
! !
− ∂α1
ϕT
!

# ∂x1 # ∂x1
− θ̂

2 3
−c1 1 "
−1 −c2

1
"
θ̂˙ "
Γ

The closed-loop adaptive system

9
Design C.
We have one more integrator, so we define the third error coordinate
and replace θ̂˙ in design B by potential update law,

z3 = x3 − α2(x1, x2, θ̂)

∂α1
ν2(x1, x2, θ̂) = Γτ2(x1, x2, θ̂).
∂ θ̂
Now the z1, z2-system is
# $ # $# $   # $
ż1 −c1 1 z1 ϕT 0
= + ∂α1 T  θ̃ +
ż2 −1 −c2 z2 − ∂x ϕ z3 + ∂α1 (Γτ2 − θ̂˙ )
1 ∂ θ̂
and
∂α1
V̇2 = −c1z12 − c2z22 + z2z3 + z2 (Γτ2 − θ̂˙ ) + θ̃T(τ2 − Γ−1θ̂˙ ).
∂ θ̂

10
z3-equation is given by
∂α2 ! T
" ∂α2 ∂α2 ˙
ż3 = u − x2 + ϕ θ − x3 − θ̂
∂x1 ∂x2 ∂ θ̂
∂α2 ! T
" ∂α2 ∂α2 ˙ ∂α2 T
= u− x2 + ϕ θ̂ − x3 − θ̂ − ϕ θ̃ .
∂x1 ∂x2 ∂ θ̂ ∂x1
Choose
1 2 1 2 1 2 1 2 1 T −1
V3(x, θ̂) = V2 + z3 = z1 + z2 + z3 + θ̃ Γ θ̃
2 2 2 2 2
we have
∂α1
2 2
V̇3 = −c1z1 − c2z2 + z2 (Γτ2 − θ̂˙ )
# ∂ θ̂ $
∂α2 ! " ∂α2 ∂α2 ˙
+z3 z2 + u − x2 + ϕTθ̂ − x3 − θ̂
∂x1 ∂x2 ∂ θ̂
- 0
∂α2
+θ̃T τ2 − ϕz3 − Γ−1θ̂˙ .
∂x1

11
Pick update law
 
- 0
. / z1
˙θ̂ = Γτ (x , x , x , θ̂) = Γ τ − ∂α2 ϕz ∂α1 ∂α2  
= Γ ϕ, ∂x ϕ, − ∂x ϕ  z2 
3 1 2 3 2 3 1 1
∂x1 z3
and control law
∂α2 ! T
" ∂α2
u = α3(x1, x2, x3, θ̂) = −z2 − c3z3 + x2 + ϕ θ̂ + x3 + ν 3 ,
∂x1 ∂x2
results in
6 7
∂α ∂α2 ˙
V̇3 = −c1z12 − c2z22 − c3z32 + z2 1 (Γτ2 − θ̂˙ ) + z3 ν3 − θ̂ .
∂ θ̂ ∂ θ̂
Notice
∂α
θ̂˙ − Γτ2 = θ̂˙ − Γτ3 − Γ 2 ϕz3
∂x1
we have
- 0
∂α2 ∂α1 ∂α2
V̇3 = −c1z12 − c2z22 − c3z32 + z3 ν3 − Γτ3 + Γ ϕz2 .
∂ θ̂ ∂ θ̂ ∂x1
% &' (
=0
Stability and regulation of x to zero follows.
12
Further insight:
   
    
ż1 −c1 1 0 z1 ϕT 0
   ∂α1
      − ∂α1 ϕT   (Γτ − ˙) 
θ̂  .
 ż2  =  −1 −c2 1   z2  +  ∂x1  θ̃ +  ∂ θ̂ 2 
 ∂α2 T  ∂α2
ż3 0 −1 −c3 z3 − ∂x ϕ ν3 − Γτ3
1 ∂ θ̂

˙ ˙ ∂α2
⇓ θ̂ − Γτ2 = θ̂ − Γτ3 − Γ ϕz3
∂x1
        
ż1 −c1 1 0 z1 ϕT 0
 ż2  =  −1 −c2 1 + ∂α1 Γ ∂α2 ϕ   z2  + 
 − ∂α1 T 
ϕ  θ̃ +  0 
∂ θ̂ ∂x1 ∂x1
ż3 0 −1 −c3 z3 − ∂α2
ϕ T ν3 − ∂α
∂ θ̂
2
Γτ3
∂x1

⇓ seletion of ν3
   
   
ż1 −c1 1 0 ϕT
  z1  
  
 ż2  =  −1 −c2 1 + ∂α1 Γ ∂α2ϕ    − ∂α1 ϕT
  z2 +

 θ̃.
 ∂ θ̂ ∂x1   ∂α∂x1 
ż3 0 −1 − ∂α1 Γ ∂α
∂x
2ϕ −c3 z3 − ∂x 2 ϕT
∂ θ̂ 1 1

13
General Recursive Design Procedure

parametric strict-feedback system:

ẋ1 = x2 + ϕ1(x1)Tθ
ẋ2 = x3 + ϕ2(x1, x2)Tθ
...
ẋn−1 = xn + ϕn−1(x1, . . . , xn−1)Tθ
ẋn = β(x)u + ϕn(x)Tθ
y = x1
where β and ϕi are smooth.

Objective: asymptotically track reference output yr (t), with

(i)
yr (t), i = 1, · · · , n known, bounded and piecewise continuous.

14
 ' ' '
Tuning functions design for tracking (z0 = 0, α0 = 0, τ0 = 0)
zi = xi − yr(i−1) − αi−1
i−1 6 7
8 ∂αi−1 ∂αi−1
αi (x̄i , θ̂, ȳr(i−1) ) = −zi−1 − ci zi − wiT θ̂ + xk+1 + (k−1)
yr
(k)
+ νi
∂xk ∂yr
k=1

∂αi−1 8 ∂αk−1
i−1

νi (x̄i , θ̂, ȳr(i−1) ) = + Γτi + Γwi zk

∂ θ̂ ∂ θ̂
k=2

τi (x̄i , θ̂, ȳr(i−1) ) = τi−1 + wi zi

8
i−1
∂αi−1
wi (x̄i , θ̂, ȳr(i−2) ) = ϕi − ϕk
∂xk
k=1
i = 1, . . . , n

x̄i = (x1 , . . . , xi ), ȳr(i) = (yr , ẏr , . . . , yr(i) )

Adaptive control law:

1 9 :
u= αn (x, θ̂, ȳr(n−1) ) + yr(n)
β(x)
Parameter update law:

θ̂˙ = Γτn (x, θ̂, ȳr(n−1) ) = ΓW z

15
Closed-loop system

ż = Az (z, θ̂, t)z + W (z, θ̂, t)Tθ̃

θ̂˙ = ΓW (z, θ̂, t)z ,
where
 
−c1 1 0 ··· 0
 
 −1

−c2 1 + σ23 ··· σ2n 

Az (z, θ̂, t) =  ... ... ... 
 0 −1 − σ23 
 ... ... ... ... 1 + σn−1,n 
 
0 −σ2n ··· −1 − σn−1,n −cn

∂αj−1
σjk (x, θ̂) = − Γwk
∂ θ̂
This structure ensures that the Lyapunov function
1 T 1
Vn = z z + θ̃TΓ−1θ̃
2 2
has derivative
n
8
V̇n = − ck zk2.
k=1

16
 Modular Design

Motivation: Controller can be combined with different identifiers.

(No flexibility for update law in tuning function design)

Naive idea: connect a good identifier and a good controller.

Example: error system

ẋ = −x + ϕ(x)θ̃
suppose θ̃(t) = e−t and ϕ(x) = x3, we have

ẋ = −x + x3e−t
;
But, when |x0| > 3,
2

1 x2
x(t) → ∞ as t → ln 2 0
3 x0 − 3/2
Conclusion: Need stronger controller.

17
Controller Design. nonlinear damping

u = −x − ϕ(x)θ̂ − ϕ(x)2x
closed-loop system

ẋ = −x − ϕ(x)2x + ϕ(x)θ̃ .
With V = 1 2
2 x , we have

V̇ = −x2 − ϕ(x)2x2 + xϕ(x)θ̃

2 32
1 1
= −x2 − ϕ(x)x − θ̃ + θ̃2
2 4
1
≤ −x2 + θ̃2 .
4
bounded θ̃(t) ⇒ bounded x(t)

18
For higher order system
ẋ1 = x2 + ϕ(x1)Tθ
ẋ2 = u
set

α1(x1, θ̂) = −c1x1 − ϕ(x1)Tθ̂ − κ1|ϕ(x1)|2x1, c1 , κ1 > 0

and define
z2 = x2 − α1(x1, θ̂)
error system

ż1 = −c1z1 − κ1|ϕ|2z1 + ϕTθ̃ + z2

∂α1 ! T
" ∂α1 ˙
ż2 = ẋ2 − α̇1 = u − x2 + ϕ θ − θ̂ .
∂x1 ∂ θ̂

19
Consider
1 2 1 2
V2 = V1 + z2 = |z|
2 2
we have
# $
1 ∂α1 ! "
∂α1 ˙
V̇2 ≤ −c1z12 + 2
|θ̃| + z1z2 + z2 u − T
x2 + ϕ θ − θ̂
4κ1 ∂x1 ∂ θ̂
# - 0$
1 ∂α1 ! " ∂α1 T ∂α1 ˙
2
≤ −c1z1 + 2
|θ̃| +z2 u+z1 − T
x2 +ϕ θ̂ − ϕ θ̃+ θ̂ .
4κ1 ∂x1 ∂x1 ∂ θ̂
controller
< < < <
< ∂α <2 < ∂α T<2 ∂α1 ! "
< 1 < < 1 < T
u = −z1 − c2z2 − κ2 < ϕ< z − g2 < < z + x2 + ϕ θ̂ ,
< ∂x1 < 2 < ∂ θ̂ < 2 ∂x1
achieves
- 0
1 1 1 ˙2
V̇2 ≤ −c1z12 − c2z22 + + 2
|θ̃| + |θ̂|
4κ1 4κ2 4g2

bounded θ̃, bounded θ̂˙ (or ∈ L2) ⇒ bounded x(t)

20
 ' '
Controller design in the modular approach (z0 = 0, α0 = 0)
zi = xi − yr(i−1) − αi−1
i−1 6 7
8 ∂αi−1 ∂αi−1
αi (x̄i , θ̂, ȳr(i−1) ) = −zi−1 − ci zi − wiT θ̂ + xk+1 + y (k)
(k−1) r
− s i zi
∂xk ∂yr
k=1
8 ∂αi−1
i−1

wi (x̄i , θ̂, ȳr(i−2) ) = ϕi − ϕk

∂xk
k=1
< <
< T <2
∂α
κi |wi |2 + gi <
i−1 <
si (x̄i , θ̂, ȳr(i−2) ) =
< ∂ θ̂ <
i = 1, . . . , n

x̄i = (x1 , . . . , xi ), ȳr(i) = (yr , ẏr , . . . , yr(i) )

Adaptive control law:

1 9 (n−1) (n)
:
u= αn (x, θ̂, ȳr ) + yr
β(x)

Controller module guarantees:

If θ̃ ∈ L∞ and θ̂˙ ∈ L2 or L∞ then x ∈ L∞

21
Requirement for identifier
error system

ż = Az (z, θ̂, t)z + W (z, θ̂, t)Tθ̃ + Q(z, θ̂, t)Tθ̂˙

where
 
−c1 − s1 1 0 ··· 0
 ... 

 −1 −c2 − s2 1 . . . 

Az (z, θ̂, t) =  0 −1 ... ... 0 
 
 ... ... ... ... 1 
 
0 ··· 0 −1 −cn − sn
   
w1T 0
   
T  w2T  T

 − ∂α1 

W (z, θ̂, t) = 
 ...  ,
 Q(z, θ̂, t) =  ∂ θ̂
...  .
   
 
wn T ∂αn−1
−
∂ θ̂

22
Since
 
1 0 ··· 0
 ... ... 

 − ∂α 1 1 
 '
W (z, θ̂, t)T =  ∂x1
... ... ...  F (x) = N (z, θ̂, t)F (x)T .
T
 0 
 ∂α ∂α 
− ∂xn−1 · · · − ∂x n−1 1
1 n−1

Identifier properties:

(i) θ̃ ∈ L∞ and θ̂˙ ∈ L2 or L∞,

(ii) if x ∈ L∞ then F (x(t))Tθ̃(t) → 0 and θ̂˙ (t) → 0.

23
 Identifier Design

Passive identifier

x
ẋ = f + F Tθ

'(
$
+
%&
+
#
−
! "
x̂
! x̂˙ = A0 − λF TF P (x̂ − x) + f + F Tθ̂

θ̂ 1 θ̂˙
" "
ΓF P

24
 +, θ̃ FP+!
)* +
!
#
−

Γ "

. /
+̇ = A0 − λF (x, u) F (x, u)P + + F (x, u)Tθ̃
T

update law
θ̂˙ = ΓF (x, u)P + , Γ = ΓT > 0 .
Use Lyapunov function

V = θ̃TΓ−1θ̃ + +TP +
its derivative satisfies
λ ˙ |2 .
V̇ ≤ −+T+ − |θ̂
λ̄(Γ)2
Thus, whenever x is bounded, F (x(t))Tθ̃(t) → 0 and θ̂˙ (t) → 0.
=∞
(+̇(t) → 0 because 0 +̇(τ )dτ = −+(0) exists, Barbalat’s lemma...)
25
Swapping identifier

x
ẋ = f + F Tθ

'(
$
> ? Ω0 +! +
Ω̇0 = A0 − λF TF P (Ω0 − x) + f +
%&
#
−

> ? ΩT! ........

Ω̇ = A0 − λF T F P Ω + F .. ..
#

1
θ̂ "
θ̂˙ ΓΩ
"
1 + ν|Ω|2

26
 '
define +̃ = x + Ω0 − ΩTθ,
. /
T
+̃˙ = A0 − λF (x, u) F (x, u)P +̃ .
Choose
1 T −1
V = θ̃ Γ θ̃ + +̃P +̃
2
we have
3 +T+
V̇ ≤ − T
,
4 1 + νtr{Ω Ω}
proves identifier properties.

27
 Output Feedback Adaptive Designs

# $
0
ẋ = Ax + φ(y) + Φ(y)a + σ(y)u , x ∈ IRn
b
y = eT
1 x,

 
0
In−1 
A =  ...

 ,
0 ··· 0
   
ϕ0,1(y) ϕ1,1(y) · · · ϕq,1(y)

φ(y) =  ... 
 ,

Φ(y) =  ... ... 
 ,
ϕ0,n(y) ϕ1,n(y) · · · ϕq,n(y)

unknown constant parameters:

a = [a1, . . . , aq ]T , b = [bm, . . . , b0]T .

28
State estimation filters

Filters:

ξ̇ = A0ξ + ky + φ(y)
Ξ̇ = A0Ξ + Φ(y)
λ̇ = A0λ + enσ(y)u

j
vj = A0λ, j = 0, . . . , m
ΩT = [vm, . . . , v1, v0, Ξ]

29
Parameter-dependent state estimate

x̂ = ξ + ΩTθ

The vector k = [k1, . . . , kn]T chosen so that the matrix

A0 = A − keT
1
is Hurwitz, that is,

P A0 + AT
0 P = −I, P = PT > 0

The state estimation error

ε = x − x̂
satisfies
ε̇ = A0ε

30
Parametric model for adaptation:

ẏ = ω0 + ω Tθ + ε2
= bmvm,2 + ω0 + ω̄ Tθ + ε2 ,
where

ω0 = ϕ0,1 + ξ2
ω = [vm,2, vm−1,2, . . . , v0,2, Φ(1) + Ξ(2)]T
ω̄ = [0, vm−1,2, . . . , v0,2, Φ(1) + Ξ(2)]T .

31
Since the states x2, . . . , xn are not measured, the backstepping design
is applied to the system

ẏ = bmvm,2 + ω0 + ω̄ Tθ + ε2
v̇m,i = vm,i+1 − kivm,1 , i = 2, . . . , ρ − 1
v̇m,ρ = σ(y)u + vm,ρ+1 − kρvm,1 .
The order of this system is equal to the relative degree of the plant.

32
Extensions

Pure-feedback systems.
ẋi = xi+1 + ϕi(x1, . . . , xi+1)Tθ, i = 1, . . . , n − 1
! "
ẋn = β0(x) + β(x)Tθ u + ϕ0(x) + ϕn(x)Tθ ,
where ϕ0(0) = 0, ϕ1(0) = · · · = ϕn(0) = 0, β0(0) )= 0.

Because of the dependence of ϕi on xi+1, the regulation or tracking

for pure-feedback systems is, in general, not global, even when θ is
known.

33
Unknown virtual control coefficients.
ẋi = bixi+1 + ϕi(x1, . . . , xi)Tθ, i = 1, . . . , n − 1
ẋn = bnβ(x)u + ϕn(x1, . . . , xn)Tθ ,
where, in addition to the unknown vector θ, the constant coefficients
bi are also unknown.

The unknown bi-coefficients are frequent in applications ranging from

electric motors to flight dynamics. The signs of bi, i = 1, . . . , n, are
assumed to be known. In the tuning functions design, in addition to
estimating bi, we also estimate its inverse 1i = 1/bi. In the modular
design we assume that in addition to sgnbi, a positive constant ςi is
known such that |bi| ≥ ςi. Then, instead of estimating 1i = 1/bi, we
use the inverse of the estimate b̂i, i.e., 1/b̂i, where b̂i(t) is kept away
from zero by using parameter projection.

34
Multi-input systems.
Ẋi = Bi(X̄i)Xi+1 + Φi(X̄i)Tθ, i = 1, . . . , n − 1
Ẋn = Bn(X)u + Φn(X)Tθ ,
. /T
T
where Xi is a νi-vector, ν1 ≤ ν2 ≤ · · · ≤ νn, X̄i = X1 , . . . , XiT ,
@i
ν
X = X̄n, and the matrices Bi(X̄i) have full rank for all X̄i ∈ IR j=1 j .
The input u is a νn-vector.

The matrices Bi can be allowed to be unknown provided they are

constant and positive definite.

35
Block strict-feedback systems.
ẋi = xi+1 + ϕi(x1, . . . , xi, ζ1, . . . , ζi)Tθ, i = 1, . . . , ρ − 1
ẋρ = β(x, ζ)u + ϕρ(x, ζ)Tθ
ζ̇i = Φi,0(x̄i, ζ̄i) + Φi(x̄i, ζ̄i)Tθ, i = 1, . . . , ρ
. /T
with the following notation: x̄i = [x1, . . . , xi]T, ζ̄i = ζ1T, . . . , ζiT ,
x = x̄ρ, and ζ = ζ̄ρ.

Each ζi-subsystem is assumed to be bounded-input bounded-state

(BIBS) stable with respect to the input (x̄i, ζ̄i−1). For this class of sys-
tems it is quite simple to modify the procedure in the tables. Because
of the dependence of ϕi on ζ̄i, the stabilizing function αi is augmented
@i−1 ∂αi−1
by the term + k=1 ∂ζ Φk,0, and the regressor wi is augmented by
k
6 7T
@i−1 ∂αi−1
− k=1 Φi ∂ζk .

36
Partial state-feedback systems. In many physical systems there are
unmeasured states as in the output-feedback form, but there are also
states other than the output y = x1 that are measured. An example
of such a system is

ẋ1 = x2 + ϕ1(x1)Tθ
ẋ2 = x3 + ϕ2(x1, x2)Tθ
ẋ3 = x4 + ϕ3(x1, x2)Tθ
ẋ4 = x5 + ϕ4(x1, x2)Tθ
ẋ5 = u + ϕ5(x1, x2, x5)Tθ .

The states x3 and x4 are assumed not to be measured. To apply the

adaptive backstepping designs presented in this chapter, we combine
the state-feedback techniques with the output-feedback techniques.
The subsystem (x2, x3, x4) is in the output-feedback form with x2 as a
measured output, so we employ a state estimator for (x2, x3, x4) using
the filters introduced in the section on output feedback.

37
 Example of Adaptive Stabilization in the Presence of a
Stochastic Disturbance

dx = udt + xdw
A B
w: Wiener process with E dw2 = σ(t)2dt, no a priori bound for σ

Control laws:

Disturbance Attenuation: u = −x − x3
Adaptive Stabilization: u = −x − θ̂x, θ̂˙ = x2

38
 Disturbance Attenuation Adaptive Stabilization
7

5
4.5

4
4
3.5

3
x 3

x 2.5

2 2
1.5
%
%&&
1 1

0.5

0 0
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
t t
3

2.5

θ̂ 1.5

0.5

t
0 0.5 1 1.5 2 2.5

39
Major Applications of Adaptive Nonlinear Control

• Electric Motors Actuating Robotic Loads

Nonlinear Control of Electric Machinery, Dawson, Hu, Burg, 1998.
• Marine Vehicles (ships, UUVs; dynamic positioning, way point
tracking, maneuvering)
Marine Control Systems, Fossen, 2002
• Automotive Vehicles (lateral and longitudinal control, trac-
tion, overall dynamics)
The groups of Tomizuka and Kanellakopoulos.

Dozens of other occasional applications, including: aircraft wing rock,

compressor stall and surge, satellite attitude control.

40
Other Books on Adaptive NL Control Theory Inspired by KKK

1. Marino and Tomei (1995),

Nonlinear Control Design: Geometric, Adaptive, and Robust

2. Freeman and Kokotovic (1996),

Robust Nonlinear Control Design: State Space and Lyapunov Techniques

3. Qu (1998),
Robust Control of Nonlinear Uncertain Systems

4. Krstic and Deng (1998),

Stabilization of Nonlinear Uncertain Systems

5. Ge, Hang, Lee, Zhang (2001),

Stable Adaptive Neural Network Control

6. Spooner, Maggiore, Ordonez, and Passino (2002),

Stable Adaptive Control and Estimation for Nonlinear Systems: Neural and
Fuzzy Approximation Techniques

7. French, Szepesvari, Rogers (2003),

Performance of Nonlinear Approximate Adaptive Controllers

41
Adaptive NL Control/Backstepping Coverage in Major Texts

1. Khalil (1995/2002),
Nonlinear Systems

2. Isidori (1995),
Nonlinear Control Systems

3. Sastry (1999),
Nonlinear Systems: Analysis, Stability, and Control

4. Astrom and Wittenmark (1995),

Adaptive Control

42