NLC Lercture45

LECTURE 4
Introduction to nonlinear control design
Enseignant: Ning Liu

NONLINEAR CONTROL PROBLEMS
Generally, the tasks of control systems can be divided into two categories: stabilisation (or regulation) and
tracking.
In stabilisation problems, a control system is to be designed so that the state of the closed-loop system will
be stabilised around an equilibrium point, e.g. position control of robot arms.
In tracking control problems, the design objective is to construct a controller, so that the system output
tracks a given time-varying trajectory.
❖ Example of stabilisation problems

Consider the pendulum. Its dynamics is:
·· · m
ml 2θ + d θ + mgl sin θ = u.
· l
Choosing x1 = θ and x2 = θ, we have:
x· 1 x2
(x· 2) (− 2 x2 − ml sin x1 + u)
Remark 1 = d g . l
ml
Our task is to bring the pendulum from an initial angle, say θ(0) =
π
, to the vertical-up position with
θ m
6
a control law u.
Nonlinear control | Besançon | 2022-2023 2

tracking.
❖ Example of tracking problems

Consider the pendulum.
x· 1 x2
(x· 2) (− 2 x2 − ml sin x1 + u)
= d g .
l
ml
θ
Its output is :
Remark 1
y(t) = x1. m
We want the output y to track a reference signal r(t) = a1sin (ω1t) + a2 sin (ω2t).

tracking.
Remark 1 Relations between stabilisation and tracking

Normally, tracking problems are more difficult to solve than stabilisation problems, because in
tracking problems the controller should not only keep the whole state stabilised but also drive
the system output toward the desired output.
Stabilisation problems can often be regarded as a special case of tracking problems, with the
desired trajectory being a constant.
❖ In what follows, we will only discuss the controller design to stabilise the system
x· = f (x, u)
at equilibrium point x = xss.

LECTURE 5
State feedback control design
Enseignant: Ning Liu

I. STATE FEEDBACK STABILISATION
❖ Recall of state feedback for linear systems
x· = A x + Bu.
With the state feedback control u = − K x, the closed system becomes:
x· = (A − BK )x.
It is asymptotically stable if and only if the matrix A − BK is Hurwitz, i.e. the eigenvalues of A − BK
is in the open left-half complex plane.
❖ Example of state feedback for nonlinear systems
Consider a nonlinear scalar system
x· = x 2 + u. (5.1)
The linearisation at the origin gives x· = u. It can be stabilised by u = − k x with k > 0.

Apply this control to the nonlinear system (5.1), we get the closed loop system:
x· = x 2 − k x. (5.2)
The linearisation of (5.2) at the origin is x· = − k x. (5.3)
According to Theorem 3.1 (Stability and local linearization), the origin is asymptotically stable.
According to Remark 2 in lecture 3, the region of attraction is {x < k}.

I. STATE FEEDBACK STABILISATION
Notions of stabilisation
Consider the nonlinear system with state feedback:

x· = f (x, u), u = ϕ(x)
Local Stabilisation: The origin of x· = f (x, ϕ(x)) is asymptotically stable (e.g. the closed-
loop system (5.2)).
Regional Stabilisation: The origin of x· = f (x, ϕ(x)) is asymptotically stable and a given
region G is a subset of the region of attraction (for all x(0) ∈ G, lim x(t) = 0) (e.g. the
t→∞
control u = − k x achieves regional stabilisation of system (5.2) with the region of
attraction {x < k}).
Global Stabilisation: The origin of x· = f (x, ϕ(x)) is globally asymptotically stable.
Semiglobal Stabilisation: The origin of x· = f (x, ϕ(x)) is asymptotically stable and ϕ(x)
can be designed such that any given compact set (no matter how large) can be included in
the region of attraction (Typically u = ϕp(x) is dependent on a parameter p such that for
any compact set G, p can be chosen to ensure that G is a subset of the region of
attraction). (e.g. any compact set {−a ≤ x ≤ b} with a > 0 and b > 0 can be included in
the region of attraction by choosing k > b).

II. LINEARIZATION
Jacobian linearization
x· = f (x, u)
f (0,0) = 0, and f is continuously differentiable in a domain Dx × Du that contains the origin
(x = 0, u = 0) with Dx ⊂ ℝn and Du ⊂ ℝm.
After the linearisation:

x· = A x + Bu,
∂f ∂f
A= |x=0,u=0 and B = |
∂x ∂u x=0,u=0
Assume (A, B) is stabilizable.
Design a matrix K such that (A − BK ) is Hurwitz with u = − K x.
Closed-loop system:
x· = f (x, − K x)
Linearization:
[ ∂x ]
∂f ∂f
x· = (x, − K x) + (x, − K x) ⋅ (−K ) = (A − BK )x.
∂u x=0
Since (A − BK ) is Hurwitz, the origin is an exponentially stable equilibrium point of the
closed-loop system.
Simply and easy.

Drawback: it has limited theoretical guarantees of stability in nonlinear operations.
Stabilizable: the pair (A, B ) is controllable, or some uncontrollable eigenvalues of A Nonlinear control | Besançon | 2022-2023 8
have negative real parts.
II. LINEARIZATION
Ex.
Consider a pendulum with dynamic equations :
·· · (5.4)
θ + sin θ + b θ = cu.
Design a state feedback control law u to stabilize the pendulum at an angle θ = δ1.
Solution
·· ·
At the equilibrium θ= δ1, we have θ = 0 and θ = 0. Thus the control at the steady state
1
uss = sin δ1.
c
·
Choose state variables x1 = θ − δ1, x2 = θ and uδ = u − uss, (5.4) becomes:
x· 1 = x2, ·
(5.5) x = f (x, u) with f (0,0) = 0
x· 2 = − [sin(x1 + δ1) − sin δ1] − bx2 + cuδ .
Linearize (5.5) at the origin yields:
(−cos(x1 + δ1) −b) (−cos δ1

B = (0)
0 1 0 1
A= = ,
−b) c
x1=0
The pair (A, B) is controllable.

cos δ1 b
Take K = (k1 k 2). If k1 > − and k 2 > − , then (A − BK ) is Hurwitz.
c c
sin δ1 sin δ1 ·
The control u = uss + uδ = − Kx = − k1(θ − δ1) − k 2θ.
c c

III. FEEDBACK LINEARISATION
Transform a nonlinear system into a (fully/partially) linear system, and then use the linear design
techniques to complete the control design.
This is not like the previous Jacobian linearization. The feedback linearisation is achieved by
exact state transformations and feedback, rather than by linear approximation, in order to cancel
the nonlinearities and impose a desired linear dynamics.
❖ Consider the nonlinear system:

x· = f (x) + G(x)u,
f (0) = 0, x ∈ ℝn, u ∈ ℝm.
Change of variables
z = T(x), defined for all x∈ D ⊂ ℝn, that transforms the system into the controller form
z· = Az + B [ψ (x) + γ(x)u] (5.6)
where (A, B) is controllable and γ(x) is nonsingular for all x ∈ D.
Input transformation
u = γ −1(x)[−ψ (x) + v] ⇒ z· = Az + Bv
Standard linear techniques
v = − Kz
Design K such that (A − BK ) is Hurwitz.
The origin z = 0 of the closed-loop system z· = (A − BK )z is globally exponentially stable.
Closed-loop system in the x-coordinates:
u = γ −1(x)[−ψ (x) − K T(x)] x· = f (x) + G(x)γ −1(x)[−ψ (x) − K T(x)]

Ex.
Consider a 2D system with the following dynamics:
x· 1
(x· 2) (−x2 cos x1 + u cos 2x1)
−2x1 + a x2 + sin x1
= (5.6)
f (0,0) = 0, and f is continuously differentiable in a domain Dx × Du that contains the origin (x = 0,

u = 0) with Dx ⊂ ℝn and Du ⊂ ℝm.
❖ Solution with Jacobian linearization
(x1 sin x1 − 2u sin 2x1 0) ( 0 0)

−2 + cos x1 a −1 a
A= |x=0,u=0 =
(cos 2x1) (1)

0 0
B= |x=0,u=0 =
The pair (A, B) is controllable if a

≠ 0.
A linear control design u = − K x with (A − BK ) Hurwitz can stabilize the system in a small region
around the equilibrium point.
It is not obvious what controller can stabilize in a larger region.

Ex.
Consider a 2D system with the following dynamics:
x· 1
(x· 2) (−x2 cos x1 + u cos 2x1)
−2x1 + a x2 + sin x1
= (5.6)
f (0,0) = 0, and f is continuously differentiable in a domain Dx × Du that contains the origin (x = 0,

u = 0) with Dx ⊂ ℝn and Du ⊂ ℝm.
❖ Solution with Feedback linearization
(a x2 + sin x1)
x1
Change of variables z = T(x) = , (5.6) now becomes:
z·1
(z·2) (−2z1 cos z1 + cos z1 sin z1 + au cos 2z1)
−2z1 + z2
= (5.7)
1 New equivalent input

Input transformation u = (v − cos z1 sin z1 + 2z1 cos z1), (5.7) now becomes:
a cos 2z1
z·1
(z2) ( ) ( 0 0) ( 2) (1)
−2z1 + z2 −2 1 z1 0 (5.8)
· = = z + v
v
The problem of stabilising the original nonlinear system (5.6) using the original control u is
transformed into the stabilising the new linear system (5.8) using the new input v.

Apply linear state feedback:
( 2)
z1
v = − (k1 k 2) z (5.9)
Eg. k1 = 0, k 2 = 2 and v = − 2z2.
Therefore (5.8) becomes:

z·1
(z·2) ( −2z2 )
−2z1 + z2
= (5.10)
The poles of (5.10) are p1,2 = − 2. So the closed loop system (5.10) is stable with the input v in (5.9).
Back to the original system,
1
u= (v − cos z1 sin z1 + 2z1 cos z1)
a cos 2z1
1
(−2z2 − cos z1 sin z1 + 2z1 cos z1)
(5.11)
=
a cos 2z1
1
= (−2a x2 − 2a sin x1 − cos x1 sin x1 + 2x1 cos x1)
a cos 2x1
The original system (5.6) is obtained from z as:

1 z converges to zero, so x converges to
x1 = z1, x2 = (z − sin z1)
a 2 zero as well.

Summary of the Ex.
Pole-placement loop Linearization loop

0 x
v = − Kz u = u(x, v) x· = f (x, u)
−
z = T(x)
Although the result is valid in a large state space, it is not valid globally. According to
π kπ
(5.11), the control law is not well defined for x1 = ± .
4 2
To implement the control law, the new states z1, z2 must be available. If they cannot be
measured directly, the original state x1, x2 must be measured and used to compute z.
If there is uncertainty in the model, e.g. a uncertain, there will be error in calculating z
and u.
1
u= (v − cos z1 sin z1 + 2z1 cos z1)
a cos 2z1
1
= (−2z2 − cos z1 sin z1 + 2z1 cos z1)
a cos 2z1
1
= (−2a x 2 − 2a sin x1 − cos x1 sin x1 + 2x1 cos x1)
a cos 2x1
Definition 5.1 Feedback linearization
A single-input nonlinear system in the form x· = f (x) + g(x)u, with f (x) and g(x) smooth on
ℝn, is said to be input-state (feedback) linearizable if there exists a region Ω in ℝn, a
diffeomorphism T : Ω → ℝn, and a nonlinear feedback law:
u = α(x) + β(x)v
Such that the new states z = T(x) and the new input v satisfy a linear time invariant relation:
z· = Az + Bv,
With
0 1 0 ⋯ 0 0
0 0 1 ⋯ 0 0
A= ⋮ ⋱ ⋱ ⋯ 0 ,B = ⋮ A linear companion form
0 0 0 ⋯ 1 0
0 0 0 ⋯ 0 1
Remark 1
The matrix A and B in the transformed linear system is in a linear companion form. From the
fact that any linear system can be transformed to this form with additional linear state
transformation and pole placement, generality does not lost with other form of feedback
linearization.
Diffeomorphism: The map T must be invertible, i.e. there is a map T −1 such that for all
x = T −1(z) for all z inside the map T over the domain D. Moreover, both T and T −1 should
be continuously differentiable.

Transform a nonlinear system into a (fully/partially) linear system, and then use the linear design
techniques to complete the control design.
This is not like the previous Jacobian linearization. The feedback linearisation is achieved by
exact state transformations and feedback, rather than by linear approximation, in order to cancel
the nonlinearities and impose a desired linear dynamics.
❖ Consider the nonlinear system:

x· = f (x) + G(x)u,
f (0) = 0, x ∈ ℝn, u ∈ ℝm.
Change of variables
z = T(x), defined for all x∈ D ⊂ ℝn, that transforms the system into the controller form
z· = Az + B [ψ (x) + γ(x)u] (5.6)
Input transformation
u = γ −1(x)[−ψ (x) + v] ⇒ z· = Az + Bv
Standard linear techniques
v = − Kz
Design K such that (A − BK ) is Hurwitz.
The origin z = 0 of the closed-loop system z· = (A − BK )z is globally exponentially stable.
Closed-loop system in the x-coordinates:
u = γ −1(x)[−ψ (x) − K T(x)] x· = f (x) + G(x)γ −1(x)[−ψ (x) − K T(x)]

EXERCISE
Back to the example of the pendulum
x· 1
(x2) (−sin x1 − bx2 − u cos x1)
x2
· = (5.7)
The objective is to stabilize the pendulum at x1 = 0 using state feedback.

(a) Design a stabilising controller using Jacobian linearization
(b) Design a stabilising controller using feedback linearization and determine its domain of validity.
SOLUTION
(a) Jacobian linearization
(−cos x1 + u sin x1 −b) (−1 −b)

0 1 0 1
A= |x=0,u=0 =
(−cos x1) (−1)

0 0
B= | x=0,u=0 =
The pair (A, B) is controllable.

A linear control design u = − K x with (A − BK ) Hurwitz can stabilize the system in a small region
around the equilibrium point.

(b) Feedback linearization

(5.7) can be directly reformulated in the controller form without applying the change of variables, i.e.
T(x) = I.
1 New equivalent input
Input transformation u = (v − sin x1 − bx2), (5.7) now becomes:
cos x1
x· 1
(x2) (0 0) (x2) (−1)
x1
= ( 2) =
x 0 1 0
· + v (5.8)
−v
Apply linear state feedback:
( 2)
x1
v = − (k1 k 2) x (5.9)
Eg. k1 = 1, k 2 = 2 and v = − x1 − 2x2.

Therefore (5.8) becomes:
x· 1
(x· 2) (−x1 − 2x2)
x2
= (5.10)
The poles of (5.10) are p1,2 = − 1. So the closed loop system (5.10) is stable with the input v in (5.9).
Back to the original system,
1
u= (v − sin x1 − bx2)
cos x1
1
(−x1 − 2x2 − sin x1 − bx2)
(5.11)
=
cos x1

IV. PARTIALLY FEEDBACK LINEARISATION
Sometimes we can only partially linearize the original system.
❖ Consider again the nonlinear system:

x· = f (x) + G(x)u,
f (0) = 0, x ∈ ℝn, u ∈ ℝm.
Change of variables
(ξ) (T2(x))
η T1(x)
z= = T(x) = ,
defined for all x ∈ D ⊂ ℝn, that transforms the system into the
η· = f0(η, ξ), (5.12)
·
ξ = Aξ + B [ψ (x) + γ(x)u],
Assume f0(0,0) = 0, ψ (0) = 0, and f0, ψ, and γ are locally Lipschitz.
State feedback control
u = γ −1(x)[−ψ (x) + v]
(5.12) becomes:
η· = f0(η, ξ),
·
ξ = Aξ + Bv .
v = − Kξ, with (A − BK ) Hurwitz.
η· = f0(η, ξ),
· (5.13)
ξ = (A − BK )ξ .

IV. PARTIALLY FEEDBACK LINEARISATION
Lemma 4.1
The origin of the cascade connection
·
η· = f0(η, ξ), ξ = (A − BK )ξ .
is asymptotically (exponentially) stable if the origin of η· = f0(η,0) is asymptotically
(exponentially) stable.
This Lemma is only a local result.
Lemma 4.2
The origin of the cascade connection
·
η· = f0(η, ξ), ξ = (A − BK )ξ .
is globally asymptotically stable if the origin of η· = f0(η, ξ) is input-to-state stable.

V. BACKSTEPPING
❖ An introduction example
Consider the system:
x· 1 = x12 − x13 + x2, (5.14)
x· 2 = u .
Design a control law for u such that the closed loop system is asymptotically stable at the origin.
View x2 as an input to x1
Suppose x2 = ϕ(x1) and stabilises the origin x1 = 0.
1
Choose x2 = ϕ(x1) = − x12 − x1 and Va(x1) = x12.
2
We obtain:
· ·
x· 1 = − x1 − x13 and Va(x1) = − x12 − x14 ≤ 0 with Va(0) = 0 ∀x1 ∈ ℝ.
Hence the origin of x· 1 = − x1 − x13 is globally asymptotically stable.
Apply the change of variables

z2 = x2 − ϕ(x1) = x2 + x1 + x12
Then the system (5.14) is transformed into:
x· 1 = − x1 − x13 + z2,
∂ϕ(x1) ·
z·2 = x· 2 − x (5.15)
∂x1 1
= u + (2x1 + 1) (−x1 − x13 + z2) .

V. BACKSTEPPING
❖ An introduction example…
1 2 1 2
Taking V(x) = x + z as the Lyapunov function candidate for the transformed system (5.15),
2 1 2 2
·
V(x) = x1x· 1 + z2 z·2
(5.16)
= − x12 − x14 + z2 [x1 + (2x1 + 1) (−x1 − x13 + z2) + u]
Apply control law

u = − x1 − (2x1 + 1) (−x1 − x13 + z2) − z2
Then (5.16) becomes:
·
V(x) = − x12 − x14 − z22 ≤ 0.
·
V(x) = 0 at x1 = 0 and z2 = 0, which implies:
x2 = ϕ(x1) = − x12 − x1 = 0.
Hence the origin is globally asymptotically stable.

V. BACKSTEPPING
Definition: Backstepping design (for n ≥ 1)

Consider the system
η· = fa(η) + ga(η)ξ
·
ξ = fb(η, ξ) + gb(η, ξ)u
With η ∈ ℝn and ξ ∈ ℝ the state variables, u ∈ ℝ the input.
fa, ga, fb and gb are smooth functions in a domain that contains the origin (η = 0, ξ = 0).
fa(0) = 0, and gb ≠ 0.
Objective: stabilize the origin using state feedback.
The system can be viewed as a cascade system as:
· η· η
u g (η, ξ ) ξ ξ
b ∫ ga(η) ∫
fa(η)
η, ξ
fb(η, ξ )
View ξ as “virtual” control input to the η· = fa(η) + ga(η)ξ.

Suppose there is ξ = ϕ (η) with ϕ(0) = 0 that stabilizes the origin of
η· = fa(η) + ga(η)ϕ (η), i.e.
· ∂V ∂V
Va(η) = a η· = a [ fa(η) + ga(η)ϕ(η)] ≤ − W(η)
∂η ∂η
With W(η) positive definite.

V. BACKSTEPPING
Now backstep to the original system:

Propose a change of variables z = ξ − ϕ(η).
Then the system becomes:
η· = fa(η) + ga(η)ϕ(η) + ga(η)z
∂ϕ
z· = fb(η, ξ) − [ f (η) + ga(η)ϕ(η) + ga(η)z] + gb(η, ξ)u
∂η a
F(η,ξ)
Propose a Lyapunov function candidate V(η, ξ) as:

1 2 1
z = Va(η) + (ξ − ϕ(η)) .
2
V(η, ξ) = Va(η) +
2 2
· ∂V
V(η, ξ) = a η· + z z·
∂η
∂V
= a η· + z [F(η, ξ) + gb(η, ξ)u]
∂η
∂V ∂V
= a [ fa(η) + ga(η)ϕ(η)] + a ga(η)z + z [F(η, ξ) + gb(η, ξ)u]
∂η ∂η
[ ∂η a ]
∂Va
≤ − W(η) + z g (η) + F(η, ξ) + gb(η, ξ)u

V. BACKSTEPPING
gb(η, ξ) [ ∂η ]
1 ∂Va
Choose u =− ga(η) + F(η, ξ) + Kz , k > 0,
·
V(η, ξ) = − W(η) − Kz 2 ≤ 0.
·
V(η, ξ) = 0 when η = 0 and z = 0 ⇒ ξ = ϕ(η)
⇒ ξ = ϕ(0) = 0.
According to Lyapunov’s theorem and LaSalle’s invariant theorem, the system is
asymptotically stable at the origin.
If all the assumptions hold globally, and Va(η) is radially unbounded, the origin is globally
asymptotically stable.

V. BACKSTEPPING
EXERCISE
Apply the backstepping to the following 3D system in order to stabilize it:
x· 1 x12 − x13 + x2
x· 2 = x3 . (5.17)
x· 3 u
Hint: the above system can be viewed as a combination of the previous example (5.14)
with x3 as the input.
SOLUTION
We first investigate the system
x· 1
(x· 2) ( )
x12 − x13 + x2
= (5.18)
x3
with x3 as the input.

From the previous example (5.14), the system (5.18) can be globally stabilised by the ‘control’
x3 = − x1 − (2x1 + 1) (x12 − x13 + x2) − (x2 + x1 + x12)
,
= ϕ(x1, x2)
1 2 1
x1 + (x2 + x1 + x12) the corresponding Lyapunov function.
2
With Va(x1, x2) =
2 2

V. BACKSTEPPING
SOLUTION Now we start to backstep…
Change of variables:
z3 = x3 − ϕ(x1, x2).
(5.17) then becomes:
x· 1 x12 − x13 + x2
(5.19)
x· 2 = ϕ(x1, x2) + z3 .
z·3 u−
∂ϕ
(x1 − x1 + x2) −
2 3 ∂ϕ
(ϕ + z3)
∂x1 ∂x 2
Lyapunov function:
1 2
V = Va + z ,
2 3
with
· ∂V ∂V
V = a x· 1 + a x· 2 + z3z·3 (5.20)
∂x1 ∂x2
[ ∂x2 ∂x1 (z + ϕ) + u]
∂Va ∂ϕ 2 ∂ϕ
= − x12 − x14 − (x2 + x1 + x12) + z3 (x1 − x13 + x2) −
2
−
∂x2 3

V. BACKSTEPPING
SOLUTION Now we start to backstep…
Choose control:
∂Va ∂ϕ 2 ∂ϕ
(x1 − x1 + x2) + (z + ϕ) − z3.
3
u=− +
∂x2 ∂x1 ∂x2 3
(5.20) becomes:
·
V = − x12 − x14 − (x2 + x1 + x12) − z32 ≤ 0.
2
·
V(x) = 0 at x1 = 0, x2 = 0 and z3 = 0, which implies x3 = ϕ(0,0) = 0.
The origin is globally asymptotically stable.
Remark 2
By recursive application of backstepping, we can stabilize the strict feedback systems of the form
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·k−1 = fk−1(x, z1, ⋯, zk−1) + gk−1(x, z1, ⋯, zk−1)zk
z·k = fk (x, z1, ⋯, zk ) + gk (x, z1, ⋯, zk )u

A short summary of nonlinear state feedback control techniques:
Linearization : approximate the nonlinearity
Feedback linearization : compensate for the nonlinearity through change of state variables and input
transformation
Backstepping: decompose the system into low-order component, then build up back to high-order.
Other state feedback stabilization techniques exist, e.g.:

Passivity-based control (to be investigated in ‘Introduction energy-based control’)
Control Lyapunov functions
29

NLC Lercture45

Uploaded by

Copyright:

Available Formats

NLC Lercture45

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

NLC Lercture45

Uploaded by

Copyright:

Available Formats

LECTURE 4

Introduction to nonlinear control design

Enseignant: Ning Liu

❖ Example of stabilisation problems

Nonlinear control | Besançon | 2022-2023 2

❖ Example of tracking problems

Nonlinear control | Besançon | 2022-2023 3

Remark 1 Relations between stabilisation and tracking

Nonlinear control | Besançon | 2022-2023 4

Enseignant: Ning Liu

❖ Recall of state feedback for linear systems

❖ Example of state feedback for nonlinear systems

Consider a nonlinear scalar system

The linearisation at the origin gives x· = u. It can be stabilised by u = − k x with k > 0.

According to Remark 2 in lecture 3, the region of attraction is {x < k}.

Nonlinear control | Besançon | 2022-2023 6

Consider the nonlinear system with state feedback:

Nonlinear control | Besançon | 2022-2023 7

After the linearisation:

Simply and easy.

(−cos(x1 + δ1) −b) (−cos δ1

The pair (A, B) is controllable.

Nonlinear control | Besançon | 2022-2023 9

❖ Consider the nonlinear system:

u = γ −1(x)[−ψ (x) − K T(x)] x· = f (x) + G(x)γ −1(x)[−ψ (x) − K T(x)]

Nonlinear control | Besançon | 2022-2023 10

f (0,0) = 0, and f is continuously differentiable in a domain Dx × Du that contains the origin (x = 0,

(x1 sin x1 − 2u sin 2x1 0) ( 0 0)

(cos 2x1) (1)

The pair (A, B) is controllable if a

It is not obvious what controller can stabilize in a larger region.

Nonlinear control | Besançon | 2022-2023 11

f (0,0) = 0, and f is continuously differentiable in a domain Dx × Du that contains the origin (x = 0,

1 New equivalent input

Nonlinear control | Besançon | 2022-2023 12

Eg. k1 = 0, k 2 = 2 and v = − 2z2.

Therefore (5.8) becomes:

The original system (5.6) is obtained from z as:

Nonlinear control | Besançon | 2022-2023 13

Summary of the Ex.

Pole-placement loop Linearization loop

Definition 5.1 Feedback linearization

Nonlinear control | Besançon | 2022-2023 15

❖ Consider the nonlinear system:

u = γ −1(x)[−ψ (x) − K T(x)] x· = f (x) + G(x)γ −1(x)[−ψ (x) − K T(x)]

Nonlinear control | Besançon | 2022-2023 16

The objective is to stabilize the pendulum at x1 = 0 using state feedback.

(a) Jacobian linearization

(−cos x1 + u sin x1 −b) (−1 −b)

(−cos x1) (−1)

The pair (A, B) is controllable.

Nonlinear control | Besançon | 2022-2023 17

(b) Feedback linearization

Eg. k1 = 1, k 2 = 2 and v = − x1 − 2x2.

Nonlinear control | Besançon | 2022-2023 18

Sometimes we can only partially linearize the original system.

❖ Consider again the nonlinear system:

Nonlinear control | Besançon | 2022-2023 19

This Lemma is only a local result.