Introduction To Nonlinear Control Lecture # 6 State Feedback Stabilization
Introduction To Nonlinear Control Lecture # 6 State Feedback Stabilization
Lecture # 6
– p. 1/5
Backstepping
η̇ = f (η) + g(η)ξ
ξ̇ = fa (η, ξ) + ga (η, ξ)u
η ∈ Rn , ξ, u ∈ R, ga (η, ξ) 6= 0
η̇ = f (η) + g(η)ξ
η̇ = f (η) + g(η)φ(η)
∂V
[f (η) + g(η)φ(η)] ≤ −W (η), ∀η∈D
∂η
– p. 2/5
z = ξ − φ(η)
Vc (η, ξ) = V (η) + 12 z 2
∂V ∂V
V̇c = [f (η) + g(η)φ(η)] + g(η)z
∂η ∂η
∂φ
+ z fa (η, ξ) + ga (η, ξ)u − [f (η) + g(η)ξ]
∂η
– p. 3/5
∂V
V̇c ≤ −W (η) + z g(η) + fa (η, ξ)
∂η
∂φ
+ ga (η, ξ)u − [f (η) + g(η)ξ]
∂η
1 ∂φ
u = −fa (η, ξ) + [f (η) + g(η)ξ]
ga (η, ξ) ∂η
∂V
− g(η) − kz , k>0
∂η
V̇c ≤ −W (η) − kz 2
– p. 4/5
Example
z2 = x2 − φ(x1 ) = x2 + x1 + x21
– p. 5/5
Vc (x) = 12 x21 + 21 z22
– p. 6/5
Example
– p. 7/5
Vc = V + 12 z32
∂V ∂V
V̇c = (x21 − x31 + x2 ) + (z3 + φ)
∂x1 ∂x2
∂φ 2 3 ∂φ
+ z3 u − (x1 − x1 + x2 ) − (z3 + φ)
∂x1 ∂x2
∂V ∂φ ∂φ
u=− + (x21 − x31 + x2 ) + (z3 + φ) − z3
∂x2 ∂x1 ∂x2
– p. 8/5
Strict-Feedback Form
ẋ = f0 (x) + g0 (x)z1
ż1 = f1 (x, z1 ) + g1 (x, z1 )z2
ż2 = f2 (x, z1 , z2 ) + g2 (x, z1 , z2 )z3
..
.
żk−1 = fk−1 (x, z1 , . . . , zk−1 ) + gk−1 (x, z1 , . . . , zk−1 )zk
żk = fk (x, z1 , . . . , zk ) + gk (x, z1 , . . . , zk )u
gi (x, z1 , . . . , zi ) 6= 0 for 1 ≤ i ≤ k
– p. 9/5
Example (Recall from Lecture # 5)
V = 12 (η 2 + ξ2 )
V̇ = η(−η + η 2 ξ) + ξu = −η 2 + ξ(η 3 + u)
– p. 10/5
Passivity Based Control
f (0, 0) = 0
T ∂V
u y ≥ V̇ = f (x, u)
∂x
Theorem 14.4: If the system is
(1) passive with a radially unbounded positive definite
storage function and
(2) zero-state observable,
then the origin can be globally stabilized by
– p. 11/5
Proof:
∂V
V̇ = f (x, −φ(y)) ≤ −y T φ(y) ≤ 0
∂x
– p. 12/5
Choice of Output
∂V
ẋ = f (x) + G(x)u, f (x) ≤ 0, ∀x
∂x
No output is defined. Choose the output as
T
def ∂V
y = h(x) = G(x)
∂x
∂V ∂V
V̇ = f (x) + G(x)u ≤ y T u
∂x ∂x
Check zero-state observability
– p. 13/5
Example
ẋ1 = x2 , ẋ2 = −x31 + u
– p. 14/5
Feedback Passivation
u = α(x) + β(x)v
such that
is passive
– p. 15/5
Theorem [31]: The system
– p. 16/5
Stabilize the system at q = qr
e = q − qr , ė = q̇
– p. 17/5
Is it zero-state observable? Set v = 0
– p. 18/5
How does passivity-based control compare with feedback
linearization?
Example 13.20
u = h(x1 ) − (k1 x1 + k2 x2 )
" #
0 1
ẋ = x
−k1 −k2
– p. 19/5
Passivity-based control:
Z x1
V = h(z) dz + 12 x22
0
V̇ = x2 h(x1 ) − x2 h(x1 ) + x2 u = x2 u
Take y = x2
With u = 0, y(t) ≡ 0 ⇒ h(x1 (t)) ≡ 0 ⇒ x1 (t) ≡ 0
u = −σ(x2 ), [σ(0) = 0, yσ(y) > 0 ∀ y 6= 0]
ẋ1 = x2 , ẋ2 = −h(x1 ) − σ(x2 )
– p. 20/5
Linearization:
" #
0 1
, k = σ ′ (0)
−h′ (0) −k
s2 + ks + h′ (0) = 0
Sketch the root locus as k varies from zero to infinity
One of the p
two roots cannot be moved to the left of
Re[s] = − h′ (0)
– p. 21/5
Cascade Connection:
– p. 22/5
T
∂W
u=− F (z, y) + v ⇒ U̇ ≤ y T v
∂z
The system
– p. 23/5
Sliding Mode Control
Example
s = a1 x1 + x2 = 0
– p. 24/5
ṡ = a1 ẋ1 + ẋ2 = a1 x2 + h(x) + g(x)u
Suppose
a1 x2 + h(x)
≤ ̺(x)
g(x)
V = 12 s2
V̇ ≤ g(x)|s|̺(x) − g(x)β(x)|s|
– p. 25/5
s < 0, u = β(x)
u = −β(x) sgn(s)
– p. 26/5
√
V̇ ≤ −g0 β0 2V
dV √
√ ≤ −g0 β0 2 dt
V
√ V (s(t)) √
2 V ≤ −g0 β0 2 t
V (s(0))
p p 1
V (s(t)) ≤ V (s(0)) − g0 β0 √ t
2
|s(t)| ≤ |s(0)| − g0 β0 t
s(t) reaches zero in finite time
Once on the surface s = 0, the trajectory cannot leave it
– p. 27/5
s=0
– p. 28/5
ẋ1 = x2 ẋ2 = h(x) − g(x)β(x)sgn(s)
u = −k sgn(s)
– p. 30/5
Chattering
B Sliding manifold
B
B
B
B
B
H
HH B
s < 0 BH Y s>0
B
B
B
H
HH B a
B HH
B YH
H
– p. 31/5
Reduce the amplitude of the signum function
ṡ = a1 x2 + h(x) + g(x)u
[a1 x2 + ĥ(x)]
u=− +v
ĝ(x)
ṡ = δ(x) + g(x)v
g(x) g(x)
δ(x) = a1 1 − x2 + h(x) − ĥ(x)
ĝ(x) ĝ(x)
δ(x)
≤ ̺(x), β(x) ≥ ̺(x) + β0
g(x)
v = −β(x) sgn(s)
– p. 32/5
Replace the signum function by a high-slope saturation
function
s
u = −β(x) sat
ε
(
y, if |y| ≤ 1
sat(y) =
sgn(y), if |y| > 1
y
sgn(y) 6 sat ε 6
1 1
- -
y ε y
−1 −1
– p. 33/5
How can we analyze the system?
With c ≥ ε
n o
c
Ω = |x1 | ≤ a1
, |s| ≤ c is positively invariant
– p. 34/5
Inside the boundary layer:
s
ẋ1 = −a1 x1 + s ṡ = a1 x2 + h(x) − g(x)β(x)
ε
in finite time
– p. 35/5
What happens inside Ωε ?
h(0) = 0 ⇒ x̄1 = 0
– p. 36/5
z1 = x1 − x̄1 , z2 = s − a1 x̄1
s
ż2 = a1 x2 + h(x) − g(x)β(x)
ε
z2 + a1 x̄1
= a1 (z2 − a1 z1 ) + h(x) − g(x)β(x)
ε
z2
ż2 = ℓ(z) − g(x)β(x)
ε
h(x) x̄1
ℓ(z) = a1 (z2 − a1 z1 ) + a1 g(x)β(x) −
a1 g(x)β(x) ε
– p. 37/5
z2
ż1 = −a1 z1 + z2 , ż2 = ℓ(z) − g(x)β(x)
ε
ℓ(0) = 0, |ℓ(z)| ≤ ℓ1 |z1 | + ℓ2 |z2 |
g(x)β(x) ≥ g0 β0
V = 12 z12 + 12 z22
z2
V̇ = z1 (−a1 z1 + z2 ) + z2 ℓ(z) − g(x)β(x)
ε
g0 β0
V̇ ≤ −a1 z12 + (1 + ℓ1 )|z1 | |z2 | + ℓ2 z22 − z22
ε
– p. 38/5
g0 β0
V̇ ≤ −a1 z12 + (1 + ℓ1 )|z1 | |z2 | + ℓ2 z22
− z22
" #T " # ε
" #
1
|z1 | a1 − 2 (1 + ℓ1
) |z1 |
V̇ ≤ −
g0 β0
|z2 | − 12 (1 + ℓ1 ) ε − ℓ2 |z2 |
| {z }
Q
g0 β0
det(Q) = a1 − ℓ2 − 14 (1 + ℓ1 )2 > 0 for small ε
ε
h(0) = 0 ⇒ lim x(t) = 0
" t→∞
#
x̄1
h(0) =
6 0 ⇒ lim x(t) =
t→∞ 0
– p. 39/5
Regular Form:
η̇ = fa (η, ξ)
ξ̇ = fb (η, ξ) + g(η, ξ)u + δ(t, η, ξ, u)
η ∈ Rn−1 , ξ ∈ R, u ∈ R
fa (0, 0) = 0, fb (0, 0) = 0, g(η, ξ) ≥ g0 > 0
Sliding Manifold:
s = ξ − φ(η) = 0, φ(0) = 0
– p. 40/5
∂φ
ṡ = fb (η, ξ) − fa (η, ξ) + g(η, ξ)u + δ(t, η, ξ, u)
∂η
(
∂φ ˆ
− ĝ1 fˆb − ∂η fa , or
u = ψ(η, ξ) + v, where ψ =
0
ṡ = g(η, ξ)v + ∆(t, η, ξ, v)
∂φ
∆ = fb − fa + δ + gψ
∂η
∆(t, η, ξ, v)
g(η, ξ) ≤ ̺(η, ξ) + κ0 |v|
– p. 41/5
∆(t, η, ξ, v)
g(η, ξ) ≤ ̺(η, ξ) + κ0 |v|
– p. 42/5
sṡ ≤ −g(η, ξ)(1 − κ0 )β0 |s| ≤ −g0 β0 (1 − κ0 )|s|
s
v = −β(x) sat , ε>0
ε
sṡ ≤ −g0 β0 (1 − κ0 )|s|, for |s| ≥ ε
The trajectory reaches the boundary layer {|s| ≤ ε} in finite
time and remains inside thereafter
η̇ = fa (η, φ(η) + s)
– p. 43/5
α1 (kηk) ≤ V (η) ≤ α2 (kηk)
∂V
fa (η, φ(η) + s) ≤ −α3 (kηk), ∀ kηk ≥ γ(|s|)
∂η
|s| ≤ c ⇒ V̇ ≤ −α3 (kηk), for kηk ≥ γ(c)
α(r) = α2 (γ(r))
– p. 44/5
V
c α(.)
0
α(c)
α(ε)
ε c |s|
– p. 45/5
Theorem 14.1: Suppose all the assumptions hold over Ω.
Then, for all (η(0), ξ(0)) ∈ Ω, the trajectory (η(t), ξ(t)) is
bounded for all t ≥ 0 and reaches the positively invariant
set Ωε in finite time. If the assumptions hold globally and
V (η) is radially unbounded, the foregoing conclusion holds
for any initial state
Theorem 14.2: Suppose all the assumptions hold over Ω
̺(0) = 0
The origin of η̇ = fa (η, φ(η)) is exponentially stale
Then there exits ε∗ > 0 such that for all 0 < ε < ε∗ , the
origin of the closed-loop system is exponentially stable and
Ω is a subset of its region of attraction. If the assumptions
hold globally, the origin will be globally uniformly
asymptotically stable
– p. 46/5
Example
|θ1 | ≤ a, |θ2 | ≤ b
x2 = −kx1 ⇒ ẋ1 = −kx1 + θ1 x1 sin x2
V1 = 12 x21 ⇒ x1 ẋ1 ≤ −kx21 + ax21
s = x2 + kx1 , k>a
ṡ = θ2 x22 + x1 + u + k(x2 + θ1 x1 sin x2 )
u = −x1 − kx2 + v ⇒ ṡ = v + ∆(x)
∆(x) = θ2 x22 + kθ1 x1 sin x2
– p. 47/5
∆(x) = θ2 x22 + kθ1 x1 sin x2
– p. 48/5
Example: Normal Form
η̇ = f0 (η, ξ)
ξ̇i = ξi+1 , 1≤i≤ρ−1
ρ ρ−1
ξ̇ρ = Lf h(x) + Lg Lf h(x) u
y = ξ1
η̇ = f0 (η, ξ1 , · · · , ξρ−1 , ξρ )
ξ̇i = ξi+1 , 1≤i≤ρ−2
ξ̇ρ−1 = ξρ
– p. 49/5
s = ξρ − φ(η, ξ1 , · · · , ξρ−1 )
s = ξρ + k1 ξ1 + · · · + kρ−1 ξρ−1
– p. 50/5
Multi-Input Systems
η̇ = fa (η, ξ)
ξ̇ = fb (η, ξ) + G(η, ξ)u + δ(t, η, ξ, u)
η ∈ Rn−p , ξ ∈ Rp , u ∈ Rp
fa (0, 0) = 0, fb (0, 0) = 0, det(G) 6= 0,
Design φ s.t. the origin of η̇ = fa (η, φ(η)) is asymp. stable
s = ξ − φ(η)
∂φ
ṡ = fb (η, ξ) − fa (η, ξ) + G(η, ξ)u + δ(t, η, ξ, u)
∂η
– p. 51/5
∂φ
ṡ = fb (η, ξ) − fa (η, ξ) + G(η, ξ)u + δ(t, η, ξ, u)
∂η
u = ψ(η, ξ) + v
where (
∂φ ˆ
− Ĝ−1 fˆb − ∂η fa , or
ψ=
0
ṡ = G(η, ξ)v + ∆(t, η, ξ, v)
– p. 52/5
Case 1:
p
∆i (t, η, ξ, v) X
≤ ̺(η, ξ) + κij |vj |
g (η, ξ)
i j=1
p
X
̺(η, ξ) ≥ 0, κij ≤ κ0 < 1 (̺ and κ0 are Known)
j=1
̺(x)
β(x) ≥ + β0 , β0 > 0
1 − κ0
– p. 53/5
si ṡi = si gi vi + si ∆i ≤ gi {si vi + |si |[̺ + κ0 max |vi |]}
1≤i≤p
vi = −β sgn(si ), 1≤i≤p
Now use
si
vi = −β sat , 1≤i≤p
ε
Read Theorem 14.1 and 14.2 in the textbook
– p. 54/5
Case 2: G(η, ξ) is a positive definite matrix, uniformly in
(η, ξ)
sT G(η, ξ)s ≥ λm sT s, λm > 0
ṡ = G(η, ξ)v + ∆(t, η, ξ, v)
W = 12 sT s
– p. 55/5
Ẇ = sT ṡ = sT Gv+sT ∆ ≤ sT Gv+̺(η, ξ)ksk+κ0 ksk kvk
s
v = −β(η, ξ)
ksk
sT Gs
Ẇ ≤ −β + ̺ksk + κ0 βksk
ksk
≤ −β(λm − κ0 )ksk + ̺ksk
≤ −̺ksk − (λm − κ0 )β0 ksk + ̺ksk
≤ −(λm − κ0 )β0 ksk
(
y
s kyk
if kyk ≥ 1
Use v = −β ϕ , ϕ(y) =
ε y if kyk ≤ 1
– p. 56/5