Nonlinear and Adaptive Control

In this course, we consider the nonlinear dynamical system

ẋ(t) = f (x(t)), x(t0 ) = x0 , t ∈ I x0 , (1)

with f (0) = 0 and x0 ∈ D ⊆ Rn , where Ix0 , (τmin , τmax ) ⊆ R denotes the maximal interval of
existence for the solution x(·).

Definition. Let D ⊆ Rn and let f : D → Rn . Then

• f is called Lipschitz continuous at x ∈ D if there exists a constant L = L(x) > 0 and a

neighborhood N of x such that

kf (x) − f (y)k ≤ Lkx − yk, (2)

for all y ∈ N ∩ D;

• f is called Lipschitz continuous on D if f is Lipschitz continuous at every point on D;

• f is called uniformly Lipschitz continuous on D if there exists L > 0 such that (2) holds with
x, y ∈ D;

• f is called globaly Lipschitz continuous if f is uniformly Lipschitz continuous on Rn .

Fact. If f (x) is continuous, then a solution to (1) exists.

Theorem. Consider (1). Assume f (·) is Lipschitz continuous on D. Then for all x0 ∈ D, there
exists τ ∈ (t0 , t1 ), where t0 ∼ t1 , such that (1) has a unique solution x : [t0 , τ ] → Rn over the
interval [t0 , τ ].
Theorem. Consider (1). Assume f (·) is globally Lipschitz continuous. Then for all x0 ∈ Rn ,
(1) has a unique solution x(−∞, ∞) → Rn over all t ∈ R.
Gronwall’s lemma. Assume there exists a continuous function x : R → R such that
Z t
0 ≤ x(t) ≤ α + βx(τ )dτ, t ≥ t0 , (3)
t0

where α, β ≥ 0. Then
x(t) ≤ αeβ(t−t0 ) , t ≥ t0 . (4)

Definition. Consider the nonlinear dynamical system (1). Then the zero solution x(t) ≡ 0 to
(1) is

• Lyapunov stable if for all ǫ > 0, there exists δ = δ(ǫ) > 0 such that

kx0 k < δ ⇒ kx(t)k < ǫ, t ≥ t0 ; (5)

• asymptotically stable if it is Lyapunov stable and there exists δ > 0 such that kx0 k < δ implies
limt→∞ kx(t)k = 0;

• unstable if it is not Lyapunov stable.

 Theorem (Lyapunov stability theorem). Consider (1). Let D ⊂ Rn be a simply connected
domain. Assume there exists a continuously differentiable function V : D → R such that

V (0) = 0, (6)
V (x) > 0, x ∈ D, x 6= 0, (7)
′
V (x)f (x) ≤ 0, x ∈ D. (8)

Then the zero solution x(t) ≡ 0 to (1) is Lyapunov stable. If, in addition,

V ′ (x)f (x) < 0, x ∈ D, x 6= 0, (9)

then the zero solution x(t) ≡ 0 to (1) is asymptotically stable.

Theorem (Krasovskii-LaSalle’s invariance principle). Consider (1) and assume Dc is a

compact invariant set with respect to (1). Assume that there exists a continuously differentiable
function V : Dc → R such that V ′ (x)f (x) ≤ 0, x ∈ Dc . Let R , {x ∈ Dc : V ′ (x)f (x) = 0} and let
M be the largest invariant set contained in R. If x0 ∈ Dc , then x(t) → M as t → ∞.

Theorem (Partial Asymptotic Stability). Consider (1). Suppose D is an invariant set

with respect to (1) and f (x) is Lipschitz continuous on D. Assume there exists a continuously
differentiable, positive-definite function V : D → R such that

V ′ (x)f (x) ≤ −W (x), (10)

where W : D → R is a continuous, nonnegative-definite function. Then there exists D0 ⊆ D such

that for all x0 ∈ D0 , x(t) → {x ∈ D : W (x) = 0} as t → ∞.
Rt
Lemma (Barbalat). Let σ : [0, ∞) be uniformly continuous and suppose limt→∞ 0 σ(τ )dτ
exists and is finite. Then σ(t) → 0 as t → ∞.

Definition (uniform continuity). A function σ : R → Rn is called uniformly continuous

if for every ε > 0 there exists δ = δ(ε) such that kσ(x) − σ(y)k < ε for all x, y ∈ R satisfying
|x − y| < δ.

Consider the nonlinear dynamical system G of the form

ẋ(t) = f (x(t)) + G(x(t))u(t), x(0) = x0 , t ≥ 0, (11)

y(t) = h(x(t)) + J(x(t))u(t), (12)

where x(t) ∈ Rn , u(t) ∈ Rm , y(t) ∈ Rℓ with f(0) = 0.

Definition. G is called dissipative with respect to the supply rate r(u, y) if the dissipation
inequality Z t
0≤ r(u(τ ), y(τ ))dτ, (13)
0
is satisfied.
G is called lossless with respect to the supply rate r(u, y) if (13) is satisfied with equality for all
t ≥ 0 and x0 = 0.
 Definition. Consider G. A continuous, nonnegative-definite function Vs : Rn → R, with
Vs (0) = 0, is called a storage function for G if Vs (·) satisfies
Z t
Vs (x(t)) ≤ Vs (x0 ) + r(u(τ ), y(τ ))dτ, t ≥ 0,
0

where x(·) is the solution to (11).

Definition. A nonlinear dynamical system G is zero-state observable if u(t) ≡ 0 and y(t) ≡ 0
implies x(t) ≡ 0.
Definition. A nonlinear dynamical system G is completely reachable if there exists a finite
ti < t0 and a square-integrable u : [ti , t0 ] → Rm such that x(t) can be driven from x(ti ) = 0 to
x(t0 ) = x0 .

Proposition.

i ) If G is passive with a continuously differentiable, positive-definite function Vs (·), then the zero
solution of the uncontrolled (u(t) ≡ 0) system G is Lyapunov stable.

ii ) If G is zero-state observable and nonexpansive with a continuously differentiable, positive-

definite function Vs (·), then the zero solution of the uncontrolled (u(t) ≡ 0) system G is
asymptotically stable.

Example (n-link manipulator).

M (q(t))q̈(t) + C(q(t), q̇(t))q̇(t) + g(q(t)) = u(t), q(0) = q0 , q̇(0) = q̇0 , t ≥ 0, (14)

y(t) = q̇(t), (15)

• M (q) > 0, q ∈ Rn ;
• C(q, q̇): accounts for centrifugal and coriolis forces, and has the property that Ṁ (q(t)) −
2C(q(t), q̇(t)) is skew-symmetric;
 T
• g(q): accounts for gravity forces such that g(q) = ∂U∂q(q) , where U (q) being a potential.
 Corollary. Let G be zero-state observable and completely reachable. Then G is passive (resp.,
exponentially passive) if and only if there exist Vs : Rn → R, ℓ : Rn → Rp , and W : Rn → Rp×m ,
and a scalar ε ≥ 0 (resp., ε > 0) such that Vs (·) is continuously differentiable, positive definite,
Vs (0) = 0, and

If, alternatively, J(x) + J T (x) > 0 holds, then G is exponentially passive (or passive) if and only if
there exist Vs : Rn → R, ℓ : Rn → Rp , and W : Rn → Rp×m, and a scalar ε > 0 (resp., ε ≥ 0) such
that Vs (·) is continuously differentiable, positive definite, Vs (0) = 0, and

Corollary. Let G be zero-state observable and completely reachable. Then G is nonexpansive

if and only if there exist Vs : Rn → R, ℓ : Rn → Rp , and W : Rn → Rp×m such that Vs (·) is
continuously differentiable, positive definite, Vs (0) = 0, and

If, alternatively, γ 2 Im − J T (x)J(x) > 0 holds, then G is nonexpansive if and only if there exist
Vs : Rn → R, ℓ : Rn → Rp , and W : Rn → Rp×m such that Vs (·) is continuously differentiable,
positive definite, Vs (0) = 0, and
 + +
G(s)

+
G(s) M1

M1
 (
;
)

 (
;
)
+
Figure 0.1: Equivalence via loop shifting

Consider the state-space realization of G(s) given by

ẋ(t) = Ax(t) + Bu(t), x(0) = x0 , t ≥ 0, (16)

y(t) = Cx(t) + Du(t), (17)

and the memoryless nonlinearity

u(t) = −σ(t, y(t)), (18)
where

σ ∈ Σ , {σ : R × Rm → Rm : σ(·, 0) = 0 and [σ(·, y) − M1 y]T [σ(·, y) − M2 y] ≤ 0, y ∈ Rm }. (19)

 
A B
Theorem. Consider the dynamical system (16)–(18) where M1 = 0. Suppose G(s) ∼
C D
is minimal and Im + M2 G(s) is strictly positive real (SPR). Then the zero solution of (16)–(18) is
globaly asymptotically stable for all σ(·, ·) ∈ Σ with M1 = 0.
 
A B
Corollary. Consider the dynamical system (16)–(18). Suppose G(s) ∼ is minimal.
C D
If det[Im + M1 G(s)] 6= 0, Re[s] ≥ 0, and (Im + M2 G(s))(Im + M1 G(s))−1 is SPR, then the zero
solution of (16)–(18) is globaly asymptotically stable for all σ(·, ·) ∈ Σ.

Theorem. Consider
 the
 dynamical system (16)–(18) with σ(t, y) replaced by σ(y) and M1 = 0.
A B
Suppose G(s) ∼ is minimal and assume there exists N = diag[N1 , . . . , Nm ] > 0 such that
C 0
det(Im + λN ) 6= 0, λ ∈ spec(A), and Im + M2 (Im + N s)G(s) is SPR, then the zero solution of (16)–
(18) is globaly asymptotically stable for all σ(·) ∈ Σ′ , {σ : Rm → Rm : σ(0) = 0 and σ T (y)[σ(y) −
M2 y] ≤ 0, y ∈ Rm }.
 Theorem. Consider the linear uncertain dynamical system

ẋ(t) = Ax(t) + Bu(t) + Dw(t)

where x(t) ∈ Rn , u(t) ∈ Rm , and w(t) ∈ Rd . Assume there exist Kg ∈ Rm×n and Ψ ∈ Rm×d such
that As , A + BKg is asymptotically stable and BΨ = D. Let Q1 , Q2 ∈ Rm×m be positive definite
and let P ∈ Rn×n satisfy
0 = ATs P + P As + R

with R > 0. Then the control law

u(t) = K(t)x(t) + Φ(t)w(t)

and the update laws

K̇(t) = −Q1 B T P x(t)xT (t)

Φ̇(t) = −Q2 B T P x(t)wT (t)

guarantees that the solution (x(t), K(t), Φ(t)) ≡ (0, Kg , −Ψ) of the closed-loop system is Lyapunov
stable and x(t) → 0 as t → ∞.

Theorem. Consider the linear uncertain dynamical system

ẋ(t) = Ax(t) + Bu(t) + Dw(t)

and the performance variable

z(t) = Ex(t).
Assume there exists Kg ∈ Rm×n such that As , A + BKg is asymptotically stable. Let Q ∈ Rm×m
be positive definite and let P ∈ Rn×n satisfy

0 = AT −2 T T
s P + P As + γ P DD P + E E.

Then the control law

u(t) = K(t)x(t)
and the update laws

K̇(t) = −QB T P x(t)xT (t)

guarantees that the solution (x(t), K(t)) ≡ (0, Kg ) of the closed-loop system is Lyapunov stable
and Z T Z T
T 2
z (t)z(t)dt ≤ γ wT (t)w(t)dt + V (x(0), K(0))
0 0
 Lemma. Consider the linear dynamical system G

ẋ(t) = Ax(t) + Bu(t),

y(t) = Cx(t).

G can be made SPR by means of

u(t) = Ky(t) + Lv(t), (20)
from v to y if and only if G is minimum phase and det CB 6= 0.
Proposition. Consider G. Let Q ∈ Rm×m be positive definite and and assume that G can be
made strictly positive real by (20). Then the control law

u(t) = K(t)y(t)

and the update law

K̇(t) = −Qy(t)y T (t)

guarantees that the solution (x(t), K(t)) ≡ (0, Kg ), where Kg ∈ Rm×m , of the closed-loop system
is Lyapunov stable and x(t) → 0 as t → ∞.