Indian Institute of Technology Bombay

Department of Aerospace Engineering

AE 666 Adaptive and Learning Control Systems
Spring 2024-2025

Tutorial 2 Solution February 21, 2025

1. If a nonlinear plant is described by the following differential equation,

n
X
ẋ = ai fi (x) + bu,
i=1

where a1 , a2 , ..., an and b are unknown, but sgn (b) is known, f1 (·), f2 (·), ..., fn (·) are known
bounded functions, and x can be measured at each instant, design a controller to match x to the
output of a time-invariant reference model that is asymptotically stable.

Solution: Let the reference model be,

ẋr = ar xr + br r,

where ar < 0. There exist constants kx and P kr , satisfying the matching condition ar = bkx and br = bkr .
The control input u be, u = k̂x x + k̂r r − ni=1 α̂i fi (x), where k̂x and k̂r denote the estimates of kx and
kr , respectively. Let the state-tracking error denoted by e is defined as e = x − xr , then the tracking
error dynamics is,
n
X n
X
ė = ai fi (x) + bk̂x x + bk̂r r − b α̂i fi (x) − ar xr − br r,
i=1 i=1
    n
X n
X
= ar e + bk̂x − ar x + bk̂r − br r + b αi fi (x) − b α̂i fi (x) , ai = bαi ,
i=1 i=1
n
X
= ar e + bk̃x x + bk̃r r − b α̃i fi (x) .
i=1

The adaptation laws are,

˙
k̂x = −sgn (b) xe,
˙
k̂r = −sgn (b) re,
α̂˙ i = sgn (b) fi (x) e, i = {1, 2, . . . , n} .
Let the Lyapunov function candidate be,
n
1 |b| |b| |b| X 2
V = e2 + k̃x2 + k̃r2 + α̃i ,
2 2 2 2
i=1
AE 666 Tutorial 2 Solution February 21, 2025

the time-derivative of the Lyapunov function along the error trajectories,

n
!
X
V̇ = e ar e + bk̃x x + bk̃r r − b α̃i fi (x) + |b| k̃x (−sgn (b) xe) + |b| k̃r (−sgn (b) re)
i=1
n
X
+ |b| α̃i (sgn (b) fi (x) e) ,
i=1
= ar e2 .

V is positive definite, V̇ is negative semi-definite, hence the closed-loop system is uniformly stable,
and the signals e, k̃x , k̃r , and α̃i are bounded. Since xr and e are bounded, x ∈ L∞ , for bounded x we
obtain bounded fi (x). Therefore, e ∈ L2 ∩ L∞ and ė ∈ L∞ , inferring limt→∞ e (t) = 0.

2. Consider a plant described by the following differential equation,

ẋ = ax + bu,
where a and b are unknowns. Let the reference model be,
ẋr = ar xr + br r.
Assume the following matching condition is satisfied,
a + br kx = ar , bkr = br ,
and use the below-mentioned control law,
u = k̂r k̂x x + k̂r r,
n o
to design adaptation laws for updating k̂x and k̂r so that the origin in the e, k̃x , k̃r space is
uniformly stable. Is the result global?

Solution: Let the state-tracking error denoted by e is defined as e = x − xr , then the tracking error
dynamics is,
ė = ax + bk̂r k̂x x + bk̂r r − ar xr − br r,
   
= ar e + a + bk̂r k̂x − ar x + bk̂r − br r,
   
= ar e + br k̂r k̂x kr−1 − kx x + br k̂r kr−1 − 1 r,
   
= ar e + br k̂r k̂x kr−1 + k̃x − k̂x x + br k̂r kr−1 − 1 r,
  
= ar e + br k̃x x + br k̂r kr−1 − 1 k̂x x + r ,
  
= ar e + br k̃x x + br kr−1 − k̂r−1 k̂r k̂x x + k̂r r ,
′
= ar e + br k̃x x + br k̃r u,
′
where k̃r = kr−1 − k̂r−1 . The adaptation laws are,
˙
k̂x = −br xe,
˙′
k̂r = −br ue,

2
AE 666 Tutorial 2 Solution February 21, 2025

where k̂r is updated as follows,

˙
k̂r = −k̂r2 br ue.
Let the Lyapunov function candidate be,
1 1 1  ′ 2
V = e2 + k̃x2 + k̃ ,
2 2 2 r
the time-derivative of the Lyapunov function along the error trajectories,
 ′
 ′
V̇ = e ar e + br k̃x x + br k̃r u + k̃x (−br xe) + k̃r (−br ue) ,
= ar e2 ,
n o
therefore, uniformly stable, however, V is not radially unbounded in e, k̃x , k̃r space hence, the result
is not global.

3. Let the error differential equation in an adaptive problem be ė = −e + ϕx. If the parameter error
ϕ is adjusted using the adaptive law ϕ̇ = −γ (t) ex, where γ (t) is a time-varying gain, what are
the conditions on γ (t) so that the origin in the (e, ϕ) space is uniformly stable?

Solution: Let the Lyapunov function candidate be,

1 1
V = e2 + ϕ2 ,
2 2γ (t)

the time-derivative of the Lyapunov function along the error trajectories,

1 1
V̇ = e (−e + ϕx) + ϕ (−γ (t) ex) − 2 γ̇ (t) ϕ2 ,
γ (t) 2γ (t)

if |γ (t)| > 0,
1
V̇ = −e2 − γ̇ (t) ϕ2 ,
2γ 2 (t)
for origin in the (e, ϕ) space to be uniformly stable, we require V > 0 and V̇ ≤ 0,

γ (t) > 0 =⇒ V > 0, γ̇ (t) ≥ 0 =⇒ V̇ ≤ 0,

therefore, γ (t) > 0 and γ̇ (t) ≥ 0 ensures uniformly stability in the (e, ϕ) space.

4. A simple model of an aircraft system can be described by the differential equation,

ẍ + 2 [ξ (t) + k (t)] ẋ + x = r (t) ,

where r (t) is a reference input. The desired response of the system is achieved when the damping
factor ξ (t) + k (t) is a constant and equals 0.7. It is known that ξ (t) varies with time due to
changes in air density. Design an adaptation law for the velocity feedback gain k (t) so that the
tracking error is asymptotically stable.

3
AE 666 Tutorial 2 Solution February 21, 2025

Solution: The state-space model of the given aircraft system,

   
0 1 0
ẋ = x+ r,
−1 −2 [ξ (t) + k (t)] 1
 T
where x = x1 x2 , x1 = x and x2 = ẋ, the reference dynamics is,
   
0 1 0
ẋr = x + r.
−1 −1.4 r 1
| {z }
Ar

The matrices P > 0 and Q > 0 satisfies, ATr P + PAr = −Q.

Let the tracking error denoted by e is defined as e = x − xr , the tracking error dynamics,
   
0 1 0 1
ė = x− x ,
−1 −2 [ξ (t) + k (t)] −1 −1.4 r
   
0 1 0 0
= e+ x,
−1 −1.4 0 1.4 − 2 [ξ (t) + k (t)]

let θ = ξ (t) + k (t) − 0.7,    

0 1 0 0
ė = e+ x.
−1 −1.4 0 −2θ
The adaptation law for k with σ-modification,
   
T p12 p11 p12
k̇ = −σk + 2e x , P= , p12 = p21 ,
p22 2 p21 p22

then θ̇ would be,

θ̇ = ξ˙ + k̇,
 
˙ T p12
= ξ − σk + 2e x
p22 2
Let the Lyapunov function candidate be,

V = eT Pe + θ2 ,

the time-derivative of the Lyapunov function along the error trajectories,

   
T T 0 ˙ T p12
V̇ = −e Qe + 2e P + 2θξ − 2σkθ + 4θe x ,
−2θx2 p22 2
= −eT Qe + 2θξ˙ − 2σkθ,
= −eT Qe + 2θξ˙ − 2σθ2 + 2σξθ − 1.4σθ,

˙
considering a valid assumption on boundedness of ξ and ξ,

V̇ ≤ −λmin (Q) ∥e∥2 − 2σθ2 + 2 |θ| ξ˙ + 2σ |ξ| |θ| + 1.4σ |θ| ,

 
≤ −λmin (Q) ∥e∥2 − |θ| 2σ |θ| − 2 ξ˙ − 2σ |ξ| − 1.4σ

4
AE 666 Tutorial 2 Solution February 21, 2025

V̇ < 0 outside a compact set,

 
 2 ξ˙ + 2σ |ξ| + 1.4σ 
D = (e, θ) : |θ| < .
 2σ 

5. Consider a plant described by the second-order nonlinear differential equation,

ẍ − ẋ + sin ẋ + 2x + 2 cos x = u,
and a reference model described by the equation,
ẍr + ẋr + xr = r.
Assume that the coefficients of the nonlinear terms sin ẋ and cos x are unknown. Design an
adaptive controller to achieve asymptotic model-following. Validate the designed adaptive
controller using numerical simulations with reference inputs (i) r (t) ≡ 0, (ii) r (t) ≡ 5, and (iii)
r (t) = 2 cos t + 3 cos 2t.

Solution: The state-space model of the system,

 θ1∗
     
0 1 0 
ẋ = x+ u + cos x sin ẋ ,
−2 1 1 θ2∗
where θ1∗ = −2 and θ2∗ = −1, the reference dynamics is,
   
0 1 0
ẋr = x + r.
−1 −1 r 1
The control law is,
u = kTx x + kr r − φT θ
 T  T  T
where φ = cos x sin ẋ and θ = θ1 θ2 , kr = 1, kx = 1 −2 , and update law for θ,
 
0
θ̇ = φeT P .
1
The tracking error and parameter estimation are shown in Figure 1, Figure 2, and Figure 3 for (i)
r (t) ≡ 0, (ii) r (t) ≡ 5, and (iii) r (t) = 2 cos t + 3 cos 2t, respectively.

6. Design a direct MRAC with σ-modification for the given system,

   
0 1 0
ẋ = x+ u + d,
a1 a2 b
where a1 , a2 , and b are unknowns, but sgn (b) is known, and d is an unstructured uncertainty
that is bounded as ∥d∥ ≤ d. The reference model is,
   
0 1 0
ẋm = x + r.
−2 −3 m 1
Comment on the performance of the system in the absence of uncertainty i.e. d = 0.

5
AE 666 Tutorial 2 Solution February 21, 2025

1 2

0.5 1

0
0

-1
-0.5
-2
-1
0 10 20 30 40 0 10 20 30 40

Figure 1: Tracking error and parameter estimation for r (t) = 0.

1
0
0.5
-1
0
-2
-0.5
-3
-1
0 10 20 30 40 0 10 20 30 40

Figure 2: Tracking error and parameter estimation for r (t) = 5.

Solution: From the given system and reference model,

       
0 1 0 0 1 0
A= , b= , Ar = , br = ,
a1 a2 b −2 −3 1
there exist controller gains kx and kr satisfying the following matching condition,

Ar = A + bkTx , br = bkr ,

since A and b are unknown, the control law is,

u = k̂Tx x + k̂r r,

where k̂x and k̂r are estimates of kx and kr , respectively.

The adaptation laws with σ-modification are as follows,
˙
k̂x = −σ k̂x − xeT Psgn (b) ,
˙
k̂r = −σ k̂r − reT Psgn (b) .
Even in the absence of uncertainty, the tracking error e does not decay to the origin.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6
AE 666 Tutorial 2 Solution February 21, 2025

1
1
0.5
0

0
-1

-0.5 -2

-1 -3
0 10 20 30 40 0 10 20 30 40

Figure 3: Tracking error and parameter estimation for r (t) = 2 cos t + 3 cos 2t.