EECS 562/AEROSP 551 – Nonlinear Systems and Control

HW #8

Due on Thursday March 28th, 2024

by 11:59 pm on Canvas

Preliminary: Recall the normalized model of the inverted pendulum on a cart:

 2
 d2 ϕ = −c(ϕ̇) sin ϕ cos ϕ + sin ϕ − cos ϕ µ̄
dt̄ 2 1+c sin2 ϕ 1+c sin2 ϕ 1+c sin2 ϕ
2 (1)
 d2 s̄ = d(ϕ̇) sin ϕ − cos ϕ sin ϕ + b
µ̄
dt̄2 1+c sin2 ϕ 1+c sin2 ϕ 1+c sin2 ϕ

Choosing the states as x1 = ϕ, x2 = ϕ̇, x3 = s̄, x4 = s̄, ˙ letting u = µ̄ and setting b = 3, c = 0.5 and d = 1.5 as in
the project, results in
   x 
ẋ1 2
2
−0.25(x2 ) sin(2x1 )+sin(x1 ) cos(x1 )
ẋ2   1+0.5 sin2 (x1 )
− 1+0.5 sin2 (x1 )
u 
 =  = f (x) + g (x) u

(2)
ẋ3   x4


2
ẋ4 1.5(x2 ) sin(x1 )−0.5 sin(2x1 ) 3
2
1+0.5 sin (x1 )
+ 2
1+0.5 sin (x1 )
u

P.1 Show that for every p ∈ R and v ∈ R,

 
0
 0 
p + vt and ū (t) = 0
x̄ (t) =  

v


is a solution of (2). Linearize the system (2) about the trajectory x̄(t), ū(t) . You should find that the lineariza-
tion is time-invariant and equal to the matrices of your project, so, in particular, you get the same linearization
for any values of p and v.

P.2 Using the state feedback gains of your project, make the trajectory x̄ (t) , ū (t) = 0 (uniformly) exponentially
stable by feeding back u (t) = k [x (t) − x̄ (t)] (see Figure 1 below). Perform a simulation that illustrates this
for x̄ (t) corresponding to p = 10 and v = −0.2. Choose initial conditions for (2) that are different from
 ⊤
x̄0 = 0 0 p v and show that x (t) converges to x̄ (t) as t −→ ∞, as long as you choose x0 sufficiently
close to x̄0 . [Same x̄ (t) and ū (t) from Prob. 1]

1
Figure 1: Block diagram of how to implement a linear feedback on the original nonlinear system when the nominal
trajectory is not the origin.

P.3 What happens if you allow p and v in x̄ (t) to vary slowly with time? In particular, let
 
0
 0 
x̄ (t) = 
 10e−εt 

d −εt
dt (10e )

for ε > 0 small, and see what happens! We note that x̄ (t) is no longer a solution of (1), but it is a “small,” slowly
-varying, perturbation of a solution of (1). Provide a printout of your simulation. Take your initial condition as
 ⊤
x (0) = 0 0 10 0 .

P.4 Prove (with mathematical analysis and not by simulation) that the feedback scheme of Problem P.2 does indeed
render the trajectory x̄ (t) (of Problem P.1) (uniformly) exponentially stable. To do this, translate x̄ (t) to the
origin via the standard transformation, z (t) = x (t) − x̄ (t), and determine the differential equation for z (t).
Then apply results from class on uniform exponential stability of the origin. We will analyze why we can “track”
slowly varying p (t) and v (t) in a later HW set.

P.5 Consider the system

ξ˙1
   
ξ2
˙ξ2 = − 1+K
α 1 ξ1
− α2 ξ2
1 |ξ1 | 1+K2 |ξ2 |

(a) Show that for all K1 ≥ 0, K2 ≥ 0, α1 > 0 and α2 > 0, the origin is (locally) exponentially stable. To
show exponential stability, you need to linearize the system. Is the NL system differentiable at the origin?
Why?
(b) Show that for all K1 ≥ 0, K2 ≥ 0, α1 > 0 and α2 > 0 the origin is GAS. To show this try the function
ˆ ξ1
x ξ2
V (ξ) = α1 dx + 2
0 1 + K1 |x| 2

2
 (c) Show that for all ϵ > 0, α1 > 0, α2 > 0 there exist K1 > 0 and K2 > 0 such that

α1 ξ1 α2 ξ2
sup + ≤ε
(ξ1 ,ξ2 )∈R2 1 + K1 |ξ1 | 1 + K2 |ξ2 |

Remark: Can you see how all of this is fitting together? Suppose that we can find a control law u = α (x1 , x2 )
that asymptotically stabilizes the pendulum for large angles, while ignoring the cart (as in the handout on Son-
tag’s feedback controller). Such a feedback will make the pendulum and cart approach a trajectory of the form
x̄ (t) from Problem P.1. From Problem P.3, we have the idea that we can “track” a slowly varying version of
x̄ (t), so it would be nice to create a slowly varying version of x̄ (t) that converges to the origin! But this is what
Problem P.5 shows us how to do...part P.5b of that problem gets us convergence to the origin from any initial
condition, and part P.5c says that we can make the trajectory “uniformly slowly varying.” We will put all of this
together in an upcoming HW and build a nice controller for the pendulum and cart.

3
Hints for Problem P.1: Plug into the ODE and see if left-hand side equals right-hand side. That is all there is to
showing it is a solution.
Hints for Problem P.2: We have not studied in lecture the stabilization of a non-trivial solution of an ODE, hence this
problem. It is quite common in practice to implement a scheme as shown in Fig. 1.
Hints for Problem P.4: See page 147 of our textbook for the change of variables being done here. When you have
shifted the solution to the origin, you should find yourself in the unusual, but fortunate, situation of z (t) satisfying
the same ODE as x (t)! In particular, the ODE for z (t) is time-invariant. Therefore, you can prove local uniform
exponential stability through linearization or though a quadratic Lyapunov function.
αξ
Hints for Problem P.5a: Compute the derivative of 1+K|ξ| for ξ > 0 and for ξ < 0; then take the limit as ξ −→ 0
and see that you get the same answer in each case.
Hints for Problem P.5c: Use the triangle inequality to simplify the problem to two scalar problems of the form
αx
supx∈R 1+K|x| ≤ 2ε