Nonlinear Systems and Control — Spring 2019

Chapter 7 - Sliding Mode Control

Problem 1 - Solution
We define the error as:  
x1 − r
e=
x2 − ṙ
where the reference derivatives are:

r(t) = sin t ṙ(t) = cos t r̈(t) = − sin t.

The error dynamics can then be written as:

ė1 = ẋ1 − ṙ = e2
(1)
ė2 = ẋ2 − r̈ = x21 + γu − r̈

We choose k1 such that s + k1 is Hurwitz, e.g. k1 = 1. Then we define the

sliding manifold as:
σ = e1 + e2 .
Note that σ = 0 ⇒ e1 = −e2 . By (1), ė1 = −e1 , thus e1 → 0, e2 → 0.
Moreover:
σ̇ = e2 + x21 + γu + sin t.
By choosing:
1
e2 + x21 + sin t + v


u=−
γ̂
we obtain:
σ̇ = γv + ∆(t, x)
with  
γ
e2 + x21 + sin(t) .


∆(t, x) = 1 −
γ̂
We assume that we have the following bound
    
 ∆(t, x)   1 1 2


 γ   γ − γ̂ x2 − cos(t) + x1 + sin(t) 
 =
  
 1 2


≤ 1−
 x2 − cos(t) + x1 + sin(t) 
γ̂
√ 
 
1 
≤ 1− |x2 | + x21 + 2 =: ρ(x)
γ̂

1 of 3
Nonlinear Systems and Control — Spring 2019

where the first inequality is for the lower bound on γ in the √ exercise text,
and in the second we use the inequality sin(t) − cos(t) ≤ 2. Note that
ρ(x) ≥ 0 ∀x ∈ <2 . 
1

2
√

Finally, designing v = − 1 − γ̂ |x2 | + x1 + 2 + β0 sgn(σ), with
β0 > 0, we achieve asymptotic tracking. Note that the uncertainty on the γ
affects the amplitude of v, hence it requires a more aggressive control.

Problem 2 - Solution
1. Take a torque τ = H(q)v + C(q, q̇)q̇ + G(q).
Due to the perfect knowledge of the matrices, the equation combines
with the model as follows:
q̈ = v
hence we choose
v = q̈d − α (q − qd ) − β (q̇ − q̇d )
and with some diagonal positive definite matrices α and β we obtain de-
coupled control and asymptotic convergence of the state to the desired
reference.
2. In the realistic case in which the model matrices are not known exactly,
we define the sliding manifold:
σ (q, q̇) = q̇ − q̇d + K (q − qd ) = q̇ − q˙r = 0
for an accordingly defined variable q̇r = q̇r (q, qd , q̇d ).
We consider now a Lyapunov function V = 21 σ T H(q)σ, which is positive
definite since H(q) is positive definite. Then:
1
V̇ = σ > H σ̇ + σ > Ḣσ
2
1
= σ > H (q̈ − q̈r ) + σ > Ḣσ
2
1
= σ > (τ − C q̇ − G − H q̈r ) + σ > Ḣσ
2
1
= σ > (τ − C (σ + q̇r ) − G − H q̈r ) + σ > Ḣσ
2
1  
= σ > (τ − C q̇r − G − H q̈r ) + σ > Ḣ − 2C σ
2
>
= σ (τ − C q̇r − G − H q̈r )

2 of 3
Nonlinear Systems and Control — Spring 2019

where the last equality follows from the skew-symmetry of Ḣ − 2C.

We design the control action τ as follows:

τ = Ĥ q̈r + Ĉ q̇r + Ĝ − βsgn(σ).

with β > 0. Hence, the derivative of the Lyapunov function reads as:

V̇ = σ > (∆H q̈r + ∆C q̇r + ∆G − βsgn(σ))

with ∆H = Ĥ − H, ∆C = Ĉ − C and ∆G = Ĝ − G. The Lyapunov

function derivative V̇ is made negative definite by choosing:

β > |∆H q̈r + ∆C q̇r + ∆G| .

Also in this case, the bigger is the uncertainty, the larger is the variable
β.

3 of 3