Control of

Mobile Robots
Module 4
Dr. Magnus Egerstedt
Professor
School of Electrical and
Control Design
Computer Engineering

How make mobile robots move in effective, safe,

predictable, and collaborative ways using modern
control theory?

School  of  Electrical  and  Computer  Engineering  

 Lecture 4.1 – Stabilizing the Point Mass
•  Given a linear system

ẋ = Ax + Bu, y = Cx

•  The dilemma: We seem to need x but all we have is y

•  Game plan:
–  Design u as if we had x
–  Figure out x from y
•  Step 1: State feedback using “pole placement”

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.1
 Back to the Point Mass
u

p � � � �
0 1 0
ẋ = Ax + Bu, A = , B=
0 0 1
� �
•  State Feedback: u = −Kx = − k1 k2 x
•  How pick the control gains?
•  Idea: Pick them in order to make the closed-loop system have
the desired eigenvalues (poles)

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.2
 Computing Eigenvalues
•  Given a matrix M, its eigenvalues satisfy the characteristic
equation
χM (λ) = det(λI − M ) = 0
� �
m1 m2
M=
m3 m4
� � � �
1 0 λ − m1 −m2
λI − M = λ −M =
0 1 −m3 λ − m4

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.3
 Computing Eigenvalues
•  Given a matrix M, its eigenvalues satisfy the characteristic
equation
χM (λ) = det(λI − M ) = 0
� �
� λ − m1 −m �
det(λI − M ) = �� 2 �
−m3 λ − m4 �

= (λ − m1 )(λ − m4 ) − m2 m3
= λ2 − (m1 + m4 )λ + m1 m4 − m2 m3 = 0

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.3
 Computing Eigenvalues
•  Given a matrix M, its eigenvalues satisfy the characteristic
equation
χM (λ) = det(λI − M ) = 0

χM (λ) = λ2 − (m1 + m2 )λ + m1 m4 − m2 m3 = 0
�
m1 + m4 (m1 + m4 )2
λ= ± − m1 m4 + m2 m3
2 4

•  But this is really annoying – is there an easier way?

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.3
 Almost Computing Eigenvalues
•  The fundamental theorem of algebra: The roots in a
polynomial are completely determined by the coefficients

χM (λ) = λ2 − (m1 + m2 )λ + m1 m4 − m2 m3 = 0

•  Let’s stop here!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.4
 Back to the Point-Mass, Again
u = −Kx ⇒ ẋ = (A − BK)x
� � � � � �
0 1 0 � � 0 1
A − BK = − k1 k2 =
0 0 1 −k1 −k2
� �
� λ −1
χA−BK (λ) = �� = λ2 + λk2 + k1
k1 λ + k2

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.5
 Desired Eigenvalues
•  Let’s pick the eigenvalues that we would like the closed-loop
system to have
λ1 , . . . , λ n
•  If these were indeed the eigenvalues, then the characteristic
polynomial would be n �
ϕ(λ) = (λ − λ1 )(λ − λ2 ) · · · (λ − λn ) = (λ − λi )
i=1
•  For the robot, let’s pick both eigenvalues at -1
ϕ(λ) = (λ + 1)(λ + 1) = λ2 + 2λ + 1
•  Now we just line up the coefficients!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.6
 Lining Up the Coefficients
χA−BK = λ2 + k2 λ + k1

ϕ(λ) = (λ + 1)(λ + 1) = λ2 + 2λ + 1

λ1 : k2 = 2 � �
λ0 : k1 = 1 K= 1 2

ẋ = Ax + Bu
u x
u = −Kx

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.1.7
 Lecture 4.2 – Pole Placement
•  Pick the control gains such that the eigenvalues (poles) of the
closed loop system match the desired eigenvalues

χA−BK (λ) = λn + an−1 λn−1 + . . . + a1 λ + a0

n
�
ϕ(λ) = (λ − λi ) = λn + bn−1 λn−1 + . . . + b1 λ + b0
i=1

λn−1 : an−1 (K) = bn−1

..
.
λ0 : a0 (K) = b0

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.1
 Questions
•  Is this always possible? No!

•  How should we pick the eigenvalues? Mix of art and science

•  Do we have to compute large determinants? No!

In MATLAB:
>> P=[lambda1,lambda2,lambda3,…];!
>> K=place(A,B,P);!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.2
 Example
� � � �
2 0 1
ẋ = x+ u
1 1 1
� �
� � 2 − k1 −k2
u = −Kx = k1 k2 x ⇒ A − BK =
1 − k1 1 − k2
� �
� λ − 2 + k1 k2 �
χA−BK (λ) = �� �
−1 + k1 λ − 1 + k2 �

= λ2 + λ(−3 + k1 + k2 ) + 2 − k1 − k2

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.3
 Example
χA−BK (λ) = λ2 + λ(−3 + k1 + k2 ) + 2 − k1 − k2
ϕ(λ) = (λ + 1)2 = λ2 + 2λ + 1

λ1 : − 3 + k1 + k2 = 2 ⇒ k1 + k2 = 5 ???
λ0 : 2 − k 1 − k 2 = 1 ⇒ k 1 + k 2 = 1

•  What’s at play here is a lack of “controllability”, i.e., the effect

of the input is not sufficiently rich to influence the system
enough
•  More on controllability later…

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.4
 How to Pick Eigenvalues?

•  Assuming that we have sufficient control authority to do pole

placement, how should the desired eigenvalues be selected?
•  No clear-cut answer
•  Some things to keep in mind:
(complex-conjugate pairs)
Im(λi ) �= 0 ⇒ ∃λj : Im(λj ) = −Im(λi ), Re(λj ) = Re(λi )
Re(λi ) < 0, ∀λi (stability)
Im(λi ) �= 0 ⇒ oscillations
Rate of convergence

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.5
 Rate of Convergence
ẋ = (A − BK)x, Re(λ) < 0, ∀λ ∈ eig(A − BK)
•  The “smallest” eigenvalue dominates the convergence rate
Im

λmin = argminλ |Re(λ)|

Re
λmin

•  But, the bigger the eigenvalues the bigger

the control gains/signals

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.6
 In MATLAB
% System matrices!
A=[2,0;1,-1]; B=[1;1];!

% Let's pick our favorite poles!

P=[-0.5+1i,-0.5-1i];!

% Pole placement!
K=place(A,B,P);!

% Compute the solution!

x=[1;1]; t=0; tf=5; dt=0.01; !

while (t<tf);!
x=x+dt.*(A-B*K)*x;!
t=t+dt;!
end; !

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.2.7
 Lecture 4.3 – Controllability
•  When can we place the eigenvalues however we want using
state feedback?
•  When is the B matrix (the actuator configuration) rich enough
so that we can make the system do whatever we want it to do?
•  The answer revolves around the concept of controllability!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.1
 A Modest Example
•  Given a discrete-time system
xk+1 = Axk + Buk , x0 = 0

•  We would like to drive this system in n steps to a particular

target state x�
•  Is this possible? �
x

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.2
 A Modest Example
x� x1 = Ax0 + Bu0 = Bu0
x2 = Ax1 + Bu1 = ABu0 + Bu1
?
x3 = Ax2 + Bu2 = A2 Bu0 + ABu1 + Bu2
..
.
0 xn = An−1 Bu0 + . . . + Bun−1
We want to solve  
un−1
� �  .. 
 
x = �
B AB ··· An−1 B  . 
 u1 
u0

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.3
 A Modest Example
 
x� un−1
� �  .. 
?  
x = �
B AB · · · An−1 B  . 
 u1 
Γ (n × mn) u0
0
•  This is possible for any target state if and only if
rank(Γ) = n

•  Turns out, this generalizes quite nicely…

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.3
 Controllability – Theorem 1
ẋ = Ax + Bu, x ∈ �n
Definition: The system is completely controllable (CC) if it
is possible to go from any initial state to any final state.
� n−1
�
Γ= B AB ··· A B controllability matrix

CC Theorem 1: The system is CC if and only if

rank(Γ) = n

rank(M)=# linearly independent rows or columns in M

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.4
 Two Systems
� � � �
2 0 1
ẋ = x+ u n=2 Γ = [B AB]
1 1 1
pole-placement not possible not CC!
� � � �
2 1 2 rank(Γ) = 1
AB = Γ=
2 1 2
� � � �
0 1 0
ẋ = + u n=2 Γ = [B AB]
0 0 1
pole-placement possible CC!
� � � �
1 0 1 rank(Γ) = 2
AB = Γ=
0 1 0

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.5
 Controllability – Theorem 2

u = −Kx, ẋ = (A − BK)x

CC Theorem 2: Pole-placement to arbitrary

eigenvalues is possible if and only if the system is CC!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.6
 Point-to-Point vs. Trajectories
� � � � In MATLAB:
0 1 0 >> G=ctrb(A,B);!
ẋ = + u
0 0 1 >> rank(G)!
ans: 2!

� x2
x x0

x1

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.3.7
 Lecture 4.4 – The Segway Robot

•  Let’s unleash our newfound powers on a complicated robotic

system

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.1
 Unicycle + Inverted Pendulum + …

The Base: The “Pendulum”:

ẋ1 = v cos ψ φ, φ̇
ẋ2 = v sin ψ
Inputs: τL , τR
ψ̇ = ω
Extra states: v, ω

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.2
 Unicycle + Inverted Pendulum + …
x = [x1 x2 v ψ ψ̇ φ φ̇]T
u = [τL τR ]T
ẋ = f (x, u)
= [v cos ψ v sin ψ v̇ ψ̇ ψ̈ φ̇ φ̈]T
1
3(mw + mb )v̇ − mb d cos φφ̈ + mb d sin φ(φ̇2 + ψ̇ 2 ) = − (τL + τR )
R
12 2 2
LINEARIZE!
2 L
((3L + )m w + mb d sin φ + I 2 ) ψ̈ + mb d sin φ cos φ ψ̇ φ̇ = (τL − τR )
2R2 R

mb d cos φv̇ + (−mb d2 − I3 )φ̈ + mb d2 sin φ cos φφ̇2 + mb gd sin φ = τL + τR

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.3
 Linearization
•  Linearize this mess around (x,u)=(0,0), and plug in the
physical parameters for the segway robot:
ẋ = Ax + Bu
   
0 0 1 0 0 0 0 0 0
 0 0 0 0 0 0 0   0 0 
   
 0 0 0 0 0 2.16 0   −1.67 −1.67 
   
A= 0 0 0 0 1 0 0  B= 0 0 
   
 0 0 0 0 0 0 0   0.029 −0.029 
   
 0 0 0 0 0 0 1   0 0 
0 0 0 0 0 72.49 0 −24.15 −24.1514

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.4
 Controllability?
>> rank(ctrb(A,B))!

ans =! Not CC!

6!

•  We already know that the unicycle dynamics gets messed up

when linearized.
•  And if we can control v and ω, that should be “enough”, so
let’s simply remove the unicycle part

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.5
 A Smaller System
x = [v ω φ φ̇]T , u = [τL τR ]T
ẋ = Ax + Bu
   
0 0 2.16 0 −1.67 −1.67
 0 0 0 0   0.029 −0.029 
A=
 0
 B= 
0 0 1   0 0 
0 0 72.49 0 −24.15 −24.15

>> rank(ctrb(A,B))!

ans =! CC!
4!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.6
 One Last Twist
•  We do not want to stabilize to (v,ω)=(0,0)
•  Let
x̃ = x − [vd ωd 0 0]T = x − δ

•  The dynamics become

x̃˙ = ẋ − δ̇ = ẋ = Ax + Bu = A(x − δ) + Bu + Aδ
•  Luckily,
Aδ = 0 ⇒ x̃˙ = Ax̃ + Bu

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.7
 Pole-Placement
•  We have a CC system that we wish to stabilize
•  After some tweaking, the following eigenvalues seem to
behave well:
λ1 = −19, λ2 = −7.5, λ3 = −5, λ4 = −0.3

u = −K x̃ ⇒ x̃˙ = (A − BK)x̃
>> P=[-19,-7.5,-5,-0.3];!

>> K=place(A,B,P);!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.8
 Now, Let’s Do It!

ω
κ=
v
curvature rather than pose

v↑

ω↓ ω↑

v↓

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.4.9
 Lecture 4.5 – Observers

•  We now know how to design rather effective controllers using

state feedback.
•  But what about y???

u ẋ = Ax + Bu y
y = Cx

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.1
 Lecture 4.5 – Observers

•  We now know how to design rather effective controllers using

state feedback.
•  But what about y???

u ẋ = Ax + Bu y
y = Cx

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.1
 Lecture 4.5 – Observers

•  We now know how to design rather effective controllers using

state feedback.
•  But what about y???

u ẋ = Ax + Bu y
y = Cx

x̂
observer

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.1
 The Predictor-Corrector
ẋ = Ax
y = Cx
First idea: Make a copy of the system
x̂˙ = Ax̂ predictor

Second idea: Add a notion of how

wrong your estimate is to the model
x̂˙ = Ax̂ + L(y − C x̂)
corrector
“Luenberger” observer

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.2
 Picking the Observer Gain

•  What we want to stabilize (drive to zero) is the estimation

error, i.e., the difference between the actual state and the
estimated state e = x − x̂
Cx
ė = ẋ − x̂˙ = Ax − Ax̂ − L(y − C x̂)
ė = A(x − x̂) − LC(x − x̂) = (A − LC)e
•  Just pick L such that the eigenvalues to A-LC have negative
real part!!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.3
 Pole-Placement, Again

ė = (A − LC)e

•  Want
Re(eig(A − LC)) < 0

•  We already know how to do this - Pole-placement!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.4
 Example
� �
−1 2 � �
ẋ = x, y = 1 −1 x
0 −2
�� � � � �
−1 2 L1 � �
ė = (A − LC)e = − 1 −1 e
0 −2 L2
� �
−1 − L1 2 + L1
= e
−L2 −2 + L2

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.5
 Example
� �
−1 − L1 2 + L1
ė = e
−L2 −2 + L2
� �
� λ + 1 + L1 −2 − L1 ��
χA−LC (λ) = ��
L2 λ + 2 − L2 �
= λ2 + λ(3 + L1 − L2 ) + 2 + 2L1 + L2
ϕ(λ) = (λ + 1)2 = λ2 + 2λ + 1
�
L1 = −2/3
L2 = 1/3

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.6
 Example
� �
˙x̂ = Ax̂ + 1 −2
(y − C x̂)
3 1
1.2 1

1 0.9

0.8
0.8

0.7
0.6

0.6
0.4
0.5

0.2
0.4

0
0.3

ï0.2
0.2

ï0.4 0.1

ï0.6 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.7
 The Block-Diagram

ẋ = Ax y
y = Cx

x̂
x̂˙ = Ax̂ + L(y − C x̂)

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.8
 Does This Always Work?
•  No!

•  Next Lecture: We need to redo what we did for control design

(controllability) for observer design (observability)

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.5.9
 Lecture 4.6 – Observability

•  Need to redo what we did for control design to understand

when we can recover the state from the output

ẋ = Ax y
y = Cx

x̂
x̂˙ = Ax̂ + L(y − C x̂)

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.6.1
 A Modest Example (Again)

•  Given a discrete time system without inputs

xk+1 = Axk
yk = Cxk
•  Can we recover the initial condition by collecting n output
values?
y0 = Cx0
y1 = Cx1 = CAx0
..
.
yn−1 = CAn−1 x0

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.6.2
 The Observability Matrix
   
y0 C
 y1   CA 
   
 y2   CA2 
 =  x0
 ..   .. 
 .   . 
yn−1 CAn−1

Ω
•  The initial condition can be recovered from the
outputs when the so-called observability matrix
has full rank.

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.6.3
 Observability – Theorem 1
ẋ = Ax, x ∈ �n
y = Cx
Definition: The system is completely observable (CO) if it
is possible to recover the initial state from the output.
 
C observability matrix
 CA 
  CO Theorem 1: The system is
Ω= .. 
 .  CO if and only if
CAn−1 rank(Ω) = n

rank(M)=# linearly independent rows or columns in M

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.6.4
 Observability – Theorem 2

x̂˙ = Ax̂ + L(y − C x̂), ė = (A − LC)e

CO Theorem 2: Pole-placement to arbitrary

eigenvalues is possible if and only if the system is CO!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.6.5
 Putting It All Together
•  We are very close to being able to design the controllers as if
we had x while all we really have is y (and hence an estimate
of x).
•  Next lecture – Putting it all together!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.6.6
 Lecture 4.7 – The Separation Principle
•  We have lots of really good building blocks:
–  Controllability
–  Observability
–  State feedback
–  Observers
–  Pole-placement

•  Now, how do we put everything together?

•  Answer: The wonderful Separation Principle

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.7.1
 A Game Plan
ẋ = Ax + Bu
Assume CC and CO
y = Cx
Step 1: Design the state feedback controller as if we had x
(which we don’t)
u = −Kx ⇒ ẋ = (A − BK)x what we design for

u = −K x̂ what we actually have

Step 2: Estimate x using an observer (that now also
contains u)
x̂˙ = Ax̂ + Bu + L(y − C x̂)
⇒ ė = (A − LC)e

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.7.2
 Does This Work?
•  Want both x and e to be stabilized
•  Analyze their joint dynamics:
ẋ = Ax − BK x̂ = Ax − BK(x − e) = (A − BK)x + BKe
e = x − x̂
ė = (A − LC)e
•  Or, when viewed together
� � � �� �
ẋ A − BK BK x
=
ė 0 A − LC e
•  The strategy works if and only if
this is an asymptotically stable system!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.7.3
 The Separation Principle
� � � �� �
ẋ A − BK BK x
=
ė 0 A − LC e

M
•  This is an (upper) triangular block-matrix. Its eigenvalues are
given by the eigenvalues of the diagonal blocks!
χM (λ) = χA−BK (λ)χA−LC (λ)
•  If we haven’t messed up in the design, we have
Re(eig(A − BK)) < 0, Re(eig(A − LC)) < 0
•  Everything works!!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.7.4
 The Separation Principle

•  This remarkable fact is known as the Separation Principle!

•  Means that we can design controllers as if we have x
•  Can design observers independent of the control actions
•  Control and observer designs can be separated!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.7.5
 The Final Block-Diagram

u ẋ = Ax + Bu y
y = Cx
u = −K x̂

x̂
x̂˙ = Ax̂ + Bu + L(y − C x̂)

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.7.6
 Lecture 4.8 – Practical Considerations
•  The separation principle means that we can decouple control
and observer design (in theory)

“In theory, theory and practice are the

same. In practice they are not.” -Yogi Berra

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.1
 Eigenvalue Selection
•  Typically, the observer should be faster than the controller,
since the controller won’t do anything useful until the state
estimate is close.
•  Note: Large observer eigenvalues give large observer gains:
–  Not a problem – The observer is a software construct!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.2
 Eigenvalue Selection

Im(λ)

Re(λ)
observer
controller

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.3
 Humanoid Robot
torque constant
1
θ̈ = (Ki − bθ̇)
J
current friction coefficient
moment of inertia
� � � �
1 0 0
ẋ = x+ u
−b/J 0 K/J
Aldebaran Nao � �
y= 1 0 x CC & CO!!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.4
 Reference Tracking
� �
ẋ = Ax + Bu θ − θd
e=
y = Cx θ̇
� �
θd
ė = Ax + Bu = Ae + A + Bu = Ae + Bu
0
� �
θd
y = Cx = Ce + C = Ce + θd
0

u = −K ê
ê˙ = Aê + Bu + L(y − C ê − θd )

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.5
 Beyond Pole Placement
•  The design structure is quite general (does not need to be
based on pole placement)
u = −K x̂ x̂˙ = Ax̂ + Bu + L(y − cx̂)

� ∞
•  LQ Optimal Control: min (xT Qx + uT Ru)dt
0

K = R−1 B T P
0 = AT P + P A + Q − P BR−1 B T P

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.6
 Beyond Pole Placement
•  The design structure is quite general (does not need to be
based on pole placement)
u = −K x̂ x̂˙ = Ax̂ + Bu + L(y − cx̂)

ẋ = Ax + Bu + DI vI
•  The Kalman Filter: y = Cx + DO vO
min lim �E(x(t) − x̂(t))�2
t→∞
T −1
L = P C T (DO DO )
T −1
0 = AT P + P A + DI DIT − P C T (DO DO ) CP

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.7
 Next Module
•  But, the real world (especially for robots) is more complex
than a simple LTI system
•  HYBRID SYSTEMS!

Magnus  Egerstedt,  Control  of  Mobile  Robots,  Georgia  Ins<tute  of  Technology  
4.8.8