Control Systems 2

Lecture 5: RHP poles and zero limitations

& how to design and ride a bike

Roy Smith

2019-3-18 5.1

Non-minimum phase behaviour (stable systems)

Right-half plane zeros

Can arise from fast and slow responses of opposite sign:

1 2 3−s
G(s) = − = .
s+1 s+5 (s + 1)(s + 5)

Amplitude
1.0
 
1
step s+1

step (G(s))
0.5

time
0 (sec)
1 2 3 4
 
2
− step s+5

−0.5

2019-3-18 5.2
Non-minimum phase behaviour
Can also be interpreted as a negative derivative response:
3 s
G(s) = −
(s + 1)(s + 5) (s + 1)(s + 5)

   
1 −t −1 −5t d 1 −t −1 −5t
g(t) = 3 e + e − e + e
4 4 dt 4 4

Amplitude
1.0

 
1
3 step (s+1)(s+5)
0.5
step (G(s))

1 2 3 4 time
0 (sec)
 
d 1
− dt step (s+1)(s+5)

−0.5

2019-3-18 5.3

Non-minimum phase systems

Some common examples

I Longitudinal aircraft dynamics
I Human digestion (energy from food)
I Investment effects on profitability
I Bicycle steering dynamics

2019-3-18 5.4
Non-minimum phase systems in feedback

Non-minimum phase response in closed-loop

NG (s) NK (s) NG (s)NK (s)

G(s) = , K(s) = , L(s) =
DG (s) DK (s) DG (s)DK (s)

L(s)
T (s) =
1 + L(s)

NG (s) NK (s)
DG (s) DK (s)
= NG (s) NK (s)
1 +D G (s) DK (s)

NG (s)NK (s)
=
DG (s)DK (s) + NG (s)NK (s)

2019-3-18 5.5

Non-minimum phase systems: r.h.p. zeros

Magnitude log ω
1 (rad/sec)
1 10 100
(s+10)
0.1 Gmp (s) = (s+1)(s+50)

(10−s)
Gnmp1 (s) = (s+1)(s+50)
0.01

0.001

1 10 100 log ω
0 (rad/sec)

(s+10)
−90 Gmp (s) = (s+1)(s+50)

−180

(10−s)
−270 Gnmp1 (s) = (s+1)(s+50)
Phase (deg.)

2019-3-18 5.6
Non-minimum phase systems: delays

Magnitude log ω
1 (rad/sec)
1 10 100
(s+10)
0.1 Gmp (s) = (s+1)(s+50)

(10−s)
Gnmp1 (s) = (s+1)(s+50)
0.01
e−0.05s (s+10)
Gnmp2 (s) = (s+1)(s+50)
0.001

1 10 100 log ω
0 (rad/sec)

(s+10)
−90 Gmp (s) = (s+1)(s+50)

e−0.05s (s+10)
−180 Gnmp2 (s) = (s+1)(s+50)

(10−s)
−270 Gnmp1 (s) = (s+1)(s+50)
Phase (deg.)

2019-3-18 5.7

Non-minimum phase systems in feedback

Delays in feedback

NG (s) NK (s) NG (s)NK (s)

G(s) = e−θs , K(s) = L(s) = e−θs
DG (s) DK (s) DG (s)DK (s)

L(s)
T (s) =
1 + L(s)

NG (s) NK (s)
e−θs D G (s) DK (s)
= NG (s) NK (s)
1 + e−θs DG (s) DK (s)

 
−θs NG (s)NK (s)
=e
DG (s)DK (s) + e−θs NG (s)NK (s)

2019-3-18 5.8
Performance limitations from delays

If G(s) contains a delay, e−θs , then T (s) also contains e−θs .

Under these circumstances the ideal T (s) ≈ e−θs ,

Magnitude
5

log ω
1 (rad/sec)
1/θ

0.1
 
 
|S(jω)| = 1 − e−jθω 

0.01

Which implies that we must have ωc < 1/θ.

2019-3-18 5.9

Controllability (summary)

Actuation constraints: from disturbances

|G(jω)| > |Gd (jω)| for frequencies where |Gd (jω)| > 1.

Actuation constraints: from reference

|G(jω)| > R up to frequency: ωr .

Disturbance rejection
ωc > ω d
or more specifically |S(jω)| ≤ |1/Gd (jω)| for all ω.

Reference tracking
|S(jω)| ≤ 1/R up to frequency: ωr .

2019-3-18 5.10
Controllability (summary)

Right-half plane zeros

For a single, real, RHP-zero: ωB < z/2.

Time delays
Approximately require: ωc < 1/θ.

Phase lag
Most practical controllers (PID/lead-lag): ωc < ω180
G(jω180 ) = −180 deg.

Unstable real pole

Require ωc > 2p.
Also require |G(jω)| > |Gd (jω)| up to ω = p.

2019-3-18 5.11

Example: controllability analysis

d
Gd (s)

y u r
+ G(s) K(s) +
−

n
+ ym

e−θs e−θd s
G(s) = k Gd (s) = kd , |kd | > 1.
1 + τs 1 + τd s
What are the requirements on k, kd , τ , τd , θ and θd in order to obtain good
performance. And how good is it?

2019-3-18 5.12
Example: controllability analysis

Objective:
|e| ≤ 1 for all |u| < 1, |d| < 1.

Disturbance rejection (satisfying actuation bound)

|G(jω)| > |Gd (jω)| for all ω < ωd .
=⇒ k > kd and k/τ > kd /τd .

Disturbance rejection
ωc > ωd ≈ kd /τd .

Delay constraints
ωc < 1/θ (assuming θ is the total delay in the loop).

2019-3-18 5.13

Example: controllability analysis

Delay and disturbance rejection requirements.

θ < τd /kd .

Plant requirements:
k > kd and k/τ > kd /τd
θ < τd /kd .

Required/achievable bandwidth
kd /τd < ωc < 1/θ.

2019-3-18 5.14
 Bicycle dynamics
U R E

By Karl J. Åström, Adapted bicycles fo

Richard E. Klein, and education and researc
Anders Lennartsson

T
his article analyzes the dynamics of b

Karl J. Åström, Adapted bicycles for cles from the perspective of cont
Models of different complexity are
hard E. Klein, and education and research sented, starting with simple ones
ending with more realistic models ge
ders Lennartsson ated from multibody software. We
sider models that capture essential behavior suc

T
self-stabilization as well as models that dem
his article analyzes the dynamics of bicy- strate difficulties with rear wheel steering.
cles from the perspective of control. relate our experiences using bicycles in con

IEEE Control Systems Magazine Models of different complexity are pre-

sented, starting with simple ones and
education along with suggestions for fun
thought-provoking experiments with pro
ending with more realistic models gener- student attraction. Finally, we describe bicy
vol. 25, no. 4, pp. 26–47, ated from multibody software. We con-
sider models that capture essential behavior such as
and clinical programs designed for children
disabilities.
self-stabilization as well as models that demon-
2005. strate difficulties with rear wheel steering. We
relate our experiences using bicycles in control The Bicycle
education along with suggestions for fun and Bicycles are used everywhere—for transportation, e
thought-provoking experiments with proven cise, and recreation. The bicycle’s evolution over
student attraction. Finally, we describe bicycles has been a product of necessity, ingenuity, materials,
and clinical programs designed for children with industrialization. While efficient and highly maneuvera
disabilities. the bicycle represents a tantalizing enigma. Learnin
ride a bicycle is an acquired skill, often obtained with s
The Bicycle difficulty; once mastered, the skill becomes subconsc
Bicycles are used everywhere—for transportation, exer-
and second nature, literally just “as easy as riding a bik
cise, and recreation. The bicycle’s evolution over time
has been a product of necessity, ingenuity, materials, and Bicycles display interesting dynamic behavior.
industrialization. While efficient and highly maneuverable, example, bicycles are statically unstable like the inv
the bicycle represents a tantalizing enigma. Learning to ed pendulum, but can, under certain conditions, be
ride a bicycle is an acquired skill, often obtained with some ble in forward motion. Bicycles also exhibit nonminim
difficulty; once mastered, the skill becomes subconscious phase steering behavior.
and second nature, literally just “as easy as riding a bike.” Bicycles have intrigued scientists ever since they appeare
Bicycles display interesting dynamic behavior. For the middle of the 19th century. A thorough presentation of
example, bicycles are statically unstable like the invert-

the forces acting between ground and wheel. Since we do

ed pendulum, but can, under certain conditions, be sta-
ble in forward motion. Bicycles also exhibit nonminimum
Geometry
history of the bicycle is given in the recent book [1]. The pa
[2]–[6] and the classic book by Sharp from 1896, which
recently been reprinted [7], are good sources for early w
not consider extreme conditions and tight turns, we The parameters that descri
phase steering behavior.
Bicycles have intrigued scientists ever since they appeared in Notable contributions include Whipple [4] and Carv
the middle of the 19th century. A thorough presentation of the [5], [6], who derived equations of motion, linearized around
assume that the bicycle tire rolls without longitudinal or
history of the bicycle is given in the recent book [1]. The papers
[2]–[6] and the classic book by Sharp from 1896, which has
LOUIS MCCLELLAN/THOMPSON-MCCLELLAN PHOTOGRAPHY
are defined in Figure 1. The ke
1066-033X/05/$20.00©2005IEEE

lateral slippage. Control of acceleration and braking is not

recently been reprinted [7], are good sources for early work.
Notable contributions include Whipple [4] and Carvallo
26
head angle λ, and trail c. T
IEEE Control Systems Magazine August

[5], [6], who derived equations of motion, linearized around the

2019-3-18
CLELLAN/THOMPSON-MCCLELLAN PHOTOGRAPHY

1066-033X/05/$20.00©2005IEEE
IEEE Control Systems Magazine
Bike parameter
considered explicitly, but we often assume that the forward shaped so that 5.15
the contact
velocity is constant. To summarize, we simply assume that
August 2005
the road is behind the exten
the bicycle moves on a horizontal plane and that the defined as the horizontal di
definitions
wheels always maintain contact with the ground. point and the steer axis wh
zero steer angle. The riding
strongly affected by the tra
improves stability but make
λ values for c range 0.03–0.08 m
Geometrically, it is conv
composed of two hinged pla
front fork plane. The frame
frame plane, while the fron
C1 plane. The planes are joined
h C2 P1 and P2 are the contact p
horizontal plane, and the po
steer axis with the horizonta

P1 P2 P3
Coordinates
a
The coordinates used to an
low the ISO 8855 standard,
b c is an inertial system with
coordinate system xyz has
P1 of the rear wheel and the
Figure 1. Parameters defining the bicycle geometry. The
is aligned with the line of c
points P1 and P2 are the contact points of the wheels with the
the horizontal plane. The x
ground, the point P3 is the intersection of the steer axis with the
point P3 , which is the inte
horizontal plane, a is the distance from a vertical line through
the center of mass to P1 , b is the wheel base, c is the trail, h is axis and the horizontal pl
the height of the center of mass, and λ is the head angle. rear wheel plane is defined
angle between the ξ -axis a
2019-3-18 5.16
vertical, and y is perpendic
left side of the bicycle so
 1 2
the horizontal plane. The
ground, the point P3 is the intersection of the steer axis with the
point P3 , which is the int
horizontal plane, a is the distance from a vertical line through
the center of mass to P1 , b is the wheel base, c is the trail, h is axis and the horizontal p
Reference frame definitions
the height of the center of mass, and λ is the head angle. rear wheel plane is defined
angle between the ξ -axis a
vertical, and y is perpendic
left side of the bicycle so
ζ ϕf obtained. The roll angle ϕ
when leaning to the right. T
η
z plane is ϕf . The steer angle
ϕ between the rear and front
ing left. The effective steer
the lines of intersection of t
the horizontal plane.
C2 δf
C1 x
Simple Second-Ord
P2 P3 Second-order models will n
P1 ψ tional simplifying assumpt
bicycle rolls on the horizo
ξ fixed position and orientati
that the forward velocity at
Figure 2. Coordinate systems. The orthogonal system ξ ηζ is For simplicity, we assume
fixed to inertial space, and the ζ -axis is vertical. The orthogo- which implies that the hea
nal system xyz has its origin at the contact point of the rear trail c is zero. We also assum
wheel with the ξ η plane. The x axis passes through the points control variable. The rotatio
P1 and P3 , while the z axis is vertical and passes through P1 . ated with the front fork then

28 IEEE Control Systems Magazine

2019-3-18 5.17

Naı̈ve analysis
only degree of freedom. All
ll so that the equations can
y z

cycle are shown in Figure 3. O

ates around the vertical axis
Vδ/b, where b is the wheel
coordinate system xyz expe-
eleration of the coordinate ϕ
e.
f the system. Consider the
wheels, the rider, and the
o the rear frame with δ = 0,
inertia of this body with
δ
D = − Jxz denote the inertia
axes. Furthermore, let the x P1 P2 x y
r of mass be a and h, respec- a
of the system with respect b

(a) (b)
dϕ VD
= J − δ. Figure 3. Schematic (a) top and (b) rear views of a naive
dt b
(λ = 0) bicycle. The steer angle is δ, and the roll angle is ϕ.
tem are due to gravity and
ngular momentum balance
typographical error: λ = 90.
2019-3-18 5.18
It follows from (1) that the transfer function from steer
angle δ to tilt angle ϕ is
Naı̈ve analysis: simple second order models

Steering angle, δ, to tilt angle, φ, transfer function

dφ dφ VD
Lx = J − Dω = J − δ Angular momentum about x
dt dt b

d2 φ DV dδ mV 2 h
J 2 − mghφ = + δ Torque balance
dt b dt b

J ≈ mh2 and D ≈ mah Inertia approximations

d2 φ g aV dδ V2
− φ= + δ Simplified model
dt2 h bh dt bh

2019-3-18 5.19

Naı̈ve analysis: simple second order models

Steering angle, δ, to tilt angle, φ, transfer function

Transfer function:
φ(s) V (Ds + mV h) aV (s + V /a)
Gφδ (s) = = 2
≈
δ(s) b(Js − mgh) bh (s2 − g/h)

r r
mgh g
poles: p1,2 = ± ≈±
J h
mV h V
zero: z1 = − ≈−
D a

2019-3-18 5.20
 considered explicitly, but we often assume that the forward shaped so that the contact
velocity is constant. To summarize, we simply assume that the road is behind the exten
the bicycle moves on a horizontal plane and that the defined as the horizontal di
Bike parameter definitions
wheels always maintain contact with the ground. point and the steer axis wh
zero steer angle. The riding
strongly affected by the tra
improves stability but make
λ values for c range 0.03–0.08 m
Geometrically, it is conv
composed of two hinged pla
front fork plane. The frame
frame plane, while the fron
C1 plane. The planes are joined
h C2 P1 and P2 are the contact p
horizontal plane, and the po
steer axis with the horizonta

P1 P2 P3
Coordinates
a
The coordinates used to an
low the ISO 8855 standard,
b c is an inertial system with
coordinate system xyz has
P1 of the rear wheel and the
Figure 1. Parameters defining the bicycle geometry. The
is aligned with the line of c
points P1 and P2 are the contact points of the wheels with the
the horizontal plane. The x
ground, the point P3 is the intersection of the steer axis with the
point P3 , which is the inte
horizontal plane, a is the distance from a vertical line through
the center of mass to P1 , b is the wheel base, c is the trail, h is axis and the horizontal pl
the height of the center of mass, and λ is the head angle. rear wheel plane is defined
angle between the ξ -axis a
2019-3-18 5.21
vertical, and y is perpendic
left side of the bicycle so
ζ ϕf obtained. The roll angle ϕ
Front fork model when leaning to the right. T
η
z plane is ϕf . The steer angle
ϕ between the rear and front
ing left. The effective steer
the lines of intersection of t
Handlebar torque, T , to tilt angle, φ, transfer function the horizontal plane.
C2 δf
C1 x
Simple Second-Ord
Model the actuation as a torque to the handlebars,P T . Second-order models will no
P2 3
P1 ψ tional simplifying assumpt
bicycle rolls on the horizon
d2 φ DV g dφ mg 2 (bh cos λ − ac sin λ)
J 2 + 2 + ξ fixed position and orientati
φ
dt V sin λ − bg cos λ dt V 2 sin λ − bg cos λ that the forward velocity at
Figure For simplicity, we assume t
DV2.b Coordinate systems.
dT The orthogonal
b(V 2 h system
− acg) ξ ηζ is
= fixed to inertial space, and the+ζ -axis is vertical.
2 sin λ − bg cos λ) dt
The orthogo- T which implies that the hea
2 sin λ − bg cos λ)
acm(V nal ac(V
system xyz has its origin at the contact point of the rear trail c is zero. We also assum
√ x axis passes through the points
The system is stable if V > Vc = bg cot λ and bh > ac tan λ control variable. The rotatio
wheel with the ξ η plane. The
P1 and P3 , while the z axis is vertical and passes through P1 . ated with the front fork then

Gyroscopic effects could be included (giving additional damping).

28 IEEE Control Systems Magazine

2019-3-18 5.22
Front fork model

Torque to steering angle transfer function

With a stabilizable bicycle going at sufficiently high speed, V ,

δ k1 (V )
= GδT (s) = ,
T 1 + k2 (V )Gφδ (s)

V (Ds + mV h) aV (s + V /a)
where, as before, Gφδ (s) = 2
≈
b(Js − mgh) bh (s2 − g/h)
 
2 mgh
k1 (V ) s − J
So, GδT (s) =
k2 (V )DV k (V )V 2 mh mgh
s2 + s + 2 − J
bJ bJ

2019-3-18 5.23

Front fork model

Torque to path deviation transfer function

If η is the deviation in path,

 
2 mgh
k1 (V ) V 2 s − J
GηT (s) =   
b k 2 (V )DV mgh V 2
s2 s2 + s + J − 1
bJ Vc2

2019-3-18 5.24
Non-minimum phase behaviour

Counter-steering
“I have asked dozens of bicycle riders how they turn to the
left. I have never found a single person who stated all the
facts correctly when first asked. They almost invariably
said that to turn to the left, they turned the handlebar to
the left and as a result made a turn to the left. But on
further questioning them, some would agree that they first
turned the handlebar a little to the right, and then as the
machine inclined to the left they turned the handlebar to
the left, and as a result made the circle inclining inwardly.”
Wilbur Wright.

2019-3-18 5.25

Non-minimum phase behaviour

Counter-steering

2019-3-18 5.26
Non-minimum phase behaviour

Aircraft control
“Men know how to construct airplanes. Men also know
how to build engines. Inability to balance and steer still
confronts students of the flying problem. When this one
feature has been worked out, the age of flying will have
arrived, for all other difficulties are of minor importance.”
Wilbur Wright, 1901.

2019-3-18 5.27

Rear-wheel steered bicycles

Klein’s Ridable Bike

c K. J. Åström, Delft, June, 2004

! 32

2019-3-18 5.28
Rear-wheel steered bicycles

Stabilization: simple model

The sign of V is reversed in all of the equations.

 
2 −s + mV h
−V Ds + mV h VD D
Gφδ (s) = =  
2
b(Js − mgh) bJ mgh
s2 − J

aV (−s + V /a)
≈
bh (s2 − g/h)
This now has a RHP pole and a RHP zero.
r s
z mV h J V h
The zero/pole ratio is: = ≈
p D mgh a g

2019-3-18 5.29

Rear-wheel steered motorbikes

NHSA Rear-steered Motorcycle

I 1970’s research program sponsored by the US National Highway Safety
Administration.
I Rear steering benefits: Low center of mass.
Long wheel base.
Braking/steering on different wheels
I Design, analysis and building by South Coast Technologies, Santa
Barbara, CA.
I Theoretical study: real(p) in range 4 – 12 rad/sec. for V of 3 – 50 m/sec.
I Impossible for a human to stabilize.

2019-3-18 5.30
Rear-wheel steered motorbikes
The NHSA
NHSA Rear-steered Rear
Motorcycle Steered Motorcycle

c K. J. Åström, Delft, June, 2004

! 37

2019-3-18 5.31

Rear-wheel steered motorbikes

NHSA Rear-steered Motorcycle

“The outriggers were essential; in fact, the only way to
keep the machine upright for any measurable period of time
was to start out down on one outrigger, apply a steer input
to generate enough yaw velocity to pick up the outrigger,
and then attempt to catch it as the machine approached
vertical. Analysis of film data indicated that the longest
stretch on two wheels was about 2.5 seconds.”
Robert Schwartz, South Coast Technology, 1977.

2019-3-18 5.32
 Rear-wheel steered motorbikes

Meeks’ bike: “Quantum Leap”

gned based about what knowledge is required to avoid this trap,

ount for sta- emphasizing the role of dynamics and control. You can
of studying spice up the presentation with the true story about the
on design NHSA rear-steered motorcycle. You can also briefly men-
at an early tion that poles and zeros in the right-half plane are crucial
ccessfully in concepts for understanding dynamics limitations. Return to
a discussion of the rear-steered bicycle later in the course
www.autoevolution.com
a recumbent when more material has been presented. Tell students how
on that has a important it is to recognize systems that are difficult to con-
ave students trol because of inherently bad dynamics. Make sure that www.robbreport.com

y, “I have a everyone knows that the presence of poles and zeros in the
e and try it.” right-half plane indicates that there are severe difficulties in
Meeks’a system
with the rear- controlling reasonandfor not
also thatriding
the polesitand zeros are
The riding influenced by sensors and actuators.
“The bike’s so expensive, it’s a concept that’s going to be shown and to ride it
verly coura- This approach, which has been used by one of the authors
and to take a chance of chipping or scratching it, it’s not worth it. All we
te of repeat- in introductory classes on control, shows that a basic knowl-
wanted to do was make sure it worked, which we did.”
ed attempts, edge of control is essential for all engineers. The approach also
a discussion. illustrates the advantage of formulating a simple dynamic
a static point model at an early stage in a design project to uncover potential
problems caused by unsuitable system dynamics.
a discussion2019-3-18 5.33

Rear-wheel steered
aligns with the bicycles
frame when the speed is sufficiently large.
the bicycle, Another experiment is to ride a bicycle in a straight path
mple experi- on a flat surface, lean gently to one side, and apply the
le and lean steer torque to maintain a straight-line path. The torque
he front fork required can be sensed by holding the handlebars with a
experiment UCSB
light bike
fingered grip. Torque and lean can also be measured
e front fork with simple devices as discussed below. The functions The UCSB Rideable Bike
KARL ÅSTRÖM

This bicycle
ravity and Figure 20. The UCSB rear-steered bicycle. This bicycle is rid-
ass of the able as demonstrated by Dave Bothman, who supervised the
heel with construction of the bicycle. Riding this bicycle requires skill
and dare because the rider has to reach high speed quickly. c K. J. Åström, Delft, June, 2004
!

2019-3-18 5.34
IEEE Control Systems Magazine 41
Rear-wheel steered bicycles

An unridable bike
Klein’s Unridable Bike

2019-3-18 5.35

c K. J. Åström, Delft, June, 2004

! 31

Rear-wheel steered bicycles

This one had another problem!

2019-3-18 5.36
Notes and references

Skogestad & Postlethwaite (2nd Ed.)

Control limitations: sections 5.7 – 5.11
Practical controllability examples: sections 5.13 – 5.15.

More on bicycles
TU Delft: http://bicycle.tudelft.nl/schwab/Bicycle/index.htm
Article: Karl J. Åström, Richard E. Klein & Anders Lennartsson,
“Bicycle dynamics and control,” IEEE Control Systems
Magazine, vol. 25, no. 4, pp. 26–47, 2005.

2019-3-18 5.37