Vous êtes sur la page 1sur 43

The Dynamics of Charles Taylor’s One-Wheeled Vehicle

by

Bayram Orazov

B.S. (University of California, Berkeley) 2006

A report submitted in partial satisfaction of the


requirements for the degree of
Masters of Science, Plan II

in

Mechanical Engineering

in the

Graduate Division
of the
University of California, Berkeley

Committee in charge:

Professor Oliver M. O’Reilly, Chair


Professor George C. Johnson

Fall 2007
The report of Bayram Orazov is approved:

Chair Date

Date

University of California at Berkeley

Fall 2007
i

c 2007 Bayram Orazov.


Copyright

All rights reserved. No reproduction without author’s permission.

For inquiries, please reach me at bayram@berkeley.edu


ii

Contents

List of Figures iii

Symbols and Abbreviations v

1 Introduction 1

2 A Simple Model 4
2.1 The Rolling Constraints . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Forces and Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Non-dimensionalization . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Motions of the Vehicle in a Straight Line 14

4 Steady Circular Motions 21

5 Conclusions 32

Bibliography 34
iii

List of Figures

1.1 One of Charles F. Taylor’s one-wheeled vehicle prototypes . . . . . . 1

2.1 Model schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3.1 Straight line stability findings . . . . . . . . . . . . . . . . . . . . . . 16


3.2 Simulations of stable straight line motion . . . . . . . . . . . . . . . . 17
3.3 Path of the center of mass during stable straight line motion . . . . . 18
3.4 Simulations of unstable straight line motion . . . . . . . . . . . . . . 20
3.5 Path of the center of mass during unstable straight line motion . . . . 20

4.1 Steady state circular motion. θ0 contours as a function of ω20 and ω30 . 22
4.2 Steady state circular motion. ω30 contours as a function of ω20 and θ0 . 23
4.3 Stability results for steady circular motion. . . . . . . . . . . . . . . . 24
4.4 Simulations of stable steady circular motion. . . . . . . . . . . . . . . 25
4.5 Path of the center of mass during stable straight line motion . . . . . 26
4.6 Simulations of unstable circular motion. . . . . . . . . . . . . . . . . 29
4.7 Path of the center of mass during unstable cornering motion . . . . . 30
iv

Acknowledgements

I express my deepest gratitude to Professor Oliver M. O’Reilly for his thorough

guidance in this work. I also wish to thank Eric S. Lew for his prior work on this

project, which made my job much easier. Jointly, we are very thankful to Mary Urry

for supplying us with background information on her father, Charles F. Taylor, and to

Bernice Yen and Tim Wheeler for their research on the patent of the vehicle analyzed

here.
v

Symbols and Variables

θ Roll angle of the vehicle

ψ Yaw angle of the vehicle

φ1 Pitch angle of the platform

ω20 Angular speed of the wheel

ω30 Yaw rate of the vehicle

σ̇ Angular speed of the rear steerable gyroscope

η̇ Angular speed of the front gyroscope


1

Chapter 1

Introduction

Figure 1.1: Charles F. Taylor on a prototype of his one-wheeled vehicle in the early
1960’s in Colorado, U.S.A. We are grateful to his daughter, Mary Urry, for supplying
this photo.

An American inventor Charles F. Taylor (1916–1997) designed, built and tested

various prototypes of a one-wheeled vehicle during the 1950’s and 1960’s. He hoped

his machine could be used as an all-terrain vehicle and he was awarded a patent for
2

it in 1964 [9]. One of the later models of his machine can be seen in Figure 1.1. This

vehicle dates to the early 1960’s and, as can be seen from the image, is stabilized by

a series of gyroscopes. In particular, it is possible for the stationary vehicle to remain

upright and at rest. Among the novel features of Taylor’s design when compared to

other single wheeled machines, such as Erich Edison-Puton’s monocycle from 1910,1

is the ability to accommodate passengers.

In this report, we propose a simple model for Taylor’s vehicle and use it to explain

some of the efficacies of his ingenious design. Specifically, we use it to show how

his machine can be stabilized in a stationary position using gyroscopes and how the

machine can execute stable steady motions. Our analysis is based on a linear stability

analysis of a multibody system. This system can be viewed as a generalization of a

rolling circular disk and we use various analyses of the rolling disk, such as [1, 2, 3,

4, 5, 6, 8], to guide our analysis. As in the rolling disk, the steady motions we are

interested in are those where the vehicle is moving in a straight line and those where

its center of mass traverses a circle.

First, we create a simplified rigid body model of the vehicle and derive the equa-

tions of motion for the system using the balance of angular momentum. Next, the

equations are linearized about straight line and circular trajectories, eigenvalues are

solved for, and linear stability criteria for each case are obtained. For several of these
1
A reproduction of the Edison-Puton monocycle can be found in the Auto und Technik Museum
Sinsheim Germany.
3

cases we find multiple zero eigenvalues and so our eigenvalue analysis needs to be

supplemented. Thus, to validate our findings, we numerically integrate the equations

of motion for various stable and unstable combinations of parameters and comment

on the results.

The developed model takes into account the instantaneous roll and yaw of the

vehicle, though we choose to ignore the vehicle’s pitch variation, as pitch should not

change during steady motions. The model also assumes that the vehicle is always

in contact with the ground, that it rolls without slip, and is independent of friction

coefficients between the ground and the wheel.


4

Chapter 2

A Simple Model

We choose to model the one wheeled vehicle as simply as possible. To this end,

a model consisting of four components will be employed. The components are a

circular disk representing the wheel, a parallelepiped representing the platform and

its attendant masses, and two gyroscopes. One of the gyroscopes is used to steer the

vehicle.

A schematic representation of the model is shown in Figure 2.1. The parallelepiped

modeling the platform has dimensions c × d × e and mass m1 , and the disk has a

radius R and mass m2 . At the rear of the parallelepiped, a gyroscope of mass m3 is

rigidly mounted. We assume that this device has the shape of a cylinder of radius R3

and height h3 . A similar gyroscope of mass m4 , radius R4 and height h4 is mounted

at the front of the vehicle. In our description here (and in the sequel), we choose
5

e3
e3 E3
η̇ Wheel σ̇
g Gyro 2 Gyro 1
O
e1 E1
X
Platform

Ground plane

Figure 2.1: Schematic of the model for the one-wheeled vehicle. The unit vectors e1
and e3 are fixed to the platform. Along with the vector e2 = e3 ×e1 , these two vectors
form a basis which is used to obtain component forms of the equations of motion.
Gyro 1 is known as the steering gyroscope.

to label quantities associated with the platform with a subscript 1, the disk with a

subscript 2, the steering gyroscope with a subscript 3 and the second gyroscope with

a subscript 4.

We define a inertial reference frame using a fixed origin O and a Cartesian basis

{E1 , E2 , E3}. As shown in Figure 2.1, we also define a basis {e1 , e2 , e3 } which is

rigidly mounted to the platform. The directions defined by these three unit vectors

are principal directions for the parallelepiped, circular disk and both gyroscopes. The

rotation of the platform is described using a set of 3-1-2 Euler angles, and so its

angular velocity vector has the representation

3
X
ω1 = ω1i ei , (2.1)
i=1
6

where

ω11 = θ̇ cos (φ1 ) − ψ̇ cos (θ) sin (φ1 ) ,

ω12 = ψ̇ sin (θ) + φ̇1 ,

ω13 = θ̇ sin (φ1 ) + ψ̇ cos (φ1 ) cos (θ) (2.2)

Here, ψ is the yaw angle of the platform, θ is the roll (or tilt) angle of the platform

and φ1 is the pitch angle of the platform. When θ = 0 and φ1 = 0, the platform has

the configuration shown in Figure 2.1.

The angular velocity vector of the disk and gyroscopes relative to the platform

are easily defined:


3
X
ω2 = ωk ek = ω 1 + φ̇2 e2 ,
k=1

ω 3 = ω 1 + σ̇ (cos (δ) e2 + sin (δ) e1 ) ,

ω 4 = ω 1 + η̇e3 (2.3)

Here, η̇ is the rotational speed of the fore-gyroscope, and δ and σ̇ are the steering

angle and rotational speed, respectively, of the aft-gyroscope.

We also assume that the center of mass X of this system of four rigid bodies is

located at the center of mass of the wheel, and has a position vector
3
X
x= xi Ei (2.4)
i=1

Relative to this X, the position vectors of the centers of mass of the gyroscopes are

x3 − x = −ae1 , x4 − x = ae1 (2.5)


2.1. THE ROLLING CONSTRAINTS 7

We can use these identities to derive simple expressions for the velocity vectors of the

centers of mass of the gyroscopes.

2.1 The Rolling Constraints

The circular disk is assumed to roll without slipping. Consequently, the following

three constraints hold at the instantaneous point of contact of the disk with the

horizontal surface (road plane):

ẋ = Rω 2 × e3 (2.6)

With the help of the identities ėk = ω 1 × ek , we find that the rolling constraints

imply that the acceleration vector of the center of mass is determined by the angular

velocities and accelerations of the disk:

ẍ = a (2.7)

where

  
2
a = R (ω̇2 + ω1 ω3 ) e1 + R (−ω̇1 + ω2 ω3 ) e2 − R ω1 + ω2 ω2 − φ̇2 e3 (2.8)

In writing down the final form of these results, we used the identities

ω1 = ω11 , ω2 = ω12 + φ̇2 , ω3 = ω13 (2.9)


2.2. MOMENTA 8

2.2 Momenta

The linear momentum G of the system is the linear momentum of its mass center:

G = mẋ, (2.10)

where m = m1 + m2 + m3 + m4 . In addition to G, to establish the equations of motion

for the system, the angular momentum H needs to be established:

4
X
H = (x3 − x) × m3 ẋ3 + (x4 − x) × m4 ẋ4 + HJ
J=1

This angular momentum has the representation

 
2 dφ2 dφ1
H = a (m3 + m4 ) − ω2 − cos(φ1 )e2
dt dt
4
X
2
−a (m3 + m4 ) (sin(φ1 )ω1 + ω3 cos(φ1 )) e3 + HJ , (2.11)
J=1

where HJ are the angular momenta of the individual components relative to their

respective centers of mass. The momenta H1 , . . . , H4 are easily expressed in terms of

moments of inertia and angular velocity components:

1
m1 2c2 + d2 + e2 ω1 − (d2 − e2 )(ω1 cos(2φ1 ) − ω3 sin(2φ1 )) e1
 
H1 =
24
 
1 2 2
 dφ1 dφ2
+ m1 d + e ω2 + − e2
12 dt dt
1
+ m1 2c2 + d2 + e2 ω3 + (d2 − e2 )(ω3 cos(2φ1 ) + ω1 sin(2φ1 )) e3
 
24
1 1 1
H2 = m2 R2 ω1 e1 + m2 R2 ω2 e2 + m2 R2 ω3 e3
4 2 4
2.3. FORCES AND MOMENTS 9

1
m3 6ω3 sin(2φ1 )a2 + 3R22 + h2 (cos(φ1 )σ̇ sin(δ) + ω1 ) e1
 
H3 =
12
1
+ m3 aω1 (a − a cos(2φ1 ) + 2R sin(φ1 )) e1
2
   
1 2 2 2
 dφ1 dφ2
+ m3 cos(δ)σ̇R2 + 2a + R2 ω2 + − − 2aRω2 sin(φ1 ) e2
2 dt dt
1
+ m3 6a2 (1 + cos(2φ1 )) ω3 + 3R22 + h2 (ω3 − σ̇ sin(δ) sin(φ1 )) e3
 
12
1
+ m3 aω1 (a sin(2φ1 ) − 2R cos(φ1 )) e3
2
1
m4 12η̇ sin(φ1 )R22 + 12a2 − h2 + 3R22 (ω3 sin(2φ1) − ω1 cos(2φ1)) e1
 
H4 =
24
1
+ m4 ω1 12a2 + 24R sin(φ1 )a + h2 + 9R22 e1

24
   
1 2 2 2
 dφ1 dφ2
+ m4 12a + h + 3R2 ω2 + − + am4 Rω2 sin(φ1 ) e2
12 dt dt
1
+ m4 12 cos(φ1 )η̇R22 + 12a2 + h2 + 9R22 ω3 + 24aRω1 cos(φ1 ) e3
 
24
1
+ m4 12a2 − h2 + 3R22 (ω3 cos(2φ1 ) + ω1 sin(2φ1 ))e3

(2.12)
24

2.3 Forces and Moments

The external forces acting on the system consist of a gravitational force −mgE3

and a reaction force F at the instantaneous point of contact P of the disk with the

horizontal surface. The force at P consists of the sum of the normal force and the

friction force. The equations of motion for the vehicle can be obtained using the

balances of linear and angular momenta:

F − mgE3 = mẍ, −Re3 × F = Ḣ (2.13)


2.3. FORCES AND MOMENTS 10

Using an idea from Routh [8], we can use (2.8) with (2.13)1 to determine F in terms

of the rotational kinematics of the system. With the help of this substitution, we

then arrive at the equations governing the angular velocities ωi :

Ḣ + mR2 (ω̇1 − ω2 ω3 ) e1 + mR2 (ω̇2 + ω1 ω3 ) e2 = mgRE3 × e3 (2.14)

In the sequel, we shall assume that the pitch angle of the platform is zero,

φ1 = 0, (2.15)

and the rotation speeds of the gyroscopes are constant:

η̇ = η̇0 , σ̇ = σ̇0 (2.16)

The equations (2.14) will feature a set of differential equations for the roll angle θ,

dθ dφ2
and the angular velocities ω1 = dt
, ω2 and ω3 . To remove dt
from the equations of

motion (2.14), we note

dψ dφ2 dψ
ω2 = sin (θ) + , ω3 = cos (θ) , (2.17)
dt dt dt

and so
dφ2
= ω2 − ω3 tan (θ) (2.18)
dt
d2 φ2
A related result for dt2
can be easily inferred.
2.4. NON-DIMENSIONALIZATION 11

2.4 Non-dimensionalization

To help arrive at a tractable set of equations, we non-dimensionalize the time t

and define a new time variable τ :


s
1 5R
t= τ (2.19)
2 g

In addition, we non-dimensionalize the masses using m2 and lengths using R. For

example,
mK R4 a
m̂K = , R̂4 = , â = (2.20)
m2 R R

The dimensionalizations we are using are identical to those in O’Reilly’s formulation

of the rolling disk [6]. After nondimensionalizing the quantities mK , R3 , R4 , a, d,


e, h3 , h4 , we subsequently remove the ˆ· and often denote dτ
by a ·˙. In addition, we

define the dimensionless total mass m:

m = m1 + 1 + m3 + m4 (2.21)

In the sequel, the following parameter values are used to generate numerical results:

m1 m3 m4
= 10, = = 0.25,
m2 m2 m2

R2 h h3 h4
= 0.25, = = = 0.25,
R R R R
a c d e
= 2.0, = 4.0, = 1.0, = 0.25 (2.22)
R R R R
2.5. EQUATIONS OF MOTION 12

2.5 Equations of Motion

Performing the aforementioned non-dimensionalizations, we find that (2.14) yields

the following ordinary differential equations:

dω1
M11 = A11 ω3 + A12 tan (θ) ω3 + A14 tan (θ) ω32 + A17 ω2 ω3

+A18 sin (θ) ,
dω2 dω3
M22 + M23 tan (θ) = A21 ω1 + A22 ω3 + A26 tan2 (θ) ω1 ω3 + A27 ω1 ω3 ,
dτ dτ
dω3
M33 = A31 ω1 + A32 tan (θ) ω3 + A35 ω1 ω2

+A36 tan (θ) ω1 ω3 , (2.23)

where m3 = m4 ,

c2 + e2 + 12 m1 + m3 h2 + 3R32 + 12 + 3 m4 R32 + 4 + 5 ,
   
M11 =

M22 = 12m + 6,

M23 = 12m3 a2 + 12m4 a2 + 6m3 R32 + 3m4 R32 + d2 + e2 m1 ,




c2 + d2 m1 + 6m4 2a2 + R32 + m3 12a2 + h2 + 3R32 + 3,


  
M33 = (2.24)

and

A11 = 6 cos(δ)σ̇0 m3 R32 ,

A12 = −6η̇0 m4 R32 ,

3m3 R32 + e2 − c2 m1 − m3 h2 − 3m4 R32 − 3 ,


 
A14 =
2.5. EQUATIONS OF MOTION 13

A17 = 12m + 6,

A18 = 15m,

A21 = 6η̇0 m4 R32 ,

A22 = −σ̇0 sin(δ)m3 h2 + 3R32 ,




A26 = − 3m4 R32 + 12 (m3 + m4 ) a2 + 6m3 R32 + m1 d2 + e2



,

A27 = − 6m3 R32 + 2m1 e2 + 12m ,




A31 = −6 cos(δ)σ̇0 m3 R32 ,

A32 = σ̇0 sin(δ)m3 h2 + 3R32 ,




A35 = −6,

A36 = − 12m4 a2 + d2 − c2 m1 + m3 12a2 − h2 + 3R32 − 3


  
(2.25)

If we were to ignore the dynamics of the gyroscopes and platform, then, as expected,

we would find that (2.23) reduce to the differential equations for a rolling disk pre-

sented in equation (7) of [6].


14

Chapter 3

Motions of the Vehicle in a

Straight Line

The foremost case to consider features a vehicle moving in a straight line at

constant speed. In this case, the yaw angle of the platform is constant and the roll

angle of platform is zero:

ψ = ψ0 , θ = 0, ω1 = 0, ω2 = φ̇2 = ω20 , ω3 = 0 (3.1)

As expected, the constant (dimensioned) forward speed of the center of mass of the

vehicle is −Rφ̇2 . Examining (2.23), we observe that this model allows the vehicle to

move in a straight line at an arbitrary speed.

To examine the stability of motion in a straight line, we consider the following


15

perturbations to the roll angle and the angular speeds:


θ = 0 + Θ, ω1 = 0 + , ω2 = ω20 + Ω2 , ω3 = 0 + Ω3 (3.2)

The linearized equations of motion are obtained in the usual manner from (2.23):
     

 1 0 0 0  dτ   0 1 0 0  Θ 
     
     
 0 M dΩ1
 11 0 0 
 dτ   18 0
  A 0 B̂24 

 Ω1 

  =  , (3.3)
     
 0 0 M dΩ2
 22 0 
 dτ
  0 A
  21 0 A22 
 Ω2 

     
     
dΩ3
0 0 0 M33 dτ
0 B̂42 0 0 Ω3

where B̂24 = (A11 + ω20 A17 ) and B̂42 = (A31 + ω20 A35 ) .

A standard analysis shows that the characteristic polynomial for this equilibrium

of (3.3) is

M22 M11 M33 λ2 − (A11 + ω20 A17 ) (A31 + ω20 A35 ) − A18 M33 λ2 = 0,

(3.4)

where λ are the roots of the characteristic polynomial. We can use (3.4) to determine

the range of gyroscope speeds η̇0 and σ̇0 and wheel speeds ω20 where the two non-

trivial eigenvalues λ1,2 will be purely imaginary. As the equilibria corresponding to

motion in a straight line always have a pair of zero eigenvalues, in order to conclude

stability from a linear eigenvalues analysis, we have supplemented this analysis with

simulations of the nonlinear equations of motion (2.23).

An examination of the terms in (3.4) reveals that the stability of the motion in

question is independent of the gyroscope speed η̇0 . The regions of stability of the
16

20000

stable

unstable
1 4
6 2
σ̇0 0
3 5
unstable
stable

−20000
−10 ω20 = φ̇2 10

Figure 3.1: Stabilization of the motion of the vehicle in a straight line using the
gyroscope rate σ̇0 for a given rotational speed ω20 . Points 1-4 lie in the stability
region, while points 5 and 6 belong to the unstable parameter space.

motion in the ω20 − σ̇0 parameter plane are shown in Figure 3.1. These regions

were determined from the eigenvalues λ1,2 and were verified using the aforementioned

simulations. It should be apparent from Figure 3.1 that the gyroscope is capable of

stabilizing the stationary vehicle.1 As evidenced by movies of prototypes of Charles

Taylor’s vehicle, this capability is borne out in the actual one-wheeled vehicle.2

dφ2 dσ0
When ω20 = dτ
is negative, the destabilizing values of dτ
are positive and vice

versa. This could be thought of as “cancelation” of angular momentum, and is similar

to the brake lockup instability in vehicles that is discussed in [7]. It is also easy

to conclude from the results shown in Figure 3.1 that the straight line stability of

the vehicle is fairly easy to ensure solely by means of the rear gyroscope, and this
1
I.e., ω20 = 0.
2
The films in question can be found at http://www.me.berkeley.edu/one wheel vehicle/.
17

gyroscope is only needed at low moving speeds. At higher speeds, the vehicle is able

to stabilize itself without gyroscopes (i.e. with σ̇0 = 0).

We now include the results of simulations of equations of motion (2.23) for six

test cases when the vehicle is moving in a straight line. Parameter values were taken

from Figure 3.1. The sample four cases illustrated below (points 1-4 in Figure 3.1)

represent stable motion, as evidenced by the θ and ψ plots, shown by solid and dashed

lines, respectively.
(1) (2)
×10−3
0.01 0 5 0

θ 0 ψ θ0 ψ

−0.01 −0.1 −5 −0.01


0 time (s) 100 0 time (s) 100
(3) (4)

0.01 0.04 0.01 0

θ 0 ψ θ 0 ψ

−0.01 0 −0.01 −0.04


0 time (s) 100 0 time (s) 100

Figure 3.2: Simulations of stable straight line motion using the gyroscope rate σ̇0
(η̇0 = 0). The time variation of the roll angle θ is shown by a solid line, while the
time variation of the yaw angle ψ is shown by a dashed line. This motion results from
a perturbation to a stable straight line motion. The parameter values used are: (1)
ω20 = 0, σ̇0 = 2000; (2) ω20 = 7, σ̇0 = −50; (3) ω20 = −4, σ̇0 = −360 and (4) ω20 = 2,
σ̇0 = 900, as taken from Figure 3.1.
18

The plots in Figure 3.2 were obtained by slightly perturbing the vehicle away

from its equilibrium state. This resulted in bounded small-angle oscillations about

the equilibrium straight line position (i.e. θ = 0, ψ = 0), which implies that the

vehicle is able to stabilize itself under the simulated conditions and retain a largely

straight line trajectory.

The position of the center of mass of the vehicle for the four above cases is shown

in Figure 3.3, simulated for 100 seconds.


(1) (2)
0.1 0.5
0

y y

0 −2.5
0 x 3 0 x 800
×10−3
(3) (4)
1 1
0 0

y y

−6 −4
−400 x 0 0 x 200

Figure 3.3: Path of the center of mass of the vehicle under perturbed stable straight
line motion, shown on an x-y plane. The parameter sets used are the same as in
Figure 3.2. The initial location of the center of mass is at the origin.

The conditions for the case in Figure 3.3 (1) set ω20 = 0, making the vehicle

stationary. However, due to the initial perturbation, the center of mass of the vehicle
19

has to move around to stabilize itself, though the traveled distance is very small.

Cases (2)-(4) show the behavior under more typical conditions, when the wheel speed

is non-zero. The vehicle inevitably travels sideways due to the variations in roll and

yaw angles. However, perfectly straight line motion should still be possible in these

cases if the vehicle operator adjusts the steering input accordingly. In situations when

no perturbations are introduced, vehicle’s path is completely straight in cases (2)-(4)

and the vehicle remains perfectly stationary in case (1).

The next two sample cases (points 5, 6 in Figure 3.1) illustrate attempts at straight

line motion when the vehicle is not able to stabilize itself after an initial perturbation

and eventually rolls over into the ground. From platform geometry, the vehicle hits

the ground when the value of the roll angle reaches θ = ±1.1071 rad. We should note

that if no perturbations were introduced, the vehicle would move in straight line.

When the value of θ reaches either the upper or the lower horizontal dashed line

in Figure 3.4, the vehicle will impact the ground and the motion will stop. However,

the simulation was allowed to continue past this point to illustrate the tendency of an

unstable vehicle to roll between the two extreme values of ± π2 . Upon reaching such

extremes, singularities occur in the equations of motion (2.23), hence the sharp spikes

in the θ plots. The path of the center of mass of the vehicle is shown in Figure 3.5,

and the dot indicates the first time the vehicle hits the ground: after 11.717 seconds

in case (5) and after 11.597 seconds in case (6)


20

(5) (6)
2 0.2 2 0.5

θ0 ψ θ0 ψ

−2 −0.2 −2 −0.5
0 time (s) 100 0 time (s) 100

Figure 3.4: Simulations of unstable straight line motion using the gyroscope rate σ̇0
(η̇0 = 0). The time variation of the roll angle θ is shown as a solid line, while the
time variation of the yaw angle ψ is shown as a dashed line. This motion results from
a perturbation to an unstable straight line motion. The parameter values used are:
(5) ω20 = 7, σ̇0 = −400 and (6) ω20 = −3, σ̇0 = 100, as taken from Figure 3.1. The
horizontal dashed lines give the limits on physically allowable values of θ = ±1.1071.

(5) (6)

0.1 100
0
0
y y

−0.25 −400
0 800 −100 0 100
x x

Figure 3.5: Path of the center of mass of the vehicle during perturbed unstable
straight line motion, shown on an x-y plane. The parameter sets used are the same
as in Figure 3.4. The initial location of the center of mass is at the origin, and the
dot indicates the first time the vehicle hits the ground.
21

Chapter 4

Steady Circular Motions

We now consider a second class of steady motions where the vehicle has a non-

constant roll angle θ0 6= 0 and constant angular speeds ω2 = ω20 and ω3 = ω30 . During

these motions, the vehicle’s center of mass traces out a circular path. We find from

(2.23) that these motions can be sustained provided the following three conditions

hold:

2
0 = A11 ω30 + A12 tan (θ0 ) ω30 + A14 tan (θ0 ) ω30 + A17 ω20 ω30 + A18 sin (θ0 ) ,

0 = σ̇0 sin(δ)m3 v 2 + 3R32 ω30 ,




0 = σ̇0 sin(δ)m3 v 2 + 3R32 tan (θ0 ) ω30



(4.1)

Setting the steering angle δ = 0 or π, we find that these equations yield a relationship

between θ0 , ω20 and ω30 . Representative examples of this relationship are displayed

in Figures 4.1 and 4.2. We observe from these figures that all roll angles can be
22

sustained provided ω20 and ω30 are chosen appropriately. Indeed, large roll angles are

possible if ω30 is sufficiently small. That is, the turning rate of the vehicle is small.

θ0 = −0.3 θ0 = 0.3
10
θ0 = 0.9
θ0 = −0.9

θ0 = −1.5 θ0 = 1.5
ω30
θ0 = 1.5 θ0 = −1.5

θ0 = 0.9
θ0 = 0.3 θ0 = −0.3 θ0 = −0.9
−10
−10 ω20 10

Figure 4.1: Steady state values of ω20 and ω30 for a given value of the roll angle θ0 .
For this figure, σ̇0 = 10, η̇0 = 5 and the steering angle δ = 0.

To ascertain stability results, we note that, as in the motion in a straight line, we

don’t possess any conserved quantities (or integrals of motion). Consequently, and in

contrast to the case of the rolling disk, we can only perform a linear stability analysis.

Linearizing the equations of motion for the case at hand we find that
     

 1 0 0 0  dτ   0 1 0 0  Θ 
     
     
 0 M dΩ1
 11 0 0 
 dτ   21 0 B23 B24
  B  Ω 
 1 
  =   (4.2)
     
 0 0 M dΩ2
22 M23 tan (θ0 )
   0 B 0 B34  Ω 
  dτ   32  2 
     
     
dΩ3
0 0 0 M33 dτ
B41 B42 0 B44 Ω3
23

2
−0.1 0.1
−1.0 1.0
10
−10 0.1 −0.1
20
−20
40
−40
θ0
40
−40
20 −20
10
−0.1 0.1 −10
1.0 −1.0
0.1 −0.1
−2
−20 ω20 20

Figure 4.2: Steady state values of the roll angle θ0 as a function of the constant speed
ω20 for a given value of the rate ω30 . For this figure, σ̇0 = 10, η̇0 = 5 and the steering
angle δ = 0.

As in the previous linearized equations, Θ denotes the perturbation to θ0 , and Ωi are

the perturbations to ωi0 . Further,

2
B21 = A14 ω30 + A12 ω30 + A18 cos (θ0 ) ,

B23 = A17 ω30 ,

B24 = A11 + (A12 + 2A14 ω30 ) tan (θ0 ) + A17 ω20 ,

A26 tan2 (θ0 ) + A27 ω30 + A21 ,



B32 =

B34 = A22 ,

B41 = 2A32 tan (θ0 ) sec2 (θ0 ) ω30 ,

B42 = A31 + A35 ω20 + A36 ω30 tan (θ0 ) ,

B44 = A32 tan (θ0 ) (4.3)


24

The characteristic polynomial associated with the equilibrium of this linear system is

0 = M11 M22 M33 λ4 − M22 (B44 M11 ) λ3

+ (B23 B42 M23 tan (θ0 ) − B24 B42 M22 − B23 B32 M33 − B21 M22 M33 ) λ2

+ (B23 B41 M23 tan (θ0 ) + B21 B44 M22 − B24 B41 M22 + B23 B32 B44 − B23 B34 B42 ) λ

−B41 B34 B23 (4.4)

This polynomial must also take into consideration the remaining non-trivial condition

for steady state motions (from (4.1)3 ):

(B24 − A14 ω30 tan (θ0 )) ω30 + A18 sin (θ0 ) = 0 (4.5)

(a) (b)
4
1 1

2 8
ω30 ω30 7
3 6
5
1

−1 −1
−10 ω20 10 −10 ω20 10

Figure 4.3: Steady state values of ω20 and ω30 for a given value of the roll angle
θ0 : (a), θ0 = −1.0, −0.6, −0.5, −0.25, −0.0001 and (b) θ0 = 0.0001, 0.25, 0.5, 0.6, 1.0.
For this figure, σ̇0 = 10, η̇0 = 5, the steering angle δ = 0 and the arrows indicate
decreasing values of θ0 . In this figure, the solid lines indicate linearly stable motions,
while the dashed curves indicate linear instability. Points 1-4 lie in the stable region,
while points 5-8 belong to the unstable region.

In Figure 4.3, many of the steady circular motions shown in Figure 4.1 are re-

visited. Now however, we compute the eigenvalues associated with the characteristic
25

polynomial and use the results to determine if the steady circular motion is stable or

unstable. The numerical results we find show that slow circular motions where ω20

is small are unstable if the roll angle θ0 is too small. However, for any angle θ0 , by

making ω20 or ω30 sufficiently large, a linearly stable circular motion of the vehicle can

be achieved. As in the straight line case, these results were verified with simulations

of the nonlinear equations of motion (2.23). The plots of several sample cases are

presented below. The motion was simulated for 100 seconds.

(1) (2)
−0.995 0 −0.59 15

θ −1 ψ θ ψ
−0.6

−1.005 −40 −0.605 0


0 time (s) 100 0 time (s) 100
(3) (4)
0.255 4 0.502 100

θ 0.25 ψ θ 0.5 ψ

0.245 0 0.498 0
0 time (s) 100 0 time (s) 100

Figure 4.4: Simulations of stable circular motion using gyroscope rates σ̇0 = 10,
η̇0 = 5 and δ = 0. The time variation of the roll angle θ is shown as a solid line, while
the time variation of the yaw angle ψ is shown as a dashed line. This motion results
from a perturbation to a stable circular motion. The parameter values used are: (1)
θ0 = −1, ω20 = −2.6226, ω30 = −0.318; (2) θ0 = −0.6, ω20 = 5.1051, ω30 = 0.1297;
(3) θ0 = 0.25, ω20 = −7.988, ω30 = 0.0371 and (4) θ0 = 0.5, ω20 = 0, ω30 = 0.9695, as
taken from Figure 4.3.
26

Figure 4.4 illustrates four cases of stable circular motion, as determined from

Figure 4.3. The plots were obtained using a small perturbation about the equilibrium

state to show that the vehicle can indeed stabilize itself with the presence of external

disturbances. It is evident that the value of the roll angle oscillates about the baseline

number θ0 but never strays far from it, indicating that such motion is sustainable.

The plots of the yaw angle ψ confirm that the vehicle travels in a circular path at a

constant rate (plot is linear). The trajectories of the center of mass of the vehicle for

the above cases are presented in Figure 4.5.

(1) (2)
150 300

y y 0
0

−150 −400
−200 0 x 200 −400 0 x 600
(3) (4)
600 1

y y 0
0

−400 −0.2
−200 0 x 800 −0.5 x0 0.5

Figure 4.5: Path of the center of mass of the vehicle during perturbed stable circular
motion, shown on an x-y plane. The parameter values used are the same as in Figure
4.4. The initial location of the center of mass is at the origin.
27

The above trajectory plots were obtained by calculating vehicle velocities in the

local ei reference frame and transforming these quantities into the Ei basis. The

positions were obtained by integrating the velocity in the ground reference frame.

The initial position for the center of mass of the vehicle was taken to be the origin.

Further, the initial forward velocity in the e1 direction has been defined using equation

(2.6), but the initial outward velocity in the e2 direction has been set to zero.

The foremost thing to notice about the plots is that the trajectories are not

completely circular, but rather spiraling in cases (1)-(3). The reason for this can be

found in the acceleration expression in equation (2.8) which is replicated below for

convenience.

  
2
a = R (ω̇2 + ω1 ω3 ) e1 + R (−ω̇1 + ω2 ω3 ) e2 − R ω1 + ω2 ω2 − φ̇2 e3

Because we maintain the roll angle θ and wheel speed ω2 to be constant, the

acceleration in the e1 direction is zero. That is, the vehicle’s forward speed is constant

as it traverses the circular trajectory. However, the acceleration in the e2 direction

is not zero under the specified conditions, since the roll angle is constant (ω˙1 = 0),

while ω2 ω3 6= 0. It is for this reason that we see the outward spiraling trajectory.

Setting the outward acceleration to zero in the simulations confirms the viability of

perfectly circular motion. For the vehicle to move in perfect circles, it would need a

(non-constant) change in the roll angle. This introduces yet another constraint for

this family of motions.


28

We can also observe that in case (1) the yaw rate is negative, while cases (2)-(3)

have a positive yaw rate. This explains the clockwise trajectory in the first case and

a counter clockwise trajectory in the other two. Further, the lower the magnitude of

the yaw rate, the bigger is the radius of the turn, as demonstrated in case (3). Case

(4) represents another interesting example. The conditions for this case set ω20 = 0,

implying no wheel speed in the e1 direction. However, the roll angle and yaw rate

are set to rather aggressive values. The only way this situation can be resolved is

if the vehicle spins about its contact point P. However, this would violate the no-

slip condition from equation (2.6). Consequently, the vehicle is forced to move in a

circular path to retain its stability. Note that unlike cases (1)-(3) the resulting path

is actually circular, since with ω20 = 0, there is no acceleration in the e2 direction.

The four cases shown in Figure 4.6 represent unstable cornering motion, simu-

lated for 100 seconds, although the vehicle impacted the ground (indicated by dashed

horizontal lines) well before the end of the simulations. Once again, the appropriate

points were taken from Figure 4.3. Note that for the given gyroscope rates, there

are no points of instability for |θ0 | ≥ 0.75 rad so long as the conditions in equation

(4.5) are met. While all four situations are inherently unstable, the vehicle hits the

ground faster in some cases than in others. Aside from the obvious differences in

parameter values, it is informative to observe where the four points lie in Figure 4.3

in terms of their proximity to the stability regions. Points (5) and (6) lie directly on

the dashed contours for their respective values of θ0 . The vehicle hits the ground after
29

(5) (6)
2 0 2 0

θ ψ θ ψ

−1 −30 −2 −10
0 time (s) 100 0 time (s) 100
(7) (8)
2 20 2 20

θ ψ θ ψ

−2 0 0 0
0 time (s) 100 0 time (s) 100

Figure 4.6: Simulations of unstable circular motion using gyroscope rates σ̇0 = 10,
η̇0 = 5 and δ = 0. The time variation of the roll angle θ is shown as a solid line,
while the time variation of the yaw angle ψ is shown as a dashed line. This motion
results from a perturbation to an unstable circular motion. The parameter values
used are: (5) θ0 = −0.6, ω20 = −2.4625, ω30 = −0.2552; (6) θ0 = 0.25, ω20 = 3.5435,
ω30 = −0.0829; (7) θ0 = 0.5, ω20 = −3.5, ω30 = 0.1598 and (8) θ0 = 0.5, ω20 = 4,
ω30 = 0.1404, as taken from Figure 4.3. The horizontal dashed lines give the limits
on physically allowable values of θ = ±1.1071.

18.937 seconds in case (5) and after 13.4789 seconds in case (6). The two numbers

are comparable and are characteristic of this subset of points.

Examining the contours in Figure 4.3 further, we see that point (7) lies close to

the boundary of the stable (solid contour) and unstable (dashed contour) regions.

We can thus speculate that point (7) is “more stable” than either points (5) or (6).

Indeed, as Figure 4.6 (7) shows, it takes the vehicle much longer to impact the ground:
30

52.867 seconds. Interestingly, the vehicle almost crashes at around 15 second mark,

but is able to avoid collision at the very last moment.

Finally, point (8) represents yet another subset of instability conditions. In Figure

4.3 point (8) lies not only in the region where one of the eigenvalues of the character-

istic polynomial is positive, but it also fails to satisfy the fundamental constrains for

cornering motion in equation (4.5) by not lying on any of the contours. The result is

a crash into the ground after only 1.464 seconds of simulation. This makes point (8)

the “most unstable” of the four instability conditions discussed here.


(5) (6)
300 500

y y
0
0
−200 −100
−200 0 x 200 −200 0 x 600
(7) (8)
400 200

0
y y
0
−600 −100
−200 0 x 400 −100 0 x 150

Figure 4.7: Path of the center of mass of the vehicle during perturbed unstable circular
motion, shown on an x-y plane. The parameter values used are the same as in Figure
4.6. The initial location of the center of mass is at the origin, and the dot indicates
the first time the vehicle hits the ground.
31

Figure 4.7 shows the trajectory of the vehicle’s center of mass for cases (5)-(8).

The dot marks the instant the vehicle first impacts the ground. These trajectories

actually possess certain circular elements, due to the presence of non-zero roll angles

and yaw rates. The above notwithstanding, there is little useful information to be

gained from these plots.


32

Chapter 5

Conclusions

The work presented here has shown how Charles F. Taylor’s one-wheeled vehicle

can use gyroscopes to stabilize itself at rest and how steady circular motions can be

executed. From here, several other aspects of the machine can be studied.

For example, it is possible to look into the transient behaviors under acceleration

and braking. Such analysis would need to take into account the variations of the pitch

angle φ1 , as well as the change in the angular speed of the wheel φ̇2 . This will alter

the presented equations of motion (2.23), which currently assume φ1 = φ̇1 = φ̈1 = 0.

However, because of the way external forces are looked at, the inherent longitudinal

weight transfer is not going to affect the analysis - an added bonus.

Another aspect that can be studied is the behavior of the vehicle when the no-slip

condition fails, something not accounted for in the current model. Because Charles
33

F. Taylor has designed his machine as an all-terrain vehicle, wheel slip is possible

under certain circumstances. Since equation (2.6) would not hold in this case, to

establish the linear acceleration of the vehicle, we would need to know something

about the contact force between the ground and the wheel, such as maximum friction

coefficients attainable. This would provide us with a way to find the reaction force at

the instantaneous point of contact, P. From there on, the moments and the equations

of motion can be looked at in a way analogous to the one presented.

The analysis of steady circular motions discussed in Chapter 4 assumed that the

transition from the straight line to circular motion has already taken place. This

transition, however, would make for an interesting behavior to study. In this case

neither the rear gyroscope’s steering angle δ nor its rate of change can be taken as

zero. This changes the equations of motion (2.23) and will also force constraints on

the allowable vehicle speed as a function of the desired turn rate.

With the industrial interest in one-wheeled vehicles, we hope that the model

presented here can also be used to design other scaled versions of the prototype and

investigate the feasibility of control schemes.


34

Bibliography

[1] Appell, P.: Sur l’intégration des équations du mouvement d’un corps pesant de

révolution roulant par une arête circulaire sur un plan horizontal. Rendiconti del

Circolo Mathematico di Palermo 14, 1–6 (1900)

[2] Chaplygin, S.A.: On a motion of a heavy body of revolution on a horizontal

plane. Regul. Chaotic Dyn. 7(2), 119–130 (2002). English translation of the

Russian paper which was published in 1897

[3] Cushman, R., Hermans, J., Kemppainen, D.: The rolling disc. In: Nonlinear Dy-

namical Systems and Chaos (Groningen, 1995), Progress in Nonlinear Differential

Equations and their Applications, vol. 19, pp. 21–60. Birkhäuser, Basel (1996)

[4] Karapetyan, A.V., Kuleshov, A.S.: Steady motions of nonholonomic systems.

Regul. Chaotic Dyn. 7(1), 81–117 (2002)

[5] Kessler, P., O’Reilly, O.M.: The ringing of Euler’s disk. Regul. Chaotic Dyn. 7(1),

49–60 (2002)
BIBLIOGRAPHY 35

[6] O’Reilly, O.M.: The dynamics of rolling disks and sliding disks. Nonlinear Dy-

namics 10, 287–305 (1996)

[7] O’Reilly, O.M., Tongue, B.H.: Some comments on vehicle instability due to brake

lockup. Journal of Sound and Vibration 194, 760–765 (1996)

[8] Routh, E.J.: The Advanced Part of a Treatise on the Dynamics of a System of

Rigid Bodies, sixth edn. Macmillan, London (1905)

[9] Taylor, C.F.: Vehicle. U.S. Patent number 3145797 issued on August 25, 1964