Beruflich Dokumente
Kultur Dokumente
Harun Yetkin† , Simon Kalouche‡ , Michael Vernier† , Gregory Colvin‡ , Keith Redmill† and Umit Ozguner†
Abstract— There are two theoretical methods by which a these studies utilize dynamic stabilization where the bicycle
two wheeled vehicle oriented in tandem can be stabilized: is actively steered to induce leans that oppose the bicycle’s
dynamic stabilization and control moment gyroscope (CMG) instabilities while moving forward at a constant velocity;
stabilization. Dynamic stabilization utilizes tactical steering
techniques to trigger a lean in the vehicle in the intended however, this attempt fails at stabilizing a static bicycle, a dif-
direction for balancing, while CMG stabilization employs the ficult task for human riders, because the passive gyroscopic
reactive precession torque of a high speed flywheel about stabilizing effect produced by the angular momentum of the
an axis that will act to balance the vehicle. Of these two, bicycle’s wheels is absent. A few different approaches have
CMG stabilization offers greater advantages for static vehicles. been pursued to achieve static stabilization, the most notable
This paper proposes a first order sliding mode controller
(SMC) design to control the CMG and stabilize a bicycle of which include: adding a rotor mounted on the crossbar [4],
at zero-forward velocity. This study also compares the SMC mounting a pendulum to balance the tilting force [5], and
method to a PID controller to validate the advantages of the using the precession effects of a gyroscopic actuator [6], [7].
SMC controller for the highly non-linear system dynamics of
static stabilization. The result of two experimental setups are
presented and discussed. The first experimental platform is a B. Research Focus
single degree of freedom (DOF) inverted pendulum and the
second is a three DOF bicycle. In this paper, a high speed flywheel with a single DOF
gimbal is used to induce the torque that will counteract
I. INTRODUCTION the moment due to gravity applied on the bicycle when
A. History of Autonomous Stabilization it deviates or tilts from its semi-stable, vertical position.
By applying a gimbal torque to a spinning flywheel, a
Attempts at autonomous stabilization of inherently un-
simultaneous, amplified reactive torque is generated about
stable vehicles have dated back to the early 20th century.
an axis orthogonal to both the flywheel’s gimbal axis and
In 1905, Louis Brennan built a Gyroscopic Monorail that
spin axis. This controllable reactive torque can be oriented
utilized a CMG system controlled by passive actuation
to act about the axis that will balance an unstable bicycle.
of several mechanisms and mechanical sensors designed
to respond to the monorail’s tilt orientation. The monorail This paper focuses on the control dynamics of a static
successfully executed test runs carrying 50 passengers along bicycle and presents the experimental test results for two
a circular path [1]; however, due to the limited accuracy separate test platforms. The first experimental test platform
of sensors and robust controllers at that time it was more was a single DOF inverted pendulum frame. Its mobility was
practical to employ inherently stable two rail systems. In constrained to rotation about just one axis with one revolute
1909 and 1911 similar projects where endeavored by Scherl joint at its base. The second experimental test platform was
and Shilovsky [2]. Shilovsky’s gyrocar was a two wheeled a three DOF bicycle that can rotate about all three principle
vehicle with the wheels oriented in tandem. The gyrocar was axes; however, due to friction between the tires and the
capable of being manually stabilized via a clutch activated ground, rotation of the bicycle will be predominantly about
CMG system, requiring the human passenger to actuate the bicycle’s balancing axis. Rotation about the axis that
the clutch appropriately to gimbal the CMG’s flywheel. would generate a ’wheelie’ would never occur because the
Lack of sensors for accurate angular position, velocity, maximum reactive torque of the flywheel used was not large
and acceleration feedback limited the autonomy of these enough to lift the front or back wheel of the bicycle off
early attempts. Today, much progress has been made in the ground. For these reasons, the three DOF bicycle can
sensor technology, motor technology, control methods, and be modeled using the same single DOF dynamic equations
autonomous controllers making the use of a CMG stabilized of motion as the pendulum. Stabilizing the bicycle adds to
2-wheeled vehicle more practical. this study by testing a body whose geometric parameters are
In the advent of computer aided programs and micro- more similar to that of a vehicle, where the bicycle is not
controllers, more research has been conducted on the self- fully constrained to rotation about just one axis.
stabilization of bicycles. Bicycle dynamics and control have The outline of this paper is as follows. In Section 2, the
been investigated in detail by Sharp [3]. The majority of reference coordinate system is introduced, the equations of
motion are derived and all the assumptions used are listed.
† The Department of Electrical and Computer Engineering, The Ohio
Section 3 provides a brief introduction to sliding mode con-
State University, Columbus, OH 43210, USA yetkin.2@osu.edu, trollers, and elaborates on the stabilization controller design.
ozguner.1@ece.osu.edu
‡ The Department of Mechanical and Aerospace Engineering, The Ohio Section 4 presents the experimental results, and Section 5
State University, Columbus, OH 43210, USA kalouche.2@osu.edu offers concluding thoughts regarding the proposed controller.
θ TABLE I: Parameters for Static Bicycle
Parameter Symbol
Bicycle mass & height of c.m mb & hb
Gyro mass & height of c.m mg & hg
Flywheel mass & height of c.m mf & hf
Flywheel spin velocity Ω
α Gravity constant g
Bicycle Inertia Ibx
z
Gyro Inertia [Igx , Igy , Igz ]
Flywheel Inertia [I fx , I fy , I fz ]
Φ
Fig. 1: Free body diagram of the bicycle model
1
Ib + mb h2b θ̇ 2
II. B ICYCLE DYNAMICS TBike = (3)
2 x
1
mg h2g θ̇ 2 + Igx θ̇ 2 cos2 α + Igy θ̇ 2 sin2 α + Igz α̇ 2
A. Coordinates and Assumptions TGyro =
2
(4)
The coordinates of the bicycle are defined as:
1
m f h2f θ̇ 2 + I fx θ̇ 2 cos2 α + I fy Ω + θ̇ 2 sin2 α + I fz α̇ 2
• θ - Roll angle TFly =
2
• α - Gimbal angle (5)
• Φ - Flywheel spin direction
Assuming a non-conservative disturbance force (d(t)) act-
Fig. 1 shows a simplified sketch of the bicycle. The bicycle ing on the horizontal x-y plane, the equation of motions were
is composed of three rigid bodies: derived from Euler-Lagrange equation
d ∂L ∂L
• The bicycle frame with the wheels − = d(t)hb sin θ (6)
• The gyroscope without the flywheel dt ∂ θ̇ ∂θ
• The flywheel giving the acceleration of the tilt angle as
Here, the flywheel and the gyro are considered two different
rigid bodies since the flywheel also spins along the y-axis K1 gsinθ + 2I1 θ̇ α̇sinαcosα − I fy Ωcosα α̇ − d(t)hb sin θ
while the gyro does not. The full list of parameters used to θ̈ =
Ibx + K2 + Igx + I fx cos2 α + Igy + I fy sin2 α
derive the equations are given in Table I. (7)
In order to derive the equations of motion for the bicycle, where
the following assumptions are considered
• Tires have zero width K1 = mb hb + mg hg + m f h f (8)
• No slipping force applied to the bicycle’s tires
K2 = mb h2b + mg h2g + m f h2f (9)
• All three rigid bodies are taken as point masses at their
center of gravity I1 = Igx + I fx − Igy − I fy (10)
From (7), it can be seen that the precession effect created
B. Equations of Motion by the motion of the flywheel has a significant role in
the upright stabilization of the bicycle. Furthermore, it can
The nonlinear dynamics of these three rigid bodies were
be concluded that as Ω increases the flywheel will be
derived from conservation of energy by using Lagrange’s
more resistant to a change in orientation and thus will be
method. Define L such that
able to generate a larger reactive torque. This means that
stabilization can be achieved at a greater initial tilt angle, or
L = T −U (1) in the presence of larger disturbances.
III. C ONTROLLER D ESIGN
where T is the kinetic energy, and U is the potential energy of
the system. The kinetic and potential energy equations were A. Brief Introduction to Sliding Mode Control
derived for the bicycle frame, the gyro, and the flywheel. Variable structure systems (VSS) have been an attrac-
tive research topic for more than half a century. The first
approach of discontinuous control was in the form of a
T = TBicycle + TGyro + TFly U = (mg hg + mb hb + m f h f )gcosθ bang-bang controller [8]. After [9], the VSS and sliding
(2) mode control theory gained significant interest from both
researchers and engineers. Utkin and Young designed a
where sliding manifold to ensure optimality by using the pole
placement technique [10]. Su, Drakunov, and Ozguner [11]
studied the problem of constructing discontinuity surfaces
from a Lyapunov point of view which is also applicable on ṡ = c1 θ̇ + θ̈ (16)
nonlinear systems. h(α)ṡ = h(α)c1 θ̇ + (K1 g + fd hb ) sin θ
The chattering phenomena, the major drawback of the (17)
+ 2I1 θ̇ sin α cos α − I fy Ω cos α u
sliding mode control method, was also studied by many
researchers and several different approaches have been taken where h(α) represents the denominator of θ̈ . Then, by
to reduce the chattering. In [12], the switching function is selecting the control in the form of
replaced with a continuous approximation in the vicinity
of sliding manifold; however, the robustness of the sliding
mode remained an issue. Young and Ozguner [13] proposed u = −|k1 θ + k2 θ̇ |sign(s) (18)
two design methods, one was based on pole placement, and
the relation between s and ṡ is obtained as
the other was based on frequency-shaped quadratic optimal
control formulation. This proved to reduce the chattering
while preserving the insensitivity of the sliding mode against
h(α)ṡ = h(α)c1 θ̇ + (K1 g + fd hb ) sin θ (19)
the uncertainties.
The major advantages of sliding mode control are [14] + 2I1 θ̇ sin α cos α − I fy Ω cos α |k1 θ + k2 θ̇ |sign(s)
(20)
• An exact model is not required to apply control to
the system since sliding mode control is insensitive to After linearizing sin θ = θ which remains valid for |θ | <
unmodeled dynamics and disturbances 30 degree, and taking the upper bounds of the nonlinear
• The complexity of the feedback design is reduced
terms and the uncertainties in (20), the equation becomes
• It is a nonlinear control method
• It can be applied to a wide range of problems in
robotics, electric drives, and vehicle and motion control H ṡ = Hc1 θ̇ + (K1 g + f0 hb ) θ (21)
B. Design of Sliding Mode Controller + 2I1 θ̇ − I fy Ω |k1 θ + k2 θ̇ |sign(s) (22)
In order to design a controller to achieve the upright
where H is the constant value of h(α) at the upper bound
stabilization of the bicycle, a nonlinear state space model
of the nonlinear terms. Global asymptotic stability will be
of the system was needed. From the equation of motion in
ensured when the reachability condition (23) is satisfied.
(7), varying the gimbal angle (α) the precession torque is
induced, which stabilizes the bicycle. However, assigning the
gimbal angle as the control input causes the relative degree ṡ < 0 ← sign(s) > 0
of the system become greater than one [15]. Hence, the rate (23)
ṡ > 0 ← sign(s) < 0
of the gimbal angle (u = α̇) is chosen as the control input
of the system dynamics, and it is integrated over time before Hence, selecting the control gains as the following will
the actual control is applied to the motor. satisfy the reachability condition and guarantee stability.
The following nonlinear state space model summarizes the [14] gives further information about the asymptotic stability
system dynamics from a control point of view: of sliding mode controllers.
= f (θ , θ̇ , α) + g(u, θ̇ ) (11)
θ̈ K g+ f h Hc1
1 0 b
k1 > and k2 > (24)
y = [θ , α] (12) 2I1 θ̇0 − I fy Ω
2I1 θ̇0 − I fy Ω
where
where, θ̇0 represents the upper bound of the rate of the tilt
K1 g sin θ + fd hb sin θ angle.
f= (13)
Ibx + K2 + Igx + I fx cos2 α + Igy + I fy sin2 α
IV. E XPERIMENTAL R ESULTS
2I1 θ̇ sin α cos α − I fy Ω cos α
g= u (14) Experimental results were obtained on two different test
Ibx + K2 + Igx + I fx cos2 α + Igy + I fy sin2 α
platforms, an inverted pendulum and a bicycle shown in
Here, fd represents the disturbance and unmodeled system Fig. 2 and Fig. 3, respectively. Fig. 2 also shows a close
dynamics. up view of the gyro and flywheel.
To find the sliding mode controller gains, the following On the inverted pendulum setup, an encoder was used to
surface equation is considered: measure the tilt angle to obtain precise tilt angle measure-
ments. On the bicycle setup, an Inertial Measurement Unit
s = c1 θ + θ̇ (15)
(IMU) was utilized in lieu of an encoder. This results in
The derivative of the surface equation is taken. oscillatory balancing of the bicycle.
TABLE II: Parameter Values for Experimental Setups
Parameter IP Setup Value Bicycle Setup Value Unit
[mb , mg , m f ] [6.2, 2.23, 2.88] [14, 2.23, 2.88] [kg]
[hb , hg , h f ] [18, 12.6, 8.3] [23, 14, 9.7] [cm]
Ω 1000 1500 [rad/sec]
Ibx 2129 10200 [kg cm2 ]
[Igx , Igy ] [39, 60] [39, 60] [kg cm2 ]
[I fx , I fy ] [28.5, 30.2] [28.5 30.2] [kg cm2 ]
GYRO
FLYWHEEL
ENCODER
(a) (b)
−2
−3
0 10 20 30 40 50 60
time [sec]
Fig. 4: Experimental results showing the bicycle is stabilized.
5 0
−5 1
0
0.5
α̇ [rad/sec]
−10
α [deg]
θ [deg]
−5
0
−15
−10 −0.5
−20
Encoder data
−1
Stability region
−15
−25
−1.5
−20 −30 −2
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25
Fig. 5: Experimental results showing the pendulum is stabilized. (a) - Tilt angle measured with encoder (b) - Gimbal angle
(c) - Control input
0.5 0.2
0
0 0.1
−10
−0.5 0
α̇ [rad/sec]
−20
−1 −0.1
α [deg]
θ [deg]
−30
−1.5 −0.2
−40
−2 −0.3
−3.5 −0.6
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 80
Fig. 6: Experimental results for first measured disturbance. (a) - Tilt angle measured with encoder (b) - Gimbal angle (c) -
Control input
4 10
5
2
5
0
0
0
−2
θ [deg]
θ [deg]
θ [deg]
−4 −5 −5
−6
−10
−10
−8
−15
−10 −15
SMC result
−12 SMC result −20
PID controller result SMC result
PID controller result PID controller result
−20
−14
−25
−16
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Fig. 7: Comparison of SMC and PID for the same initial tilt angles. (a) - Initial tilt angle is −16 degrees (b) - Initial tilt
angle is −21 degrees (c) - Initial tilt angle is −26 degrees