This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Samir Bouabdallah

Autonomous Systems Laboratory

Swiss Federal Institute of Technology

Lausanne, Switzerland

Email: samir.bouabdallah@epﬂ.ch

Pierpaolo Murrieri

Interdepartmental Center ”E. Piaggio”

University of Pisa

Pisa, Italy

Email: p.murrieri@ing.unipi.it

Roland Siegwart

Autonomous Systems Laboratory

Swiss Federal Institute of Technology

Lausanne, Switzerland

Email: roland.siegwart@epﬂ.ch

Abstract—Recent progress in sensor technology, data pro-

cessing and integrated actuators has made the development of

miniature ﬂying robots fully possible. Micro VTOL

1

systems

represent a useful class of ﬂying robots because of their strong

capabilities for small-area monitoring and building exploration.

In this paper we describe the approach that our lab

2

has taken

to micro VTOL evolving towards full autonomy, and present the

mechanical design, dynamic modelling, sensing, and control of

our indoor VTOL autonomous robot OS4

3

.

I. INTRODUCTION

Autonomous ﬂying robots have gained enormous commer-

cial potential during the last years. Recent developments in

high density power storage, integrated miniature actuators and

MEMS

4

technology sensors have made autonomous minia-

turized ﬂying robots possible. This new situation has opened

the way to several, complex and highly important applications

for both military and civilian markets. Military applications

currently represent the lion’s part of the unmanned ﬂying

vehicle market, and this industrial sector is growing strongly.

Depending on the ﬂying principle and the propulsion mode,

one can classify aircraft vehicles in multiple categories as

shown in ﬁgure 1. In the motorized heavier-than-air category,

a new generation of MAV

5

with a wingspan less than 15cm

and less than 100 grams in mass has emerged. Generally

these MAVs are fully equipped with stabilization sensors

and miniature cameras. The Black Widow

6

MAV is a 15cm

span, ﬁxed-wing aircraft with an embedded color camera. It

ﬂies at 48 km/h with an endurance of 30 minutes, and a

maximum communication range of 2km. In the same category,

bird/Insect-like MAVs seem to be the perfect solution for fast

navigation in narrow spaces and perhaps the best approach to

miniaturization. The Micromechanical Flying Insects (MFI)

project at UC Berkeley [1] uses biomimetic principles to

develop a ﬂapping wing MAV. This project represents one

of the most promising endeavors towards autonomous MFIs.

In the motorized lighter-than-air category, Floreano’s group

at ASL

7

is applying biomimetic and evolutionary methods to

1

Vertical Take-Off and Landing

2

Autonomous Systems Lab

3

Omnidirectional Stationary Flying Outstretched Robot

4

Micro Electromechanical Systems

5

Micro Aerial Vehicle

6

www.aerovironment.com

7

Autonomous Systems Laboratory, EPFL, Switzerland

Fig. 1. Aircraft general classiﬁcation depending on the ﬂying principle and

the propulsion mode.

indoor ﬂying robots for autonomous vision-based navigation,

and has achieved noteworthy results with the ”Blimp” [2].

The state of the art in micro helicopters is not far behind,

and considerable efforts are being made, especially in control

and miniaturization. Mesicopter [3], an ambitious project

currently underway, is exploring the science of millimeter and

centimeter-size vehicles in spite of unfavorable scaling laws.

The project’s driving application is the deployment over large

areas or planets of a huge number of micro vehicles providing

atmospheric and meteorological data.

A. Helicopters vs Other Flying Principles

Compared with the other ﬂying principles discussed above,

VTOL systems have speciﬁc characteristics which allow the

execution of applications that would be difﬁcult or impossible

otherwise. Table I gives a non-exhaustive comparison between

the different ﬂying principles from the miniaturization point of

view. From this table, one can easily conclude that the VTOL

systems like helicopters or blimps have an unquestionable

advantage compared to the other concepts. This superiority

is thanks to their unique ability for vertical, stationary and

low speed ﬂight. The key advantage of blimps is the ”auto-

lift” and the simplicity of control which can be essential for

critical applications such as space exploration [4]. However,

VTOL vehicles with different conﬁgurations probably repre-

sent currently the most promising ﬂying concept seen in terms

of miniaturization.

II. THE OS4 PROJECT

The OS4 project, initiated at the Autonomous Systems

Laboratory (EPFL), focuses on micro VTOL vehicles evolving

towards a full autonomy in indoor environments. The long

TABLE I

FLYING PRINCIPLES COMPARISON FOCUSED ON ABILITY TO

MINIATURIZATION. (1=BAD, 3=GOOD)

Airplane Helicopter Bird Autogiro Blimp

Power cost 2 1 1 2 3

Control cost 2 1 1 2 3

Payload/volume 3 2 2 2 1

Maneuverability 2 3 3 2 1

DOF 1 3 3 2 1

Stationary ﬂight 1 3 2 1 3

Low speed ﬂy 1 3 2 2 3

Vulnerability 2 2 3 2 2

VTOL 1 3 2 1 3

Endurance 2 1 2 1 3

Miniaturization 2 3 3 2 1

Indoor usage 1 3 2 1 2

Total 20 28 26 20 26

term goal is to allow indoor navigation using various concepts.

The approach advocated for this project is to simultaneously

work on design and control. This original approach makes

it possible to simplify control by design changes, and vice

versa. A Quadrotor conﬁguration vector has been chosen as a

starting platform for the preliminary experiments. This vector

conﬁguration considerably simpliﬁes the vehicle design and

intrinsically reduces the gyroscopic effects. The project started

with the dynamic modelling and the development of a static

method for propulsion group evaluation and optimization. In

addition, a test bench has been designed to experiment and

tune the ﬁrst controllers.

A. Quadrotor Conﬁguration

The Quadrotor concept has been around for a long time. The

Breguet-Richet Quadrotor helicopter Gyroplane No.1 built in

1907 is reported to have lifted into ﬂight [5]. One can describe

the vehicle as having four propellers in cross conﬁguration.

The two pairs of propellers (1,3) and (2,4) as described in

ﬁgure 2, turn in opposite directions. By varying the rotor

speed, one can change the lift force and create motion. Thus,

increasing or decreasing the four propeller’s speeds together

generates vertical motion. Changing the 2 and 4 propeller’s

speed conversely produces roll rotation coupled with lateral

motion. Pitch rotation and the corresponding lateral motion,

result from 1 and 3 propeller’s speed conversely modiﬁed.

Yaw rotation is more subtle, as it results from the difference

in the counter-torque between each pair of propellers. In spite

of the four actuators, the Quadrotor still an under-actuated and

dynamically unstable system.

1) Advantages and Drawbacks: Although disadvantages,

such as space and energy requirements for the Quadrotor,

spring more quickly to mind than the system’s advantages,

this concept offers a better payload and is potentially simpler

to build and to control. This could be a decisive advantage.

Table II gives a rapid idea about Quadrotor’s advantages and

drawbacks.

III. QUADROTOR DYNAMIC MODELLING

The ﬁrst step before control development is an adequate dy-

namic system modelling [6]. Especially for lightweight ﬂying

1

2

3

4

Fig. 2. Quadrotor concept motion description, the arrow width is proportional

to propeller rotational speed.

TABLE II

QUADROTOR MAIN ADVANTAGES & DRAWBACKS.

Advantages Drawbacks

Rotor mechanics simpliﬁcation Weight augmentation

Payload augmentation High energy consumption

Gyroscopic effects reduction

systems, the dynamic model ideally includes the gyroscopic

effects resulting from both the rigid body rotation in space,

and the four propeller’s rotation. These aspects have been often

neglected in previous works. However, the main effects acting

on a helicopter [7] are described brieﬂy in table III.

TABLE III

MAIN PHYSICAL EFFECTS ACTING ON A HELICOPTER

Effect Source Formulation

Aerodynamic effects - Propeller rotation

- Blades ﬂapping CΩ

2

Inertial counter torques - Change in propeller

rotation speed J

˙

Ω

Gravity effect - Center of mass position

Gyroscopic effects - Change in orientation

of the rigid body Iθψ

- Change in orientation JΩθ, φ

of the propeller plane

Friction - All helicopter motion C

˙

φ,

˙

θ,

˙

ψ

Let us consider earth ﬁxed frame E and body ﬁxed frame

B, as seen in ﬁgure 3. The center of mass and the body

ﬁxed frame origin are assumed to coincide. Using Euler angles

parametrization, the airframe orientation in space is given by

a rotation R from B to E, where R ∈ SO3 is the rotation

matrix. The dynamics of a rigid body under external forces

applied to the center of mass and expressed in the body ﬁxed

frame as shown in [6] and [8] are in Newton-Euler formalism:

¸

mI

3x3

0

0 I

¸

˙

V

˙ ω

+

¸

ω ×mV

ω ×Iω

=

¸

F

τ

(1)

Where I ∈

(3x3)

the inertia matrix, V the body linear

speed vector and ω the body angular speed.

In the frame system ﬁgure 3, the equations of motion for

the helicopter can be written as [11]:

Fig. 3. Quadrotor conﬁguration, frame system with a body ﬁxed frame B

and the inertial frame E.

˙

ζ = ν

m˙ ν = RF

b

˙

R = Rˆ ω

J ˙ ω = −ω ×Jω + τ

a

(2)

The ﬁrst-level approximate model (3) of the Quadrotor can

be rewritten as:

˙

ζ = ν

˙ ν = −ge

3

+ R

e3

(

b

m

¸

Ω

2

i

)

˙

R = Rˆ ω

I ˙ ω = −ω ×Iω −

¸

J

r

(ω ×e

3

)Ω

i

+ τ

a

(3)

where :

Symbol deﬁnition

ζ position vector

R rotation matrix

ˆ ω skew symmetric matrix

φ roll angle

θ pitch angle

ψ yaw angle

Ω rotor speed

I

x,y,z

body inertia

J

r

rotor inertia

τ

a

torque on airframe body

b thrust factor

d drag factor

l lever

The torque applied on the vehicle’s body along an axis is

the difference between the torque generated by each propeller

on the other axis.

τ

a

=

¸

lb(Ω

2

4

−Ω

2

2

)

lb(Ω

2

3

−Ω

2

1

)

d(Ω

2

2

+ Ω

2

4

−Ω

2

1

−Ω

2

3

)

¸

(4)

The full Quadrotor dynamic model with the x,y,z motions

as a consequence of a pitch or roll rotation is:

**¨ x = (cos φsin θ cos ψ + sin φsinψ)
**

1

m

U

1

¨ y = (cos φsin θ sinψ −sin φcos ψ)

1

m

U

1

¨ z = −g + (cos φcos θ)

1

m

U

1

¨

φ =

˙

θ

˙

ψ(

Iy−Iz

Ix

) −

Jr

Ix

˙

θΩ +

l

Ix

U

2

¨

θ =

˙

φ

˙

ψ(

Iz−Ix

Iy

) +

Jr

Iy

˙

φΩ +

l

Iy

U

3

¨

ψ =

˙

φ

˙

θ(

Ix−Iy

Iz

) +

1

Iz

U

4

(5)

Then, the system’s inputs are posed U

1

, U

2

, U

3

, U

4

and Ω

a disturbance, obtaining:

U

1

= b(Ω

2

1

+ Ω

2

2

+ Ω

2

3

+ Ω

2

4

)

U

2

= b(Ω

2

4

−Ω

2

2

)

U

3

= b(Ω

2

3

−Ω

2

1

)

U

4

= d(Ω

2

2

+ Ω

2

4

−Ω

2

1

−Ω

2

3

)

Ω = Ω

2

+ Ω

4

−Ω

1

−Ω

3

(6)

A. Rotor Dynamics

The rotors are driven by DC-motors with the well known

equations [10]:

L

di

dt

= u −Ri −k

e

ω

m

J

dωm

dt

= τ

m

−τ

d

(7)

As we use a small motor with a very low inductance, the

second order DC-motor dynamics may be approximated by:

J

dωm

dt

= −

k

2

m

R

ω

m

−τ

d

+

km

R

u

(8)

By introducing the propeller and the gearbox models, the

equation (8) may be rewritten:

˙ ω

m

= −

1

τ

ω

m

−

d

ηr

3

Jt

ω

2

m

+

1

kmτ

u

with :

1

τ

=

k

2

m

RJt

(9)

The equation (9) can be linearized around an operation point

˙ w

0

to the form ˙ w

m

= −Aw

m

+ Bu + C with:

A =

1

τ

+

2dw0

ηr

3

Jt

, B =

1

kmτ

, C =

dω

2

0

ηr

3

Jt

(10)

Symbol Deﬁnition

u motor input

k

e

back EMF constant

k

m

torque constant

ω

m

motor angular speed

τ

m

motor torque

τ

d

motor load

τ motor time-constant

R motor internal resistance

r gear box reduction ratio

η gear box efﬁciency

J

t

total inertia

Fig. 4. OS4 test bench for stabilization strategies testing, 3DOF are locked,

the cross is made with carbon rods and the ﬂying system weight is about

240g. 1)RS232 to I2C translator, 2)Motor modules, 3)3D captured universal

joint, 4)Micro IMU, 5)Propulsion group.

Fig. 5. OS4 test bench block diagram

IV. OS4 TEST BENCH DESIGN

The development of a control system for a ﬂying robot

requires the development of an adequate test bench at least for

the preliminary experiments. This can help lock some number

of degrees of freedom in order to reduce control complexity

and avoid system damage.

From a PC and through a standard RS232 port, one can send

orders to the test bench. The RS232 to I2C module translates

the serial signals to the I2C bus motor modules. These modules

integers a P.I.D regulator on a PIC16F876 microcontroller and

are capable of open or closed loop operation in position, speed

or torque control. The MT9-B

8

IMU

9

estimates with a kalman

ﬁlter the 3D orientation data and gives the calibrated data of

acceleration and angular velocity. It weights about 33g and

communicates at 115kbps. The captured motion from the 3D

universal joint

10

can be decoded to extract absolute orientation

information, thanks to the micro optical encoders in each axis.

The vehicle is thus lightweight, about 235g for all the ﬂying

system. The OS4 test bench has 4 propulsion group, each

composed of a 29g motor

11

including magnetic encoders, a 6g

gear box and a 6g propeller. To design the propulsion group,

a test, evaluation and comparison method was developed.

8

www.xsens.com

9

Inertial Measurement Unit

10

www.forcedimension.com

11

1724 motor from: www.minimotor.ch

A. Propulsion group evaluation and design procedure

Finding the highest thrust to weight ratio is one of the most

important challenges in micro VTOL design. Our approach

is ﬁrstly to specify the application requirements in terms of

thrust, energy and overload allowed. Secondly is to build a

propeller and motor data bank and then ﬁnd the best combi-

nation. Finally we compare the results to the requirements.

For the propeller data-bank, we use a speciﬁc test bench

to extract thrust and drag coefﬁcients through experiments

where we measure tension, current, thrust and rotational speed.

Designing a ﬂying robot is an iterative process and one has to

ﬁx starting conditions. For our development, we have chosen

to start from the determination of the vehicle’s approximate

size which allows the propeller selection from the data base

according to its size. Using the evaluation tool, one can easily

select the appropriate motor. Finally, we use the well known

motor equations to determine the optimal reduction ratio for

our propulsion group.

V. CONTROL OF THE VTOL SYSTEM

The model (5), developed in the previous sections, can be

rewritten in a state-space form

˙

X = f(X, U) by introducing

X = (x

1

...x

12

)

T

∈

12

as state vector of the system as

follows:

x

1

= x

x

2

= ˙ x

1

= ˙ x

x

3

= y

x

4

= ˙ x

3

= ˙ y

x

5

= z

x

6

= ˙ x

5

= ˙ z

x

7

= φ

x

8

= ˙ x

7

=

˙

φ

x

9

= θ

x

10

= ˙ x

9

=

˙

θ

x

11

= ψ

x

12

= ˙ x

11

=

˙

ψ

(11)

From (11) and (5) we obtain:

f(X, U) =

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

¸

x

2

(cos x

7

sin x

9

cos x

11

+ sinx

7

sin x

11

)

U1

m

x

4

(cos x

7

sin x

9

sinx

11

−sinx

7

cos x

11

)

U1

m

x

6

−g + (cos x

7

cos x

9

)

1

m

U

1

x

8

x

12

x

10

Iy−Iz

Ix

−

JR

Ix

x

10

Ω +

l

Ix

U

2

x

10

x

12

x

8

Iz−Ix

Iy

+

JR

Iy

x

8

Ω +

l

Iy

U

3

x

12

x

10

x

8

Ix−Iy

Iz

+

l

Iz

U

4

¸

(12)

It is worthwhile to note inside the dynamic of the latter

system how the angles and their time derivatives do not

depend on translation components; on the other hand the

translations depend on angle (and not on angular velocities).

We can ideally imagine the overall system described by (12)

as constituted by two subsystems, the angular rotations and

the linear translations, see ﬁgure 6. The angular rotations

subsystem has as state the restriction X

α

of X to the last

6 components which regard the roll, pitch, yaw and their time

derivative. The dynamics of these variables are described by

f

α

(X, U) which corresponds to the last 6 components of the

mapping (12). Note that the mapping f

α

(X, U) is function

only of X

α

and of (U

2

, U

3

, U

4

)

T

, and does not depend on

translation components. On the other hand, the translations

subsystem (with state X

∆

) regards the ﬁrst 6 element of the

state X, which are the x, y, z and their time derivative; in

this case too the dynamics are described by the ﬁrst 6 rows

f

∆

(X, U) of the mapping (12). Conversely to the previous

case, the translations subsystem mapping f

∆

(X, U) is not

independent of the angle variables but depends only on roll,

pitch and yaw and not on their time derivative.

Fig. 6. Connection of the two ideal subsystems of the overall dynamical

system described by mapping (12). From the angular rotations subsystem the

roll, pitch and yaw are obtained and become with U

1

inputs for the following

translation subsystem.

A. Control of the Angular Rotations Subsystem

Due to its complete independence from the other subsystem,

it is interesting to consider ﬁrst the control of the angular ro-

tations subsystem. In particular, in this subsection we consider

the stabilization of the OS4 angles in a particular conﬁguration

X

d

α

= (x

d

7

, 0, x

d

9

, 0, x

d

11

, 0)

T

.

Let us consider the Lyapunov Function V (X

α

) which is C

1

and positive deﬁned around the desired position X

d

α

.

(x

7

−x

d

7

)

2

+ x

2

8

+ (x

9

−x

d

9

)

2

+ x

2

10

+ (x

11

−x

d

11

)

2

+ x

2

12

2

(13)

The time derivative of (13),

˙

V = (∇V )

T

f

α

, in the case of a

perfect cross VTOL (I

x

= I

y

) is drastically reduced to:

˙

V = (x

7

−x

d

7

)x

8

+ x

8

l

Ix

U

2

+ (x

9

−x

d

9

)x

10

+ x

10

l

Iy

U

3

+

+(x

11

−x

d

11

)x

12

+ x

12

l

Iz

U

4

(14)

Equation in which does not appear the perturbation term with

Ω. By simply choosing:

U

2

= −

Ix

l

(x

7

−x

d

7

) −k

1

x

8

U

3

= −

Iy

l

(x

9

−x

d

9

) −k

2

x

10

U

4

= −I

z

(x

11

−x

d

11

) −k

3

x

12

,

(15)

with k

1

, k

2

and k

3

positive constants, we obtain for (14):

˙

V = −x

2

8

lk1

Ix

−x

2

10

lk2

Iy

−x

2

12

k3

Iz

, (16)

which is only negative semi-deﬁned. By Lyapunov theorem

[12] is now ensured the simple stability for equilibrium. By

Lasalle invariance theorem we can ensure also that starting

from a level curve of the Lyapunov function deﬁned in (13)

where V (X

α

) is constant, the state evolution is constrained

inside the region bounded by the level curve. This is very

useful when trying to avoid particular conﬁguration; it is

simply necessary to start with a level curve not containing

these points and apply the previous deﬁned controls. We can

also ensure the asymptotic stability by applying the Lasalle

theorem because the maximum invariance set of (angular

rotations) subsystem under control (15) contained in the set

S = {X

S

α

∈

6

:

˙

V |

X

S

α

= 0} is restricted only to the

equilibrium point.

By the latter consideration we can ensure an asymptotical

stability starting from a point in a set around the equilib-

rium. To ensure the global stability it is sufﬁcient that the

lim

|Xα|→∞

V (X

α

) = ∞, which is our case.

B. Height Controller

Let us consider the simple task for the VTOL to hover

at a particular height z = z

d

. The dynamic of the height is

described by lines 5 and 6 of system (12), that is:

˙ x

5

˙ x

6

=

x

6

−g + cos x

7

cos x

9

U1

m

(17)

Using the considerations in the previous paragraph V-A, we

ensure that starting from an initial condition where V (X

α

) <

π

2

, the angles and their velocities are constrained in this

hypersphere of

6

. In this case cos x

7

cos x

9

= 0 during

all the trajectories of the system under previous control law.

If the latter condition is satisﬁed we can linearize system

(17) by simply compensating the weight force by U

1

=

mg

cos x7 cos x9

+

m

ˆ

U1

cos x7 cos x9

, where

ˆ

U

1

is an additional term. By

the latter law (17) becomes:

˙ x

5

˙ x

6

=

x

6

ˆ

U

1

, (18)

By a simple state-space linear stabilization law

ˆ

U

1

= k

4

x

5

+

k

5

x

6

we can stabilize the height by placing the poles of the

subsystem in any position in the complex left half plane.

VI. SIMULATIONS

Before implementation on the real system, we performed

several simulations on Matlab. The controller’s task was to

stabilize the height while compensating the initial error on

the roll, pitch and yaw angles. The real system suffers from

undesired but unavoidable delays and actuator saturation. The

delays are mainly due to RS232 communications and the

actuator time constant. To emulate this lacks, two Simulink

discrete-step delay blocks have been introduced in the feed-

back loop and on the actuators. Saturation level depends on

the chosen actuators. The motors work in our application

Fig. 7. Simulation: the system has to maintain the height of 2 meters although

the noise on the actuators.

with a maximum angular velocity of 600 rad/sec; a saturation

block has been placed between the controller and the delay.

Finally, the overall system has been simulated at 30Hz using

a discrete time solver in order to model the behavior of the

digital controller. In the simulation, see ﬁgure 7, the task is

to hover although an added normal gaussian noise of variance

4 rad/sec on each angular velocity. The height is taken with

an added zero mean error.

VII. EXPERIMENTS

In order to validate the control law developed in the previous

section, we implemented the controller and we performed

several experiments on the real system. The task was to control

the vehicle orientation thus, the Roll, the Pitch and the Yaw

angles was controlled, see ﬁgure 8, while the height was ﬁxed

by the test bench.

In spite of the test bench limitations in term of delays and

errors introduced by the tethering system, the experimental

results obtained show that the proposed controller works well

especially for the yaw angle.

VIII. CONCLUSION AND FUTURE WORK

In this paper, we presented a survey of existing ﬂying micro-

vehicles and made a comparison to micro VTOLs in terms of

miniaturization. We introduced the OS4 project and discussed

the undergoing developments of ﬂying robots at ASL. This

includes dynamic modelling, vehicle design optimization and

control. As it can be seen from the experimental plots, the

controller introduced prove the ability to control the orien-

tation angles. Our next goal is to enhance the control with

position controller and to develop a fully autonomous vehicle.

The positive results obtained in this development towards

autonomous micro-VTOL reinforce our conviction that these

systems have potential as candidates for the miniaturized ﬂying

micro-vehicles emergence.

0 500 1000 1500

−10

−8

−6

−4

−2

0

2

4

6

8

10

ROLL

steps

d

e

g

r

e

e

s

0 500 1000 1500

−10

−8

−6

−4

−2

0

2

4

6

8

10

PITCH

steps

d

e

g

r

e

e

s

0 500 1000 1500

−10

−8

−6

−4

−2

0

2

4

6

8

10

YAW

steps

d

e

g

r

e

e

s

Fig. 8. Experiments: the controller has to stabilize the system by maintaining

the roll, pitch and yaw angels to zero.

ACKNOWLEDGMENT

The authors would like to thank Jean-Christophe Zufferey

for fruitful discussions and advices on ﬂying robots, Georges

Perrenoud for the realization of the test bench mechanical

components and all the students who worked or are working

on this project.

REFERENCES

[1] X. Deng, L. Schenato and S.S Sastry, Attitude Control for a Microme-

chanical Flying Insect Including Thorax and Sensor Models ICRA

2003, Teipei, Taiwan 2003.

[2] J.C Zufferey, D. Floreano Evolving Vision-Based Flying Robots. Pro-

ceedings of the 2nd International Workshop on Biologically Motivated

Computer Vision, LNCS 2525, pp. 592-600, Berlin, Springer-Verlag,

2002.

[3] I. Kroo, F.B. Prinz, The Mesicopter: A Meso-Scale Flight Vehicle.

http://aero.stanford.edu/mesicopter/.

[4] A. Elfes, S.S. Bueno and al, Robotic Airship for Exploration of planetary

Bodies with an Atmosphere Autonomy Challenges. Autonomous Robots

Journal: Kluwer Academic Publishers, 2003.

[5] J.G. Leishman , The Breguet-Richet Quad-Rotor Helicopter of 1907.

http://www.enae.umd.edu/AGRC/Aero/Breguet.pdf.

[6] S. Sastry, A mathematical introduction to robotic manipulation . Boca

Raton, FL, 1994.

[7] P. Mllhaupt, Analysis and Control of Underactuated Mechanical

Nonminimum-phase Systems. Phd thesis, Department of Mechanical

Engineering, EPFL, 1999.

[8] A. Chriette, Contribution ` a la commande et ` a la mod´ elisation des

h´ elicopt` eres : Asservissement visuel et commande adaptative.. Phd

Thesis, 2001.

[9] P. Pounds, R.Mahony, Design of a Four-Rotor Aerial Robot. Australasian

Conference on Robotics and Automation, Auckland, Australia, 2002.

[10] E. Jucker, Equations fondamentales des micromoteurs courant continu

avec rotor sans fer. Bulletin technique Portescap, La Chaud-de-Fonds,

1974.

[11] R. Olfati-Saber, Nonlinear Control of Underactuated Mechanical Sys-

tems with Application to Robotics and Aerospace Vehicles. Phd thesis,

Department of Electrical Engineering and Computer Science, MIT, 2001.

[12] S. Arimoto, Control Theory of Non-linear Mechanical Systems. Oxford

Science Publications, 1996.

φ ˙ ˙ ˙ C φ. Changing the 2 and 4 propeller’s speed conversely produces roll rotation coupled with lateral motion. turn in opposite directions. Table II gives a rapid idea about Quadrotor’s advantages and drawbacks. the main effects acting on a helicopter [7] are described brieﬂy in table III. as seen in ﬁgure 3. This could be a decisive advantage. 1) Advantages and Drawbacks: Although disadvantages. Pitch rotation and the corresponding lateral motion.Change in orientation of the propeller plane . TABLE II Q UADROTOR MAIN A DVANTAGES & D RAWBACKS . θ. TABLE III M AIN PHYSICAL EFFECTS ACTING ON A HELICOPTER Effect Aerodynamic effects Inertial counter torques Gravity effect Gyroscopic effects Source . The approach advocated for this project is to simultaneously work on design and control. These aspects have been often neglected in previous works. 2. as it results from the difference in the counter-torque between each pair of propellers.Blades ﬂapping . Thus. and vice versa. By varying the rotor speed. increasing or decreasing the four propeller’s speeds together generates vertical motion. the airframe orientation in space is given by a rotation R from B to E.Propeller rotation . The center of mass and the body ﬁxed frame origin are assumed to coincide. The two pairs of propellers (1. and the four propeller’s rotation. Using Euler angles parametrization. Quadrotor Conﬁguration The Quadrotor concept has been around for a long time.4) as described in ﬁgure 2.3) and (2. the equations of motion for the helicopter can be written as [11]: .All helicopter motion Formulation CΩ2 ˙ JΩ Iθψ JΩθ. III. In spite of the four actuators. The dynamics of a rigid body under external forces applied to the center of mass and expressed in the body ﬁxed frame as shown in [6] and [8] are in Newton-Euler formalism: mI3x3 0 0 I ˙ V ω ˙ + ω × mV ω × Iω = F τ (1) Where I ∈ (3x3) the inertia matrix. result from 1 and 3 propeller’s speed conversely modiﬁed. a test bench has been designed to experiment and tune the ﬁrst controllers. Advantages Rotor mechanics simpliﬁcation Payload augmentation Gyroscopic effects reduction Drawbacks Weight augmentation High energy consumption term goal is to allow indoor navigation using various concepts. Yaw rotation is more subtle. where R ∈ SO3 is the rotation matrix.TABLE I F LYING PRINCIPLES COMPARISON FOCUSED ON ABILITY TO MINIATURIZATION . However. One can describe the vehicle as having four propellers in cross conﬁguration. A Quadrotor conﬁguration vector has been chosen as a starting platform for the preliminary experiments. In addition. the arrow width is proportional to propeller rotational speed. the dynamic model ideally includes the gyroscopic effects resulting from both the rigid body rotation in space. This original approach makes it possible to simplify control by design changes.1 built in 1907 is reported to have lifted into ﬂight [5]. 3=G OOD ) Power cost Control cost Payload/volume Maneuverability DOF Stationary ﬂight Low speed ﬂy Vulnerability VTOL Endurance Miniaturization Indoor usage Total Airplane 2 2 3 2 1 1 1 2 1 2 2 1 20 Helicopter 1 1 2 3 3 3 3 2 3 1 3 3 28 Bird 1 1 2 3 3 2 2 3 2 2 3 2 26 Autogiro 2 2 2 2 2 1 2 2 1 1 2 1 20 Blimp 3 3 1 1 1 3 3 2 3 3 1 2 26 1 4 2 3 Fig. ψ Friction Let us consider earth ﬁxed frame E and body ﬁxed frame B.Center of mass position .Change in propeller rotation speed . In the frame system ﬁgure 3. The Breguet-Richet Quadrotor helicopter Gyroplane No. Quadrotor concept motion description. (1=BAD . QUADROTOR DYNAMIC MODELLING The ﬁrst step before control development is an adequate dynamic system modelling [6]. spring more quickly to mind than the system’s advantages. this concept offers a better payload and is potentially simpler to build and to control. The project started with the dynamic modelling and the development of a static method for propulsion group evaluation and optimization. the Quadrotor still an under-actuated and dynamically unstable system. Especially for lightweight ﬂying systems. V the body linear speed vector and ω the body angular speed. one can change the lift force and create motion. such as space and energy requirements for the Quadrotor. This vector conﬁguration considerably simpliﬁes the vehicle design and intrinsically reduces the gyroscopic effects.Change in orientation of the rigid body . A.

the second order DC-motor dynamics may be approximated by: − τd + km R u Symbol ζ R ω ˆ φ θ ψ Ω Ix. 3. C= 2 dω0 ηr 3 Jt (10) The torque applied on the vehicle’s body along an axis is the difference between the torque generated by each propeller on the other axis. the equation (8) may be rewritten: 1 1 2 ωm = − τ ωm − ηrdJt ωm + km τ u 3 ˙ with : (9) 2 1 = km τ RJt + 2dw0 ηr 3 Jt . U3 . obtaining: U1 = b(Ω2 + Ω2 + Ω2 + Ω2 ) 1 2 3 4 U2 = b(Ω2 − Ω2 ) 2 4 U3 = b(Ω2 − Ω2 ) (6) 3 1 U4 = d(Ω2 + Ω2 − Ω2 − Ω2 ) 2 4 1 3 Ω = Ω 2 + Ω4 − Ω1 − Ω3 A. the system’s inputs are posed U1 .y. B= 1 km τ . U2 . frame system with a body ﬁxed frame B and the inertial frame E.z motions as a consequence of a pitch or roll rotation is: Symbol u ke km ωm τm τd τ R r η Jt Deﬁnition motor input back EMF constant torque constant motor angular speed motor torque motor load motor time-constant motor internal resistance gear box reduction ratio gear box efﬁciency total inertia . U4 and Ω a disturbance.y.z Jr τa b d l deﬁnition position vector rotation matrix skew symmetric matrix roll angle pitch angle yaw angle rotor speed body inertia rotor inertia torque on airframe body thrust factor drag factor lever (8) The equation (9) can be linearized around an operation point w0 to the form wm = −Awm + Bu + C with: ˙ ˙ A= 1 τ By introducing the propeller and the gearbox models. 1 x = (cos φ sin θ cos ψ + sin φ sin ψ) m U1 ¨ y = (cos φ sin θ sin ψ − sin φ cos ψ) 1 U ¨ m 1 z = −g + (cos φ cos θ) 1 U ¨ m 1 Fig. lb(Ω2 − Ω2 ) 4 2 2 lb(Ω3 − Ω2 ) τa = 1 2 2 d(Ω2 + Ω4 − Ω2 − Ω2 ) 1 3 (4) The full Quadrotor dynamic model with the x. Rotor Dynamics The rotors are driven by DC-motors with the well known equations [10]: di L dt = u − Ri − ke ωm (7) dωm J dt = τm − τd J dωm = − dt 2 km R ωm φ = θψ( Iy −Iz ) − ¨ ˙ ˙ Ix ¨ ˙ ˙ Iz −Ix θ = φψ( Iy ) + ¨ ˙ ˙ I −I ψ = φθ( xIz y ) + Jr ˙ Ix θΩ + l Ix U 2 (5) Jr ˙ Iy φΩ 1 Iz U 4 + l Iy U 3 (3) where : As we use a small motor with a very low inductance. The ﬁrst-level approximate model (3) of the Quadrotor can be rewritten as: ˙ ζ=ν b ν = −ge3 + Re3 ( m ˙ Ω2 ) i ˙ = Rˆ R ω I ω = −ω × Iω − Jr (ω × e3 )Ωi + τa ˙ ˙ ζ=ν mν = RFb ˙ ˙ R = Rˆ ω J ω = −ω × Jω + τa ˙ (2) Then. Quadrotor conﬁguration.

Secondly is to build a propeller and motor data bank and then ﬁnd the best combination. U ) = − JR x10 Ω + Ix x10 + JR x 8 Ω + Iy x12 + l Iz U 4 l Ix U 2 l Iy U 3 x10 x8 Ix −Iy Iz .minimotor. 2)Motor modules. The captured motion from the 3D universal joint10 can be decoded to extract absolute orientation information. a 6g gear box and a 6g propeller. To design the propulsion group. Using the evaluation tool. developed in the previous sections.x12 ) ∈ as state vector of the system as follows: x1 = x x2 = x 1 = x ˙ ˙ x3 = y x4 = x 3 = y ˙ ˙ x5 = z x6 = x 5 = z ˙ ˙ (11) x7 = φ ˙ x8 = x 7 = φ ˙ x9 = θ ˙ x10 = x9 = θ ˙ x11 = ψ ˙ x12 = x11 = ψ ˙ From (11) and (5) we obtain: x2 (cos x7 sin x9 cos x11 + sin x7 sin x11 ) U1 m x4 (cos x7 sin x9 sin x11 − sin x7 cos x11 ) U1 m x6 1 −g + (cos x7 cos x9 ) m U1 x8 x12 x10 x12 x8 Iy −Iz Ix Iz −Ix Iy Fig. the cross is made with carbon rods and the ﬂying system weight is about 240g.A. 4)Micro IMU. thanks to the micro optical encoders in each axis. each composed of a 29g motor11 including magnetic encoders.com 9 Inertial 11 1724 10 www. It weights about 33g and communicates at 115kbps. thrust and rotational speed... For the propeller data-bank. Finally.forcedimension. 4. 5)Propulsion group. Our approach is ﬁrstly to specify the application requirements in terms of thrust.I. one can send orders to the test bench. speed or torque control. we use the well known motor equations to determine the optimal reduction ratio for our propulsion group. we have chosen to start from the determination of the vehicle’s approximate size which allows the propeller selection from the data base according to its size. CONTROL OF THE VTOL SYSTEM The model (5). OS4 test bench for stabilization strategies testing. 1)RS232 to I2C translator. Fig.D regulator on a PIC16F876 microcontroller and are capable of open or closed loop operation in position. current. energy and overload allowed. 8 www. V. 3)3D captured universal joint. 5. The RS232 to I2C module translates the serial signals to the I2C bus motor modules. The vehicle is thus lightweight. For our development.ch (12) It is worthwhile to note inside the dynamic of the latter system how the angles and their time derivatives do not depend on translation components. Designing a ﬂying robot is an iterative process and one has to ﬁx starting conditions. OS4 TEST BENCH DESIGN The development of a control system for a ﬂying robot requires the development of an adequate test bench at least for the preliminary experiments. evaluation and comparison method was developed. 3DOF are locked. on the other hand the f (X. OS4 test bench block diagram IV. can be ˙ rewritten in a state-space form X = f (X. U ) by introducing T 12 X = (x1 . about 235g for all the ﬂying system. These modules integers a P. From a PC and through a standard RS232 port. Propulsion group evaluation and design procedure Finding the highest thrust to weight ratio is one of the most important challenges in micro VTOL design.xsens.com Measurement Unit motor from: www. The OS4 test bench has 4 propulsion group. Finally we compare the results to the requirements. we use a speciﬁc test bench to extract thrust and drag coefﬁcients through experiments where we measure tension. a test. one can easily select the appropriate motor. This can help lock some number of degrees of freedom in order to reduce control complexity and avoid system damage. The MT9-B8 IMU9 estimates with a kalman ﬁlter the 3D orientation data and gives the calibrated data of acceleration and angular velocity.

This is very useful when trying to avoid particular conﬁguration. In particular. yaw and their time derivative. see ﬁgure 6. The real system suffers from undesired but unavoidable delays and actuator saturation. SIMULATIONS Before implementation on the real system. U ) is function only of Xα and of (U2 . we performed several simulations on Matlab. 7 9 11 Let us consider the Lyapunov Function V (Xα ) which is C 1 d and positive deﬁned around the desired position Xα . it is interesting to consider ﬁrst the control of the angular rotations subsystem. If the latter condition is satisﬁed we can linearize system (17) by simply compensating the weight force by U1 = ˆ mg m U1 ˆ cos x7 cos x9 + cos x7 cos x9 . the angles and their velocities are constrained in this hypersphere of 6 . V = ( V )T fα . To ensure the global stability it is sufﬁcient that the lim|Xα |→∞ V (Xα ) = ∞. the translations subsystem mapping f∆ (X. y. in this case too the dynamics are described by the ﬁrst 6 rows f∆ (X.translations depend on angle (and not on angular velocities). 0. 6. x5 ˙ x6 ˙ = x6 −g + cos x7 cos x9 U1 m (17) A. The delays are mainly due to RS232 communications and the actuator time constant. 0)T . B. pitch and yaw angles. 8 I 10 I 12 I (16) which is only negative semi-deﬁned. By Lasalle invariance theorem we can ensure also that starting from a level curve of the Lyapunov function deﬁned in (13) where V (Xα ) is constant. it is simply necessary to start with a level curve not containing these points and apply the previous deﬁned controls. the state evolution is constrained inside the region bounded by the level curve. The motors work in our application . that is: Fig. To emulate this lacks. we ensure that starting from an initial condition where V (Xα ) < π 2 . with k1 . the angular rotations and the linear translations. U4 )T . U ) of the mapping (12). The controller’s task was to stabilize the height while compensating the initial error on the roll. Saturation level depends on the chosen actuators. We can also ensure the asymptotic stability by applying the Lasalle theorem because the maximum invariance set of (angular rotations) subsystem under control (15) contained in the set S ˙ S S = {Xα ∈ 6 : V |Xα = 0} is restricted only to the equilibrium point. U ) is not independent of the angle variables but depends only on roll. 11 (15) 2 ˆ By a simple state-space linear stabilization law U1 = k4 x5 + k5 x6 we can stabilize the height by placing the poles of the subsystem in any position in the complex left half plane. which is our case. where U1 is an additional term. In this case cos x7 cos x9 = 0 during all the trajectories of the system under previous control law. On the other hand. Height Controller Let us consider the simple task for the VTOL to hover at a particular height z = z d . U ) which corresponds to the last 6 components of the mapping (12). Control of the Angular Rotations Subsystem Due to its complete independence from the other subsystem. we obtain for (14): 1 3 ˙ V = −x2 lkx − x2 lky2 − x2 kz . xd . By the latter law (17) becomes: x5 ˙ x6 ˙ = x6 ˆ U1 . pitch and yaw are obtained and become with U1 inputs for the following translation subsystem. Connection of the two ideal subsystems of the overall dynamical system described by mapping (12). We can ideally imagine the overall system described by (12) as constituted by two subsystems. By simply choosing: U2 = − Ilx (x7 − xd ) − k1 x8 7 I U3 = − ly (x9 − xd ) − k2 x10 9 U4 = −Iz (x11 − xd ) − k3 x12 . (x7 − 2 xd ) 7 Using the considerations in the previous paragraph V-A. 0. which are the x. in the case of a perfect cross VTOL (Ix = Iy ) is drastically reduced to: ˙ V = (x7 − xd )x8 + x8 Ilx U2 + (x9 − xd )x10 + x10 Ily U3 + 9 7 +(x11 − xd )x12 + x12 Ilz U4 11 (14) Equation in which does not appear the perturbation term with Ω. k2 and k3 positive constants. in this subsection we consider the stabilization of the OS4 angles in a particular conﬁguration d Xα = (xd . the translations subsystem (with state X∆ ) regards the ﬁrst 6 element of the state X. From the angular rotations subsystem the roll. The dynamic of the height is described by lines 5 and 6 of system (12). and does not depend on translation components. Conversely to the previous case. pitch. xd . two Simulink discrete-step delay blocks have been introduced in the feedback loop and on the actuators. By Lyapunov theorem [12] is now ensured the simple stability for equilibrium. Note that the mapping fα (X. z and their time derivative. (18) + x2 8 + (x9 − 2 xd ) 9 + x2 10 + (x11 − 2 xd ) 11 + x2 12 (13) ˙ The time derivative of (13). The dynamics of these variables are described by fα (X. U3 . The angular rotations subsystem has as state the restriction Xα of X to the last 6 components which regard the roll. pitch and yaw and not on their time derivative. By the latter consideration we can ensure an asymptotical stability starting from a point in a set around the equilibrium. VI.

10 8 6 4 degrees ROLL 10 8 6 4 2 degrees PITCH 2 0 −2 −4 −6 −8 0 −2 −4 −6 −8 −10 0 500 steps YAW 1000 1500 −10 0 500 steps 1000 1500 10 8 6 4 2 degrees 0 −2 −4 −6 −8 Fig. Attitude Control for a Micromechanical Flying Insect Including Thorax and Sensor Models ICRA 2003. This includes dynamic modelling.S Sastry. Elfes. 2001. the experimental results obtained show that the proposed controller works well especially for the yaw angle. Boca Raton. 1996. We introduced the OS4 project and discussed the undergoing developments of ﬂying robots at ASL. Olfati-Saber. [4] A. Arimoto. VII. the Pitch and the Yaw angles was controlled. Finally. 1994. Chriette.S. As it can be seen from the experimental plots. In the simulation. Phd thesis. Floreano Evolving Vision-Based Flying Robots. Prinz. S. [10] E. we presented a survey of existing ﬂying microvehicles and made a comparison to micro VTOLs in terms of miniaturization. VIII. Contribution a la commande et a la mod´ lisation des e ` ` e e Phd h´ licopt` res : Asservissement visuel et commande adaptative. . 1974. 2002. Schenato and S. the controller introduced prove the ability to control the orientation angles. Nonlinear Control of Underactuated Mechanical Systems with Application to Robotics and Aerospace Vehicles. Simulation: the system has to maintain the height of 2 meters although the noise on the actuators. [5] J.umd. Berlin. [3] I. La Chaud-de-Fonds. http://aero. Australia. R EFERENCES [1] X. Equations fondamentales des micromoteurs courant continu avec rotor sans fer. [8] A. The height is taken with an added zero mean error.stanford. Analysis and Control of Underactuated Mechanical Nonminimum-phase Systems.pdf. EXPERIMENTS In order to validate the control law developed in the previous section. pitch and yaw angels to zero.edu/AGRC/Aero/Breguet. Mllhaupt. Control Theory of Non-linear Mechanical Systems. 2001. 1999. Experiments: the controller has to stabilize the system by maintaining the roll. Autonomous Robots Journal: Kluwer Academic Publishers. Thesis. Jucker. The Breguet-Richet Quad-Rotor Helicopter of 1907. 7. Australasian Conference on Robotics and Automation. [11] R. we implemented the controller and we performed several experiments on the real system. 592-600. In spite of the test bench limitations in term of delays and errors introduced by the tethering system. MIT. Leishman . Bulletin technique Portescap. LNCS 2525. http://www. EPFL. 8. L. Department of Mechanical Engineering. Bueno and al. Phd thesis.enae. 2003. CONCLUSION AND FUTURE WORK In this paper.. with a maximum angular velocity of 600 rad/sec. see ﬁgure 7.B. Proceedings of the 2nd International Workshop on Biologically Motivated Computer Vision. [2] J. the overall system has been simulated at 30Hz using a discrete time solver in order to model the behavior of the digital controller. 2002. Sastry. The Mesicopter: A Meso-Scale Flight Vehicle. Taiwan 2003. [12] S. the Roll. R. Robotic Airship for Exploration of planetary Bodies with an Atmosphere Autonomy Challenges. The positive results obtained in this development towards autonomous micro-VTOL reinforce our conviction that these systems have potential as candidates for the miniaturized ﬂying micro-vehicles emergence. Our next goal is to enhance the control with position controller and to develop a fully autonomous vehicle. pp. A mathematical introduction to robotic manipulation . −10 0 500 steps 1000 1500 Fig. [6] S. Auckland. Department of Electrical Engineering and Computer Science. Georges Perrenoud for the realization of the test bench mechanical components and all the students who worked or are working on this project. Springer-Verlag. Oxford Science Publications. Teipei.G. while the height was ﬁxed by the test bench. D. a saturation block has been placed between the controller and the delay.Mahony. vehicle design optimization and control. the task is to hover although an added normal gaussian noise of variance 4 rad/sec on each angular velocity. Design of a Four-Rotor Aerial Robot. [9] P. The task was to control the vehicle orientation thus. Pounds. ACKNOWLEDGMENT The authors would like to thank Jean-Christophe Zufferey for fruitful discussions and advices on ﬂying robots. see ﬁgure 8. Kroo. [7] P. Deng. F.edu/mesicopter/. FL.C Zufferey.

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd