MikroKopter Documentation

Cyphy Laboratory

Inkyu Sa
September 19, 2011

Contents

I. Introduction

A. Purpose

B. Coordinate systems

C. Position control

D. MK acceleration scale factor measurement

E. SI units and log message protocol

1. SI units

2. log message protocol

F. Complex dynamics of the MK quadrotor

11

G. System latency

12

H. Velocity estimation

15

1. X axis acceleration

15

2. Y axis acceleration

16

3. Z axis acceleration

17

I. ROS implementation

22

J. Yaw control

23

K. Altitude control

25

1. PID altitude control

26

L. Flight time calculation

26

M. Battery discharge dynamics

28

N. Platform price comparison.

28

O. Precision comparison

29

P. Ground truth and estimation.

30

1. VICON coordinate system and experiment setup

31

II. References

34

References

35

A. Matlab implementation

36

I.

INTRODUCTION
A.

Purpose

The objectives of this document are to create written history while implementing the
quadrotor using Mikrokopter (MK). In addition it will be a good opportunity to share this
knowledge with other researchers who are interested in using the MK platform.

B.

Coordinate systems

The MK doesnt specify a scientific coordinate system. It only has pitch, roll and yaw
angles. The red bar is the front and the opposite is the rear. Pitch (Nick) angle results
in forwards and backwards movements. In contrast, roll angle results in left and right
movements. When we push all the way up the pitch stick (right stick on mode2), StickNick
value (516) is transmitted to the FlightControl(FC). This implies moving forward with the
fastest speed. It is the maximum value of pitch angle. When we pull all the way down
the pitch stick, StickNick value (-592) is transmitted to the FC. This implies moving
backward with the fastest speed. On the other hand, when pushed all the way right, the
roll stick, StickRoll value (-516) is transmitted to the FC. When pushed all the way left, the
roll stick, StickRoll value (512) is transmitted to the FC. Commonly the forward direction
is the x-axis of robots and the right direction is the positive y-axis. Pitch angle implies
rotation around the y-axis and the roll angle implies rotation around x-axis according to the
right-hand rule. Positive pitch angle causes the vehicle to move backwards- the opposite of
MK notation. Positive roll angle causes it to move right-again the opposite of MK notation.

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

AccZ
yaw

MK frame

NickAngle
AccRoll

AccNick
RollAngle

FIG. 1: The MK coordinate frame

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

Body frame

World frame
xis
x-a
yax

z-axis

x-axis
roll,

is

Note
All coordinate frames is
ruled by right hand.
y-axis
pitch,

yaw,
z-axis

FIG. 2: The defined coordinate frame

C.

Position control

In order to control the quadrotor, a traditional PD controller is implemented. The inputs for the controller are desired position and velocity and the outputs are pitch and roll
commands. The laser range finder provides x,y positions and a yaw angle every 100ms and
these PD controllers generate outputs at the same rate. The unit of the position and angle
are represented in SI units, such as metres and radian.

d W W
( x x)
dt
d
W
= Kp (W y W y) + Kd (W y W y)
dt

= Kp (W x W x) + Kd

(1)
(2)

notes the pitch input in world frame and W x is the reference input which the quadrotor

should follow.

x is the current position of the quadrotor in the world frame and this data

is provided from the laser scan matcher. The desired output to the MK must be in the
range 128. Inputs are divided by the scale factor(120).
Then command angles are with respect to the world frame. The world and body frames
are related by the yaw angle transformation.

RW () =

cos

sin

sin

cos

where is the angle of the world frame with respect to the body frame.

B
W

= R()

B
W

(3)

(4)

Finally, the commands in the body frame are

= cos W sin W

(5)

(6)

= sin W + cos W
5

Pitch

90

60

50

45

15

15

45

50

60

90

1st

602

522

458

428

152

-122

-431

-457

-531

-617

2nd

602

525

468

432

154

-123

-433

-462

-532

-618

3rd

602

523

470

422

153

-122

-432

-463

-532

-617

Mean

602

523.3

465.3

427.3

153

-122.3

-432

-460.6

-531.6

-617.3

TABLE I: MK AccNick Acceleration versus pitch true angle measurement ruled by right
hand.
D.

MK acceleration scale factor measurement

MK doesnt provide either SI units or documents explaning MK units. These data are
unscaled measurements. Hence we try to figure out what the actual units are. This data is
measured by an external accelerometer and MikroKopter Tool V1.72.

a. Pitch acceleration The contents of TABLE I are the pitch acceleration data from
the MK. The unit of the measurement is unknown.
Fig. 3 shows the linear line fitting with the measured data. In the x-direction the equation
of motion is
x = k sin

(7)

Where x refers to the acceleration data provided from MK quadrotor, k is the scale factor
and is the pitch angle. Finally, k = 606.3 units/g can be calculated and the acceleration
value also can be obtained in units of 1g.

b.

Roll acceleration The contents of TABLE II are the roll acceleration data from

the MK. The unit of measurement is unknown as well.

Fig. 4 shows the linear line fitting with the measured data. According to the equation 7,
the k = 603.44 units/g can be computed for the roll acceleration.
6

FIG. 3: MK pitch acceleration data versus true pitch angle.

Roll

90

60

50

45

15

15

45

50

60

90

1st

612

522

458

436

162

-167

-430

-462

-522

-587

2nd

611

521

456

443

161

-162

-432

-463

-515

-588

3rd

613

523

457

439

160

-165

-431

-463

-517

-588

Mean

612

522

457

439.3

161

-164.6

-431

-462.6

-518

-587.6

TABLE II: MK roll acceleration versus true roll angle measurement ruled by right hand
c.

Z acceleration The drifting problem of raw ACCZ ADC measurement was found

when vibration came from four rotors and it was thought that error accumulated when
analog data was converted into digital. Hence the preprocessed MK default ACCZ value
is exploited to estimate the z-velocity. The MK provides vibration robust filtered data
but has an issue that it doesnt know what the unit is. In order to use this acceleration
7

FIG. 4: MK AccRoll acceleration versus true roll angle.

data for the complementary filter, the unit of data should be addressed. A experiment was
conducted with a XSENS sensor. We put a XSENS on the quadrotor and logged XSENS
and MK ACCZ data concurrently at the same sample rate(20 Hz), then matched the two
data manually. Figure 5 shows the result.
ACCZxsens (t)[m/s2 ] =

ACCZmiko (t)
+ 9.8
15.5

(8)

FIG. 5:
E.

SI units and log message protocol

1.

SI units

All units are used in the system following the SI units. All angle measurements are
represented in Radian and Metre as a unit of all distance and angle measurements and
Second is a unit of time. m/s2 is a unit of all acceleration.
2.

log message protocol

There are two log messages. One is logging attitude of the quadrotor such as pitch
angle and acceleration along x axis. Another measures data using the laser range finder like
position, velocity and estimation of velocity. The reason why we have two log messages is
because of different sampling rates of the laser range finder and the MK quadrotor. Position
data based on the laser range finder comes into every 100ms and attitude of the quadrotor
9

Order

Notation

Description

Unit

g W Vel.x

World velocity along x axis.

m/s

g B Vel.y

Body velocity along x axis.

m/s

g B esti Vel.x

Body estimated velocity along x axis.

m/s

g W Vel.y

World velocity along y axis.

m/s

g B Vel.y

Body velocity along y axis.

m/s

g B esti Vel.y

Body estimated velocity along y axis.

m/s

g CurrentPosition.x

current x position.

g CurrentPosition.y

current y position.

g CurrentPosition.theta

current angle.

radian

10

index

Counter increasing +1 every 100ms

number

TABLE III: 100ms loop log message format

is measured every 50ms.
a. 100ms sampling time TABLE III shows the 100ms sampling time log message
protocol.

10

Order

Notation

Description

AnglePitch

Pitch angle

radian

AngleRoll

Roll angle

radian

AngleYaw

Yaw angle

radian

AccX

X acceleration with respect to the body frame

m/s2

AccY

Y acceleration with respect to the body frame

m/s2

AccZ

Z acceleration with respect to the body frame

m/s2

ControlPitch

Pitch angle commanded to the quadrotor (-128 128)

unknown

ControlRoll

Roll angle commanded to the quadrotor (-128 128)

unknown

ControlYaw

Yaw angle commanded to the quadrotor (-128 128)

unknown

10

StickPitch

Transmitter pitch stick value (-128 128)

unknown

11

StickRoll

Transmitter roll stick value (-128 128)

unknown

12

StickYaw

Transmitter yaw stick value (-128 128)

unknown

13

StickThrust

Transmitter thrust stick value (0 255)

unknown

14

index

Counter increasing +1 every 50ms

Unit

number

TABLE IV: 50ms loop log message format

b.

50ms sampling time TABLE IV shows the 50ms sampling time log message

protocol.

F.

Complex dynamics of the MK quadrotor

In this section, dynamics of the MK quadrotor are presented.

11

FIG. 6: Measured angle versus commanded angle

The lag can be resolved by looking at the vehicle velocity, in the vehicle frame, and
correlating that with the pitch angle. Taking the cross correlation, Fig. 8 shows a lag of
approximately 4 time steps which implies that the MK responses are 200ms later.

G.

System latency

In order to find out the system latency between the Gumstix and the ground station,
the network time protocol (NTP) server is exploited. NTP allows the setting up of two
systems with less than 1ms time error. Fig. 9 shows synchronised systems and delay.
After synchronisation, time duration was measured using ROS time stamp function. This
function provides s accuracy and it is sufficient to measure the time period.
Fig.

10 shows 100ms time delay of the laser range finder.

Fig.

11 also shows 5ms

delay of the ICP scan matcher. The WIFI latency can be determined by subtraction of

12

300

250

Number

200

150

100

50

0
30

20

10

10

20

30

40

pitch angle[degree]

FIG. 7: Pitch angle histogram

x 10
4

3.95

3.9

3.85

3.8

3.75

3.7

3.65

3.6
20

15

10

10

15

20

FIG. 8: Cross correlation

Current T ime Received T ime. The WIFI latency is in the order of 7ms.
In summary, figure 12 shows the total system latency in the system.

As shown in

the Fig. 8, the MK has about 200ms natural response time and WIFI, laser range finder
and scan matcher have 167ms delay.
13

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

Desktop Time

Gumstix Time, 1.4ms delay and -1.3ms offset.

FIG. 9: Time synchronization using NTP server

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

100ms delay

Stop streaming

FIG. 10: Laser time delay

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

FIG. 11: ICP time delay and wifi latency

14

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

System latency

157ms 167ms
1

50ms

Serial communication with

MK over Xbee.

5ms 10ms

ICP processing time

2 ms 7 ms

Laser scanner
inherited delay

Transfering Laser
data over WIFI

100ms

time (ms)

Publish ROS topic, 2s

Send extern control

signal 7s

Running PID for

position control,

Parsing Data, 1s

2s

500s
get data from the
read buffer. 28s

10ms delay to make

sure receiving. 10ms

Request debug data, 11s

Expand

10.852ms 11.783ms (including debug print time)

50 ms

0 ms
ROS thread

20Hz

FIG. 12: The total system latency

H.

Velocity estimation

It is necessary to use a velocity estimator because of the complex dynamics and latency
of the system. A complementary filter is exploited and this requires accelerations of each
axis in order to estimate velocities.

1.

X axis acceleration

the velocity in body frame. The acceleration along the x axis is

ax = B x cos g sin
B

x =

ax + g sin
cos
15

(9)
(10)

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

This is corrected
coordinate frame.
Q

ax

s
co

in
gs

Qf rame

Quadrotor

gs

in

FIG. 13: X axis acceleration

2.

Y axis acceleration

The acceleration along the y axis is

ay = B y cos + g sin

(11)

ay g sin
cos

(12)

y =

x, w y, w vx and w vy refer to the position and velocity in world frame respectively. is

measured orientation in world frame. b vx and b vy .

16

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

This is corrected
coordinate frame.

Qf rame

ay

Quadrotor

g sin

y
Q

y c

os

gs

in

FIG. 14: Y axis acceleration

3.

Z axis acceleration

The acceleration along the z axis is

az = cos cos (g + Q z)
az
Q
z =
g
cos cos

17

(13)
(14)

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

This is corrected
coordinate frame.

Qf rame

Quadrotor

az = cos cos (g + Q z)
Q

18

a. A complementary filter
vx (t + 1) = vx (t) + T ( k ( B vx vx (t)) + Q x)

(15)

vy (t + 1) = vy (t) + T ( k ( B vy vy (t)) + Q y)

(16)

vz (t + 1) = vz (t) + T ( k ( B vz vz (t)) + Q z)

(17)

where T is the sample interval, and k is the filter gain.

1.5
laser world position x
acc world position x

world position x from accx

0.5

0.5

1.5

10

20

30

40

50

60

70

80

90

time [sec]

FIG. 16: X position in world frame for laser-based localizer

Fig. 16 is the one sample predicted position using Equation 18. We assume that the
quadrotor is hovering and the velocity of it is almost constant.

19

x(t + 1) = x(t) + v(t) T

(18)

= x(t) +

(19)

x(t) x(t 1)
T
T

= 2x(t) x(t 1)

(20)

Fig. 17 shows the velocity in body frame calculated derivative of the x position and
Equation 4.

estimated velocity x
body velocity x

0.5

0.4

0.3

velocity x [m/s]

0.2

0.1

0.1

0.2

0.3

0.4

0.5

10

20

30

40

50

60

70

time [sec]

FIG. 17: x velocity versus estimated x velocity in body frame

Fig. 17 shows the estimated velocity in body frame using Equation 15 and 16.

20

80

Fig. 18 plots the command data which generated by Equation 21.

80
pitch cmd
stick pitch
60

pitch cmd vs stick pitch

40

20

20

40

60

10

20

30

40

50

60

70

80

90

time [sec]

FIG. 18: Pitch command generated with P velocity controller

d W
( vx (t) W vx (t))
dt
d
W
(t) = Kp (W vy (t) W vy (t)) + Kd (W vy (t) W vy (t))
dt
d
W
(t) = Kp gas (W vz (t) W vz (t)) + Kdgas (W vz (t) W vz (t))
dt
W

(t) = Kp (W vx (t) W vx (t)) + Kd

In the Equation 21 we assume that

W
vx (t)

and

21

W
vy (t)

are zero to hovering.

(21)
(22)
(23)

FIG. 19: This plot shows estimated velocity along z axis. Green, Red and Blue lines stand
for ACCZ from MK, body velocity and estimated velocity. ACCZ is sampled at 20 Hzand
estimated velocity is computed at the same rate. The body velocity is calculated based on
laser range finder height data and has a sample rate of 10 Hz. Note that body velocity is
smoothed.
I.

ROS implementation

It is easily achieved to integrate systems with ROS. All messages come into the ROS
core and are redistributed to nodes through the TCP/IP protocol. This feature allows
implementation of the software packages as modules, ROS stacks or packages. Fig. 20 shows
simple implementation based on the ROS. In this figure, dark gray boxes denotes the ROS
nodes which act as an individual process. The canonical scan matcher receives laser range
data and produces 2D positions using the ICP algorithm. It publishes /pose2D topic every
100ms and the data structures are shown in TABLE V and TABLE VI. A Miko Serial
node functions are as obtaining IMU data from the flight control board while a quadrotor
is flying and publishing /mikoImu topic every 50ms. These two topics can be recorded to
create a log file and be played using rosbag package. Finally, a complementary Filter node
22

Name

geometry msgs/Pose2D

Description

This expresses a position and orientation on a 2D manifold.

Data type

definition

Unit

Description

float64

x position in world frame

float64

y position in world frame

float64

theta

radian

theta angle in world frame

TABLE V: /Pose2D topic data type

computes body and estimated velocity given position and IMU data with Equation 15 and
16. With this complementary filter, the position control of the quadrotor can be achieved.
The demonstration footage can be found on the Youtube.

10
0m
s(
W
IFI
)

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

Canonical
scan matcher

Miko Serial

m
co

an

/po

se

ko
mi

2D

Im

rosbag
play

rosbag
record

Complementary
Filter

rosbag

IMU data

pos(1) imu(1) imu(2) pos(2) imu(3) imu(4)

kC
/m

Quadrotor
Platform

tim

ate

loc

ity

md

Es

e
dv

Position

Velocity

P controller

PID controller

FIG. 20: ROS complementary filter implementation

J.

Yaw control

Yaw control of the quadrotor is relatively easier than pitch, roll and height control. Gyro
data is drifting overtime which implies the vehicle is going to change its heading. Therefore
23

Name

sensor msgs/mikoImu

Description

This expresses a attitude of a quadrotor.

Data type

definition

Unit

Description

float64

pitchAngle

radian

pitch angle in body frame

float64

rollAngle

radian

roll angle in body frame

float64

yawAngle

radian

yaw angle in body frame

float64

AccX

m/s2

Acceleration along x axis in body frame

float64

AccY

m/s2

Acceleration along Y axis in body frame

float64

AccZ

m/s2

Acceleration along Z axis in body frame

TABLE VI: /mikoImu topic data type

Name

sensor msgs/mikoCmd

Description

This expresses command input to the quadrotor.

Data type

definition

Unit

Description

int32

pitch

pitch cmd -128 128

int32

roll

roll cmd -128 128

int32

yaw

yaw cmd -128 128

int32

throttle

throttle cmd 0 512

TABLE VII: /mikoCmd topic data type

we need to use the absolute measurement data which is less changed as time passes. Laser
range data can provide the yaw angle with minimum error as its variance is smaller than
a gyro sensor. A traditional PID controller is used to hold the position. The equation 24
shows this controller and figure 21 presents the output and yaw angle. In this figure, the
pilot gives the artificial angle using the transmitter and the controller generates the output
to keep the initial angle 0.

24

u(t) = Kp e(t) + Ki

e( )d + Kd
0

d
e(t)
dt

(24)

where e(t) =

20

Yaw(Deg)

10

10

20

30

10

20

30

40

50

60

70

80

time(sec)

FIG. 21: Yaw angle and yaw command output using PID controller.

K.

Altitude control

In order to implement the height controller, it is a compulsory requirement to make

sure safety. Once the vehicle gets a wrong command such as putting negative sign on the
unsigned char variable, it flies to the sky at a full speed. In fact, this can be a very
frightening moment. To avoid this kind of accident, the MK limits the FC code to prevent
the user from controlling the gas value if the control cmd value is bigger than the stick gas.
To achieve the altitude control, the FC code requires modification by enabling the external
control with the saturation function that limits the range from 0 to 130. This number is the
threshold of the gas value which lets the vehicle fly with the current payload and is obtained
empirically.
25

1.

PID altitude control

u(t) = Kp e(t) + Ki

e( )d + Kd
0

d
e(t)
dt

(25)

where e(t) = z z and z denotes the desired position along z axis and z is the current
position. In addition, the height measurement is coupled with and . The projected
estimation of the height can be determined as the following equation.

z = cos cos z

(26)

FIG. 22: Height measurement data using PID control with a goal at 0.6 metre. The input
for the PID controller is the height measurement from laser hat and output is throttle
control command. The MK uses unsigned char data type to control quadrotor range
from 127. Note that average height measurement is within 5cm. Rarely does maxim
error goes up up to 10cm owing to ground effect and aerodynamics, but the quadrotor
quickly converges into the goal.

L.

Flight time calculation

Flight time is essential to the MAV research area. Many research platforms can only fly
for up to 10 minutes or less. However, the MK platform is able to fly up to 41 minutes
unofficially. The video demonstration can be found on vimeo. Detailed information can be
26

found in the article of MikroKopter wiki page. In order to estimate the flight time given
payload, we need to know total weight of a vehicle. In our case, the quadrotor weight is
around 1,040g including frames, laser range finder, Gumstix, XBee, laser hat and battery.
Therefore 260g is the thrust per motor.

FIG. 23: The Ampere Versus payload characteristic of ROXXY2827-35 motor.

From figure 23, we could notice that a motor consumes around 1.7A. This implies that
there is a total current of 6.8A = 1.7 4.
B
(27)
C
Where t is the total flight time in units of hour, B and C are battery capacity and Total
t=

current respectively. A Lipo battery, 2.2Ah, 4 cells, was used and Figure 27 shows 0.323
hour = 19.4 minutes flight time. This flight time is an ideal value that specifies the absolute
limit. These factors should be taken account.
The thrust characteristic was measured with a 100% loaded LiPo. The thrust decreases
slightly when the voltage drops in flight.
27

 Turbulence
Losses due to regulations
Other losses (e.g. in cables, etc.)
For these reasons, a scale factor, 0.9, was multiplied. This parameter is chosen empirically
and should be changed as the battery efficiency goes down.
In conclusion, the estimated total flight time is 17.4 minutes and this value also can be
found in Figure 24. This graph is validated by comparison of our previous work as described
in Figure 25
Moreover, extra power consumption electric pars are used such as a laser range finder,
a Gumstix, a XBee module and a USB HUB. These parts also consume energy and reduce
the total flight time. The power consumption of the Gumstix is order of 2.5W 5W and
the laser range finder spends 2.5W. Hence, the extra power consumption could be said to be
less than 0.5A. Finally, from the calculation our quadrotor can fly up to 16.28 minutes,
0.9

2.2
6.8+0.5

M.

Battery discharge dynamics

We logged Lipo batteries, (3cell, 2200mAh), to compensate the trust offset. Interestingly,
each battery showed a different discharge curve. (see Figure 26).

N.

Platform price comparison.

The advantage of the MK is a competitive price. As described in Table VIII, it is a lot

cheaper than a similar level of research platform. Although MK Basicset is needed to put all
parts together before flight, there is detailed documents on the MK web. In addition, it is
feasible to modify the platform for users demands. The pelican from Asecnding Technology
is 6.41 times more expensive than MK quadrotor.

28

<
=
0
>
1
2
?
3
@
;4
5
6
/7
8
9
:.A
-B
+
,C
*#
D
$
%
)"E
(F
!&
G
'K
JH
L
M
IN
~
O
}P
|m
Q
u
w
v{zyxtR
sln
ko
p
q
rS
T
jU
g
h
V
ia
b
d
ce
W
f_
`^]X
\Y
Z
[

FIG. 24: Flight time Versus payload characteristic graph. Total weight of the vehicle is
1,040g and this graph shows the flight time with 640g+238g (battery) payload. Therefore,
the payload is 162g. In addition, the hovering graph is appropriate for the indoor
application.
O.

Precision comparison

In this section, a variety of accuracy and precisions of state estimations are presented
when a quadrotor performs hovering or following given trajectory. As MAVs have fast
dynamics, filtering techniques such as complementary filter and Kalman Filter are often
used to estimate the state of the vehicle. Table IX shows a summary of precision from 5
different research groups.[2][5][1][3][4]
29

FIG. 25: Payload Versus Flight time table. This table was measured by us.
Mikrokopter Price(USD) Asecnding Tech Price(USD)
platform

MK Basicset

987.46

Pelican

7435.09

Transmitter

6CH

169.99

7CH

499.49

Receiver

7 CH

79.99

Total

1237.44

Ratio

7934.58
6.41

TABLE VIII: This table shows prices comparison between the MK Quadrotor set and the
similar level platform of Ascending technology, Pelican. These prices come from
MikroKopter , HobbyKing and Ascending Tech.
P.

Ground truth and estimation.

In this section, we compare ground truth from the VICON system to our dynamic state
estimation. We estimate position x, y, z and velocity Vx , Vy . in addition, The MK estimates
pitch(), roll() and yaw() angles.
30

170
battery new1
battery new2
battery old

160
150

batt adc value

140
130
120
110
100
90
80
0

100

200

300

400
500
time (sec)

600

700

800

900

FIG. 26: Lipo battery discharge curves. X axis denotes time and y axis is a battery ADC
reading value.
1.

VICON coordinate system and experiment setup

Figure 28 shows VICON coordinate and experiment configuration. Since the maximum
range of the laser range finder attached to the quadrotor is 4m, we need to put structures
within the 4m range. VICON system tracks infrared beam reflected by each makers and
provides position and orientation of the makers. Figure 29 shows the place where makers
on the quadrotor are attached. By computing error of Euclidean distance between two
markers,Equation 28 VICON system shows 5mm position error. We could find the distance
error increased after taking off because of frame vibration.

e = einitial

(xmaker1 xmaker2 )2 + (ymaker1 ymaker2 )2 + (zmaker1 zmaker2 )2

where einitial denotes the very first distance between markers.

31

(28)

FIG. 27: Precision comparison of state-of-the-art platforms.

FIG. 28: This photo shows the VICON system coordinate and setup. This system is
equipped with 14 cameras to track markers and provides position, normal and orthogonal.
A x mark on the ground is the initial position of VICON system. The coordinate
transformation is required from VICON coordinate to MK coordinate as described in 2.

32

FIG. 29: Two makers are attached in order to calculate position error of VICON system.

FIG. 30: Position error between two markers. Note that the quadrotor starts taking off and
increasing position error due to frame vibration around 22sec. Mean is 0.0002 metre.

33

TABLE IX:
University MIT(RANGE)[2]

GRASP Lab.[5]

ETHZ(ASL)[1]

Year

2010

2011

2011

Precision

Hovering

Hovering

Hovering

(metre)

RMS X= 0.06

Std X= 0.0849

RMS X=0.12

(metre)

RMS Y= 0.06

Std Y= 0.0911

RMS Y=0.12

(metre)

RMS Z= 0.02

Std Z= 0.0554

RMS Z=0.01

Trajectory following

Trajectory following

(metre)

RMS X=0.07

Std X=0.0247

(metre)

RMS Y=0.07

Std Y=0.0323

(metre)

Std Z=0.007

Platform

Pelican

Pelican

Pelican

Sensor

Laser 40Hz

Laser 40Hz

1000Hz IMU
10Hz mono Vision
Fusion

University

Freiburg[3]

Oxford(Chris Kemp) [4]

QUT

Year

2009

2006

2011

Precision

Hovering

(metre)

Std X=0.075368

(metre)

Std Y=0.080849

(metre)

a.

Error Z=0.1

Error Z=0.1

Std Z=0.031913

Position estimation Figure 31 and Figure 32 show ground truth and position, ve-

locity estimation.

II.

REFERENCES

34

FIG. 31: Blue is the ground truth and red is the laser position estimation. The x-axis is
time in second, the y-axis is position in unit of metre.

[1] Markus Achtelik, Michael Achtelik, Stephan Weiss, and Roland Siegwar. Onboard imu and
monocular vision based control for mavs in unknown in- and outdoor environments.

In

ICRA2011, 2011.
[2] A. Bachrach, A. de Winter, Ruijie He, G. Hemann, S. Prentice, and N. Roy. RANGE - robust
autonomous navigation in GPS-denied environments. In Robotics and Automation (ICRA),
2010 IEEE International Conference on, pages 1096 1097. MIT, may 2010.
[3] S. Grzonka, G. Grisetti, and W. Burgard. Towards a navigation system for autonomous indoor
flying. In Robotics and Automation, 2009. ICRA 09. IEEE International Conference on, pages
2878 2883, may 2009.
[4] Christopher Kemp. Visual Control of a Miniature Quad-Rotor Helicopter. PhD thesis, Univer-

35

FIG. 32: Blue is the ground truth and red is the velocity estimation from complementary
filter. The x-axis is time in second, the y-axis is velocity in unit of m/s.
sity of Cambridge, 2006.
[5] Shaojie Shen, Nathan Michael, and Vijay Kumar. Autonomous multi-floor indoor navigation
with a computationally constrained mav. In 2011 IEEE International Conference on Robotics
and Automation, 2011.

Appendix A: Matlab implementation

To implement a complementary filter, a simulation was conducted with MATLAB. The

data that was used for simulation was obtained by manual pilot. All data was converted
into SI units, metres and radians. The follow is the MATLAB code.
1

clear

all ;

d a t a=i m p o r t d a t a ( d a t a M a n u a l C o o r d i n a t e , , , 1 ) ;
3
sa m p ti m e = 0 . 0 5 ; % s e c o n d /50ms s a m p l i n g t i m e
5 W x=d a t a . d a t a ( : , 1 0 ) / 1 0 0 0 ;

% u n i t=m e t r e X p o s i t i o n

W y=d a t a . d a t a ( : , 1 1 ) / 1 0 0 0 ;

% u n i t=m e t r e Y p o s i t i o n

36

FIG. 33: Blue is ground truth and red is angle estimation by accumulating angular rates of
gyro sensor. The x-axis is time in second, the y-axis is the angle in unit of degree. This
data comes from the MK.

7 t i m e=d a t a . d a t a ( : , 1 5 ) sa m p t im e ; % u n i t=s e c o n d
numData=max ( d a t a . d a t a ( : , 1 5 ) ) +1;

% +1 b e c a u s e i n d e x

starts

from 0

9 Yaw=d a t a . d a t a ( : , 1 2 ) p i / 1 8 0 ; % r a d i a n
ACC X=d a t a . d a t a ( : , 4 ) ; % u n i t =1g = 9 . 8m/ s 2
11 ACC Y=d a t a . d a t a ( : , 5 ) ; % u n i t =1g = 9 . 8m/ s 2
MK PITCH=d a t a . d a t a ( : , 1 ) /10 p i / 1 8 0 ; % r a d i a n
13 MK ROLL=d a t a . d a t a ( : , 2 ) /10 p i / 1 8 0 ; % r a d i a n
s t i c k p i t c h=d a t a . d a t a ( : , 1 3 ) ;
15

s t i c k r o l l =d a t a . d a t a ( : , 1 4 ) ;
g =9.81; % unit 1g

17 %k = 0 . 2 5 ; % f i l t e r
k =0.5; % f i l t e r
19

gain

gain

d e s i r e d v x =0; % d e s i r e d

velocity

along X axis

u n i t=m/ s

d e s i r e d v y =0; % d e s i r e d

velocity

along X axis

u n i t=m/ s

21 Kp=30;
Kd=30;
23

25

27 %Down s a m p l i n g t h e x p o s i t i o n
%The s a m p l i n g

rate

of

the

data because

laser

it

s c a n matcher

oversampled .
is

100ms and t h e

37

log

data

is

29 %sampled a t 50ms r a t e .

31

h a l f s i z = numData / 2 ;
W x 100ms= z e r o s ( h a l f s i z , 1 ) ;

33 W y 100ms= z e r o s ( h a l f s i z , 1 ) ;
for

i =1 :

half siz

W x 100ms ( i ) = W x( i 2 1) ;

35

W y 100ms ( i ) = W y( i 2 1) ;
37 end

39 %Do i n t e r p o l a t i o n

to

c r e a t e smooth g r a p h .

W x =i n t e r p ( W x 100ms , 2 ) ;
41 W y =i n t e r p ( W y 100ms , 2 ) ;

43 %Array

size

align

w i t h numData !

%I t s n o t f a n c y but works .
45 W x mine= z e r o s ( numData , 1 ) ;
W y mine= z e r o s ( numData , 1 ) ;
47

for

i =1 : numData1
W x mine ( i ) = W x ( i ) ;

49

W y mine ( i ) = W y ( i ) ;
end

51

53 %W x mine ( 1 6 5 9 )=W x ( 1 6 5 8 ) ;
%W y mine ( 1 6 5 9 )=W y ( 1 6 5 8 ) ;
55

57 %W v x= d i f f (W x) . / d i f f ( t i m e )

59
%=================================================
61 % S t e p 1 .

Calculate

world frame

velocity

using

laser

position

data

%=================================================
63 W x v = z e r o s ( numData ,

1) ;

W y v = z e r o s ( numData ,

1) ;

65 %f o r n =2:numData
W x v= d i f f ( W x mine ) / sa mp ti m e ;
67

W y v= d i f f ( W y mine ) / sa mp ti m e ;
%

69

W x v ( n ) =(W x mine ( n )W x mine ( n1) ) / sa mp t im e ;

W y v ( n ) =(W y mine ( n )W y mine ( n1) ) / sa mp t im e ;

%W y v ( n ) =(W y( n+1)W y( n ) ) / sa mp t im e ;

71 %end

73 %=================================================
% Step2 .

Covert world frame

velocity

t o t h e body f r a m e u s n i g

75 % m a t r i x
%%=================================================
77
b x v = z e r o s ( numData ,

1) ;

79 b y v = z e r o s ( numData ,

1) ;

f o r n =1:numData1
b x v ( n )=c o s (Yaw( n ) ) W x v ( n )s i n (Yaw( n ) ) W y v ( n ) ;

81

b y v ( n )=s i n (Yaw( n ) ) W x v ( n )+c o s (Yaw( n ) ) W y v ( n ) ;

83 end

85 %=================================================
% Step3 .

velocity

complementary

filter

87 %%=================================================

38

rotation

89 b v h a t x = z e r o s ( numData ,

1) ;

b v h a t y = z e r o s ( numData ,

1) ;

91 ddotX = z e r o s ( numData ,

1) ;

ddotY = z e r o s ( numData ,

1) ;

93 b v h a t x ( 1 ) =0;
b v h a t y ( 1 ) =0;
95
f o r n =1:numData1
ddotX ( n ) = (ACC X( n ) g s i n (MK PITCH( n ) ) ) / c o s (MK PITCH( n ) ) ;

97

ddotY ( n ) = (ACC Y( n ) g s i n (MK ROLL( n ) ) ) / c o s (MK ROLL( n ) ) ;

b v h a t x ( n+1) = b v h a t x ( n ) + sa m p ti m e ( k ( b x v ( n )b v h a t x ( n ) )+ddotX ( n ) ) ;

99

b v h a t y ( n+1) = b v h a t y ( n ) + sa m p ti m e ( k ( b y v ( n )b v h a t y ( n ) )+ddotY ( n ) ) ;
101 end

103
%=================================================
105 % S t e p 5 .

Velocity P controller

%%=================================================
107 p i t c h c m d = z e r o s ( numData ,
r o l l c m d = z e r o s ( numData ,
109

1) ;
1) ;

e r r o r p i t c h = z e r o s ( numData ,
e r r o r r o l l = z e r o s ( numData ,

111

1) ;
1) ;

p o s t e r r o r p i t c h = z e r o s ( numData ,
p o s t e r r o r r o l l = z e r o s ( numData ,

1) ;
1) ;

113
e r r o r p i t c h ( 1 ) =0;
115

e r r o r r o l l ( 1 ) =0;

117

f o r n =2:numData
e r r o r p i t c h ( n )=d e s i r e d v x b v h a t x ( n ) ;
e r r o r r o l l ( n )=d e s i r e d v y b v h a t y ( n ) ;

119

121

%p i t c h c m d ( n ) = Kp e r r o r p i t c h ( n )+Kd ( e r r o r p i t c h ( n ) e r r o r p i t c h ( n1) ) ;

123

%r o l l c m d ( n ) = Kp e r r o r r o l l ( n )+Kd ( ( e r r o r r o l l ( n ) e r r o r r o l l ( n1) ) / sa mp ti m e ) ;

p i t c h c m d ( n ) = Kp e r r o r p i t c h ( n ) ;

r o l l c m d ( n ) = Kp e r r o r r o l l ( n ) ;
125 end

127 %=================================================
% Step6 .

Acceleration

integration

129 %%================================================

131 %f o r n =1:numData1
%

v e l i n t x ( n )= sa mp t im e / 2 ( cumsum (ACC X( n+1) )+cumsum (ACC X( n ) ) ) 9 . 8 ;

133 %end

135

v e l i n t x = cumsum ( d e t r e n d ( ddotX ) sa m p ti m e ) ;

137

p o s i n t x = cumsum ( d e t r e n d ( v e l i n t x ) s am p ti m e ) ;

139

141

143
%=================================================
145 % S t e p 4 .

Plotting

%%=================================================

39

147
%C r e a t e a m u l t i l i n e
149 %x l a b e l ( { f i r s t
%C r e a t e a b o l d

label

for

l i n e ; second
label

for

t h e xa x i s

using a multiline

cell

array :

contains a single

quote :

l i n e })

t h e ya x i s

that

151 %
%y l a b e l ( George s

Popularity , f o n t s i z e ,12 , fontweight , b )

153

155 %=================================================
% Fig1 .

laser

world

position

vs acc world

position

157 %%=================================================
figure (1) ;
159 p l o t ( time , W x mine , r ) ;
x l a b e l ( time

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( world

p o s i t i o n x [m] ,

f o n t s i z e ,12) ;

161 h o l d on ;

163 p l o t ( time , p o s i n t x , b ) ;
x l a b e l ( time

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( world

p o s i t i o n x from a c c x ,

f o n t s i z e ,12) ;

165
legend ( l a s e r

world

p o s i t i o n x , acc world

position x ) ;

167 %f i g u r e ( 2 ) ;
%p l o t ( time , W x v , b ) ;
169

171 %f i g u r e ( 2 ) ;
%p l o t ( time , b x v , m ) ;
173 %x l a b e l ( t i m e

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( body v e l o c i t y x [m/ s ] ,

f o n t s i z e ,12) ;

175 %f i g u r e ( 4 ) ;
%p l o t ( time , b y v ) ;
177
%=================================================
179 % F i g 2 . body v e l o c i t y

v s CF e s t i m a t e d

velocity

%%=================================================
181
figure (2) ;
183 p l o t ( time , b v h a t x , r ) ;
x l a b e l ( time

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( v e l o c i t y x [m/ s ] ,

f o n t s i z e ,12) ;

185
h o l d on ;
187 p l o t ( time , b x v , b ) ;

189 l e g e n d ( e s t i m a t e d

191

axis ([0

v e l o c i t y x , body v e l o c i t y x ) ;

t i m e ( end ) 0.6

0.6]) ;

%h o l d on ;
193 %p l o t ( time , v e l i n t x , g ) ;

195 %=================================================
% Fig3 .

pitch

input vs

stick

pitch

197 %%=================================================

199

figure (3) ;
p l o t ( time , p i t c h c m d , r ) ;

201

x l a b e l ( time

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( p i t c h cmd v s

h o l d on ;
203 p l o t ( time , s t i c k p i t c h , b ) ;

205 l e g e n d ( p i t c h cmd , s t i c k

pitch ) ;

40

stick

pitch ,

f o n t s i z e ,12) ;

207 %f i g u r e ( 7 ) ;
%p l o t ( time , r o l l c m d , r ) ;
209 %x l a b e l ( t i m e

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( r o l l cmd v s

stick

roll ,

f o n t s i z e ,12) ;

%h o l d on ;
211 %p l o t ( time , s t i c k r o l l , b ) ;

213 %f i g u r e ( 8 ) ;
%p l o t ( time , b v h a t y , r ) ;
215 %x l a b e l ( t i m e

[ sec ] ,

f o n t s i z e , 1 2 ) , y l a b e l ( v e l o c i t y y [m/ s ] ,

217 %h o l d on ;
%p l o t ( time , b y v , b ) ;
219
%l e g e n d ( e s t i m a t e d

v e l o c i t y y , body v e l o c i t y y ) ;

41

f o n t s i z e ,12) ;