Comprehensive Simulation of Quadrotor UAVs

Using ROS and Gazebo

Johannes Meyer1 , Alexander Sendobry1 , Stefan Kohlbrecher2, Uwe Klingauf1 ,

and Oskar von Stryk2
1
Department of Mechanical Engineering, TU Darmstadt, Germany
2
Department of Computer Science, TU Darmstadt, Germany

Abstract. Quadrotor UAVs have successfully been used both in re-

search and for commercial applications in recent years and there has
been significant progress in the design of robust control software and
hardware. Nevertheless, testing of prototype UAV systems still means
risk of damage due to failures. Motivated by this, a system for the com-
prehensive simulation of quadrotor UAVs is presented in this paper. Un-
like existing solutions, the presented system is integrated with ROS and
the Gazebo simulator. This comprehensive approach allows simultane-
ous simulation of diverse aspects such as flight dynamics, onboard sensors
like IMUs, external imaging sensors and complex environments. The dy-
namics model of the quadrotor has been parameterized using wind tunnel
tests and validated by a comparison of simulated and real flight data. The
applicability for simulation of complex UAV systems is demonstrated us-
ing LIDAR-based and visual SLAM approaches available as open source
software.

1 Introduction
Quadrotor UAVs have successfully been used both in research and for commercial
applications in recent years. Impressive results have been shown using quadrotor
aircraft of various sizes and in different scenarios. The inherently instable nature
of quadrotor flight can lead to loss or damage of UAVs easily, especially when
evaluating prototype soft- or hardware. The lack of a simulation environment for
quadrotor UAVs that covers realistic flight dynamics, camera and range sensors
and an easy integration with existing robotic middleware solutions motivated
this work. We present a comprehensive framework to simulate our quadrotor,
that has been developed during the last few years. It is based on the Gazebo open
source simulator and the Robot Operating System (ROS), that has become a de
facto standard in robotics research and facilitates integration of contributions by
other researchers. Common sensors for autonomous robots like LIDAR devices,
RGB-D and stereo cameras are already available for Gazebo and can be attached
to the robot, while plugins for other, more UAV-specific sensors like barometers,
GPS receivers and sonar rangers have been added as part of this work.
The remainder of this paper is organized as follows: After the discussion of
related work in section 2, section 3 presents the simulation model considering

I. Noda et al. (Eds.): SIMPAR 2012, LNAI 7628, pp. 400–411, 2012.
c Springer-Verlag Berlin Heidelberg 2012
 Comprehensive Quadrotor Simulation 401

geometry, flight dynamics and control and how the model is implemented in
Gazebo. Comparative results from flight tests and simulation runs as well as a
demonstration of applicability for evaluating high-level algorithms are presented
in section 4.

2 Related Work

As we aim at a comprehensive approach for simulation of quadrotor UAV sys-

tems, we provide an overview both of simulation/ground truth tracking ap-
proaches as well as quadrotor control approaches. Most approaches for quadrotor
simulation focus on vehicle dynamics for controller design, often using special-
ized tools like Matlab/Simulink [20]. Sometimes other tools like the Flightgear
open source simulation framework are used for visualization [18,9]. Using such
approaches, testing of sensor-based high level control and behaviors is not pos-
sible or requires significant additional implementation effort. Quadrotor UAVs
can be simulated using USARsim [7], but a recently published ROS integration
[2] is of limited scope. In [1], the use of a simulator also providing sensor data is
mentioned, but not made available for testing.
Several authors have proposed dynamics models for the simulation of quadro-
tor aircraft which are based on the same flight mechanical principles [10,3,9,11,20].
While dealing with different aspects in detail, none of them considers motor and
propeller dynamics, aerodynamics, external disturbances (e.g. wind), and noisy
sensor signals and state estimation in an integrated fashion.
Recently, external optical tracking for the acquisition of ground truth data has
been used with great success [8,17]. The installation of such systems however is
costly and often not feasible due to space constraints. Even if such a system is
available, testing of multi-UAV control approaches in simulation is advantageous,
as potential collisions or other faults incur no cost in simulation.

(a) (b)

Fig. 1. Mesh-based quadrotor model: (a): Model shown rendered in Blender. (b) Model
used in Gazebo. A Hokuyo UTM-30LX laser scanner is mounted below the main body.
402 J. Meyer et al.

3 Model Description and Simulation

As we aim at comprehensive simulation of all relevant components including low
level sensing, system dynamics and control, we provide an overview of these parts
independently. Gazebo provides a multi-robot simulation environment including
dynamics simulation, which is provided by the ODE or bullet physics engines.
While the simulator considers gravity, contact forces and friction by its own, it
does not cover aerodynamics and propulsion systems that are especially required
for aerial vehicles. A plugin systems enables the user to add custom controllers
for simulated robots and sensors or to influence the environment.

3.1 Geometry
The robot geometry has been modeled using the open source software Blender.
To be able to provide different colors (both texture or material based) for the
model, the visual geometry is provided using the COLLADA format, while the
collision geometry is modeled as a .stl mesh. The model is designed to have a
low polygon count and still retain the relevant aspects of quadrotor geometry.
As a trade-off between visual fidelity, collision detection and dynamics modeling
needs, the propellers are modeled as discs.

3.2 Dynamics Model

One of the main advantages of the quadrotor concept is the simplicity of its
propulsion and steering system, consisting only of four independent motors and
propellers with fixed pitch, where each pair of opposite propellers rotates in one
direction to avoid yaw torque during roll and pitch movements. As a result,
the overall system dynamics are mainly determined by the thrust and torque
induced by the individual motor/propeller units.

Flight Dynamics. The movement of a rigid body can be described by the sum
of all forces F and torques M acting on the vehicle:

p˙n = v n (1a)
v˙n = m−1 Cnb F (1b)
ω˙b = J −1 M (1c)
Here, p and v are the position and velocity of the body’s center of gravity in
n n

the (inertial) navigation coordinate system, ω b is its angular rate given in body
coordinates and Cnb is the rotation matrix that transforms a vector from body
(index b) to navigation coordinates (index n).
The mass m and inertia J of the quadrotor need to be known and have
been estimated by weighing the individual components and using the geometric
model. The force vector F comprises motor thrust F M , drag forces F d and the
gravity vector F g . The torque vector M is divided into propulsion torque M M
and drag moments M d . Drag forces and moments are given by:
 Comprehensive Quadrotor Simulation 403

F d = −C d,F · Cbn · |v n − v nw | (v − v w ) (2a)

 
M d = −C d,M · ωb  ω b (2b)
with the diagonal drag coefficient matrices C d,F and C d,M and the wind vector
v nw . Finally, the gravity force is given by
 T
F g = m · Cbn · 0 0 ge . (3)

With these forces and torques resulting from self motion of any system in space
and the propulsion forces and torques described in the following section, the
vehicle movement can be obtained by solving equations (1).

Motor Dynamics. The propulsion system of our quadrotor UAV consists of

four brushless DC motors. The dynamic behavior of a brushless DC motor has
been derived from [12] with some simplifications. Assuming a very low inductance
of the motor coils, the current rise time can be neglected. The motor dynamic
behavior therefore simplifies to a P T1 element and is described by Eqs. (4) - (6).
In steady state the induced anchor voltage UA depends on the rotation speed
ω M and the anchor current IA :

UA = RA IA + Ψ ωM (4)

The electromagnetic torque Me for each motor is given by

Me = Ψ IA (5)

With the mechanical torque Mm and the motor inertia JM the change in rotation
speed can be calculated through:
 
1 1 Ψ
ω̇M = · (Me − Mm ) = · · (UA − Ψ ω M ) − Mm (6)
JM JM RA

The nonlinear term Mm describes the torque resulting from bearing friction as
well as load friction (i.e. drag) of the airscrew. It can be written as Mm = kT · T
where T is the thrust of a single airscrew [15] which is a a broad simplification
of a former approach [21] without loss of accuracy.

Thrust Calculation. In contrast to the former approach we now use a nonlinear

quadratic approximation for thrust calculation, similar to [9]. This approach
has been selected based on wind tunnel tests (cf. Fig. 2(a)) and is sufficiently
accurate as to not require the use of more complex thrust models [10]. With
the dynamic expression of the motor’s rotational speed ω M from equation (6)
it is straightforward to calculate the thrust force T for a single motor-airscrew
combination:

T = CT,0 ω2M + CT,1 v1 ω M ± CT,2 v12 (7)

404 J. Meyer et al.

·10−5 F1
4
M1
F4
M4
3
xb
thrust coefficient CT

2 yb F2

F3 mg
M2
1 zb

M3
xn
0 yn

−1
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
performance factor J zn

(a) (b)

Fig. 2. (a) Thrust coefficient CT of an airscrew as a function of its performance factor

J. Crosses mark wind tunnel measurements, while the solid line represents the ap-
proximation. (b) Sketch of the quadrotor to show the different coordinate systems and
naming conventions.

Dividing the above equation by ω 2M and using the performance factor J = v1 /ωM
the thrust coefficient CT (J) is given by:

CT (J) = CT,0 + CT,1 J ± CT,2 J 2 (8)

where the parameters CT,i have been identified in wind tunnel test. A negative
v1 (meaning a falling quadrotor) results in a positive prefix of CT,2 . In Fig. 2(a)
the polynomial approximation of CT (J) is shown. For a quadrotor helicopter the
free stream velocity v1 in general is different for each of the rotors. It can be
calculated through geometric inspection of the vehicle shown in Fig. 2(b):
   
(v1 )i = − [0 0 1] · v b + ω b × ei · lM (9)

with the unit vectors

 T  T  T  T
e1 = 1 0 0 , e2 = 0 1 0 , e3 = − 1 0 0 , e4 = − 0 1 0

for the four different motors. lM is the distance between the geometric centers
of motors and quadrotor. With the coordinate system conventions shown in
Fig. 2(b) the following expression for the overall wrench of the quadrotor can be
determined:
⎡ ⎤ ⎡ ⎤
0 (F4 − F2 ) · lM
FM = ⎣
b
0 ⎦ MM = ⎣
b
(F1 − F3 ) · lM ⎦ (10)
−Σi=1
4
Fi −M1 + M2 − M3 + M4

The four single forces Fi are calculated by solving equation (7) while the moments
Mi are obtained through combining equation (4) and (5). The incorporation of
 Comprehensive Quadrotor Simulation 405

blade flapping effects which can be used to aid state estimation [4] is subject of
future work.
We implemented two plugins that calculate propulsion and drag forces acting
on the aircraft given the internal state of the vehicle, the four motor voltages
and the wind vector. The current wind can be specified as constant vector or
provided by an external model or from real log data. Gazebo then applies the
calculated forces and torques to the quadrotor body for each simulation step.

3.3 Sensor Simulation

As attitude, position and velocity cannot be measured directly, accurate models
are needed to simulate the signals from various sensors needed for estimating the
state of the UAV. These sensors have been implemented as independent Gazebo
plugins and can be attached to the model by including them in the robot URDF
description. The plugins accept parameters covering the error characteristics and
the WGS84 position, altitude and orientation of the Gazebo reference frame in
the world coordinate system wherever necessary.

Error Model. All sensors share a common first order Gauss Markov error
model [5], permitting simulation of sensors with different error characteristics.
Each simulated measurement y(t) at time t is given by
y = ŷ + b + wy (11a)
1
ḃ = − b + wb (11b)
τ
where ŷ is the true value or vector, b is the current bias and wy and wb are
independent, zero-mean white Gaussian noise variables. wy is additive noise
acting directly on the measurement and wb describes the characteristics of the
random drift with time constant τ .

Inertial Measurement Unit. The inertial measurement unit (IMU) is the

most important sensor for the stabilization of quadrotor flight as it measures
the angular velocities and accelerations of the vehicle body in the inertial frame.
Integration of these values provides a good reference of attitude and speed over
short time intervals with fast response times, but is not suitable for long-term
reference due to the significant drift of available low-cost sensors. Also note that
an observer onboard the vehicle cannot distinguish gravity from other external
forces and therefore the acceleration of the body in the world frame cannot be
measured directly without knowing the orientation of the body.

Barometric Sensor. For simulating the static pressure at the present altitude,
we use the International Standard Atmosphere (ISA) model as defined by the
International Civil Aviation Organization (ICAO), which describes the pressure,
temperature and density of the earth’s atmosphere under average conditions at
mid latitudes. The elevation of the simulation reference frame above mean sea
level and the simulated pressure (only required for the output of pressure values
in hPa) at sea level can be specified as parameters.
406 J. Meyer et al.

Ultrasonic Sensor. For controlling the height during the takeoff and landing
phases and for switching on and off the motors, the range estimate from an down-
ward pointing ultrasonic sensor is used. This device transmits short ultrasound
impulses and returns the distance corresponding to the first echo returned from
the ground or an object within it’s field of view. Available ultrasound sensors
have a maximum range of about 3 to 6 meters. The simulated ultrasonic sensor
uses the Gazebo ray sensor interface to determine ray-casting based distances to
world geometry. The distance value returned is the minimum of all rays (9 by
default).

Magnetic Field Sensor. The earth magnetic field serves as a reference for the
heading or yaw angle of the quadrotor. As using a single axis compass would lead
to significant errors with increasing roll and pitch angles, three-axis magnetome-
ters are commonly used for UAVs. With the assumption that the earth-fixed
magnetic field vector is constant within the area of operation, it is straightfor-
ward to calculate the body-fixed vector given the declination, inclination and
field magnitude. Deviation errors through interference from parts of the robot
itself are covered by the generic error model.

GPS Receiver. Pseudo range measurements and the resulting position and
velocity solution are influenced by different factors like the satellite ephemeris
errors, atmospheric errors or receiver errors [19]. These error sources are approx-
imated using the Gauss-Markov error model, with the parameters of our uBlox
receiver module having been determined experimentally. To reproduce the in-
terdependency of position and velocity errors we use the noise-affected velocity
measurement error (v GPS − v̂ GPS ) instead of wb in Eq. (11b) for the integration
of the position error. A more detailed consideration of GPS errors and especially
multipath effects in the vicinity of buildings is left for future work. To calculate
WGS84 coordinates from the simulated position and velocity in Cartesian coor-
dinates we use a simple equirectangular projection that is based on a flat world
assumption. This projection is accurate enough in the vicinity of the chosen
reference point and outside the polar regions.

3.4 State Estimation and Control

Although state estimation and control are not specific to simulation, both com-
ponents are required to close the loop between simulated sensor signals and the
resulting motor voltages required to stabilize and control the quadrotor.
For estimating the state of the system we use an Extended Kalman Filter
(EKF) to fuse all available measurements to a single navigation solution con-
taining the orientation, position and velocity of the vehicle as well as observable
error states like the IMU bias errors. This approach is usually referred to as
integrated navigation.
Our controller is implemented as a set of cascaded PID controllers, with the
inner loop controlling the attitude, yaw rate and vertical velocity and an outer
loop controlling the horizontal velocity, heading and altitude (Fig. 3). This
 Comprehensive Quadrotor Simulation 407

vx,d
b
PI ax,d PID
- θd -
v̂xn vx Mψ θ̂
MTψ g
pitch

ay,d φd
PI PID My
vy,d
b - - U1
v̂yn Mx
vy φ̂ roll U2
vz,d Fz Mixer
U3
vz,d Mz
pz,d PI PD U4
- -
p̂z height v̂z vz
ωz,d
ωz,d
ψd PI PD
- -
ψ̂ yaw ω̂z yaw rate

Fig. 3. The controller is realized through separate cascaded PID controllers controlling
roll and pitch movement, yaw rate and vertical velocity

approach assumes that each axis and the altitude can be controlled indepen-
dently, which is valid for moderate deviations from the hovering state. The out-
put of the inner loop are commanded torques and vertical thrust, which are
translated to motor voltages either by using a static mixture matrix or by feed-
ing them into an inverted model of the propulsion system presented in section
subsection 3.2.
For simulation we use exactly the same implementation as on the real quadro-
tor. It is based on the Open Robot Control Software (Orocos) toolchain [6],
which provides interfaces to ROS and executes tasks satisfying hard realtime
constraints on the onboard PC system. This software-in-the-loop approach offers
great flexibility for testing advanced control algorithms before the deployment on
the real vehicle and therefore minimizes the risk of damage or loss dramatically.
Implementation details can be found in previous publications [16].

4 Experiments

Different aspects of simulation are validated using experiments in this section.

We also show examples of comprehensive simulation scenarios using the flight
dynamics model as well as leveraging existing ROS open source software.

4.1 Validation of Dynamics Model

To validate the dynamics model, we let both the real and simulated UAV per-
form a test trajectory consisting of transitions between different velocities. All
measurable variables of the real quadrotor show the same characteristics as the
corresponding simulated counterparts. The power spectrum densities (PSD) of
408 J. Meyer et al.

2
100
|P (f )| in rad/s

vx in m/s
0
10−1

−2
ωx
10−2
100 101 0 5 10 15 20 25 30 35

2
100
|P (f )| in rad/s

vy in m/s
0
10−1

−2
ωy
10−2
100 101 0 5 10 15 20 25 30 35

2
100
|P (f )| in rad/s

vz in m/s

0
10−1

−2
ωz
10−2
100 101 0 5 10 15 20 25 30 35
f in Hz time in s

Fig. 4. Diagrams of simulated and measured angular and translational velocity. Dotted
lines represent measurements while solid lines are simulated data. The left side shows
the PSD of the angular rates. On the right side the estimated velocity both in simulation
and reality with the commanded speeds (dashed line) is shown.

(a) (b)

Fig. 5. Indoor SLAM simulation: (a): Screenshot of the GUI. On the left the Gazebo
simulation environment is visible. On the top right the view of the forward facing
camera is shown, with LIDAR point cloud and map data projected into the image.
A top down ortho view is visible on the bottom right (b) Final map generated after
teleoperation of the UAV through the scenario.
 Comprehensive Quadrotor Simulation 409

(a) (b)

Fig. 6. Visual SLAM simulation: (a): Calibration of camera system in simulation. (b)
Screenshot of PTAM being used for visual SLAM on a quadrotor hovering above a
simulated NIST standard arena for response robots.

the angular rates and and the velocities are shown in Fig. 4. The controller
and a dead time of about 15 ms cause the quadrotor to oscillate slightly with a
frequency of about 3 Hz which is easily visible in the frequency domain. Differ-
ences in velocity between simulation and reality are mainly due to a gusty wind
of about 5 m/s which was apparent during the outdoor tests. In simulation, we
therefore defined a constant wind of 5 m/s.

4.2 Example Scenarios

In this section, different example scenarios are shown, demonstrating the com-
prehensive nature of quadrotor simulation and the interfacing with other open
source ROS software. Instructions for reproducing all presented scenarios are
provided on the hector quadrotor 1 website on ros.org.

In- and Outdoor Flight Scenarios. We flew the simulated quadrotor through
two example indoor and outdoor worlds to evaluate the quality of high-level
sensor data. Using the estimated state or ground truth data, the quadrotor pose
can be visualized along with sensor data.
To demonstrate the applicability for indoor SLAM simulation, we deploy a
previously developed SLAM approach [14] on the quadrotor UAV. The Willow
Garage office environment is part of the Gazebo ROS package, demonstrating
the applicability and interoperability of the quadrotor simulation with existing
Gazebo environments. The quadrotor UAV is teleoperated using a gamepad for
this demonstration. As shown in Fig. 5(a), sophisticated visualization including
projection of visualization data into camera images is possible by leveraging
available ROS tools like rviz. The final map learned is shown in Fig. 5(b) and of
comparable quality to those learned in real world scenarios.
A video of outdoor flying is available online2 .
1
http://www.ros.org/wiki/hector_quadrotor
2
http://www.youtube.com/watch?v=9CGIcc0jeuI
410 J. Meyer et al.

Visual SLAM. To demonstrate simulated image based state estimation, we de-

ploy a modified version [22] of the original PTAM system [13] for visual SLAM.
As demonstrated in Fig. 6(a), checkerboard-based calibration of camera param-
eters can also be performed in simulation. Fig. 6(b) shows a screenshot of the
PTAM GUI while the simulated quadrotor UAV hovers above an example sce-
nario, successfully tracking features in the image and estimating the aircraft
pose. It should be noted that the default camera simulation in Gazebo is of
limited fidelity as it does not exhibit effects like motion blur.

5 Conclusion

We presented a framework for the simulation of quadrotor UAV systems em-

ploying ROS and the Gazebo simulator. The tight integration with existing (and
future) ROS tools permits the comprehensive simulation of quadrotor UAVs in-
cluding low level sensing, flight dynamics and external sensing using any sensor
available for Gazebo simulation. The level of detail can be adapted depending
on the application, e.g. by using ground truth data for control or bypassing the
propulsion model.

Acknowledgments. This work has been funded by the Research Training

Group 1362 “Cooperative, Adaptive and Responsive Monitoring in Mixed-Mode
Environments” of the German Research Foundation (DFG).

References

1. Achtelik, M., Bachrach, A., He, R., Prentice, S., Roy, N.: Autonomous navigation
and exploration of a quadrotor helicopter in GPS-denied indoor environments. In:
Robotics: Science and Systems Conference (2008)
2. Balakirsky, S.B., Kootbally, Z.: USARSim/ROS: A Combined Framework for
Robotic Control and Simulation. In: ASME 2012 International Symposium on
Flexible Automation (ISFA 2012). ASME (2012)
3. Bouabdallah, S., Siegwart, R.: Full control of a quadrotor. In: IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems (IROS), pp. 153–158 (Novem-
ber 2007)
4. Bristeau, P., Callou, F., Vissière, D., Petit, N., et al.: The navigation and con-
trol technology inside the AR Drone micro UAV. In: 18th IFAC World Congress,
Milano, Italy, pp. 1477–1484 (2011)
5. Brown, R., Hwang, P., et al.: Introduction to random signals and applied Kalman
filtering. Wiley, New York (1992)
6. Bruyninckx, H.: Open robot control software: the OROCOS project. In: IEEE
International Conference on Robotics and Automation (ICRA), vol. 3, pp. 2523–
2528. IEEE (2001)
7. Carpin, S., Lewis, M., Wang, J., Balakirsky, S., Scrapper, C.: USARSim: a
robot simulator for research and education. In: IEEE International Conference
on Robotics and Automation (ICRA), pp. 1400–1405 (2007)
 Comprehensive Quadrotor Simulation 411

8. Ducard, G., D’Andrea, R.: Autonomous quadrotor flight using a vision system
and accommodating frames misalignment. In: IEEE International Symposium on
Industrial Embedded Systems (SIES), pp. 261–264. IEEE (2009)
9. Goel, R., Shah, S., Gupta, N., Ananthkrishnan, N.: Modeling, Simulation and
Flight Testing of an Autonomous Quadrotor. In: IISc Centenary International Con-
ference and Exhibition on Aerospace Engineering, ICEAE, Bangalore, India, pp.
18–22 (2009)
10. Hoffmann, G.M., Huang, H., Wasl, S.L., Tomlin, E.C.J.: Quadrotor helicopter flight
dynamics and control: Theory and experiment. In: AIAA Guidance, Navigation,
and Control Conference (2007)
11. Huang, H., Hoffmann, G., Waslander, S., Tomlin, C.: Aerodynamics and control
of autonomous quadrotor helicopters in aggressive maneuvering. In: IEEE Inter-
national Conference on Robotics and Automation (ICRA), pp. 3277–3282 (May
2009)
12. Isermann, R.: Mechatronische Systeme: Grundlagen (German Edition). 1. Auflage
1999, 1. korrigierter Nachdruck - Studienausgabe edn. Springer (December 1999)
13. Klein, G., Murray, D.: Parallel tracking and mapping for small AR workspaces. In:
6th IEEE and ACM International Symposium on Mixed and Augmented Reality
(ISMAR), pp. 225–234. IEEE (2007)
14. Kohlbrecher, S., Meyer, J., von Stryk, O., Klingauf, U.: A Flexible and Scalable
SLAM System with Full 3D Motion Estimation. In: IEEE International Symposium
on Safety, Security and Rescue Robotics (SSRR). IEEE, Kyoto (2011)
15. Leishman, G.: Principles of Helicopter Aerodynamics, 2nd edn. Cambridge
Aerospace Series. Cambridge University Press (April 2006)
16. Meyer, J., Strobel, A.: A flexible real-time control system for autonomous vehicles.
In: 41st International Symposium on Robotics (ISR) and 6th German Conference
on Robotics (ROBOTIK). VDE (2010)
17. Michael, N., Mellinger, D., Lindsey, Q., Kumar, V.: The GRASP Multiple Micro-
UAV Testbed. IEEE Robotics Automation Magazine 17(3), 56–65 (2010)
18. Qiang, Y., Bin, X., Yao, Z., Yanping, Y., Haotao, L., Wei, Z.: Visual simulation sys-
tem for quadrotor unmanned aerial vehicles. In: 30th Chinese Control Conference,
pp. 454–459 (July 2011)
19. Rankin, J.: An error model for sensor simulation GPS and differential GPS. In:
Position Location and Navigation Symposium, pp. 260–266. IEEE (1994)
20. Rodić, A., Mester, G.: The Modeling and Simulation of an Autonomous Quad-
Rotor Microcopter in a Virtual Outdoor Scenario. Acta Polytechnica Hungar-
ica 8(4) (2011)
21. Sendobry, A.: A Model Based Navigation Architecture for Small Unmanned Aerial
Vehicles. In: European Navigation Conference. Royal Institute of Navigation (RIN)
(November 2011)
22. Weiss, S., Scaramuzza, D., Siegwart, R.: Monocular-SLAM–based navigation for
autonomous micro helicopters in GPS-denied environments. Journal of Field
Robotics 28(6), 854–874 (2011)