Vous êtes sur la page 1sur 72

JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO.

1, JANUARY 2009 43

Developing a Low-Cost Autonomous Indoor


Blimp
P. González1, W. Burgard2, R. Sanz1 and J.L. Fernández1

inspection of power lines and pipelines, mineral and


Abstract— This paper describes the design of an autonomous archaeological site prospection, law enforcement and
blimp-based robot and its navigation system. The robot was based telecommunication relay systems [1]. Blimps are well-suited
on a commercial kit and its dimensions were suitable for use in for these applications because their ability to remain stationary
indoor environments. Our main goal was to develop a simple and
safe model for evaluating different autonomously controlled
for long periods of time in the air enables data to be gathered.
navigation techniques. Due to the special requirements of this Blimps can also be used for research purposes in a variety of
application, two specific electronics boards to control the blimp applications including ecological, biodiversity and climate
and to communicate with the PC ground station were designed research and monitoring in different environments [2].
and two different altitude controllers and also a controller to Our primary interest was the development of a low-cost
maintain distance from obstacles were implemented. Finally, blimp designed to operate autonomously in indoor
comparative results on both altitude controllers are presented.
environments where different control strategies and navigation
Index Terms— Blimp, autonomous robot, robot navigation, paradigms are tested and evaluated. The design of a blimp
altitude controller, control avoidance controller, fuzzy logic imposes certain restrictions, primarily because of its limited
control. payload capability, given that a blimp relies on its neutral
buoyancy to stay afloat. A key challenge was to build an
I. INTRODUCTION electronic board that was sufficiently light to be carried on
board the blimp. Electronic components were selected to fit
N on-rigid airships, also known as blimps, are basically
unmanned aerial vehicles (UAVs) that use gas (usually
helium) balloons. In contrast to a rigid airship, a blimp has no
our main navigation requirements including limited
autonomous navigation capabilities.
internal structure to maintain the shape of its hull envelope. This paper is structured as follows. Section 2 discusses
Rather, its shape is maintained by a higher pressure of the gas. related works. The commercial blimp selected and other main
The only rigid components are the driving elements, the fins components are described in detail in Section 3. Section 4
and the gondola attached to the envelope. describes the design and implementation of the navigation
Unmanned blimp robots can be used in both indoor and system, which basically consists of a fuzzy logic obstacle
outdoor environments. The buoyancy force provides an avoidance controller and an altitude controller. Two different
energy-free form of lift, offering a non-traditional approach to control techniques were experimented with for the altitude
long-duration missions for which conventional aircrafts are not controller: a simple proportional–integral–derivative (PID)
well-suited. Miniaturization of sensors and actuators and the linear controller and a fuzzy logic controller. The experimental
development of long-duration batteries have also opened up results for these systems are also compared in this section.
opportunities for further progress in the development of these Finally, Section 5 summarizes our conclusions.
small-scale autonomous vehicles.
The first rigid airships, which were constructed in the early II. RELATED WORKS
20th century, consisted of a balloon with a metal frame Several researchers have recently developed autonomous
covered by fabric and filled with a gas (helium or hydrogen). robotic systems based on blimps and studied appropriate
These airships were mainly used in wars for military aerial control paradigms. Much of this research is devoted to large-
exploration and transportation. Nowadays, however, they are scale systems, with payloads of kilograms and therefore
mainly used for advertising and aerial filming. Nevertheless, capable of including a number of sensors (such as cameras)
they have great potential in terms of applications such as and remaining airborne for long periods of time. Elfes and
search and rescue missions, traffic monitoring, urban planning, colleagues [1] presented arguments that favor blimps over
airplanes and helicopters as ideal platforms for standard aerial
This work was partially funded by Spanish Ministerio de Ciencia y exploration missions. Kantor and colleagues [2] discussed the
Tecnología (DPI2005-06210) and the Xunta de Galicia (PGIDIT06PXIC use of solar energy as a renewable source of power for airships
303194PN).
1 using an outdoor blimp. Hygounenc and colleagues [3]
Dept. of Systems Engineering and Automation, University of Vigo,
focused on flight control and terrain mapping issues in
Spain.
2 cooperation between ground and aerial robots.
Dept. of Computer Science, University of Freiburg, Germany.
44 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

Other authors described the use of blimps in indoor Due to the limitation on payload, onboard hardware could
environments. Motoyama and colleagues [4] designed an not be equipped with sensors to measure absolute vehicle
autonomously controlled indoor blimp and an action-value position and this limited the autonomous navigation
function for motion planning based on a potential field capabilities of our vehicle (as described below).
method, evaluating its effectiveness in a simulated The main characteristics of the selected blimp components
environment [5]. Geoffrey and colleagues [6] used a and the design of other elements are described immediately
commercial indoor blimp, concluding that the vertical motor below.
was severely underpowered in tele-operated control. These
A. Blimp selection
researchers used a commercial wireless board to send the
sensor measurements to the ground computer and a separate Payload, which depends on balloon volume, is a critical
board to control the motors. A PC running Linux was used to aspect that constrains the choice of other onboard components
process sensor data and send control signals to the blimp and a such as the battery.
wireless communication unit (WCU) was used for sensor data The blimp that we selected was a hobby radio-controlled
communication and blimp-side servo control. Hydrogen and (RC) blimp from Plantraco, which comes with an RC
helium were used to increase the payload capacity. transmitter than can be connected via USB to a PC and
We used the same type of blimp as Geoffrey and colleagues controlled via a connection to a TCP socket. It has a 52” non-
[6], but with a different vertical motor, as a more powerful rigid hull made of a light material, achieving 200 grams of
motor and a big propeller were necessary to control altitude so estimated payload capacity (Figure 1) [9] [10]. This size is
as to have full altitude control in each indoor environment in very appropriate for indoor laboratory applications. It can fly
which the vehicle was tested. We also developed a lighter in a corridor with people and enter rooms through standard-
specific board to control the motors and sensors and thus had sized office doors. The payload capacity of this blimp was
more payload capacity for additional components. We only considered adequate for our purposes.
used helium, because hydrogen is flammable and so is very The blimp has a tri-turbofan gondola with three light DC
dangerous in indoor environments. micro-motors: a vertical motor allows altitude to be regulated
An important navigation problem is automatic control of and the other two motors control speed and rotation. Since the
altitude and of horizontal movement. If the blimp can be original vertical motor does not have enough power to
maintained at a specific altitude, it can be moved in a properly control altitude, we replaced it with a Futaba 3003
horizontal plane. Kadota and colleagues [7] used PID servo motor and a bigger propeller blade. With all these
controllers to control blimp altitude and horizontal movement, changes the gondola components weighed only 55 grams,
arguing that blimp trajectory could be unstable in the vertical leaving 74 grams free for other hardware components
direction. We also used PID and fuzzy logic controllers to including the battery (Table I).
control blimp altitude and evaluated the performance of the TABLE I
control systems developed using the controllers in two BLIMP COMPONENT WEIGHTS
different environments. Component Weight (g)
A second important navigation problem for blimps (and for Envelope (52"x37") 60.5
autonomous mobile robots in general) is obstacle detection and Gondola components 55.0
collision avoidance. Green and colleagues [8] used an infrared Fins and propellers 10.5
sensor to detect obstacles: when the collision avoidance system Onboard hardware 39.0
detects an obstacle, the blimp turned 180 degrees to avoid Battery 35.0
collision. The collision avoidance system does not take Total blimp weight 200.0
account of blimp dynamics, however; consequently, we used
an ultrasonic sensor to measure the distance to potential
obstacles and implemented a fuzzy logic controller to avoid
collisions.

III. BLIMP DESIGN


The design of an autonomous blimp has certain restrictions
arising from the assembled hardware. Various issues were
evaluated simultaneously in terms of making appropriate
choices. For a blimp system, the higher the volume of the
envelope—that is, the higher the ascending force—the higher
the possible payload. However, a blimp for indoor applications
has to be fairly small. For our blimp system, the goal was to
minimize both the size of the blimp and the weight of the
necessary onboard hardware.
Fig. 1. The commercial blimp selected for this research, with a tri-
GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 45

turbofan gondola at its base. Balloon size was 52”. The vertical motor and the onboard wireless communication link (see Figure 3).
controls altitude and the side motors control horizontal movements.
The main components are:
Power supply unit. It consists of a power regulator which
The choice of battery was another key aspect because of the provides two stabilized voltage levels: 8V and 5V and a
weight issue: a very light battery that still provided enough battery circuit which prevents total discharge of the battery.
current and autonomy was required. We selected a 350mA Microcontroller unit (MCU). It is based on a PIC
lithium polymer battery based on a new technology. Weighing microcontroller, which remains the best balance of cost,
only 35 grams, it supplies a maximum current of 3A and processing power, complexity, and power consumption. The
provides around 40 minutes of autonomy for the blimp. The PIC 16F873 microcontroller from Microchip appears to be a
nominal voltage level of 11V was too high for the electronic good choice. It has USART, analog ports, and I2C ports
components, and so a power management circuit to provide required to connect sensors and motor drivers. Some other
the required voltages (8V and 5V) was designed. A circuit to inputs and outputs are used to communicate with the wireless
control voltage level was also added to avoid damage to the transceiver.
battery when the voltage was under 9V.
B. Ultrasonic sensors
Two ultrasonic sensors provided the autonomous blimp with
information on the environment. An ultrasonic, lightweight
SRF05 sensor (with a resolution of 1 mm and a very narrow
beam) was mounted facing downwards at the bottom of the
gondola to measure the distance from the blimp to other
objects. Sensor measurements were integrated by means of a
Kalman filter which sequentially estimated blimp altitude. The
other ultrasonic sensor, located in the forward-facing part of
the balloon, was used by the obstacle avoidance controller. In
this case, we selected a lightweight SRF10 ultrasonic sensor
with an operating range of up to 6 meters that could be
connected to the microcontroller via a standard I2C bus
interface.
C. Electronic components
An electronic board was custom-designed for this
application because no commercial board met with our
requirements (Figure 2). An appropriate communication
system was also designed and implemented. The main reasons
motivating this approach were the following:
Fig. 2. Block diagrams of the gondola onboard and PC interface cards.
• A new motor speed control was necessary, firstly, because
the blimp could not be controlled using the original circuit, Motor drivers. They are necessary to control the speed of
which only worked at maximum speeds, and secondly, each motor. The drivers are based on a L293B circuit, which
because we could not control the new vertical motor with offers 1A per channel and can modulate the voltage and
the original board. control the motor speed. The pulse-width modulation (PWM)
• Good wireless communication coverage with a lightweight is directly controlled by the microcontroller.
circuit was desirable, so lighter components for Wireless communication unit (WCU). It is employed to
communications between the vehicle and the PC ground transmit data between the blimp and the PC ground station. It
station were selected, given that a wireless access point or consists of a serial wireless modem that works in the 400 MHz
a wireless router could not be used because of the limited band, and it is based on an ER400TRS transceiver. It also
payload capability. includes a buffer to protect the transceiver. Data are
• A bidirectional communication link to send and receive transferred in real time at a speed of 19,200 bps. The WCU
data from the blimp was required in order to be able to sends sensor data to the ground PC and receives servo
send data to and from the blimp and the PC ground station positions from the ground PC to control the blimp (Figure 2).
(in other words, we needed to close the control loop). PC interface card. The blimp was remotely controlled using
For the above reasons, two specific boards were designed: a a customized computer board that also contained a WCU. We
gondola onboard card and a PC interface card. also developed a software package module with generic
functions that enabled easy control of the blimp by the
Gondola onboard card. This board included all the programmer.
electronic components necessary to control the three motors
46 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

PID and fuzzy logic controllers. PID parameters can be


experimentally adjusted on the basis of well-known methods
and fuzzy logic controllers can be easily tuned on the basis of
trial and error.
Our control system was composed of two different
controllers that would ensure safe autonomous navigation in
indoor environments: an altitude controller and a collision
avoidance controller. Both controllers were implemented in
the PC ground station using the distance measurements sent by
the WCU, with computed control signals sent back to the
onboard microcontroller.
The reference blimp altitude was specified in the program
user interface, with the altitude control algorithm endeavoring
to automatically maintain the blimp at this altitude. Controlling
Fig. 3. Electronic components of the gondola onboard card. vertical motion reduced blimp movements by one degree of
freedom. Two different altitude controllers were implemented
The ground computer could both write and read from the and tested.
WCU using these functions. The design of the function The collision avoidance controller enabled control of the
interface was such that the programmer was not required to horizontal movements of the blimp so as to avoid frontal
deal with serial port characteristics and the communication collisions. Only a fuzzy controller was implemented for this
protocol between the blimp and the computer. The interface purpose because the results for PID controllers were poor.
had ten user-friendly functions so that the programmer could When the front sensor did not detect an obstacle in its path, the
develop software to move the blimp in few minutes. controller navigated the blimp along a straight line; when an
obstacle was detected, the controller kept the blimp at a certain
D. Onboard software distance from the obstacle.
The onboard software processed the sensor measurements
A. PID altitude controller
and sent them to the PC ground station. These data were
received by the PC ground station and used to compute new The current altitude of the blimp was measured by the
control signals for the gondola micromotors. The WCU in the SRF05 ultrasonic sensor and then sent to the PC. Sensor errors
PC interface card sent control data to the onboard WCU, were corrected with a Kalman filter, thus obtaining the
which were used by the microcontroller to set the estimated altitude. The filter parameters depended on sensor
corresponding commands for the micromotors. characteristics, the dynamic model of the blimp and the
previous measurements, but in our case they were tuned
IV. BLIMP NAVIGATION SOFTWARE experimentally. We did not make use of the Kalman filter
variance because of real time restrictions in controlling motor
The navigation software provided limited autonomous
action.
operation of our blimp due to:
The first altitude controller implemented was a PID-type
• The lack of odometry. controller. Control actions were calculated as:
• Misalignment caused by wind gusts and temperature t de(t )
changes. u PID = K p e(t ) + K i ∫ e(t )dt + K d (1)
• The non-linear nature of propeller action.
0 dt
• Blimp operation in three dimensions.
where Kp, Ki and Kd were parameters experimentally calculated
The control system for our autonomous blimp was designed
using the Zieger-Nichols method and uPID was the command
basically to keep the blimp moving in a straight line if its path
signal to the vertical propeller, responsible for up-and-down
was clear, maintaining the desired horizontal speed and
movements of the aerial vehicle.
reference altitude. When a head-on obstacle was detected the
control system attempted to maintain a certain distance from it.
Assuming a low speed, the control problem could be
decoupled in two sub-systems describing motions in both
vertical and horizontal planes [11].
Although some blimp control approaches are based on the
vehicle mathematical model (see [11] and [12], for example),
we preferred not to implement an analytical controller because
the blimp has complex dynamics due to its nonlinear
characteristics and environmental influences (air gusts,
temperature, altitude references, etc). We implemented both
GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 47

a)

Fig 5. Experimental environment 1: a corridor in Building 79, Freiburg


University.

The experiments have shown diverse results using the same


controller in these two environments. An overall good
performance can be observed during the experiments carried
out in environment 1. However, the same PID controller in
environment 2 has shown an oscillating behavior with an
important deviation from the mean. Although it can reach the
b)
altitude reference, there is a significant error as it is shown in
Fig 4. Behavior of the PID altitude controller when the altitude Figure 4b.
reference is set to 1 meter. a) Experiments in environment 1. b) Note that the controller response in the second environment
Experiments in environment 2.
could be improved by tuning PID parameters. In fact, these
Blimp altitude in Figure 4 is depicted in red, while average parameters had to be recalculated to take account of changes
values are represented in blue. The figure shows the blimp PID in environmental conditions while the blimp was navigating.
altitude controller behavior for two different scenarios. The
first was a corridor in Building 79 of Freiburg University
(Figure 5), while the second was a computer laboratory in the
same university (Figure 6). The computer laboratory had a
door that, when opened, could alter the draught inside. Eight
similar experiments were performed in both environments, all
commencing with the blimp lying on the ground and then
moving to its target position, one meter above the ground.

Fig 6. Experimental environment 2: a computer laboratory in Freiburg


University.

B. Fuzzy altitude controller


A non-linear altitude controller based on fuzzy logic was
also implemented. Fuzzy logic uses fuzzy sets to model
48 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

designer knowledge about the system to control, with


knowledge representation modeled using fuzzy rules. This The fuzzy logic controller is characterized by a set of
kind of controller has several advantages because it does not linguistic variables and fuzzy if-then rules. All input and
need to recalculate parameters when environmental conditions output linguistic variables have a finite number of linguistic
changes. values with membership functions that are empirically defined
A fuzzy controller is composed of a knowledge base and an after exhaustive simulation studies. The linguistic values
inference engine. The knowledge base contains rules and representing the linguistic variable altitude error and vehicle
linguistic variable descriptions, while the inference engine speed are very negative (VN), negative (N), zero (Z), positive
generates a control action as a function of state variable values (P) and very positive (VP). Linguistic values describing the
in a given time instant. output linguistic variable are very negative 3 (VN3), very
The altitude fuzzy logic controller in our blimp had two negative 2 (VN2), very negative (VN), negative (N), zero (Z),
inputs: altitude error and estimated current vertical speed. positive (P), very positive (VP), very positive 2 (VP2) and very
Altitude error was the difference between the desired altitude positive 3 (VP3). Their corresponding membership functions
and current altitude (Figure 7). A change in altitude error are shown in Figure 8.
indicated whether the aerial vehicle was approaching the Fuzzy rules describe the controller behavior in terms of
reference altitude or moving away from it. The controller relationships between input and control variables. A rule is
output was the vertical motor command. Note that the fuzzy usually of the type:
controller structure differed from the PID controller in that the If x1 is A1 and x2 is A2 then y is B,
latter has only one input, which ensures better and more where xi and y are, respectively, input and control linguistic
accurate altitude control. variables, and Ai and B are linguistic terms.
The altitude controller can be described with a small set of
rules, for example:
If altitude error is negative and vertical speed is very
positive then motor command is positive.
Figure 9 depicts, in tabular form, the fuzzy rule set used to
generate the motor command. These rules endeavor to
maintain the blimp at a specific height and represent human
expertise on how to control the system. Note that the rule
descriptions were developed on the basis of multiple
experiments.
Figure 10 depicts one of the situations described in the table
of Figure 9. In this situation, the blimp is descending and the
altitude error is significant. In the position shown in the figure,
the input variables have the linguistic labels VN for velocity
and VP for altitude error, respectively. So, the command signal
Fig. 7. Variable relations regarding altitude control of the blimp. Zref is the value must be VP3, as shown in the fuzzy rule table depicted
altitude reference; Z is the estimated height using a Kalman filter. Other
variables are shown on the left. in Figure 9.
Note that one set of rules describes when motor action must
be null. These rules have a dead band to avoid continuous
motor action.
Experimental results for the fuzzy altitude controller are
depicted in Figures 11a and 11b. The blimp showed only slight
oscillations in both environments and deviations from the
mean, shown in blue, were very small. A significant number of
tests carried out in different circumstances led to similar good
results in both environments.

Fig 8. Fuzzy linguistic variables for the fuzzy altitude controller.


GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 49

Fig 9. Fuzzy rules for altitude control.

b)
Fig. 11. Behavior of the fuzzy altitude controller when the altitude
reference is set to 1 meter. a) Experiments in environment 1. b)
Experiments in environment 2.

C. Collision avoidance controller


The collision avoidance system should cause the vehicle to
stop reliably when the frontal distance sensor detects an
obstacle in the vicinity. In such circumstances, the horizontal
navigation speed is changed by the collision avoidance
controller. Only the frontal sensor no longer detects an
obstacle close to the blimp, horizontal blimp speed is reset to a
certain value. Different approaches can be used to implement
the collision avoidance controller.
We again chose a fuzzy logic approach, based on a PID
controller which demonstrated oscillating behaviors in almost
Fig. 10. Behavior of the fuzzy altitude controller when the blimp all the experiments carried out in both environments. The goal
descends.
of this controller was to keep the blimp at a safe distance from
frontal obstacles. Controller inputs were frontal distance and
estimated speed and controller output was a speed index for
the horizontal motors.
Fuzzy rules for the collision avoidance controller are shown
in Figure 12. Figure 13 depicts a possible situation in which
the blimp is very close to an obstacle detected by the frontal
sensor. The blimp is heading straight for a vertical wall; speed
is V=VN and the distance error value is E=VP (near the wall).
In this situation, which is critical for the blimp as it is likely to
collide with the wall, the fuzzy collision avoidance controller
transmits the maximum control command (A=VP3) to the
motors, thereby transmitting the maximum opposite power to
ensure that the blimp avoids the collision. However, if this
speed is used for a long time the blimp develops inertia and it
a) becomes impossible to stop it in the desired reference. When
the speed is reduced so as to be comprised within the linguistic
variable V=N, the blimp does not need maximum power and
control action is reduced to A=VP2. These few rule examples
illustrate how the controller works.
50 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

sensors were used for navigation; the small payload also did
not permit inclusion of an inertia sensor or miniature wireless
camera usually used for global position calculation [6].
We also implemented onboard and PC programs to control
the blimp with the PC ground station, with the communication
software module capable of receiving onboard sensor data and
sending commands to the blimp (as well as facilitating future
application developments).
Two different controllers were implemented. An altitude
Fig. 12. Fuzzy rules for the collision avoidance controller.
controller maintained the blimp at a certain distance from the
floor and a second controller avoided obstacles in the path of
the blimp. The altitude controller was implemented using two
different approaches: PID and fuzzy logic. A first experiment
was designed to compare the two controllers operating in the
same conditions, so as to determine which controller was
better suited to controlling blimp altitude.
In the first scenario, both controllers were tested on an
alternating basis in one environment (a corridor), an
experiment that was repeated eight times. Similarly, the blimp
Fig. 13. Behavior of the fuzzy collision avoidance controller when an was tested in a second environment (a computer room) with
obstacle is detected. the same perturbations on the blimp for both controllers.
Comparative results show that the fuzzy logic controller
produced balanced behavior in either of the two environments.
The PID altitude controller performed slightly better than the
fuzzy logic controller in environment 1 (see Figure 14a) but its
performance in environment 2 was significantly poorer than
that of the fuzzy logic controller. A better PID controller in
other environments would require online adjustment of PID
parameters. Using a fuzzy logic controller, on the other hand,
does not affect the behavior of the blimp too much and the
controller parameters do not require modification.
A second fuzzy logic controller, designed to avoid head-on
collision with obstacles while the vehicle was navigating in the
indoor environment, showed good experimental performance.

Fig. 14. Behavior of the blimp in a corridor. The blue line is the safety
distance from the floor.

Figure 14 shows the blimp distance to a vertical wall while


it is navigating in environment 1. Initially, it maintained a
constant horizontal speed. When it approached a wall at the
end of this corridor, the obstacle avoidance controller reduced
the speed accordingly. In the experiment, the vehicle
maintained a security distance of about 0.2 meters away from
the obstacle without crashing at any time.

V. CONCLUSIONS
a)
In this paper we have described the construction of a small
blimp based on a commercial kit, with the main goal of
designing and implementing different control navigation
techniques. It was necessary to change the vertical motor of
the commercial blimp and develop onboard control hardware,
including a wireless radio connection with a PC ground
station. Due to the vehicle’s small payload only two distance
GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 51

[2] G. Kantor, D. Wettergreen, J. P. Ostrowski, and S. Singh, “Collection of


environmental data from an airship platform,” in Proceedings SPIE Vol
4571, Sensor Fusion and Decentralized Control in Robotic Systems,
2001.
[3] E. Hygounenc, I.-K. Jung, P. Soueres and S. Lacroix, “The autonomous
blimp project of LAAS-CNRS: Achievements in flight control and
terrain mapping,” The International Journal of Robotics Research, vol.
23, n. 4-5, pp. 473-511, 2004.
[4] K. Motoyama, H. Kawamura, M. Yamamoto, and A. Ohuchi,
“Development of autonomous blimp robot with intelligent control”, in
Entertainment Computing: Technologies and Applications, Eds. R.
Nakatsu and J. Hoshino, Kluwer Academic Pub. 2003, pp. 191-198.
[5] K. Motoyama, H. Kawamura, M. Yamamoto, and A. Ohuchi, “Design
of evaluation function in motion planning for autonomous balloon
b) robot”, in Proceedings of 2003 Asia Pacific Symposium on Intelligent
Fig. 13. Comparative results for the PID and fuzzy logic controllers. a) and Evolucionary Systems, 2003.
Experiments in environment 1. b) Experiments in environment 2. [6] G. A. Hollinger, Z. A. Pezzementi, A. D. Flurie, and B. A. Maxwell,
“Design and construction of an indoor robotic blimp for urban search
and rescue tasks”, Swarthmore College Senior Design Thesis, Spring
Note that we could not measure the absolute position of our 2005.
vehicle with the onboard sensor due to the low payload of the [7] H. Kadota, H. Kawamura, M. Yamamoto, T. Takaya, and A. Ohuchi,
“Vision-based positioning system for indoor blimp robot”, Proceedings
blimp. This limited blimp navigation in a desired direction of the 4th International Conference on Advanced Mechatronics-
without global path planning. For this reason we are Toward Evolutionary Fusion of IT and Mechatronics, 2004.
developing a new design with a bigger balloon than described [8] W. E. Green, K. W. Sevcik and P. Y. Oh, “A competition to identify key
in this paper that will enable new vehicle control functions to challenges for unmanned aerial robots in near-earth environments”, in
Proceedings. 12th International Conference on Advanced Robotics
be incorporated [13]. ICAR '05, July 2005.
[9] Balloon shop. http://www.heo.com.
ACKNOWLEDGMENTS [10] Plantraco shop. http://www.plantraco.com.
[11] S. van der Zwaan, M. Perrone, A. Bernardino, and J. Santos-Victor,
We would like to thank the people of the University of “Control of an aerial blimp based on visual input”, in Proceedings of
Freiburg (Germany) for their assistance during Pablo the 8th International Symposium on Intelligent Robotic Systems
SIRS’00, 2000.
González’s visit to the Autonomous Intelligent Systems [12] N. Rooz and E. N. Johnson, “Design and modeling of an airship station
Laboratory. holding controller for low cost satellite operations,” in Proceedings of
the AIAA Guidance, Navigation, and Control Conference and Exhibit,
REFERENCES August 2005.
[13] A. Rottmann, M. Sippel, T. Zitterell. W. Burgard, L. Reindl, and C.
[1] A. Elfes, S. Siquiera, M. Bergerman, and J. Guimaraes, “A semi- Scholl.” Towards an Experimental Autonomous Blimp Platform”, in
autonomous robotic airship for environmental monitoring missions,” in Proceedings of 3rd European Conference on Mobile Robots (ECMR
Proceedings of the 1998 IEEE. International Conference on Robotics '07), September 2007.
and Automation, 1998.
52 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/322308785

Design and Autonomous Control of a Solar-Power Blimp

Conference Paper · January 2018


DOI: 10.2514/6.2018-1588

CITATION READS
1 115

3 authors:

Changhuang Wan Nathaniel Kingry


The Ohio State University Collins Aerospace
7 PUBLICATIONS   4 CITATIONS    8 PUBLICATIONS   22 CITATIONS   

SEE PROFILE SEE PROFILE

Ran Dai
The Ohio State University
65 PUBLICATIONS   351 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Wearable Technology View project

All content following this page was uploaded by Changhuang Wan on 24 January 2019.

The user has requested enhancement of the downloaded file.


Design and Autonomous Control of
a Solar-Power Blimp

Changhuang Wan∗, Nathaniel Kingry† and Rai Dai‡


Iowa State University, Ames, IA, 50011, USA

In this paper, we present the design, modeling, and control law development of a solar-
powered blimp. Blimps offer flexibility and low cost to perform long endurance missions,
such as agricultural, surveillance and search and rescue missions. While their benefits
are apparent, blimps are subject to a variety of perturbations during flight. A six DOF,
nonlinear kinematic and dynamics model for a customized blimp is developed. These
nonlinear equations are then linearized to allow for developing real-time control laws. The
developed model and controller is simulated and then experimentally tested in an indoor
environment.

Nomenclature
A, B State matrix and input matrix of the linear model, respectively
CIB Rotation matrix from body frame to inertial frame
FΩ Rotation matrix for angular velocity
Jb Jacobian Matrix
F Force vector
T Torque vector
x Blimp state vector
u Blimp control vector
M General total mass matrices
m Mass, kg
B Buoyancy of the blimp
G Gravity of the blimp
φ roll angle
θ pitch angle
ψ yaw angle
U linear velocity vector expressed in body frame
Ω angular velocity expressed in body frame

I. Introduction
In the past decades, there has been extensive research in many areas of unmanned aerial vehicles (UAVs).
UAVs have played an indispensable role due to the development in a variety of engineering technology fields
and the rise of challenging demands.1 Blimps are a type of light-than-air UAVs which rely on their neutral
buoyancy, rather the aerodynamic lift, to stay afloat in the air. Therefore, light-than-air UAVs intrinsically
have better stability over other platforms in terms of maintaining the lofting status.2 Unlike rigid airships,
blimps maintain their shape through the high pressure acting on their envelope rather than the internal
∗ Graduate Student, Aerospace Engineering Department, Iowa State University, Ames, IA, 50011 Email: cwan@iastate.edu
† Graduate Student, Aerospace Engineering Department, Iowa State University, Ames, IA, 50011 Email: Nethank-
ing@iastate.edu
‡ Assistant Professor, Aerospace Engineering Department, Iowa State University, Ames, IA, 50011 Email: dairan@iastate.edu

1 of 12

American Institute of Aeronautics and Astronautics


structure.3 The main advantages of blimps compared with other flying robots are the low power consumption,
the safety for taking off and landing, and the long endurance capability. For scenarios where the vehicle is
required to remain aloft over a specific area for a long duration, blimps offer lower power consumption and
long endurance over airplanes and helicopters, which may require refueling to maintain the same endurance.2
Therefore, blimps have great potential for applications in many fields, such as advertising, search and rescue,
traffic monitoring, inspection, exploration, law enforcement, and telecommunication relay systems.3, 4
However, the challenges of applying blimps in real world missions lay on the load capacity, dynamic mod-
eling and the control design. The payload capacity of blimps is usually limited, especially for indoor blimps
featured by small size, low altitude, and slow flight velocity.5, 6 For the light-than-air UAVs, aerodynamic
virtual mass and inertia have significant effects on controlling blimps, which is different from conventional
fixed-wing aircraft. Due to the mass and inertia of the air, virtual mass and inertia effects are changing in the
dynamics of the blimps.1, 7 Furthermore, blimps are very sensitive to wind and temperature disturbances,
which makes them even more difficult to control.
Different from traditional fixed-wing aircraft, in the blimp control system, the trajectory control is the
external circle and attitude control is the inner circle. The basic requirement for a stable flight is effective
and smooth attitude control.5 Traditional control and advanced control methods are two existing strategies
developed for controlling blimps.8 Classical control strategies are characterized by easy implementation and
reliable control performance.9 Many advanced control methods, such as neural network,10 have been applied
to improve control performance when considering the complicated and unstable flight environment.
Recent studies of blimps focus on two parts, design and control law development. Geoffrey et al.11
constructed a lighter-than-air indoor blimp and realized blimp wandering motion without collision. Authors
in6 proposed a novel structure for an indoor quad-rotor blimp which will improve the balance performance
and increase the lift force of existing indoor blimps. Motoyama12, 13 built an autonomously controlled indoor
blimp and used action-value function for motion planning and evaluated its effectiveness using simulation.
Takanori Fukaot14 used back stepping techniques to design the image-based tracking controller of a blimp.
The literature15 examined the fuzzy logic approach for the designing of speed, heading, and height controllers.
Combination of Proportional, Integral, and Derivative (PID) and fuzzy logic controllers has been developed
to control the blimp altitude and evaluation of the control system performance has been conducted in two
different environments.3
In this work, a solar-powered blimp is developed. Solar energy is used to power motors and/or charge the
onboard battery, which greatly extends the endurance of a blimp.16 A PID controller is applied to control
the system states, including location and attitude of the blimp. In addition, results from experimental tests
are presented to verify the effectiveness of the blimp control system.
The paper is organized as follows: In §II, major components for building a solar-powered blimp are
described. The kinematics and dynamic equations are analyzed in §III. In §IV, the nonlinear dynamic model
is linearized and then PID controller for this linear model is developed. §V describes the experimental
environment and the results. We then conclude the paper with final remarks in §VI.

II. Blimp Configuration


In this section, the system architecture of the solar-powered blimp is presented. Several constraints exist
when designing an autonomous solar-power blimp. Generally, as the volume of the envelope increases, so
does the payload capacity. For a small-size blimp, which is typically restricted in size, the design strategies
focus on minimizing the weight of the on-board hardware to allow for a smaller envelope size. For a solar-
powered blimp, it must be capable of measuring the power consumption, power gain, and battery levels at
all times. The demonstration blimp, shown in Figure 1, was developed with several key aspects to allow for
simulation and experimental verification for the work presented in this paper.
As it’s shown in Figure 1, a flying toy ‘shark’, which is an inflatable balloon, was selected as the envelope.
The shape of the blimp is approximated by an ellipsoid with [Ra , Rb , Rc ] be the three semi-major axises.
The volume of helium filled into the blimp is about Vb = 190.5cm3 . According to Archimedes Principle, the
maximum weights of payload is calculated by

mload = Vb · (ρair − ρhelium ) = 202.1g (1)


3 3
where ρhelium = 169.3g/cm and ρair = 1226.0g/cm .
Figure 2 shows the system block diagram of the customized blimp. It is based on a balloon controlled by

2 of 12

American Institute of Aeronautics and Astronautics


an arduino mini pro board. The blimp is equipped with an 6W solar panel and a Maximum Power-Point
Tracker (MPPT) to convert the recovered energy to the vehicle’s operating voltage. System power flow is
monitored by two INA219B voltage/current (V/C) sensors. The two sensors are place within the system
so that the energy flow can be recorded in real time. One is placed between the MPPT and the battery,
which measures the total energy in and out of the battery. The second is placed between the MPPT and the
load, which is comprised of all of the system electronics, including Arduino mini pro, wireless synapse for
communication, two motor board driver boards, and three light DC micro-motors. A vertical motor installed
at the bottom allows altitude to be regulated and the other two motors placed at each side of the blimp can
control speed and yaw angle. The power consumption of three motors is about 2.7W in total. The layout of
the three motors are shown in Figure 3.
Due to the limitation of weight, the selection of battery is a key issue. First, the battery is required to
provided enough power while also being light. For the demonstrator blimp presented in this work, a single
cell, 1200 mAh Li-Po battery with 3.7 V output voltage was selected, weighing only 20 grams. Solar panels
is another issue. The solar panels are supposed to be flexible since the surface of blimp is not flat, and the
solar panels should supply enough power for the entire system. Two pieces of MPT 6-75 lightweight, thin,
flexible solar panels were selected17 for the customized blimp. The maximum wattage is about 6 W, which
is enough for the blimp system. The major components and their weights are listed in Table 1.

Figure 1: Autonomous Blimp Envelope

Figure 2: Autonomous blimp system block diagram

III. Blimp Dynamic Modeling


In this section, the formulations of the blimp kinematics and dynamics are presented. Since the dynamic
model of a blimp is nonlinear, the linearization is applied to simplify the model. Moreover, the shape of
a blimp is assumed to be an ellipsoid with major semi-axis Ra , minor semi-axis Rb and Rc . The blimp is
assumed to be non-deformable, which means it can be handled as a rigid body.

A. Coordinate Frames
In order to describe the states of the blimp, several reference frames shown in figure 4 are defined below
Body Coordinate Frame, FB : the origin locates at the center of the blimp, the XB axis is directed
along the axis of symmetry, YB axis to the left, and ZB axis to the up. It is used to analyze the linear
velocity and angular velocity during the flight.
Inertial Coordinate Frame, FI : also known as inertial reference system, is used to locate the blimp
relative to the earth. The origin locates at the surface of the Earth.

3 of 12

American Institute of Aeronautics and Astronautics


Figure 3: Layout of three motors and distribution of markers

Table 1: Blimp Components and Weights Distributions

Component Weights (g)


Envelope(140×78×60 cm3 ) 82.0
Motors and Propellers(×3) 7.0 × 3 = 21.0
Electronic Boards(Arduino,Wireless Synapse.etc) 42.6
Battery 20.0
Markers(×6) 2.0 × 6 = 12.0
Wires 12.1
Mounts 5.0
Solar panels(×4) 2.3 × 2 = 4.6
Total weight 200.3

Airflow Coordinate Frame, FA : also named speed coordinate system, is used to describe the relative
motion between the atmosphere and the blimp. It is used to analyze the aerodynamic force and torque. The
origin is located at center of the body (CB).
Local Horizon Frame, FH : This frame moves with the airship, but its axes remain parallel to the earth
fixed inertial frame axes and its origin CB is the center of volume of the vehicle.

B. Kinematics
Under the assumption of a rigid body, the blimp’s equations of motion is composed of translational and
rotational motion.

1. Translational Kinematics Equations


Based on the definitions of coordinates shown in the Figure 4, the translational motion of the center of a
blimp can be described by its position vector ~r.

~r˙ = V
~I (2)
~I denotes the inertial velocity vector in form of
where V
~I = V
V ~rel + V
~W , (3)

where V ~rel is the velocity vector of the blimp relative to the surrounding air. It can be expanded as V
~rel =
T
[u, v, w] = U ∈ < 3×1
in FB , where u, v, and w are the blimp’s velocity components in its body axis x-y-z.

4 of 12

American Institute of Aeronautics and Astronautics


Zh

Zb

Xh CV
CB

CG
ZI
r Position vector

Yh XI OI

YI

Figure 4: Definitions of Coordinate Frames

~W represents the wind velocity vector and it can be decomposed into V


V ~W = [Vx , Vy , Vz ]T = W ∈ <3×1
W W W
in FI . Then in the FI frame, Eq.(3) can be expressed as
     
ẋ u Vxw
ẏ  = CIB  v  + Vyw  (4)
     

ż w Vzw

where CIB is the rotation matrix from body frame to inertial frame.

2. Rotational Kinematic Equations


The Rotational motion of a blimp can be described by the orientation of frame FI relative to the local
horizon frame FI . It is defined in terms of Euler angles, roll (φ), pitch (θ), and yaw (ψ). Using the 3-2-1
convention of Euler angles, the rotational matrix CHB from FB to FH is written as
 
cθcψ cθsψ −sθ
CHB = sφsθcψ − cφsψ sφsθsψ + cφcψ sφcθ , (5)
 

cφsθcψ + sφsψ cφsθsψ − sφcψ cφcθ

where c ≡cos and s ≡ sin. Let Ω denotes the angular velocity expressed in FB . According to the definitions
of frames FI and FH , CIB = CHB .  
p
Ω = q  . (6)
 

r
Then, the rotational kinematic equation can be written as
    
φ̇ 0 sinψsecθ cosψsecθ p
 θ̇  = F · Ω = 0 cosψ −sinψ  q  . (7)
    

ψ̇ 1 sinψtanθ cosψtanθ r

5 of 12

American Institute of Aeronautics and Astronautics


Equations (4) and (6) can be organized in one matrix, expressed as
   
ẋ u
 ẏ  " # v 
   

  
  = CIB 03×3 w .
 ż   
(8)
 φ̇ 
  03×3 FΩ  p

 θ̇  q
   

ψ̇ r

C. Force Analysis
In this section, analytic expressions for the forces and torques acting on the Blimp are derived. There are
several types of forces acting on the blimp, shown in Figure 5, including the gravity force G, the buoyancy
force B, the aerodynamic forces, and the propulsion force generated by motors.

1. Gravity and Buoyancy


In the inertial frame FI , the gravity and buoyancy forces are expressed as
GI = [0 0 − mg]T
BI = [0 0 Vb · (ρair − ρhelium )]T . (9)

2. Propulsion Force
Propulsion force has a positive correlation with motor rotational speed and it is also related to the position
and direction of the propellers. It is assumed that the relationship between propulsive force by one propeller
Fp and rotational speed np is a polynomial function, written as:
Fp = η1 · n2 + η2 · n + η3 , (10)
where ηi , i = 1, 2, 3, are the coefficients and constants. The motor’s torque Tp is linear about Fp such that
Tf p = λ · Fp . (11)
The locations of three propellers in FB are
     
x p1 x p2 xp3
Lp1 =  yp1  Lp2 =  yp2  Lp3 =  yp3  (12)
     

zp1 zp2 zp3


. So the torque vector produced by propulsion forces can be expressed as
TXb = Tp1 + Tp2
TYb = Fp1 · zp1 + Fp2 · zp2 − Fp3 · xp3 (13)
TZb = −Fp1 · yp1 + Fp2 · yp2 + Tf p3
.

3. Aerodynamics
Since the blimp is light and operates at a low speed, the added mass or virtual mass can’t be ignored.
Referring to literature,2 the force produced by aerodynamics with virtual mass and inertia is formulated as
AA = A0 − A1 U̇ − A2 Ω̇, (14)
where A0 is the standard aerodynamic force in the body frame, expressed as
   
Fax 1/2ρair Vb2 Sref cN " #
2 U
A0 = Fay  = −  1/2ρair Vb Sref cT  − DU , (15)
   
2 Ω
Faz 1/2ρair Vb Sref cL

6 of 12

American Institute of Aeronautics and Astronautics


where cN , cT , and cL are the aerodynamic force coefficients in the body frame, Sref is considered as Θ2/3 ,
and DU is the translational portion of the Coriolis-centrifugal coupling matrix. Additionally, Vb is the blimp
velocity determined by p
Vb = u2 + v 2 + w2 . (16)
The aerodynamic force coefficients are function of angle of attack α and side-slip angle β. They are expressed
as

cL = c1 · cos(α) · sin(2β)
cN = c1 · cos2 (β) · sin(2α) (17)
cT = 0

Similar to the aerodynamic force, the aerodynamic torque Ti , i = 0, 1, 2 can be expanded as

TA = T0 − T1 Ω̇ − T2 U̇ , (18)

where T0 is the standard aerodynamic torque vector in body frame, expressed as


   
Tax 1/2ρair Vb2 Sref cn " #
2 U
T0 = Tay  = − 1/2ρair Vb Sref cm  − DΩ (19)
   

Taz 1/2ρair Vb2 Sref cl

where cn , cm , and cl are the aerodynamic torque coefficients in the body frame and DΩ is the rotational
portion of the Coriolis-centrifugal coupling matrix.

D. Dynamics
Based on the force analysis, applying Newton’s Laws, the dynamic model is expressed as

M~v˙ =
X
~ ex = F
F ~G +F ~B +F~P +F~A +F~C (20)

where M is the general mass and inertia matrix containing rigid-body inertia and added-mass terms, ~v is
the velocity vector including translational and rotational velocities, F~ G is the general gravity force, F
~ B is the
~
general buoyancy force containing buoyancy force and its tongue, FP is the force vector of the propulsion, F ~C
~
is the vector of Coriolis effects, and FA is the aerodynamic force vector. Since the blimp is in symmetric shape,
Eq.(20) is described in the body frame FB . Similar with kinematic analysis, translational and rotational
dynamic equations are also discussed separately in this section.

1. Translational Dynamic Equations


According to literature,2 using Newton’s second law and Euler’s equations, the three dimensional transla-
tional equations of the blimp are written as
X
× × 2
(mI3×3 + A1 )U̇ = m[SΩ (U − CBI Ẇ )] + CBI (G + B) + A0 + Fpi − m(SΩ ) ρCB − (mSρ×CB + A2 )Ω̇, (21)

where I3×3 is the identity matrix, G + B denotes the sum of gravity and buoyancy force vector expressed in
inertial frame, and ρCB = [rx , ry , rz ]T denotes the vector from the origin of the body frame to the location
×
of rigid body empty center of mass. SΩ is the skew-symmetric form of the vector Ω, expressed as
   
0 −r q 0 −rz ry
×  ×
SΩ = r 0 −p , SρCB =  rz 0 −rx  . (22)
  

−q p 0 −ry rx 0

7 of 12

American Institute of Aeronautics and Astronautics


B
L

Fp1 Tp1

Fp2 CB(CV)
Tp2

CG

Tp3
Fp3 G

Figure 5: Forces analysis for a blimp

2. Rotational Dynamic Equations


Using Newton’s second law and Euler’s equations, the rotational dynamic equations are formulated as
X
× ×
(Jb + T1 )Ω̇ = T0 + TG + Tpi SΩ Jb Ω + mSρ×CB (−SΩ U + CBI Ẇ ) + (mSρ×CB − T2 )U̇ , (23)

where Jb is the inertia matrix of the rigid (time-invariant mass) with respect to body frameFB , T0 represents
the aerodynamic torque vector in the body frame FB , which is a function Pof airspeed, and control surface
deflections, TG denotes the torque vector due to gravity in the FB , and Tpi denotes of the torque from
propulsion in FB .
Equations (8), (21, and (23) are the 6-DOF nonlinear model of the blimp system. By defining the state
vector x = [u, v, w, p, q, r, x, y, z, φ, θ, ψ]T and the control vector u = [n1 , n2 , n3 ]T , the nonlinear function can
be rewritten as
ẋ = F1 (x, u), (24)
where F1 is a set of functions about x, u.
For this customized blimp, roll angle control is inherent stable due to its configuration. Thus, the dynamic
model can be simplified by removing the roll and pitch controllers. Let X = [u, v, w, r, x, y, z, ψ]T , then the
simplified model is reformulated as
Ẋ = F2 (X, u). (25)

IV. Model Linearization and Controller Design


A. Linearization
The simplified model in Eq.(25) is a nonlinear model, which is difficult for real-time autonomous control
purposes. In this section, model in Eq.(25) is linearized. If the movement of the blimp is divided into desired
movement and disturbance movement, Taylor-series expansion is used to conduct the linearization of the
dynamic model.

8 of 12

American Institute of Aeronautics and Astronautics


Let X0 be the desired movement and δX be the disturbance, the state vector X = X0 + δX and
u = u0 + δu, then the linear model of the blimp system can be written in the form of
δX˙ = AδX + Bδu, (26)
where A and B are the Jacobian matrices, and they can be obtained by partial differential functions of F2 ,
expressed as
∂F2
A= (27)
∂X
∂F2
B= . (28)
∂u

B. PID controller
A PID controller is a classical closed-loop feedback control method. PID controller is well-known for its
simple structure and robust performance. In the section, PID method will be applied to the flight control of
the blimp. Define the observation model as
y = CX + Du (29)
where C = I3×3 and D = 03×3 .
Let Xd be the desired state variables, and Xp be the present state of blimp. The difference between Xd
and Xp is eX = Xd − Xp . So the PID controller is expressed as
Z
u = Kp · eX + Ki · eX dt + Kd · ėX (30)

where Kp , Ki , Kd are the proportional, integral, derivative gain matrices, respectively. The flowchart of
system controller is shown in Figure 6.

Figure 6: The Control Diagram of the Blimp System

V. Experimental Results
A. Experimental Environment
For the experiment tests, we constructed a time-invariant test area comprised of uniform terrain in an indoor
laboratory with 1.31×3.8 m2 area. More details of the environments can be found in the our previous work.18
In the indoor testing environment, the temperature of the experimental space is around 15.0 ◦ C with air
pressure of 1 atm. The helium density is 169.3 g/m3 at 15.0 ◦ C and the air density [g/m3 ] is 1226.0 at
15.0◦ C.
In Figure 7, the experiment flow chart of the solar-powered blimp system is presented. The test
includes three parts, the customized blimp, base station, and Vicon motion capture system.19 The Vicon
system is used to obtain the information about markers’ location put on the blimp and motion. Through
Vicon, we can obtain the location, orientation, and velocities of the blimp. All information, such as location
and power level of the blimp can be stored in the base station. Based on the information, the new control
vector is calculated using the proposed PID. Then the control commands are sent to blimp through wireless
communications. In this phase, the blimp is designed to operate in a straight path with a constant velocity
under wind disturbances.

9 of 12

American Institute of Aeronautics and Astronautics


Figure 7: Experiment Flow Chart of the Solar-Powered Blimp System

B. Results
This section presents results obtained from experimental testbed in the indoor testing environment. Given
the desired straight path through this area. The trajectory of the center of the blimp in three dimensional
space is shown in Figure 8. The red path is the desired straight path, and the blue line is the measured
trajectory. Figure 9 shows the time history of the control vector. Note that the maximum amplitude of
PWM for each motor was set as 200. Results with power consumption and management will be integrated
in the final version. We will also present outdoor test results in the final manuscript.

VI. Conclusion and Future work


In this paper, the design and control of an autonomous solar-powered blimp is presented. The key
components of the customized blimp are described. Furthermore, the six DOF nonlinear dynamic model of
the demonstration blimp is established and a PID controller is developed. Experimental results verify that
the developed blimp is capable of gathering energy and maintaining stability. Future research will explore
methods to optimize the blimp’s operation in outdoor environments.

References
1 Atmeh, G. and Subbarao, K., “Guidance, Navigation and Control of Unmanned Airships under Time-Varying Wind for
Extended Surveillance,” Aerospace, Vol. 3, No. 1, 2016, pp. 8.
2 Nakpiam, J., “Control of Airship Motion in the Presence of Wind,” 2012.
3 González, P., Burgard, W., Sanz Domı́nguez, R., and López Fernández, J., “Developing a low-cost autonomous indoor

blimp,” 2009.
4 Al-Rashedi, N. and Gerke, M., “3D Off-Line Path Planning for Autonomous Airships in Restricted Known Environments.”

Proceeding of the 18th International Conference on Process Control, 2011.


5 Miao, J., Zhou, J., Nie, Y., and Yang, X., “Yaw controller design of stratospheric airship based on phase plane method,”

Chinese Journal of Aeronautics, Vol. 29, No. 3, 2016, pp. 738–745.


6 Li, Y. and Wang, G., “Quad-Rotor Airship Modeling and Simulation Based on Backstepping Control,” International

Journal of Control and Automation, Vol. 6, No. 5, 2013, pp. 369–384.


7 Takaya, T., Kawamura, H., Minagawa, Y., Yamamoto, M., and Ouchi, A., “Motion Control in Three Dimensional Round

System of Blimp Robot,” SICE-ICASE, 2006. International Joint Conference, IEEE, 2006, pp. 1291–1294.
8 Liu, Y., Pan, Z., Stirling, D., and Naghdy, F., “Control of autonomous airship,” Robotics and Biomimetics (ROBIO),

2009 IEEE International Conference on, IEEE, 2009, pp. 2457–2462.


9 Kahale, E., Castillo Garcia, P., and Bestaoui, Y., “Autonomous path tracking of a kinematic airship in presence of

unknown gust,” Journal of Intelligent & Robotic Systems, 2013, pp. 1–16.
10 Park, C.-S., Lee, H., Tahk, M.-J., and Bang, H., “Airship control using neural network augmented model inversion,”

Control Applications, 2003. CCA 2003. Proceedings of 2003 IEEE Conference on, Vol. 1, IEEE, 2003, pp. 558–563.
11 Hollinger, G. A., Pezzementi, Z. A., Flurie, A. D., and Maxwell, B. A., “Design and construction of an indoor robotic

blimp for urban search and rescue tasks,” Swarthmore College Senior Design Thesis, 2005.
12 Motoyama, K., Kawamura, H., Yamamoto, M., and Ohuchi, A., “Development of autonomous blimp robot with intelligent

control,” Entertainment Computing, Springer, 2003, pp. 191–198.


13 Motoyama, K., Kawamura, H., Yamamoto, M., and Ohuchi, A., “Design of evaluation function in motion planning for

autonomous balloon robot,” Proceedings of 2003 Asia Pacific Symposium on Intelligent and Evolucionary Systems, 2003.
14 Fukao, T., Fujitani, K., and Kanade, T., “Image-based tracking control of a blimp,” Decision and Control, 2003. Pro-

ceedings. 42nd IEEE Conference on, Vol. 5, IEEE, 2003, pp. 5414–5419.
15 Avenant, G. C., Autonomous flight control system for an airship, Ph.D. thesis, Stellenbosch: University of Stellenbosch,

2010.
16 Kantor, G., Wettergreen, D., Ostrowski, J. P., and Singh, S., “Collection of environmental data from an airship platform,”

Intelligent Systems and Advanced Manufacturing, International Society for Optics and Photonics, 2001, pp. 76–83.

10 of 12

American Institute of Aeronautics and Astronautics


(a) The trajectory of blimp in 3-dimension space

(b) The trajectory of blimp shown in x-y plane

(c) The trajectory of blimp shown in y-z plane

Figure 8: The trajectory of the blimp

11 of 12

American Institute of Aeronautics and Astronautics


Figure 9: The PWM histories for the three motors installed on the blimp

17 “Lightweight,
Thin, Flexible Solar Panels,” 2014.
18 Kaplan,
A., Uhing, P., Kingry, N., and Dai Adam, R., “Integrated path planning and power management for solar-
powered unmanned ground vehicles,” Robotics and Automation (ICRA), 2015 IEEE International Conference on, IEEE, 2015,
pp. 982–987.
19 “Affordable Motion Capture for Any Application,” 2014.

12 of 12

American Institute of Aeronautics and Astronautics

View publication stats


See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/258650311

Airship Research and Development in the Areas of Design, Structures,


Dynamics and Energy Systems

Article · June 2012


DOI: 10.5139/IJASS.2012.13.2.170

CITATIONS READS

46 1,184

3 authors, including:

Alessandro Ceruti Pier Marzocca


University of Bologna RMIT University
54 PUBLICATIONS   332 CITATIONS    299 PUBLICATIONS   2,312 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Flutter control View project

Research at Clarkson University View project

All content following this page was uploaded by Pier Marzocca on 11 July 2017.

The user has requested enhancement of the downloaded file.


Review Paper
Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)
DOI:10.5139/IJASS.2012.13.2.170

Airship Research and Development in the Areas of Design, Structures,


Dynamics and Energy Systems
Casey Stockbridge*
Department of Mechanical and Aeronautical Engineering, Clarkson University, Potsdam, NY 13699, USA

Alessandro Ceruti**
DIEM, Department of Mechanical, Nuclear, Aviation, and Metallurgical Engineering, University of Bologna, Italy

Pier Marzocca***
Department of Mechanical and Aeronautical Engineering, Clarkson University, Potsdam, NY 13699, USA

Abstract
Recent years have seen an outpour of revived interest in the use of airships for a number of applications.Present day developments
in materials, propulsion, solar panels, and energy storage systems and the need for a more eco-oriented approach to flight are
increasing the curiosity in airships, as the series of new projects deployed in recent years show; moreover, the exploitation of
the always mounting simulation capabilities in CAD/CAE, CFD and FEA provided by modern computers allow an accurate
design useful to optimize and reduce the development time of these vehicles.The purpose of this contribution is to examine the
different aspects of airship development with a review of current modeling techniques for airship dynamics and aerodynamics
along withconceptual design and optimization techniques, structural design and manufacturingtechnologies and, energy
system technologies. A brief history of airships is presented followed by an analysis of conventional and unconventional airships
including current projects and conceptual designs.

Key words: Airships, Design, Structures, Aerodynamics, Dynamics, Energy Systems

Nomenclature
A State matrix
B Control matrix
CD Drag coefficient
CL1, CL2 Aerodynamic coefficients in roll moment equation
CM1, CM2, CM3, CM4 Aerodynamic coefficients in pitching moment equation
CN1, CN2, CN3, CN4 Aerodynamic coefficients in yaw moment equation
CX1, CX2 Aerodynamic coefficients in axial force equation
CY1, CY2, CY3, CY4 Aerodynamic coefficients in lateral force equation
CZ1, CZ2, CZ3, CZ4 Aerodynamic coefficients in normal force equation
E Identity matrix
Ereq Solar energy required for airship
F Normal force, Fineness ratio
F0 All external forces acting on body
I0 Inertia matrix taken about the origin of the body frame

This is an Open Access article distributed under the terms of the Creative Com- *** Graduate Student
mons Attribution Non-Commercial License (http://creativecommons.org/licenses/by- *** Assistant Professor, e-mail: alessandro.ceruti@unibo.it
nc/3.0/) which permits unrestricted non-commercial use, distribution, and reproduc- *** Associate Professor, *Corresponding Author,e-mail: pmarzocc@clarkson.edu
tion in any medium, provided the original work is properly cited.

Received: June 15, 2012 Accepted: June 25, 2012


Copyright ⓒ The Korean Society for Aeronautical & Space Sciences 170 http://ijass.org pISSN: 2093-274x eISSN: 2093-2480

3)(170-187)(20120615).indd 170 2012-07-25 오후 3:47:23


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

I'0 Added inertia matrix


k2 − k1 Added mass factor for ellipsoids
L Roll moment
M Bending moment, pitching moment
m Airship mass
ms Airship structural mass
mp Airship payload mass
M' Added mass matrix
N Yaw moment
p Roll rate
P0 Power required by payload
q, qo Pitch rate, Dynamic pressure
r Yaw rate
r G Position vector
rG× Skew symmetric matrix of the position vector
S Hull cross-sectional area, Total hull surface area
T All external torques acting on body
tday Number of seconds in a day
u Axial velocity, Gust velocity
Ū Average velocity
V, v Airship speed, lateral velocity
Vmax Maximum volume of the airship
vo Linear velocity
v̇0 Linear velocity derivative
w Normal velocity
X Force in the x-direction (axial force)
x State vector
ẋ State space equation
Y Force in the y-direction (lateral force)
Z Force in the z-direction (normal force)
α Angle of attack
β Sideslip angle
δAIL Differential elevator and rudder deflection
δELV Symmetric elevator deflection
δELVL Left elevator deflection
δELVR Right elevator deflection
δRUD Symmetric rudder deflection
δRUDB Bottom rudder deflection
δRUDT Top rudder deflection
η Control vector
ηp Propulsion system efficiency
ρ Air density
ρA0 Air density at sea level
ρHO Lifting gas density at sea level
σp Ratio between air density at design altitude and at sea level
μ Propeller pitch angle
τ Throttle
ω Angular velocity of the body-fixed frame
ω Angular velocity
ω̇ Angular velocity derivative

171 http://ijass.org

3)(170-187)(20120615).indd 171 2012-07-25 오후 3:47:23


Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

1. Introduction Airships offer advantages over conventional air cargo


transport because they do not require any power to stay aloft
1.1. General Information since all the necessary lift is acquired from the buoyancy
of lifting gases. This significantly reduces the power
An airship is a “lighter-than-air” aircraft which unlike requirements and fuel consumption for transportation
traditional fixed-wing and rotary-wing aircraft uses buoyancy and thus reduces the overall operating costs. Compared to
forces as its main source of lift instead of conventional ground and sea transportation, airships require greater fuel
lifting surfaces such as wings and blades. These buoyancy consumption but have a significantly lower travel time. These
forces are produced by lifting gases contained within the advantages and disadvantages are visible in Figure 1 below
airships envelope which have a density less than that of the which places airships in a unique niche in the transportation
atmosphere. The most common types of airships have the industry and shows their economic potential.
classical “teardrop” shape or axi-symmetric design and can
be classified as either non-rigid, semi-rigid, or rigid. 1.2. History

The history of airships has its beginnings in the eighteenth


century with the first recorded flight of a non-rigid dirigible
by Jean-Pierre Blanchard it 1784. The airship consisted of a
balloon fitted with a hand powered propeller for propulsion.
Attempts at adding propulsion to balloons continued into
the nineteenth century with Henri Giffard who was the first
person to make an engine powered flight. In 1852, he flew
27 kilometers in a steam powered airship. Twenty years later
in 1872, Paul Haenlein flew an airship over Vienna that was
powered by an internal combustion engine, the first time
such an engine was used to power an aircraft.
Fig. 1. Airship Efficiency vs. Conventional Transport Systems [1]. In the 1890s Count Ferdinand von Zeppelin began

[1] Stockbridge, C.M., “Stability and Control


of Unconventional Airships, Design and
Experimental Investigations,” MS Thesis,
Mechanical and Aeronautical
Engineering Department, Clarkson
University, Potsdam NY, 2012.

Fig. 2 History of
Fig.Airship
2. History ofDevelopment.
Airship Development. (a) Rigid
(a) Rigid Airships.
Airships. (b)
(b) Semi-rigid Semi-rigid
Airships. (c) Non-rigidAirships.
Airships [2]. (c) Non-rigid
Airships [2].

DOI:10.5139/IJASS.2012.13.2.170 172
2. Conventional Airships

2.1. Non-rigid Airships


3)(170-187)(20120615).indd 172 2012-07-25 오후 3:47:24
Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

experimenting with rigid airships. This led to the launch termed by the sound that the envelope of the airship makes
of the famous Zeppelins and the “Golden Age of Airships”. when you tap it with your finger [3]. Most often, non-rigid
During the first half of the twentieth century airships gained airships use helium as their lifting gas to fill internal ballonets
popularity for passenger transport and military uses such located inside the ship’s outer envelope which provide both
as tactical bombing, reconnaissance, surveillance, and balance and the aircrafts external shape. Ballonets are also
communications.During World War I, Germany, France, used to balance volume changes of the lifting gas due to both
Italy, and Britain all used airships for various military altitude and temperature change and are also associated
operations.The Norge, an Italian semi-rigid airship became with pitch control. This assures that the overpressure of the
the first confirmed aircraft to fly over the North Pole. The USS gas can be maintained and speed and steering ability are not
Shenandoah was the first American built rigid airship. It was affected. Sometimes instead of using lighter than air gases,
operated by the United States Navy and first flew in 1923. The these airships will use heated air as their lifting medium.
Shenandoah was the first airship to fly across North America These are termed hot-air airships. The only rigid components
and was the first dirigible to use helium as a lifting gas. of theseconfigurations are the engines, fins, and the gondola
In 1937, moments before landing, the Hindenburg, a or car that hangs from the blimp’s belly [3]. Historically, these
hydrogen filled rigid airship burst into flames, killing 36 aircraft launched “lighter-than-air”, where they received
people onboard and becoming one of the most well-known enough buoyancy from their internal gases to lift them off
and widely remembered airship disasters of all time. The the ground. Modern non-rigid airships however usually
public’s confidence in airships was shattered by this disaster. lift off overweight so they need to lift their nose and apply
This along with the onset of World War IIbrought the use propulsive forces or angle the engines downward to achieve
of airships for passenger transport to a halt.Airships also takeoff.Non-rigid airships are the most commonly used form
saw deployment during the Second World War and were of airships today because of their ease of construction and
predominantly used by the United States Navy for patrol and storability. Fig. 3 below shows a typical non-rigid airship
convoy escorts for ships to detect enemy U-boats.In 2.3.the
Rigid Airships with its internal structural layout.
years
since the war, airships have seen a decline in popularity
Unlike non-rigid and semi-rigid airships that maintain their shape by the internal pressure of
and usage. In present day, airships are typically used for 2.2. Semi-Rigid Airships
lifting
advertising, sightseeing, surveillance, and research. Figure gases, rigid airships retain their shape from an internal structural framework on which the aircraft’s
2 below shows a timeline of airship development starting in Semi-rigid airships are similar to blimps in that they
outer envelope is connected.
have no With the internal
internal frame to framework, a rigidenvelopes.
support their airship has They
the capability
do of being built
the 1850s with Henri Giffard’s first engine powered dirigible
and ending in the 1960s. have, however,
much larger than a non-rigid rigid dirigible
or semi-rigid objects because
on themthere
thatisgive them of
no chance some
kinking in the hull due
backbone. A stiff keel runs along the length of the airship
to aerodynamic forces forand moments. weight
distributing Inside the internal
and framework,
attaching fins andthe airship
engines is filled
[3]. with multiple gas
2. Conventional Airships The keel
cells holding the lifting gases.also provides
Because of thestructural integrity
size of most during using
rigid airships, flight multiple gas cells
maneuvering. Similar to non-rigid airships, the shape of the
nose and apply propulsive forces or angle the engines downward to achieve takeoff.Non-rigid
minimizes the chances airships are
2.1. Non-rigid Airships hullofisamaintained
catastrophe in the event
largely that
by an one is compromised.
overpressure A typical
of the lifting gas. rigid airship with
the most commonly used form of airships today because of their ease of its construction Light
layoutframework
and storability.
internal structural visible inatFig.
is Fig. the4.nose and the tail may also contribute
A non-rigid airship, more commonly known as a blimp to the hull’s outer shape. For small types the lifting gas is
3 below shows a typical
usesnon-rigid airship pressure
higher internal with its internal structural
from its layout.
lifting gases to maintain
both its shape and structural integrity.The word blimp was

Fig. 3.Typical Non-rigid Airship [2]. Fig. 4.Typical Rigid Airship[2].


Fig. 3. Typical Non-rigid Airship [2]. Fig. 4. Typical Rigid Airship[2].

2.2. Semi-Rigid Airships 3. Unconventional Airships


173 http://ijass.org
Semi-rigid airships are similar to blimps in that they have no3.1. Heavyframe
internal Lift Vehicles (HLVs)
to support their

A numberA of
envelopes. They do have, however, rigid objects on them that give them some backbone. Heavy
stiff Lift Vehicle (HLV) concepts have been proposed for cargo and passenger
keel runs

transportthat
along the length of the airship for distributing weight and attaching fins and enginesare[3].The
efficientkeel
andalso
cost effective. These vehicles have excellent fuel economy, which make

provides structural integrity during


3)(170-187)(20120615).indd themairships,
173 flight maneuvering. Similar to non-rigid viable the
alternatives to conventional
shape of the hull transportation methods over short distances. A number
2012-07-25 of
오후 3:47:24
others [5-12] have been performed confirming the effectiveness of airship lifting platforms for transport.

Capable of transporting payloads ranging from 1 to 1,000 tons these heavy lift airships have enormous
Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)
economically potential. Typical operating altitudes for such vehicles areusually less than 15,000 feetat

low flight speeds between about 80mph and 120 mph.


sometimes held in the hull itself, while larger types tend to the event that one is compromised. A typical rigid airship with
Many projects
use and endeavors
separate gas involving the development
cells, which of these
mitigates the heavy lift airships
consequences itshave beenstructural layout is visible in Fig. 4.
internal
of a single gas cell failure and helps reduce the amount of
proposed in recent years. One undertaking worth taking note of is DARPA’s Walrus HULA (Hybrid
overpressure needed [4].Semi-rigid airships fell out of favor
Ultra Large Aircraft)
forprogram. The after
many years Walrustheprogram aimedthe
1930s until to development
develop and evaluate 3. Unconventional Airships
of the a very large airlift
Zeppelin NTs which are some of the more recent and popular
vehicle concept designed to control lift in all stages of air or ground operations including the
3.1.ability
Heavy to Lift Vehicles (HLVs)
semi-rigids in use today.
off-load payload without taking on-board ballast other than surrounding air. The Walrus operational
A number of Heavy Lift Vehicle (HLV) concepts have
2.3. Rigid Airships
vehicle was intended to carry a payload of 500-1,000 tons up to 12,000 nautical miles, in lessbeen than 7proposed
days for cargo and passenger transportthat are
Unlike non-rigid and semi-rigid efficient and cost effective. These vehicles have excellent
and at a competitive cost [13].Some examples of other airships
heavy liftthat maintain
vehicle theircan be seen below.
concepts
shape by the internal pressure of lifting gases, rigid airships fuel economy, which make them viable alternatives to
The SkyHook JHL-40 seentheir
retain in Fig. 5 is from
shape a jointanproject between
internal SkyHook
structural conventional
and Boeing which will be
framework transportation methods over short distances. A
on which the aircraft’s outer envelope is connected. With number of feasibility and comparative studies by the National
capable of carrying 40 tons. It will be 302 feet in length and use four helicopter rotors to lift its payload
the internal framework, a rigid airship has the capability Aeronautics and Space Administration (NASA) and others
and propel itself, making it the largest helicopter in the world [14].
of being built much larger than a non-rigid or semi-rigidTheAeros Pelican is [5-12]
shown have6 been performed confirming the effectiveness of
in Fig.
dirigible because there is no chance of kinking in the hull airship lifting platforms for transport. Capable of transporting
which is a proposed 60 ton payload vehicle. The SkyFreighter from Millennium Airships can be seen in
due to aerodynamic forces and moments. Inside the internal payloads ranging from 1 to 1,000 tons these heavy lift airships
Fig. 7, aconcept proposed for both commercial and military applications.Additional
framework, the airship is filled with multiple gas cells holding have can
concepts enormous
be economically potential. Typical operating
the lifting gases. Because of the size of most rigid airships, using altitudes for such vehicles areusually less than 15,000 feetat
viewed in a review of heavy lift systems by Ardema [15].
multiple gas cells minimizes the chances of a catastrophe in low flight speeds between about 80mph and 120 mph.
Many projects and endeavors involving the development
of these heavy lift airships have been proposed in recent
years. One undertaking worth taking note of is DARPA’s
Walrus HULA (Hybrid Ultra Large Aircraft) program. The
Walrus program aimed to develop and evaluate a very
large airlift vehicle concept designed to control lift in all
stages of air or ground operations including the ability to
off-load payload without taking on-board ballast other
than surrounding air. The Walrus operational vehicle was
intended to carry a payload of 500-1,000 tons up to 12,000
nautical miles, in less than 7 days and at a competitive cost
[13].Some examples of other heavy lift vehicle concepts can
be seen below. The SkyHook JHL-40 seen in Fig. 5 is a joint
Fig. 5.SkyHook JHL-40 [16].
Fig. 5. SkyHook JHL-40 [16]. project between SkyHook and Boeing which will be capable
10  of carrying 40 tons. It will be 302 feet in length and use four
 
helicopter rotors to lift its payload and propel itself, making
Fig. 6.Aeros Pelican [17].
it the largest helicopter in the world [14]. TheAeros Pelican is

Fig. 6.Aeros Pelican [17]. Fig. 7.Millennium Airships SkyFreighter [18].


Fig. 6. Aeros Pelican [17]. Fig. 7. Millennium Airships SkyFreighter [18].

3.2. High Altitude Airships (HAAs)

Many concepts for High Altitude Airships (HAA) have also been proposed for intelligence
DOI:10.5139/IJASS.2012.13.2.170 174
gathering, surveillance and reconnaissance, and communications which will offer cheaper alternatives to

satellites. Fig. 8 below shows a HAA concept under development by Lockheed Martin which would

operate at altitudes of about 60,000 ft. In recent years the topic of high altitude and stratospheric airships

11 
3)(170-187)(20120615).indd 174 2012-07-25 오후 3:47:24
 
Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

shown in Fig. 6 which is a proposed 60 ton payload vehicle. HiSentinel stratospheric airship program to design a family
The SkyFreighter from Millennium Airships can be seen in of high altitude, long endurance airships for unmanned
Fig. 7, aconcept proposed for both commercial and military military operations. These airships, such as the HiSentinel 80
applications.Additional concepts can be viewed in a review operate with the use of solar power at altitudes ranging from
of heavy lift systems by Ardema [15]. 13-15 miles above the Earth [24].

3.2. High Altitude Airships (HAAs) 3.3. Hybrid Airships

Many concepts for High Altitude Airships (HAA) have Hybrid airships are aircraft that combine lighter than air
also been proposed for intelligence gathering, surveillance technology of aerostats and heavier-than-air technology
has become very popular and received much attention. A great deal of work has been put into this subject
and reconnaissance, and communications which will offer of traditional fixed-wing or rotary-wing aircraft.Hybrid
for the modeling and analysis
cheaper of these airships
alternatives [19-23].
to satellites. Fig.Most of these
8 below showsairships
a HAAare remote operated
aircraft can offer many advantages over traditional airship
concept under development by Lockheed Martin which configurations. For example, adding a pair of wings to the
aircraft or unmanned aerial vehicles because of their high operating altitudes. The U.S. Army Space and
would operate at altitudes of about 60,000 ft. In recent years main vehicle body helps to produce substantial aerodynamic
Missile Defense Command
the topicdeveloped
of highthealtitude
HiSentinel
andstratospheric airship
stratospheric program
airships hasto designlift,
a family
improveof vehicle stability, decrease drag, as well as
become very popular and received much attention. A great increase payload capability [26]. Two examplesof winged
high altitude, long endurance airships for unmanned military operations. These airships, such as the
deal of work has been put into this subject for the modeling airship concepts can be seen below; the WB-1010 (Fig. 9)
and analysis of these airships [19-23]. Most of these airships
HiSentinel 80 operate with the use of solar power at altitudes ranging from 13-15 miles above andthe theEarth
Airship One (Fig. 10).No hybrid airship has been
are remote operated aircraft or unmanned aerial vehicles built for production but several manned and unmanned
[24].
because of their high operating altitudes. The U.S. Army experimental vehicles have been flown demonstrating the
Space and Missile Defense Command developed the technology.
payload capability [26]. Two examplesof winged airship concepts can be seen below; the WB-1010 (Fig.

3.4.
9) and the Airship OneUnconventional BodyhasShapes
(Fig. 10).No hybrid airship been built for production but several manned and

unmanned experimental vehicles of


A number have been flown
designs demonstrating
that the technology.
display geometries that stray
from the conventional double ellipsoid, axisymmetric hull
shape have been proposed in recent years. For instance,
the P-791 seen in Fig. 11 is an experimental hybrid airship
that exhibits a triple hull design. Developed by Lockheed
Martin and first flown in 2006, this design maximizes the hull
volume and lifting gas capacity to maximize lifting capability.
Similar designs have been proposed butt with a double hull.
Another hull trait that is new to airship design is the use of an
airfoil shaped cross sectional area along the longitudinal axis
to produce dynamic lift. In effect, this makes the hull itself a
payload capability [26]. Two examplesof winged airship concepts can be seen below; the WB-1010 (Fig.
lifting surface similar to a wing.
Fig. 9.WB-1010 [27]
9) and the Airship One (Fig. 10).No hybrid airship has been built for production but several mannedThe
and Renault Zep’lin in Fig. 12 is a more radical concept

unmanned experimental vehicles have


Fig. been flown demonstrating
8.Lockheed Martin HAA the[25].
technology.
Fig. 8.Lockheed Martin HAA [25].

3.3. Hybrid Airships

Hybrid airships are aircraft that combine lighter than air technology of aerostats and heavier-than-

air technology of traditional fixed-wing or rotary-wing aircraft.Hybrid aircraft can offer many advantages

over traditional airship configurations. For example, adding a pair of wings to the main vehicle body

helps to produce substantial aerodynamic lift, improve vehicle stability, decrease drag, as well as increase

Fig. 9.WB-1010
12  [27] Fig. 10.Airship One [28].
  Fig. 9. WB-1010 [27] Fig. 10. Airship One [28].

3.4. Unconventional Body Shapes


http://ijass.org
A175
number of designs that display geometries that stray from the conventional double ellipsoid,

axisymmetric hull shape have been proposed in recent years. For instance, the P-791 seen in Fig. 11 is an

13 
 
3)(170-187)(20120615).indd 175 2012-07-25 오후 3:47:25
Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

with an unconventional shape. Its unique hull design not was edited, in which the approach to size estimation is
only acts as a storage vessel for lifting gases to provide different: this author proposes a design methodology based
aerostatic lift but as a sail for additional lift, propulsion, and on preliminary design calculations, evaluation of static and
directional control. dynamic bending moments, gas pressure stresses, design of
cars for power systems, passengers, and flight crew, gas cells,
and finally tail cones, stabilizing surfaces, and mast mooring
4. Airship Design and Optimization gears. This book provides one of the first examples of a
complete list of formulas to be used for the initial estimation
Some of the earlier scientific and technical documents of size and horsepower for a given performance and proposes
related to airship design can be found as technical reports a well coded “step by step” embodiment process to deploy
from NACA and the Royal Aeronautical Society (RAeS). the entire design process in a systematic way.
experimental hybrid airship that exhibits a triple hull design. Developed by Lockheed Martin and first
One of the most famous is a report by Lamb [31] focusingon A resume of the design experiences of the years up to the
the study of inertia coefficients of an ellipsoid moving flown inin2006,
a this1940sdesign canmaximizes
be foundthe
in ahull volume manual
technical and lifting
[36]gas capacity
in which theto maximize lifting
fluid: these coefficients were needed to keep into account formulas and methodologies developed for the design are
capability. Similar designs have been proposed butt with a double hull. Another hull trait that is new to
added masses in airship design. The increased interest in summarized in a very practical and “design oriented” way.
airships of the first years of the 20th century wasairship design is theThe
supported use Hindenburg
of an airfoil shaped
accidentcross
andsectional area along
the interest the longitudinal
towards vehicles axis to produce
by studies on airship design focusing their attention on with higher speed deadened the interest in airships, and in
dynamic lift. In effect, this makes the hull itself a lifting surface similar to a wing.
aerodynamics and weight which was considered at the 1962, the US Navy program for airshipsstoped. The design
moment the two most critical issues in the design of process of airships is kept going by Kostantinov [37] who
airships. Two reports by Tuckerman, the first dealing with collected the formulas and experiences in the field of airships
the determination of forces on an airship hull [32], and the and merged the up to date aerodynamic and structural
second focusing on inertia factors [33] show the need for a research in a comprehensive paper. Since the 1970s, airships
precise loads assessment in order to design a lightweight and blimps are designed for advertising purposes or touristic
structure able to sustain the stresses due to lifting gas and adventure trips: Goodyear in USA and Zeppelin in Germany
dynamic pressure. The experience and knowledgein airships are good examples of such activities. The increasein personal
gained during the period 1900-1927 by pioneering designers computers and the computational load available made
like General Umberto Nobile in Italy, Count Zeppelin and possible the solutions of complex equations and the large
NikolausBasenach in Germany, and Goodyear in the USA are number of simulations that can be ran simultaneously,
experimental hybridreported
airship that
in aexhibits
book byaThomas
triple hull design. Developed
Blakemore and Watters by Lockheed
Pagon Martin and firstto experimental data (as for the studies of CFD
compared
flown in 2006, this[34] where
design all the subsystems
maximizes the hull volume of anand
airship
liftingaregasconsidered relatedlifting
capacity to maximize to the German LOTTE). Also the airship design field
one by one. The approach followed in [34] to solve the weight was affected by these new capabilities: the work of Lutz et
capability. Similar estimation
designs have(which
been proposed
can bebutt with a double
considered the hull.
mostAnother
criticalhull
in trait that is new
al. [38] to of the fist describing the optimization of the
is one Fig.11.Lockheed Martin P-791 [29].
airship design) is based upon comparison with already
airship design is the use of an airfoil shaped cross sectional area along the longitudinal axis to produce airship by means of evolutionary algorithms
shape of an
designed and built airships where a wide list of tables in
dynamic lift. In effect,
whichthisthe
makes the hull itself aofa
characteristics lifting surface
large numbersimilar
of to a wing. are
airships
listed. In the same year a book by Charles P. Burgess [35]

Fig.11.Lockheed Martin P-791 [29]. Fig. 12.Renault Zep’lin Solar Powered Airship [30].
Fig.11. Lockheed Martin P-791 [29]. Fig. 12. Renault Zep’lin Solar Powered Airship [30].
14 
 

DOI:10.5139/IJASS.2012.13.2.170 176

3)(170-187)(20120615).indd 176 2012-07-25 오후 3:47:25


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

and stochastic methods: the airship design process can the design process for a HAP; moreover, a list of tables shows
make now use of the new available computing capabilities. how the change of design parameters (like the purity of
Khoury and Gillet [39] present a book in which a chapter is helium, or the sunlight hour related to the season of the year)
devoted to Design Synthesis. Airship design now focusesits affect the lift. The most interesting part of this study reports
attention on the integration of sub-systems and trade-off the effects of technology advances on airship performance
considerations. Moreover, the design process is divided in parameters: by this way the designer can have an idea of how
Conceptual, Preliminary and Detailed phases. Flowcharts the payload can be increased with an increase in propeller
are presented to drive and support the designer in the weight/massratio and solar cells efficiency, or a decrease
Conceptual Design phase, in the trade off analysis, and of envelope area weight and batteries capacity/mass ratio.
in the trade study process. The airship is considered as a Also, Chen [46] presents a similar work of sensitivity analysis
system, and the mutual interactions between subsystems arriving at similar results in terms of influence of weights
(condensed in the airship sizing matrix) is considered the key and efficiency on the design: a design flowchart is presented
of success for a good and balanced design. Also, sensitivity here also to assure the equilibrium between lift and weight
analysis and parametric weight estimation (derived from due to solar panels, structure, batteries, and propulsion
the aircraft conceptual methodologies [40]) are introduced systems. The multidisciplinary approach to design, which is
in this comprehensive book.In the end of the 20th century, a consequence of a concurrent engineering approach, has
flight simulation is proposed not only for the training of been applied also to airship design: the work of Ram and Pant
pilots [41], but also to check the design results and to verify [47] presents the aerodynamic and structural optimization
the behavior of the airship,even inthe conceptual design of an airship using variable thickness fabrics and a low drag
phase. The availability of new film materials, efficient solar shape.
panels, and the need for high altitude observation platforms As the new reprint of the book Airship Technology [48]
focused the attention of designers to High Altitude Platforms reports, in addition to the classical interest related to materials
(HAP). In a study by Mueller et al. [42] the design of a HAP [49], solar panels, and unconventional configurations, one
is presented in a parametric way: data like weight of the of the challenges for the future is the design of multi gas,
envelope and efficiency of solar panels are not kept fixed. The multi chamber airships [50] seems to be a solution for cost
design process output is not a defined sizing, but a series of reduction and lower environmental impact.
graphs which the designer can use to dimension the airship
with materials available at the moment. With this method,
the design can be updated if new materials or technology 5. Structures
become available. The basic formula used to evaluate the
volume of the airship [42] is:
5.1. Structural Design and Analysis
(1) Structural analysis in an important area in airship
design because airships experience deformation under
where Vmax is the maximum volume of the airship, ms is aerodynamic and aerostatic loads. Predictions of loads,
the structural mass of the airship, given by the sum of the stress distribution, and bending moments are useful in both
envelope and remaining structural elements masses, mp is the design of airships and the preventions of catastrophic
the mass of the payload, σP is the ratio between the density failures. Structures technology is discussed in several review
of air at sea level (ρA0) and at the max altitude of service (ρA), works by Burgess et al. [51], Hess [52], and Liao et al. [53]
and ρH0 is the density of the gas filled inside the envelope. along with structural problems in the construction of lighter-
Also, in the work of Wei et al. [43]andNickol et al.[44] the than-air vehicles. The structural properties of the Navy’s
attention is focused on the trade-off analysis, on the sensitivity rigid airship USS Los Angeles were summarized in a work
analysis, and on how the airship would be impactedby a new by Althoff [54]. Burgess also wrote a significant report for the
technology or change in mission requirement. In the latter Navy analyzing forces on an airship in wind gusts [55].
of these two papers, the design is based upon the proposal Structural strength was a significant issue in the
of several configurations, each one evaluated in the mission development of early rigid airships. For example, Evans
through a Life Cycle Cost Analysis approach where a design [56] computed the force distribution due to aerostatic,
is considered good if it presents a cheap operational cost and aerodynamic, and inertial forces for the Shenandoah rigid
a low cost for environmental impact and final dismissing. airship and demonstrated the possibility of catastrophic
The work of Yu and Lu [45] presents a flowchart describing failure due to the structural bending moment [57].For rigid

177 http://ijass.org

3)(170-187)(20120615).indd 177 2012-07-25 오후 3:47:25


Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

airships several equations for calculating maximum design adhesion, maximum material life vs. ease of field repair, and
moment on the hull and methods for distributing design minimum price vs. all other demands [65].
moment over the length of the hull were developed in [58, In present day, the internal framework in semi-rigid and
59]. One such equation from Goodyear [58] shows: rigid airshipsis typically made of aircraft-grade aluminum
that is riveted. The nose cone is usually made of wood,
(2) plastic, or metal and is then laced to the airship’s envelope.
The gondola is commonly manufactured with metal [66].
where is the fineness ratio, is the gust velocity, is the airship As seen in the rest of the aviation industry new composite
speed, is the dynamic pressure, and is the airship volume. materials and carbon fibers are also making their way into
Li et al. [60] developed a linear model for flexible airships the construction of airships as building materials mainly for
which was used to study structural flexibility effects on airship the use of gondolas and crew cabins.
flight dynamics and aerodynamics [53]. In this model, the One of the major design challenges for airships is the
equations of motion of an elastic airship are derived by the use of materials for the construction of ballonets or airbags
Lagrangian formulation and the airship is modeled as a free- and envelopes to prevent leakage of the lifting gases but
free Euler-Bernoulli beam where the bending deformations provide flexibility.So airship envelopes are normally made
are represented by shape functions chosen as the free-free from Dacron and Mylar or other polyester fabric materials.
normal modes [60]. They are sometimes made of Tedlar, a polyvinyl film, which
Recent advances in computational tools such as Finite is bonded with Hytrel, a thermoplastic polyester elastomer
Element Analysis (FEA) software has allowed for extensive which provides the flexibility of rubber and the strength
structural analysis to be performed on airships with a high of plastic. These fabrics help protect the envelope from
degree of accuracy. Hunt [61, 62] performed static structural ultraviolet light. The ballonets are normally made from leak-
analysis of an aerostat with the use of NASTRAN to come resistant polyurethane plastic [66].In a paper by Miller and
up with an idealized structural model by analyzing stress Mandel [65], the design requirements of airship envelopes
distribution and shape deformation. Similarly in Smith and materials and material development and qualification
[63], Boeing used the FEA package ABAQUS to develop information is examined.
an internal loads model for the Skyhook HLV aircraft seen Kang et al. [67] studied the material characterization
in Fig. 5.Bessert and Frederich [64] presented a method to of a film-fabric laminate developed for a stratospheric
investigate the nonlinear aeroelastic behavior of an airship airship envelope consisting of a single plain woven fabric
due to geometric and material nonlinearities. They tested layer impregnated in a polymer matrix laminated with
their method on the CL-160 airship using ABAQUS and thin films. They performed uni-axial tests to obtain tensile
VSAERO (an aerodynamic solver based on the aerodynamic properties and finite element analyses to obtain effective
panel method using potential flow theory) and found that it tensile properties. McDaniels et al. [68] of the Cubic Tech
could handle all the nonlinearities present in the different Corporation examined the use and development of non-
models, and could deliver the required derivatives which woven flexible laminates for lighter-than-air vehicles. They
would be difficult to obtain from free-flight or wind tunnel concluded that the use of these flexible laminates achieved
experiments [64]. a significant weightsavings over woven fabrics of similar
strengths by eliminating strength and modulus loss andother
structural deficiencies caused by crimping of yarns during
5.2. Materials
the weaving process. Theabsence of crimp in non-woven
Materials have made a huge leap forward in the fabrics results in a linear elastic response that allows for
development of airships, making them lighter, stronger, and easein predicting material properties and simplification of
more efficient.There are very particular demands on materials structural models [69].
when it comes to airships construction. They need to exhibit Two other important material manufacturers for
proper properties of strength, weight, air-tightness, weather envelope construction worth taking note of are ILC Dover
and UV stability, conductivity, and non-flammability. and Contitech. ILC Dover is the world’s largest producer of
However, different requirements including performance, modern airship envelopes and has been producing materials
cost, risk, and service life have to be considered. Therefore for the construction of aerostats since the 1970s. They have
the material becomes a delicate balance between often served in the production of materials for customers such
competing demands such as highest tensile strength vs. as the American Blimp Corporation, Lockheed Martin, and
lowest possible mass, maximum tear strength vs. maximum Skyhook. Contitech is one of the globe’s leading specialists

DOI:10.5139/IJASS.2012.13.2.170 178

3)(170-187)(20120615).indd 178 2012-07-25 오후 3:47:25


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

in rubber and plastics technologyand through innovative over time the ship will gain significant positive buoyancy
manufacturing processes and extremely rigorous finishing [69]. As this is undesirable from both a control and structural
quality have developed materials for the special applications viewpoint, the airship must have a mechanism for buoyancy
of airships. compensation. Hydrogen-filled airships can simply vent
excess hydrogen into the atmosphere to compensate for
5.3. Lifting Gases the weight of fuel burned. Since helium is more expensive,
helium-filled airships are constructed with an apparatus
In terms of today’s airships, helium is by far the on the engine exhaust to condense and recover the water
predominately used source for lifting gases. Unlike hydrogen, it contains. The water is then stored to compensate for the
helium is an inert gas so it is not flammable. This is the weight of fuel burned [70]. These water condensers can be
main reason hydrogen isn’t commonly used today, being
Two other important material manufacturers for heavy envelope construction
and provide worth
additional taking
drag beingnote of areon the
mounted
responsible for many disasters of early airships such as the skin of the airship.
Hindenburg.
ILC DoverHowever, with the use
and Contitech. ILCof Dover
helium is
comes many
the world’s largest producer of modern airship envelopes and
tradeoffs which are explored by Gordon et al. [69], Ghanmi et
al.has
[70],been producing
andLinner materials
[71]. For example,for the isconstruction
helium expensive andof aerostats since the 1970s. They have served in the
6. Aerodynamics
supplies are limited whereas hydrogen is both inexpensive
production
and of materials
can be easily generated.for customers
Helium is alsosuch as the American
less buoyant BlimpTunnel
6.1. Wind Corporation,
and FlightLockheed
Tests Martin, and
than hydrogen and has about 7% less lifting capability
Skyhook.
[71]. Another Contitech is one
trade-off that ofsothetechnically
is not globe’s leading
obvious specialists
in in rubber
Few and plastics
publications technologyand
are available concerning through
the modeling
buoyancy compensation. When an airship takes off with of airship aerodynamics. A large amount of the available
innovative
neutral manufacturing
buoyancy the aerostaticprocesses
lift produced and
by extremely
the helium rigorous finishing with
literaturedeals quality have developed
empirical data and materials
resultswithout
is equal to the total weight of the vehicle. As fuel is burned addressing or going into too much depth about modeling
enfor the special
route, however,applications of airships.
the total weight of the airship decreases techniques. Literature that is available for viewing on the
but the aerostatic lift remains the same. If nothing is done, aerodynamics of airships mainly deals with the classicaxi-

Table 1. Summary of Construction Materials for Recent Airship Projects


Table 1. Summary of Construction Materials for Recent Airship Projects
Airship Model Manufacturer Classification Envelop Material Frame Material Lifting
Gas
SkyCat-220 Advanced Non-rigid Heat-bonded, high-tensile The payload Helium
Technologies laminated fabric, module is
GroupLtd. incorporating a Mylar film formed from
that provides the gas barrier. Kevlar
composite
material

Zeppelin LZ ZLT Zeppelin Semi-rigid Laminate of polyester Carbon-fiber Helium


N07 Luftschifftechnik basecloth and poly-vinyl frames and
GmbH & Co KG. fluoride(PVF or Tedlar) film aluminumlonger
ons braced by
aramide cables
Goodyear GZ- Goodyear & Non-rigid 2 ply polyester fabric coated Steel framed and Helium
22 Lockheed Martin with neoprene rubber composite
skinned gondola
Zeppelin NT ZLT Zeppelin Semi-rigid High-strength multilayer Aluminum and Helium
Luftschifftechnik laminate high-strength,
GmbH & Co KG. External layer : Tedlar lightweight
(protective film UV-proof) carbon-fiber
Intercellular layer: Polyester
fabric (tear-resistant)
Internal layer : Polyurethane
(weldable and leak-proof)
Skyship 600 Airship Industries Non-rigid Kevlar Gondola made Helium
from Composite
Materials

5.3. Lifting Gases


179 http://ijass.org

In terms of today’s airships, helium is by far the predominately used source for lifting gases.

Unlike hydrogen, helium is an inert gas so it is not flammable. This is the main reason hydrogen isn’t

commonly used today, being responsible for many disasters of early airships such as the Hindenburg.
3)(170-187)(20120615).indd 179 2012-07-25 오후 3:47:25
Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

symmetric elliptical bodies of revolution and little is available particularly near the aft section of the body where viscous
for unconventional body shapes. effects become more important. In Allen and Perkins [87],
With the lack in development of aerodynamic models a term related to the cross flow drag was added to Monk’s
for airships in the early years, flight behaviors were mainly results to correct for the effects of viscosity and additionally
analyzed usingflight test and wind tunnel testing. A large accounted for axial forces.Similar to Allen and Perkins,
wealth of information was collected for wind tunnel test in Hopkins [88] proposed a method in which the transverse
the 1920s and 1930s for scaled airship models. Jones et al. forces on the forward portion of the hull could be calculated
[72-74]at the Aeronautical Research Committee (ARC) in from potential flow theory and the transverse forces on the
Britain performed tests on models of theBritish airships R-29 remaining portion of the hull could be calculated by relating
and R-101and the rigid German airship L33. The National the local transverse force for the body to the cross flow drag.
Advisory Committee for Aeronautics (NACA) in the United
States published multiple reports on wind tunnel testing of 6.3. Semi-Empirical Approaches
the airship Akron [75-79] where experimental measurements
were taken of the flow in the boundary layer, pressure Jones and DeLaurier [89] further developed this model by
distribution on the hull, and ground handling forces. Zahm accounting for interference between the hull and fins with
[80] also presented a NACA report on the air forces, moments, a semi-empirical approach, defining hull-fin interference
and damping on a model for the airship Shenandoah. factors with experimental data in the analytical model. In
Pannell et al. [81-83] published multiple reports at ARC their steady state model, the airship is separated into two
around 1920 regarding the results of flight tests for the aerodynamic regions, where the forces and moment on the
British airships R-26, R-29, and R-33. Testson these airships airship hull are evaluated from the nose of the hull to the
were performed to determine turn radius at various rudder point where the hull and fins intersect each other, and where
deflections and the drag forces. However, accuracy of these the hull and fins are evaluated together after this intersection
measurements was poor due to the limited technology of the point.
time. Modern flight test have proven more accurate results The aerodynamic model below was developed by Mueller
and measurements with more advanced instrumentation et al. [42] following the procedures outlined by Jones and
as seen in [84, 85], where flight tests were performed on the DeLaurier [89] for an axisymmetric airship hull with four
Skyship-500 airship in the Patrol Airship Concept Evaluation equally sized fins which also incorporates the effects of
(PACE) program measuring the responses to inputs of rudder and elevator deflection. The equations for the
elevator, rudder, and throttle. aerodynamic forces and moments can be seen below where
a complete list of the aerodynamic coefficients shown in the
equations can be found in [42]. A schematic of this steady
6.2. Potential Flow Theory
state aerodynamic model can be viewed in Fig. 13.
Some of the earliest works for analytical airship
(4)
aerodynamics models date back to the 1920s and are based
on potential flow theory such as the work performed by Monk (5)
[86] used for the investigation of the Navy’s ZR-1. In his work,
Monk derived the normal aerodynamic force distribution
over an airship hull modeled as an elongated surface of
revolution using a slender body assumption: (6)

(3)
(7)

where is the cross sectional area of the hull, is the velocity,


is the angle of attack, and is the added mass factor for
ellipsoids derived in work performed by Lamb [31] which (8)
correct the effects of the hull’s finite length. Monk further
concluded that an airship has an unstable pitching moment
about non-zero angles of attack. (9)
Potential flow theory offered some basis for aerodynamic
predictions but neglected effects of viscosity on the hull,

DOI:10.5139/IJASS.2012.13.2.170 180

3)(170-187)(20120615).indd 180 2012-07-25 오후 3:47:27


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

6.4. Shape and Drag Optimization model, and a variationalmultiscale LES model. Their
results showed that all three models predicted the primary
More recent developments in airship aerodynamics have longitudinal vortex shed from an airship body but that only
been the use of shape optimization in airship hulls to reduce thevariationalmultiscale LES model predicted the secondary
drag and increase performance. Since power requirements vertical flow structure observed in experimental studies. In
are directly related to drag, optimization is a powerful tool a paper by Wang et al. [94], Fluent was used to accurately
for maximizing an airship’s potential and minimizing fuel simulate the motion of stratospheric airships with a model
consumption. Extensive research has been performed that calculated the aerodynamics of an airship based on a
by Lutz and Wagner [90, 91] in developing a method for panel method and an engineering estimation approach.
numerical shape optimization of axisymmetric bodies in
incompressible flow at zero incidence.Specific aerodynamic
optimizations of bodies of revolution for prescribed 7. Dynamics
Reynolds number regimes were performed. This resulted in
minimized drag at maximized volume for Reynolds number 7.1. Equations of Motion (6 DOF)
regimes relevant for airship application [90]. Fig. 14 shows
their resulting minimized drag curves for the optimized body An airship is most commonly modeled as a rigid body
shapes in each design regime. with six degrees of freedom, three translational and three
A similar paper was proposed by Nejati and Matsuuchi rotational which results in six nonlinear equations that
[92] using genetic algorithms for shape optimization which represent the motion of the airship. The equations can be
showed that the method of using genetic algorithms for seen in matrix form below:
optimization could minimize the drag coefficient faster for
different Reynolds number regimes. (10)

6.5. CFD Analysis

Computational Fluid Dynamics software tools have


allowed researchers to examine the flow interactions with
airship bodies to a certain degree of accuracy without the where is the total mass of the system, is the skew symmetric
need for expensive wind tunnel testing or full scale flight matrix of the position vector, is a 3x3 identity matrix, is the
test saving a great deal of time and money. Such commercial inertia matrix taken about the origin of the body frame, is the
programs like Fluent, Star-CCM+, and FLOW-3D have made added mass matrix, and is the added inertia matrix.The right
these capabilities widely available. hand side consists of all the external forces and torques acting
El Omari et al. [93] published an important paper on the on the body. These are made up by the weight, buoyancy
challenges of turbulence modeling in airship CFD studies force, aerodynamic forces and moments, and propulsive
where
� � �� ���� cos�� ���they investigated three�turbulence models�based on � � forces.
��� ������ The
(9)added mass and inertia matrices are functions of
sin���� � ��� sin���� ��� sin��� sin����� ���� ��
statistical and Large Eddy Simulation (LES) approaches;a
standard high Reynolds k–ε model, a Smagorinsky LES

Fig. Fig. 14. Drag Curve of Optimized Body Shapes [90].


Fig.13.Schematic
13. Schematicof
ofSteady-State
Steady-State Aerodynamic Model[42].
Aerodynamic Model [42]. Fig. 14. Drag Curve of Optimized Body Shapes [90].

6.4. Shape and Drag Optimization A similar paper was proposed by Nejati and Matsuuchi [92] using genetic algorithms for shape
181 http://ijass.org
optimization
More recent developments in airship aerodynamics have been the which
use of shape showed that
optimization in the method of using genetic algorithms for optimization could minimize

the drag coefficient


airship hulls to reduce drag and increase performance. Since power requirements faster
are directly for different
related to Reynolds number regimes.

drag, optimization is a powerful tool for maximizing an airship’s potential and minimizing fuel
3)(170-187)(20120615).indd 181 2012-07-25 오후 3:47:27
6.5. CFD
consumption. Extensive research has been performed by Lutz and Wagner [90,Analysis
91] in developing a
Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

the vehicle shape [42]. Yamaski and Goto [98] conducted a series of flight tests
The equations of motion (9) can be linearized and on a full scale blimp with feedback systems for stabilizing
naturally decoupled into two separate modes consisting yawing and pitching motions, and a sensor system to
of the longitudinal and lateral systems. To linearize the measure the motion and control outputs. Data from the
equations of motion trim conditions are computed for tests were analyzed to yield parameter values including
equilibrium flight and the velocity components of the linear added mass effects and stability derivatives. A comparison
models are treated as perturbations about the trimvelocity. was made between the experimental values and estimated
The linearized models can be represented below in state values obtained using analytical formulas where they were
space form: determined to be consistent.
Mueller et al. [42, 99] developed a comprehensive set of
(11)
modeling, analysis and control designtools for airships at
Princeton Satellite Systems Inc. with the help of the Missile
whereand are the state and control matrices and and are
Defense Agency. An integrated guidance and control system
the state and control vectors. The longitudinal state vector
was designed for a high altitude airship where a simple
consists of the pitch rate , axial velocity , and normal velocity
control law design provided robust feedback control of the
. These states are affected by the control vector consisting
airship’s angular rates and velocity.
of throttle , propeller pitch angle , and symmetric elevator
deflection . The lateral state vector consists of the roll rate ,
yaw rate , and lateral velocity . These states are affected by
8. Energy Systems
the control vector consisting of symmetric rudder deflection
, and differential elevator and rudder deflection [42]. A
Recent advances and growing interest in solar power
complete and detailed derivation of the linearized equations
technology indicates that solar energy systems will play a
of motion can be viewed in [95].
greater role in energy production for airships of the future.
In a paper by Lubkowski et al. [100], an analysis is performed
7.2. Stability& Control
of several different solar power technologies to evaluate
Although most of the airship lift is generated by the the trade-offs of the cost against the power consumption,
aerostatic forces, the aerodynamic characteristics determine efficiency, and reliability. These technologies included
the stability of the aircraft [57]. As seen previously, Monk photovoltaic flat panels, thin film photovoltaic panels,
determined that the body of an airship experiences an trough solar concentrators, stirling dish solar concentrators,
unstable pitching moment due to the added mass terms. This and luminescent solar concentrators. Their results show that
also cause the yaw rotations to destabilize, but the viscous flat solar technologies such as thin film, luminescent solar
effects acting on the aft end of the ship including the tail fins concentrators, and photovoltaic flat panels ranked highest.
tend to be stabilizing along with other aerodynamic forces Gawale and Pant [101] present a methodology of determining
normal to the centerline of the airship.In a stability analysis the initial sizing of a stratospheric airship including the
performed by Cook et al. [96], it was determined that the required volume and dimensions based on parameters such
longitudinal modes of the airship are comprised of the surge as operating altitude, speed, payload, available irradiance,
mode caused by axial aerodynamic drag, the heave-pitch solar cell efficiency, atmospheric conditions, and propulsion
subsidence mode cause by normal aerodynamic drag, and system efficiency.
the oscillatory pitch-incidence mode which is caused by the The solar energy required by an airship with an elliptic
center of gravity being located under the center of volume. axisymmetric shape can be computed as
The lateral-directional modes of the airship are comprised of
the sideslip subsidence mode, the yaw subsidence mode and (12)
the oscillatory roll pendulum mode. Approximate models for
these modes are derived and expressed in terms of concise where, is the average velocity, is the total surface area
aerodynamic stability derivatives in [96]. of the hull, is the drag coefficient, is the propulsion system
Kornienko [97] conducted aninvestigationof the stability efficiency, is the number of seconds in a day, and represents
and controllability of an airship under different flightand the power required by the payload [42].This shows that
configuration conditions with a linearized flight model. The required energy is directly related to parameters such
basic dynamical characteristics of the research airship Lotte as the size of the airship and mass of the payload. Solar
were determined from the flight data gathered. Similarly, power generation technologies do present some trade-offs.

DOI:10.5139/IJASS.2012.13.2.170 182

3)(170-187)(20120615).indd 182 2012-07-25 오후 3:47:27


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

For adequate power generation, solar cells need to cover array that generates power from solar energy and transmits
a substantial amount of the hull’s surface, significantly it to electrolyte cells for hydrogen and oxygen production.
increasing the weight of the vehicle and contributing to the The fuel cell would then use the hydrogen and oxygen to
hull’s skin friction, increasing the overall drag coefficient. generate power for propulsion. This cycle can be seen below
Also, solar energy is only available during the daytime which in Fig. 16. Their test results show that the RFC was found to
requires additional systems for energy storage for power meet design performance requirements.Mitlitsky et al. [104]
supply during the nighttime hours. However, solar energy also explore the use of RFC systems for high altitude long
is renewable and can be harvested during flight eliminating endurance missions.
the need to carry fuel on board and thus landing to refuel, Colozza and Dolce [103] at NASA’s Glenn Research
extending flight operations. The diagram below shows a Center studied the technologies needed to build renewable
typical comparison between the solar energy available electrical power systems for long duration observation
during the day and the energy required for engine power aircraft, including photovoltaic sources, energy storage
to refuel, extending flight operations. The diagram below
consumption demonstrating the need for efficient and high
shows a typical comparison between the solar
systems, electrical propulsion systems, waste heat rejection,
capacity energy storage systems. structural attachments, and mechanical modules to house
energy available during the day and the energy required for engine power consumption demonstrating the
Fuel cells offer a feasible and practical solution to this the equipment. Their studies concluded that long-duration,
problem. Fujihara and Eguchi[102] present a report in which coast-observing, stratospheric airships using renewable
need for efficient and high capacity energy storage systems.
a Regenerative Fuel Cell (RFC) is built and tested for airship energy systems were feasible provided appropriate
nighttime operations. The RFC testbed consists of a solar cell technology investments were made [103].

Fig. 15.Energy Storage


Fig. 15. Energy Storage Power Diagram
Power Diagram [101].[101].

Fuel cells offer a feasible and practical solution to this problem. Fujihara and Eguchi[102]

present a report in which a Regenerative Fuel Cell (RFC) is built and tested for airship nighttime

operations. The RFC testbed consists of a solar cell array that generates power from solar energy and

transmits it to electrolyte cells for hydrogen and oxygen production. The fuel cell would then use the

hydrogen and oxygen to generate power for propulsion. This cycle can be seen below in Fig. 16. Their

test results show that the RFC was found to meet design performance requirements.Mitlitsky et al. [104]

also explore the use of RFC systems for high altitude long endurance missions.
Fig. 16. Fig.
Solar RFCRFC
16. Solar Cycle
Cycle Mode During
Mode During Day/Night
Day/Night [102]. [102].

http://ijass.org
Colozza and Dolce [103] at NASA’s Glenn183
Research Center studied the technologies needed to

build renewable electrical power systems for long duration observation aircraft, including photovoltaic

sources, energy storage systems, electrical propulsion systems, waste heat rejection, structural
3)(170-187)(20120615).indd 183 2012-07-25 오후 3:47:28
Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

9. Conclusions Transportation, Washington, D.C., January 1977.


[11] Ardema, M.D., “Feasibility of Modern Airships -
Many advances have been made in airship research and Preliminary Assessment”, Journal of Aircraft, Vol. 14, No.
development in recent years because of the revived interest in 11, 1977, pp. 1140-1148.
using airships for applications such as commercial transport, [12] Carson, B.H., “An Economic Comparison of Three Heavy
research platforms, and military operations. Better analytical Lift Airborne Systems”, Proceedings of the Interagency
techniqueshave been developed for aerodynamic, dynamic, Workshop on Lighter Than Air Vehicles, University Press
and structural modeling which have been discussed in this of the Pacific, Honolulu, Hawaii, 1975, pp. 75-85.
paper. These have enabled for more reliable and functional [13] Defense Industry Daily, “Walrus/HULA Heavy-
airship designs that can meet the demands of today’s Lift Blimps Rise, Fall Rise?” URL: http://www.
applications. The development of new technologies for defenseindustrydaily.com/walrus-heavylift-blimp-
construction materials and energy systems have also made getting-off-the-ground-01103[cited 30 Dec. 2009]
modern airships and dirigibles more efficient, economical, [14] Skillings, J., ”Boeing touts heli-blimp for heavy lifting”,
and environmentally friendly. URL: CNET News.com [cited 9 July 2008]
[15] Ardema, M.D., “Vehicle Concepts and Technology
Requirements for Buoyant Heavy-Lift Systems”, Ames
References Research Center, NASA Technical Paper 1921, 1981.
[16] Naujokas, J., “Boeing Teams with Canadian Firm to
[1] Stockbridge, C.M., “Stability and Control of Build Heavy-Lift Rotorcraft”, URL:http://www.boeing.
Unconventional Airships, Design and Experimental com/news/releases/2008/q3/080708c1_pr.html [cited 8
Investigations,” MS Thesis, Mechanical and Aeronautical May 2012]
Engineering Department, Clarkson University, Potsdam [17] “Aeroscraft”, Aeros. 2011, URL: http://www.aerosml.
NY, 2012. com/imagegallery.html [cited 8 May 2012]
[2] Ardema, M.R.,   “Airship”,   AccessScience, McGraw-Hill [18] “The Airship “SkyFreighter””, Millennium Air Ship Inc.,
Companies, 2008, URL: http://www.accessscience. URL: http://www.millenniumairship.com/products.
com[cited 8 May 2012] htm [cited 8 May 2012]
[3] ORACLE ThinkQuest Education Foundation, “Non- [19] Chen, L., An, J., and Yang, C., “Exploring some key
rigid Airships”, 2005, URL:http://library.thinkquest. problems in modeling a stratospheric airship”, Journal
org/05aug/01843/[cited 8 May 2012] of Northwestern Polytechnical University, Vol. 25, No. 3,
[4] Nobile, U., ”Semi-Rigid v. Rigid Airships”, FLIGHT, 2007, pp. 387-391.
January 22, 1922. [20] Jin, O., Qu, W., and Xi, Y., “Stratospheric verifying airship
[5] Bloetscher, F., “Feasibility Study of Modern Airships, modeling and analysis”, Journal of Shanghai Jiaotong
Phase I, Final Report, Vol. 1, Summary and Mission University, Vol. 37, No. 6, 2003, pp. 956-960.
Analysis”, NASA CR-137692, 1975. [21] Chen, W., Xiao, W., Kroplin, B., and Kunze, A., “Structural
[6] Faurote, G.L., “Feasibility Study of Modern Airships, performance evaluation procedure for large flexible
Phase I, Final Report, Vol. 111, Historical Overview”, airship of HALE stratospheric platform conception”,
NASA CR-137692(3), 1975. Journal of Shanghai Jiaotong University, Vol. 12E, No. 2,
[7] Jones, B., Grant, D., Rosenstein, H., and Schneider, J., 2007, pp. 293-300.
“Feasibility Study of Modern Airships, Final Report, [22] Wang, H., Song, B., Liu, B., and An, W., “Exploring
Phase I, Vol. I”, NASA CR-137691, 1975. configuration design of high altitude airship”,
[8] Goodyear Aerospace Corporation, “Feasibility Study XibeiGongyeDaxueXuebao (Journal of Northwestern
of Modern Airships, Phase 11, Vol. 11, Airport Feeder Polytechnical University), Vol. 25, No. 1, 2007, pp. 56-60.
Vehicle”, NASA CR-151920, 1976. [23] Eguchi, K., Yokomaku, Y., and Mori, M., “Overview of
[9] Goodyear Aerospace Corporation, “Feasibility Study of Stratospheric Platform Airship R&D Program in Japan”,
Modern Airships, Phase 11, Executive Summary”, NASA 14th Lighter-Than-Air Technical Committee Convention
CR-2922, 1977. & Exhibition, Akron, OH, July 2001.
[10] Huston, R.R., and Ardema, M.D., “Feasibility of Modern [24] Perry, W.D., “Sentinel in the Sky: An autonomous
Airships, Design Definition and Performance of airship offers long-duration, high-altitude capabilities”,
Selected Concepts”, 13thAnnual Meeting and Technical Technology Today, 2010.
Display Incorporating the Forum on the Future of Air [25] Boyle, A., “Airship groomed for flight to edge of space”,

DOI:10.5139/IJASS.2012.13.2.170 184

3)(170-187)(20120615).indd 184 2012-07-25 오후 3:47:28


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

URL: http://www.msnbc.msn.com/id/5025388/ns/ of an Aerodynamic Model and Control Law Design for


technology_and_science-space/t/airship-groomed- a High-Altitude Airship”, AIAA Unmanned Unlimited
flight-edge-space/#.T5VI59W3OW8 [cited 15 May 2012] Conference, Chicago, IL, September2004.
[26] Fink, D., “Hybrid heavy lift vehicle under study”, Aviation [43] Wei, Y., Yong, L., Wenjun, W., and Wei, Z., “Stratospheric
Week, July 1974. airship optimization method and design parameters
[27] Allen, D., “Beautiful Concept Airship Looks 85 Years sensitivity analysis”, Journal of Astronautics, Vol.28, No.6,
into the Future”, Gizmodo, URL: http://gizmodo.com/ 2007, pp. 1524-1528.
wb_1010-klm-design-competition/[cited 15 May 2012] [44] Nickol, C.L., Guynn, M.D., Kohout, L., and Ozoroski, T.A.,
[28] Justa, A., “Amazing futuristic airships for sustainable air “High altitude long endurance air vehicle analysis of
travel”, Green Diary, URL:http://www.greendiary.com/ alternatives and technology requirements development”,
entry/amazing-futuristic-airships-for-sustainable-air- 45th AIAA Aerospace Sciences Meeting, Vol. 18, 2007, pp.
travel/[cited 15 May 2012] 12653-12669.
[29] Harrison, J., “Lockheed’s LEMV  Consolation”, URL: [45] Yu, D., and Lu, X., “Configurations analysis for high-
http://edgefighter.com/2010/06/17/lockheeds-lemv- altitude/ long-endurance airships”, Aircraft Engineering
consolation/[cited 8 May 2012] and Aerospace Technology, Vol. 82, No. 1, 2010, pp. 48-
[30] Schwartz, A., “Flying Sailboat Soars Through the Sky On 59.
Solar Power”, URL: http://inhabitat.com/flying-sailboat- [46] Chen, Q., Zhu, M., and Sun, K., “Analysis to Effects on
soars-through-the-sky-on-solar-power/[cited 15 May Conceptual Parameters of Stratospheric Airship with
2012] Specified Factors”, Journal of Computers, Vol. 6, No.5,
[31] Lamb, H., “The inertia-coefficients of an ellipsoid 2011, pp. 1055-1062.
moving in a fluid”, Reports and Memoranda, No. 623, [47] Ram, C.V., and Pant, R., “Multi-disciplinary Shape
October, 1918. Optimization of Aerostat Envelopes”, Journal of Aircraft,
[32] Tuckerman, L. B., “Notes on Aerodynamic Forces on Vol. 47, Issue3, 2010, pp. 1073-1076.
Airship Hulls”, Naca report, No. 129, 1923. [48] Khoury, G.A., Airship Technology Second Edition,
[33] Tuckerman, L. B., “Inertia Factors of Ellipsoids for use in Cambridge University Press, New York, NY, 2012.
Airship Design”, Naca report, No. 210, 1926. [49] Miller, J.I., and Hahon, M., “Analysis and Design of
[34] Blakemore, T.L., and Watters Pagon, W., Pressure Airship, Robust Helium Aerostats”, Journal of Aircraft, Vol. 44,
University Press of the Pacific, Honolulu, Hawaii, 2003. No. 5, 2007, pp. 1447-1458.
[35] Burgess, C.P., Airship Design, University Press of the [50] Gupta, P., “A multi-chamber. multi gas configuration
Pacific, Honolulu, Hawaii, 2004. for robust and high performance nonrigid airship”, 8th
[36] United States War Department, Airship Aerodynamics: International Airship Convention, Bedford, England,
Technical Manual, University Press of the Pacific, October 2010.
Honolulu, Hawaii, 2003. [51] Burgess, C.P., “The Strength of Rigid Airships”, Journal
[37] Konstantinov, L., “The Basics of Gas and Heat Airship of Royal Aeronautical Society, Vol. 28, No. 162, 1924, pp.
Theory”, Montgolfier, Aeroplast, Kyiv, Ukraine, 2003. 327-448.
[38] Lutz, T., and Wagner, S., “Drag Reduction and Shape [52] Hess, T.E., “Structures Technology for Lighter-
Optimization of Airship bodies”, Journal of Aircraft, Vol. Than-Air Vehicles”, Department of the Navy: Naval
35, No. 3, 1998, pp. 345-351. Air Development Center, Air Vehicle Technology
[39] Khoury, G.A., and Gillet, J.D., Airship Technology, Department, Technical Memorandum No. VT-TM-1891,
Cambridge University Press, Cambridge, UK, 1999. March 1977.
[40] Raymer D., “Aircraft Design: A Conceptual Approach, [53] Liao, L., and Pasternak, I., “A review of airship structural
Fourth Edition”, AIAA Education Series, AIAA, Reston, research and development”, Progress in Aerospace
VA, 2006. Sciences, Vol. 45, No. 4-5, 2009, pp. 83-96.
[41] Ramos, J.J.G, Maeta, S.M., Bergerman, M., Bueno, S.S., [54] Althoff, W.F., USS Los Angeles: The Navy’s Venerable
Mirisola, L.G.B., and Bruciapaglia, A., “Development Airship and Aviation Technology, Potomac Books Inc.,
of a VRML/Java Unmanned Airship Simulating Washington, D.C., 2004.
Environment”, IEEE/RSJ International Conference on [55] Burgess, C.P., “Forces on Airships in Gusts”, Bureau of
Intelligent Robots and Systems, Kyongju, Korea, October Aeronautics, Navy Department. Report No. 204, 1924.
1999. [56] Evans, J.R., and DeLaurier, J.D., “The Shenandoah flies
[42] Mueller, J.B., Zhao, Y., and Paluszek, M., “Development again: a computer simulation”, AIAA lighter-than-air

185 http://ijass.org

3)(170-187)(20120615).indd 185 2012-07-25 오후 3:47:28


Int’l J. of Aeronautical & Space Sci. 13(2), 170–187 (2012)

systems technology conference, Annapolis, MD, July CORA TM 2010-011, January 2010.
1981. [71] Linner, M., “Hydrogen and Helium”, American Chemical
[57] Yuwen, L., Nahon, M., and Sharf, I., “Airship dynamics Society, Lakehurst, NJ, May 1937.
modeling: A literature review”, Progress in Aerospace [72] Jones, R., Williams, D.H., and Bell, A.H., “Experiments
Sciences, Vol. 47, No. 3, 2011, pp. 217-239. on a Model of the Airship R.29”, ARC RM-714, London,
[58] Brewer, W.H., ”Lighter-Than-Air Structural Design HIS MAJESTY’S STATIONERY OFFICE, 1920.
Techniques for Near Term Applications”, AIAA Lighter [73] Jones, R., and Bell, A.H., “Experiments on a Model of the
Than Air Technology Conference, Snowmass, CO, July Airship R.101”, ARC RM-1168, London, HIS MAJESTY’S
1975. STATIONERY OFFICE, 1926.
[59] Burgess, C.P., ”The Longitudinal Strength of Rigid [74] Pannel, J.R, and Jones, R., “Experiments on a Model of
Airships”, Bureau of Aeronautics, Memo No. 361, July the German Rigid Airship L33, ” ARC RM-361, London,
1944. HIS MAJESTY’S STATIONERY OFFICE, 1917.
[60] Li, Y., Dynamics Modeling and Simulation of Flexible [75] Freeman, H.B., “Measurements of Flow in the Boundary
Airships, McGill University, Department of Mechanical Layer of a 1/40-Scale Model of the U.S. Airship ‘Akron’”,
Engineering, Montreal, Canada, 2008. NACA TR-430, Langley Field, VA, April1932.
[61] Hunt, J.D., “Structural Analysis of Aerostat Flexible [76] Freeman, H.B., “Force Measurements on a 1/40-Scale
Structure by the Finite Element Method”, Journal of Model of the U.S. Airship Akron”, NACA TR-432, Langley
Aircraft, Vol. 19, No. 9, 1982, pp. 674-678. Field, VA, May1932.
[62] Hunt, J.D., “Structural Analysis of the Light Weight Hard [77] Freeman, H.B., “Pressure Distribution Measurements
Nose on the 71M Aerostat”, 10th AIAA Lighter-Than-Air on the Hull and Fins of a 1/40 Scale Model of the U.
Systems Technical Conference, Scottsdale, AZ, September S. Airship Akron”, NACA TR-443, Langley Field, VA,
1993. June1932.
[63] Smith, Jr., F.A., “Advanced Finite Element Analysis for the [78] Silverstein, A., and Gulick, B.G., “Ground-handling
Skyhook-Boeing HLV Aircraft”, 2009 SIMULIA Customer Forces on a 1/40-Scale Model of the U. S. Airship ‘Akron’”,
Conference, London, England, 2009. NACA TR-566, Langley Field, VA, April1936.
[64] Bessert, N., and Frederich, O., “Nonlinear airship [79] McHugh, J.G., “Pressure-distribution Measurements at
aeroelasticity”, Journal of Fluids and Structures, Vol. 21, Large Angles of Pitch on Fins of Different Span Chord
No. 8, 2005, pp.731-742. Ratio on a 1/40-scale Model of the U. S. Airship ‘Akron’”,
[65] Mandel, M., and Tim, M., “Airship Envelopes: NACA TR-604, Langley Field, VA, April1937.
Requirements, Materials and Test Methods”, URL: [80] Zahm, A.F., “Air Forces, Moments and Damping on
https://imageserv5.teamlogic.com/mediaLibrary/93/ Model of Fleet Airship Shenandoah”, NACA TR-215, U.S.
Airship_Envelopes_Requirements__Materials_and_ Government Printing Office, January1926.
Test_Methods.pdf[cited 15 May 2012] [81] Pannell, J.R., and Frazer, R.A., “Account of Some
[66] Petrusso, A., “Airship- How products are made”, Experiments on Rigid Airship R.26”, ARC RM-674,
URL: http://www.madehow.com/Volume3/Airship. Langley Field, VA, January1920.
html[cited 15 May 2012] [82] Pannell, J. R., and Bell, A.H., “Experiments on Rigid
[67] Kang, W., Suh, Y., Woo, K., and Lee, I. “Mechanical Airship R.29”, ARC RM-675, Langley Field, VA, January
property characterization of film- fabric laminate for 1920.
stratospheric airship envelope”, Composite Structures, [83] Pannell, J. R., and Frazer, R.A., “Experiments on Rigid
Vol. 75, No. 1-4, 2006, pp. 151-155. Airship R.33”, ARC RM-668, Langley Field, VA, 1919.
[68] McDaniels, K., Downs, R.J., Meldner, H., Beach, C., and [84] Bailey, D.B., “Patrol Airship Concept Evaluation
Adams, C., “High Strength to Weight Ratio Non-Woven (PACE)”, Final Report, NADC- 85019-60, Warminster, Pa,
Technical Fabrics for Aerospace Applications”, Cubic March1985.
Tech Corp., Mesa, AZ, 2009. [85] Jex, H.R., and Gelhausen, P., “Pre- and post-flight-test
[69] Gordon, W.O., and Holland, C., “Back to the Future: models versus measured skyship-500 control responses”,
Airships and the Revolution in Strategic Airlift”, Air Force 7th AIAA lighter-than-air technology conference,
Journal of Logistics, Vol. 29, No. 3, 2005, pp. 47-58. Monterey, CA, August 1987.
[70] Ghanmi, A., and Abderrahmane, S., “Airships for military [86] Munk, M.M., “The aerodynamic forces on airship hulls”,
logistics heavy lift”, Canadian Operational Support NACA TR-184, Springfield, VA, January1924.
Command Operational Research & Analysis, DRDC [87] Allen, H.J., and Perkins, E.W., “A Study of Effects

DOI:10.5139/IJASS.2012.13.2.170 186

3)(170-187)(20120615).indd 186 2012-07-25 오후 3:47:28


Casey Stockbridge Airship Research and Development in the Areas of Design, Structures, Dynamics, and Energy Systems: ...

of Viscosity on Flow over Slender Inclined Bodies 290.


of Revolution”, NACA TR-1048, Moffett Field, CA, [97] Kornienko, A., “System identification approach for
August1951. determining flight dynamical characteristics of an
[88] Hopkins, E.J., “A Semi-empirical Method for Calculating airship from flight data”, PhD thesis, University of
the Pitching Moment of Bodies of Revolution at Low Stuttgart, Stuttgart, Germany, 2006.
Mach Numbers”, NACA RM-A51C14, Moffett Field, CA, [98] Yamasaki, T., and Goto, N., “Identification of blimp
May1951. dynamics via flight tests”, Transactions of the Japan
[89] Jones, S.P., and DeLaurier, J.D., “Aerodynamic Estimation Society for Aeronautical and Space Sciences, Vol. 46, No.
Techniques for Aerostats and Airships, ”Journal of 153, 2003, pp. 195-295.
Aircraft, Vol. 20, No. 2, 1983, pp. 120-126. [99] Mueller, J., “Guidance, Navigation and Control of High-
[90] Lutz, Th., and Wagner, S., “Drag Reduction and Shape Altitude Airships”, Princeton Satellite Systems Inc.,
Optimization of Airship Bodies”, Journal of Aircraft, Vol. Princeton, NJ, 2006.
35, No. 3, 1998, pp. 345-351. [100] Lubkowski, S., Jones, B., Rojas, E., and Morris,
[91] Lutz, Th., and Wager, S., “Numerical Shape Optimization D., “Trade-off Analysis of Regenerative Power
of Natural Laminar Flow Bodies”, 21st Congress of Source for Long Duration Loitering Airship”, Systems
International Council of the Aeronautical Sciences, and Information Engineering Design Symposium,
Melbourne, Australia, September 1998. Charlottesville, VA, April 2010.
[92] Nejati, V., and Matsuuchi, K., “Aerodynamics Design and [101] Gawale, A.C., and Pant, R.S., “Initial Sizing and
Genetic Algorithms for Optimization of Airship Bodies”, Sensitivity Analyses of Stratospheric Airships for
JSME International Journal, Series B, Vol. 46, No. 4, 2003, Psuedolite Based Precision Navigation System”, AIAA
pp. 610-617. 5th ATIO and 16th Lighter-Than-Air Systems Tech
[93] El Omari, K., Schall, E., Koobus, B., and Dervieux, and Balloon Systems Conferences, Arlington, Virginia,
A., “TURBULENCE MODELING CHALLENGES September 2005.
IN AIRSHIP CFD STUDIES”, Monografías del [102] Fujihara, T., and EguchiK., “Experimental Work
SeminarioMatemáticoGarcía de Galdeano, No. 31, 2004, on Solar Regenerative Fuel Cells for SPF Airship”,
pp. 545-554. Stratospheric Platforms Project Center, JAXA, Tokyo,
[94] Wang, X.-L., Ma, Y., and Shan, X.-X., “Modeling of Japan, 2002.
Stratosphere Airship”, Advances in Theoretical and [103] Colozza, A., and Dolce, J.L., “High-Altitude, Long-
Applied Mechanics., Vol. 2, No. 3, 2009, pp. 123 – 142. Endurance Airships for Coastal Surveillance”, National
[95] Cook, M.V., “The linearized small perturbation equations Aeronautics and Space Administration, Glenn Research
of motion for an airship”, College of Aeronautics Reports, Center, Cleveland, OH, February 2005.
WP8, Cranfield Institute of Technology, Cranfield, UK, [104] Mitlitsky, F., Colella, N. J., Myers, B., and Anderson,
January1990. C. J., “Regenerative Fuel Cells for High Altitude Long
[96] Cook, M.V., Lipscombe, J.M., and Goineau, F., “Analysis Endurance Solar Powered Vehicles”, Intersociety Energy
of the stability modes of the non-rigid airship”, The Conversion Engineering Conference, Atlanta, GA, August
Aeronautical Journal, Vol. 104, No. 1036, 2000, pp. 279- 1993.

187 http://ijass.org

3)(170-187)(20120615).indd
View publication stats 187 2012-07-25 오후 3:47:28
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/221066533

Autonomous blimp control using model-free reinforcement learning in a


continuous state and action space

Conference Paper · October 2007


DOI: 10.1109/IROS.2007.4399531 · Source: DBLP

CITATIONS READS

24 95

4 authors, including:

Christian Plagemann Peter Hilgers


Google Inc. OWIS GmbH
64 PUBLICATIONS   1,868 CITATIONS    3 PUBLICATIONS   30 CITATIONS   

SEE PROFILE SEE PROFILE

Wolfram Burgard
University of Freiburg
677 PUBLICATIONS   42,556 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multi-model Hypothesis Group Tracking View project

Distributed and Decentralised Estimation of Non-linear Systems View project

All content following this page was uploaded by Wolfram Burgard on 09 July 2015.

The user has requested enhancement of the downloaded file.


Autonomous Blimp Control using Model-free Reinforcement Learning
in a Continuous State and Action Space
Axel Rottmann∗ Christian Plagemann∗ Peter Hilgers† Wolfram Burgard∗

Abstract— In this paper, we present an approach that applies


the reinforcement learning principle to the problem of learning
height control policies for aerial blimps. In contrast to previous
approaches, our method does not require sophisticated hand-
tuned models, but rather learns the policy online, which makes
the system easily adaptable to changing conditions. The blimp
we apply our approach to is a small-scale vehicle equipped
with an ultrasound sensor that measures its elevation relative Fig. 1. Blimp used to evaluate our approach (left image). The right image
to the ground. The major problem in the context of learning shows the gondola with two rotors for pitch and thrust control. They can
control policies lies in the high-dimensional state-action space be rotated by 180 degrees.
that needs to be explored in order to identify the values of
all state-action pairs. In this paper, we propose a solution
to learning continuous control policies based on the Gaussian
process model. In practical experiments carried out on a real is specified by virtual rewards given to the system when
robot we demonstrate that the system is able to learn a policy certain system states are reached. Our approach employs a
online within a few minutes only. Kalman filter to estimate the ground clearance based on noisy
distance measurements obtained from an ultrasound range
I. I NTRODUCTION
sensor. Based on these height estimates, we apply Monte
Compared to other flying vehicles, aerial blimps have the Carlo reinforcement learning in combination with Gaussian
advantage that they operate at relatively low speed, that they processes to represent the Q-function over the continuous
do not need to move in order to keep their altitude, and state-action space. To evaluate our approach, we imple-
additionally are not overly sensitive to control errors like, mented it on the blimp depicted in Figure 1. Experimental
e.g., helicopters. In this paper, we investigate the problem results demonstrate that our approach can quickly learn a
of learning to control the height of an autonomous blimp policy that shows the same performance as a manually tuned
online and without pre-defined physical models. The goal of PD2 -T2 controller.
this work is to allow the blimp to autonomously learn the The paper is organized as follows. After discussing related
actions necessary to maintain its height from scratch after it work in the following section, we will describe the properties
has been switched on. We have several demands to such a of our blimp in Section III. In Section IV we will then
learning approach: introduce our approach to learning the control policy of the
1) It should be able to learn the control policy directly blimp. After that, we will present how to learn such a policy
on the blimp, i.e., without the need for simulation or online on the blimp in Section V. Finally, we will present
human intervention, our experimental results obtained on a real robot and in
2) it should be able to learn the policy within a few simulation in Section VI.
minutes on the real blimp, and
3) it should be able to deal with the continuous II. R ELATED W ORK
state-action space given by the current estimate for the The problem of controlling a blimp has been stud-
height and the vertical velocity. ied intensively in the past. For example, Varella Gomes
The first requirement assures that the blimp adapts its be- and Ramos [19] as well as Hygounenc et al. [8] de-
havior to the current properties of the environment and its scribe the physical principles of airship operations and use
own dynamics such as the amount of helium contained in the the non-linear dynamics to control several flight phases
envelope. The second property is necessary to ensure that the from takeoff to landing. Zhang and Ostrowski [21] use
blimp can actually learn new policies from scratch whenever a vision-guided blimp combined with a PID controller.
needed and still being able to fulfill its mission. The third Fukao et al. [6] discuss image-based tracking control for
requirement is important to deal with the continuity of the an indoor blimp. Wyeth and Barron [20] present a low level
underlying state-action spaces. reactive controller and Rao et al. [12] use a fuzzy controller.
Our approach is based on the reinforcement learning Compared to these approaches, our algorithm does not
principle, i.e., the goal to be achieved by the controller assume prior knowledge about the dynamics or pre-defined
∗ University of Freiburg, Department for Computer Science, D-79110
controllers. Instead, it learns the control policy online and
Freiburg. † University of Freiburg, Department of Microsystems Engineer- does not require a-priori information about the payload, the
ing, D-79110 Freiburg temperature, or the air pressure. Furthermore, autonomous
blimps have been presented in different application scenarios, IV. R EINFORCEMENT L EARNING
for instance by Jung and Lacroix [10], Fukao et al. [5], and Reinforcement learning is based on the idea that an agent
Green et al. [7]. interacts with a potentially unknown environment and gets
The approach that is most closely related to the one rewarded or penalized according to the actions it performs.
described in this paper has recently been presented by In general, the agent receives rewards for actions that are
Ko et al. [11]. They also deal with the problem of learning beneficial in certain states for achieving a long-term goal.
to control an autonomous blimp and choose a similar set of The agent thus seeks to behave in a way that maximizes
methods for this task. In contrast to our approach, however, its numerical reward. A reinforcement learning task can
they make extensive use of models of the dynamics, both be defined by a tuple {S, A, δ, r}, consisting of states S,
for describing an analytical motion model as well as for actions A, a transition function δ : S × A → S, and a reward
the controller itself. Gaussian processes are used to learn function r : S ×A → R, which defines the immediate reward
the residuals of these models in a supervised manner using to be received when executing action a in state s. The goal
motion capture data as ground truth. In contrast to their is to determine a policy function π : S → A, which maps
work, we aim at learning the control policy directly, without each state s to an action a, such that the future expected
representing an explicit motion model or based on simula- reward is maximized. As described by Sutton and Barto [18],
tion. Using only the real hardware as source of information the expected long time reward in state st can be expressed
about the dynamics, our system learns a non-parametric recursively as
representation for the Q-Function while the system is flying. ∞
X
Additionally, Gaussian processes have been applied to Rπ (st ) = γ i rt+i , (1)
the reinforcement learning problem in various ways. En- i=0
gel et al. [2] approached the problem from the viewpoint
where γ ∈ (0, 1] is a discount factor. In general, the sequence
of temporal difference learning (GPTD) and later extended
of rewards rt+i is obtained by starting in state st and then
this scheme to be able to deal with stochastic state transitions
iteratively applying a policy π for selecting the next action
to improve action selection and to learning Q-values without
at = π(st ).
an explicit transition model (GPSARSA) [3]. Their approach
was successfully applied to the problem of learning complex A. On-Policy Monte Carlo Control
manipulation policies [4]. Continuous state spaces have also Several approaches have been proposed to solve the re-
been considered by Smart and Kaelbling [17]. Rasmussen inforcement learning problem by maximizing the expected
and Kuss [13] present a different approach, in which the long time reward as stated in (1). In this paper we apply the
system dynamics and the value function are learned using Monte Carlo (MC) method, which has the advantage that
separate Gaussian processes. In contrast to their work, we it allows the agent to learn directly from experience while
show that our approach can be directly applied on a real interacting online with a completely unknown environment.
robot and be utilized to learn a policy online. This enables us to learn without prior knowledge and also in
situations in which no simulation environment is available.
III. T HE AUTONOMOUS B LIMP Formally, we seek to learn the state-action value function
Q(s, a) : S×A → R, representing the expected future reward
The approach presented in this paper has been targeted for
when selecting action a in state s. To gather training data, we
the system depicted in the left image of Figure 1. It is based
iteratively generate episodes e1 , . . . , eT . Each episode et =
on a commercial 1.8 m blimp envelope, which is described 
(s1 , a1 ), . . . , (sN , aN ) consists of state-action pairs, which
in detail in [16]. The blimp is steered by three motors. One
are in turn selected using a given policy. An episode ends
motor is mounted in the tail fin to control the yaw. The other
when a pre-defined goal state is reached, which typically
two are mounted on each side of the gondola to control the
yields maximum reward or a maximum episode length is
pitch and thrust. The latter two are also attached to a shaft
exceeded. After the execution of an episode, the Q-values
which can be rotated up to 180 degree by a servo (see right
are derived by averaging over the expected long time rewards
image of Figure 1). The gondola includes an Intel XScale
PXA270 based system-on-a-chip with 600 MHz and 128 MB Q(s, a) = R(s, a) , (2)
RAM. An on-board 32 MB flash memory serves as storage
where R(s, a) is the average long time reward if action a is
for the Linux operating system. This board is used to control
executed in state s. In this scheme, the best policy is given
the speed of the motors, to change the position of the servo,
by the maximum over the Q-value function
and to process the sensor data. For our experiments we used
a downward-facing ultrasound sensor at the bottom of the π(s) = arg max Q(s, a) . (3)
gondola. The measurements of this sensor are integrated a

on-board by a Kalman filter which sequentially estimates A commonly applied strategy for generating the training
the height and the vertical velocity of the vehicle. The total episodes e1 , . . . , eN is the so-called ǫ-greedy approach. Un-
weight of the gondola and hardware is about 200 grams der this scheme one chooses a random action with probability
and the complete system is powered via a 3.7 V/1500 mAh ǫ ∈ (0, 1] and, otherwise, selects the currently best action as
lithium polymer battery. defined by (3).
TABLE I
B. Learning the Q-Function
M ONTE C ARLO LEARNING USING G AUSSIAN PROCESSES TO
A crucial choice for any learning task is the type of APPROXIMATE THE Q- FUNCTION .
representation that is to be used for the core concepts. In the
reinforcement learning task outlined above, we seek to learn • Input: amin , amax , ǫ, γ
the Q-function from sampled action sequences. Most existing • Repeat until convergence:
approaches to the problem represent the Q-function using a 1) Generate an episode (s1 , a1 ), . . . , (sN , aN ) of state-action
discrete approximation over the space of state-action pairs. pairs using the ǫ-greedy policy.
2) For each pair (sn , an ) appearing in the episode calculate
This, however, can lead to discretization errors or, when fine- the expected long time reward
grained grids are used, requires a huge amount of memory
N −n
and a time-consuming exploration process. Therefore, func- R(sn , an ) =
X
γ k rsn+k ,an+k ,
tion approximation techniques that directly operate on the k=0
continuous space such as neural networks [1], [15], kernel where γ ∈ (0, 1] and r represents the immediate reward.
methods [9], or Gaussian processes [13], [3] have been pro- 3) Add all R(sn , an ) to the training set D of the Gaussian
posed as powerful alternatives to the discrete approximations process GP and optimize the hyper-parameters θ and σn .
of the continuous Q-function. From a regression perspective, • The best policy π for a state s′ is given by
these techniques seek to model the dependency π(s′ ) = arg max GP (s′ , a),
a

qi = Q(si , ai ) + ǫi (4) where a ∈ [amin , . . . , amax ].

for the unknown function Q and independent and identically,


normally distributed noise terms ǫi , given a training set D =
{(xi , qi )}D
i=1 of state-action pairs xi = (si , ai ) ∈ S × A and To avoid the blimp having to wait until the parameters have
estimates of their Q-value qi . been updated, we realized a thread-based implementation
Gaussian processes can be seen as a generalization of which performs the optimization in parallel to the learning
weighted nearest neighbor regression [14], where the de- task. In the simulation experiments, however, the optimiza-
pendency of a function value on its local neighborhood tion is carried out after each round of 100 episodes.
is described using parameterized covariance functions. We
imply in this work the squared exponential V. O NLINE P OLICY L EARNING ON A R EAL B LIMP
For the task of learning online to control an autonomous
 
d 2
X (xj − y )
j  blimp, several specific constraints have to be met. First, we
k(x, y) = σf2 exp − 2 , (5)
j=1
2ℓj need to specify the reward function, which implicitly defines
how we want the blimp to behave and react. Second, we need
where ℓ is the length-scale and σf the signal variance. In to define how the episodes are created during continuous
our model, the inputs x and y with x, y ∈ S × A are online learning without restarting the blimp. Finally, we need
d-dimensional vectors, which are constructed by concate- to define the state and action space and need to explicitely
nating the corresponding state and action vectors. Given a constrain the state space to limit exploration.
set of data samples D from the Q-function and the hyper- We consider the task of stabilizing the blimp at a given
parameters θ = (σf , ℓ1 , . . . , ℓd ) of the covariance function, altitude h∗ without knowing the specific dynamics or any
arbitrary Q-values q ∗ can be predicted for a new input parameters of the environment. In practice, altitude control is
location x∗ by already a complex task as blimps are very sensitive and their
−1 behavior highly depends on outer influences such as payload,
q ∗ = k∗ K + σn2 I q, (6)
battery level, temperature, and air flow. In general, it is hard
where K ∈ RD×D is the covariance matrix for the train- to determine a globally suitable policy applicable to several
ing points, Kij = k(xi , xj ), k∗ ∈ RD , represents the conditions. Therfore, we seek to learn the best policy for the
covariances between training points and the query point current conditions while the blimp is in operation. For this
x∗ , ki∗ = k(x∗ , xi ), and q = (q1 , . . . , qD ) represents the task, we propose to use the on-policy MC approach to rein-
Q-values from the training set. For details about Gaussian forcement learning. To deal with the continuous Q-function,
process regression, we refer to the book of Rasmussen and we apply Gaussian processes as approximation technique.
Williams [14] or the numerous other introductory works The resulting algorithm is outlined in Table I.
available. The learning task consists of two parts, i.e., the We define the state space S by st = (dt , vt ), where
gathering of training data D, which has been described in the dt = ht − h∗ represents the distance to the goal altitude h∗ .
previous section, and the adaptation of the hyper-parameters The terms ht and vt are the estimated height and velocity
θ of the covariance function as well as the global noise values calculated by the Kalman filter. The velocity v is
parameter σn . In the Gaussian process framework, these negative, if the blimp descends. The action space A is
parameters can be optimized by maximizing the marginal naturally limited by the capabilities of the motors, i.e., by
data likelihood of the observed training data D. In our current an interval at ∈ [amin , amax ].
system we optimize these parameters using gradient descent. To apply MC learning to the blimp, we need to specify
how the episodes are created. In our application, we have to the experiments is to demonstrate that our approach learns
sample the episodes e1 , . . . , eN from a sequence of measure- to control a real blimp online and within a short time-frame
ments (s1 , a1 ), . . . (sT , aT ) obtained while the blimp is mov- without any prior information about the dynamics of the
ing trough the environment. For each state-action pair (st , at) vehicle. Furthermore, we describe results indicating the
in the sequence an episode et = (st , at ), . . . , (st+p , at+p ) improvement obtained by utilizing Gaussian processes for
is generated consisting of the p successor states. The length function approximation without loosing precision. Finally,
p of an episode is defined by the factor γ. An episode ends we compare the behavior of our policy learned online with
if the factor γ p is smaller than a given threshold θ which has a manually tuned PD2 -T2 controller.
been set to 0.1 in our current implementation. The expected The real-world experiments described in this section have
long time reward for each state-action pair is finally given been conducted with the small-scale blimp described in
by Section III. The challenge is to control the altitude without
p knowing the specific dynamics of the blimp or any param-
X
π
R (st , at ) = γ i rt+i . (7) eters of the environment. In practice, height control already
i=0 is a complex task as blimps are very sensitive to outer
influences like payload or air flow. Additionally, it has to be
To guarantee convergence to the optimal policy, the learn-
taken into account that the resolution of the ultrasound sensor
ing task would either have to be endlessly restarted at random
is 1 cm and since the velocity is very small the estimation
initial states to assure that all state-action pairs are visited,
is susceptible to errors. Therefore, the estimated state of
or, like in our work, a continuous policy would have to
the blimp is extremely noisy and accordingly the results of
ensure that all state-action pairs are visited with a non-zero
potential actions can hardly be predicted.
probability.
To evaluate the progress of the learning processes we
An additional condition induced by continual movement
derived the dynamics of the blimp based on standard physical
of the blimp is the limitation of the exploration space. In our
aeronautic principles [19]. Since we only took the movement
case, in which the blimp is never restarted in well-defined
in the vertical direction into account, the effects to all other
starting states, the system may move outside of the safe
dimensions were ignored. Afterwards, the produced forces of
working environment and, for example, hit the floor or
the motors depending on the control signals were determined
ceiling. Our experiments were performed in a factory build-
on a measuring system and the parameters of the dynamics
ing with a ceiling height of 5 m. We avoid collisions, by
were optimized based on several test flights. We applied
artificially guaranteeing that the selected action at satisfies
the resulting dynamics to move the blimp in simulation as
the equation dt · at < 0.0 if the absolute value of the current
well as to determine a policy we can use as a base-line in
distance dt is greater than 1.5 m.
our experiments. This policy was computed using dynamic
Finally, to stabilize the blimp at a given altitude h∗ , we
programming [18] with a transition function based on the
have to specify the conditions of the learning task that
dynamics. Please note, that this policy is optimal with respect
the agent reaches the goal when it tries to maximize the
to the dynamics, but not in general for all conditions. As can
expected long time reward. A common strategy is a boolean
be seen from the experiments, this policy is well-suited to
reward function, means, if the goal is reached r+ is received
evaluate the learning progress.
otherwise r− , where r+ > r− . In our specific task, we
In all experiments only the two motors mounted at the
found that such a boolean reward function requires a longer
gondola were used. They were oriented upwards to control
learning phase than a reward function that is proportional to
the altitude and the maximum values amin and amax for the
the distance to the goal
actions were set to −0.8 and 0.8 respectively. These values
represent the engine speed in percent. We also analyzed
r(st , at ) = −|dt |. (8) different time intervals between the processed measurements
and obtained the best results for a time delay of 1 second.
A common method to learn the optimal policy is the so Whereas shorter intervals made it nearly impossible to pre-
called actor-critic [18] which separates the policy from the dict the effects of an action, larger intervals complicated the
value function. One structure is used to select the action learning in environments of limited height like our factory
whereas the other represent the actual learned progress. building as the blimp typically quickly reaches the maximum
This guarantees that the policy is fixed while learning the height or gets too close to the floor.
Q-values. In our current system, we use one structure to
simultaneously represent the values and the current ǫ-greedy A. Simulated Learning
policy. In experiments not reported here, we compared these In the first experiments we analyzed whether the integra-
two methods and achieved no significant difference in the tion of a Gaussian process for function approximation yields
resulting policies. an improvement of the learning process. To perform this
experiment, we simulated the movement of the blimp based
VI. E XPERIMENTS on the dynamics with additional noise and learned control po-
The approach described above has been implemented and lices using the standard MC approach and our MC approach
tested in simulation as well as on a real robot. The goal of with Gaussian processes. Although the time-complexity of
90 200
monte carlo online learning
error to base-line policy 80 monte carlo using GPs 180

error to base-line policy


160
70
140
60 120
50 100
40 80
30 60
40
20
20
10 0
0 -20
0 10000 20000 30000 40000 50000 0 50 100 150 200 250 300
time (sec) time (sec)

Fig. 2. The sum of squared differences between the present best actions Fig. 3. Learning rate of our Monte Carlo learning approach using Gaussian
and the base-line policy for the standard Monte Carlo learning approach processes applied online on the real blimp.
and our approach using Gaussian processes.
4
online learning
inverting the covariance matrix in Gaussian processes can 3

altitude (m)
be speed up by the fact that they are symmetric and positive
definite, the number of samples D has to be limited in the 2
case of long runs. Therefore, we average two samples if their
1
euclidian distance is smaller than a given threshold of 0.01.
Thus, we follow the idea of MC but still adhere the property 0
of continuous spaces. Incidentally, for short-term runs the 1 ε
reduction of the samples is needless since the calculation of
the inverse matrix with 1, 000 samples requires less than one 0
second. 0 50 100 150 200 250 300
Figure 2 plots the sum of the squared differences to the time (sec)
base-line policy averaged over multiple runs. As can be
Fig. 4. The upper graph shows the altitude of the blimp while it learns to
seen from the figure, our approach using Gaussian processes stabilize at 2 m. The lower graph plots the schedule of the ǫ-greedy policy.
for approximating the Q-function converges significantly
(α = 0.05) faster and also provides a significantly better pol-
icy up to 50, 000 seconds. This illustrates that the approxima- rate is very instable which is due to the fact that the low
tion provided by the Gaussian process is extremely beneficial number of samples typically introduce a high variance in the
when only a few states have been visited. Furthermore, it Gaussian process. As the number of samples increases, the
also yields a good performance in the long-term without learning rate stabilizes and converges quickly to an accurate
sacrificing precision despite the reduction of the sample estimate, which is close to the base-line policy.
space. Due to space restrictions we only plot the graph up to Figure 4 shows a typical evolution of the height of the
time-step 50, 000. Note that even up to 1, 000, 000 time-steps vehicle and the schedule of the ǫ-greedy policy for one
the error of the standard MC approach never fell below the learning experiment. As can be seen, at the beginning the
error obtained with our method. blimp is exploring the states and with descending ǫ the
blimp is gradually exploiting the current policy and finally
B. Learning Online on a Real Blimp stabilizing at the goal of 2 m. Figure 5 depicts the final policy
This experiment is designed to demonstrate that our MC learned during this run, which lasted 300 seconds.
approach to reinforcement learning using Gaussian processes
for approximating the Q-function can be applied online on a C. Comparison to a Manually Tuned PD2 -T2 Controller
real blimp. It also illustrates that our approach allows us to The final experiment is designed to illustrate the benefit
efficiently learn on a completely model-free, real system with of online learning on real systems compared to a manually
unknown dynamics. To perform this experiment we ran the tuned PD2 -T2 controller, which is a PD controller with
blimp in a factory building with a vertical exploration space two D terms and two delay elements. This PD2 -T2 has
of 5 m. Figure 3 illustrates the average learning rate over been developed based on the dynamics and characteristics of
multiple runs compared to the base-line policy calculated our blimp system and the parameters of the controller have
from the blimp dynamics. As can be seen from the graph, been optimized in simulation. Additionally, we integrated
our approach is able to learn a good policy already after only a virtual PWM module to deal with the dead-zone of the
a few states have been visited. Note, that every second only non-linear actor function. Figure 6 shows the behavior of the
one new state is inspected. At the beginning, the learning blimp using the PD2 -T2 and the controller learned with our
2.5

best action 2
0.8
0.6
0.8 0.4

altitude (m)
0.4 0.2 1.5
0
0 -0.2
-0.4
-0.4 -0.6 1
-0.8
-0.8
0.4
0.2 0.5
-1 0 policy learned online
-0.50 -0.2 velocity (m/s2)
0.5 manually tuned controller
distance to goal (m) 1-0.4 0
0 20 40 60 80 100
time (sec)

Fig. 5. Policy learned online on the real blimp with our Monte Carlo Fig. 6. Altitude progress obtained by applying the online learned policy
learning approach using Gaussian processes within three minutes. and the policy of the manually tuned controller while the blimp stabilizes
at 2 m.

approach over a time period of three minutes. As can be seen,


the policy learned online stabilizes the blimp as efficient as [4] Y. Engel, P Szabo, and D. Volkinshtein. Learning to control an octopus
arm with gaussian process temporal difference methods. In Advances
the PD2 -T2 controller and yields a similar behavior. It can in Neural Information Processing Systems (NIPS), 2005.
also be observed that the controller learned online is not [5] T. Fukao, K. Fujitani, and T. Kanade. An autonomous blimp for a
penalized by the reward function for overshooting and thus surveillance system. In Proc. of the IEEE/RSJ Int. Conf. on Intelligent
Robots and Systems (IROS), 2003.
behaves different to approach the goal. [6] T. Fukao, K. Fujitani, and T. Kanade. Image-based tracking control of
a blimp. In Proc. of the IEEE Conf. on Decision and Control, 2003.
VII. C ONCLUSIONS AND F URTHER W ORK [7] W. Green, K. Sevcik, and P. Oh. A competition to identify key
In this paper we presented an approach to control the challenges for unmanned aerial robots in near-earth environments. In
Proc. of the Int. Conf. on Advanced Robotics (ICAR), 2005.
height of a blimp using reinforcement learning. Our ap- [8] E. Hygounenc, I-K. Jung, P. Soueres, and S. Lacroix. The autonomous
proach applies Monte Carlo reinforcement learning and blimp project at LAAS/CNRS: Achievements in flight control and
utilizes Gaussian processes for dealing with the continu- terrain mapping. In International Journal of Robotics Research, 2003.
[9] N. Jong and P. Stone. Kernel-based models for reinforcement learning
ous state-action space. Our approach does not require any in continuous state spaces. In ICML workshop on Kernel Machines
pre-defined models about the dynamics of the blimp and is and Reinforcement Learning, June 2006.
able to learn a policy online on a real blimp. As a result, [10] I-K. Jung and S. Lacroix. High resolution terrain mapping using low
altitude aerial stereo imagery. In Int. Conf. on Computer Vision, 2003.
our approach can be applied whenever the blimp is switched [11] J. Ko, D. Klein, D. Fox, and D. Hähnel. Gaussian processes and
on and does not require any tuning of parameters whenever reinforcement learning for identification and control of an autonomous
internal or external conditions change. blimp. In Proc. of the IEEE Int. Conf. on Robotics & Automation
(ICRA), 2007.
Our approach has been implemented and evaluated on a [12] J. Rao, Z. Gong, J. Luo, and S. Xie. A flight control and navigation
real blimp and has been compared to policies learned based system of a small size unmanned airship. In Proc. of the IEEE
on prior information of the system dynamics. Experimental Int. Conf. Mechatronics and Automation, 2005.
[13] C. E. Rasmussen and E. Kuss. Gaussian processes in reinforcement
results demonstrate that our approach can quickly learn an learning. In Advances in Neural Information Processing Systems
effective policy and also performs equally well as a manually (NIPS), 2004.
tuned PD2 -T2 controller. [14] C.E. Rasmussen and C. Williams. Gaussian Processes for Machine
Learning. MIT Press, 2006.
ACKNOWLEDGMENT [15] M. Riedmiller. Neural fitted Q iteration - first experiences with a
data efficient neural reinforcement learning method. In Proc. of the
This work has partly been supported by the DFG within European Conference on Machine Learning (ECML), 2005.
the Research Training Group 1103, by the EC under con- [16] A. Rottmann, S. Sippel, T. Zitterell, W. Burgard, L. Reindl, and
C. Scholl. Towards an experimental autonomous blimp platform. In
tract number FP6-IST-34120-muFly (action line: 2.5.2.: mi- Proc. of the Europ. Conf. on Mobile Robots (ECMR), 2007.
cro/nano based subsystems) and BP6-004250-CoSy, and by [17] W. Smart and L. Kaelbling. Practical reinforcement learning in
the German Ministry for Education and Research (BMBF) continuous spaces. In Proc. of the Int. Conf. on Machine Learning
(ICML), 2000.
through the DESIRE project. [18] R.S. Sutton and A.G. Barto. Reinforcement Learning: An Introduction.
MIT Press, Cambridge, MA, 1998.
R EFERENCES [19] S. Varella Gomes and J. Ramos. Airship dynamic modeling for
[1] R. Coulom. Reinforcement Learning Using Neural Networks, with autonomous operation. In Proc. of the IEEE Int. Conf. on Robotics &
Applications to Motor Control. PhD thesis, Institut National Poly- Automation (ICRA), 1998.
technique de Grenoble, 2002. [20] G. Wyeth and I. Barron. An autonomous blimp. In Proc. of the IEEE
[2] Y. Engel, S. Mannor, and R. Meir. Bayes meets Bellman: The Gaussian Int. Conf. on Field and Service Robotics (FSR), 1997.
Process approach to temporal difference learning. In Proc. of the [21] H. Zhang and J. Ostrowski. Visual servoing with dynamisc: Control
Int. Conf. on Machine Learning (ICML), 2003. of an unmanned blimp. In Proc. of the IEEE Int. Conf. on Robotics
[3] Y. Engel, S. Mannor, and R. Meir. Reinforcement learning with & Automation (ICRA), 1999.
gaussian processes. In Proc. of the Int. Conf. on Machine Learning
(ICML), 2005.

View publication stats


JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, JANUARY 2009 43

Developing a Low-Cost Autonomous Indoor


Blimp
P. González1, W. Burgard2, R. Sanz1 and J.L. Fernández1

inspection of power lines and pipelines, mineral and


Abstract— This paper describes the design of an autonomous archaeological site prospection, law enforcement and
blimp-based robot and its navigation system. The robot was based telecommunication relay systems [1]. Blimps are well-suited
on a commercial kit and its dimensions were suitable for use in for these applications because their ability to remain stationary
indoor environments. Our main goal was to develop a simple and
safe model for evaluating different autonomously controlled
for long periods of time in the air enables data to be gathered.
navigation techniques. Due to the special requirements of this Blimps can also be used for research purposes in a variety of
application, two specific electronics boards to control the blimp applications including ecological, biodiversity and climate
and to communicate with the PC ground station were designed research and monitoring in different environments [2].
and two different altitude controllers and also a controller to Our primary interest was the development of a low-cost
maintain distance from obstacles were implemented. Finally, blimp designed to operate autonomously in indoor
comparative results on both altitude controllers are presented.
environments where different control strategies and navigation
Index Terms— Blimp, autonomous robot, robot navigation, paradigms are tested and evaluated. The design of a blimp
altitude controller, control avoidance controller, fuzzy logic imposes certain restrictions, primarily because of its limited
control. payload capability, given that a blimp relies on its neutral
buoyancy to stay afloat. A key challenge was to build an
I. INTRODUCTION electronic board that was sufficiently light to be carried on
board the blimp. Electronic components were selected to fit
N on-rigid airships, also known as blimps, are basically
unmanned aerial vehicles (UAVs) that use gas (usually
helium) balloons. In contrast to a rigid airship, a blimp has no
our main navigation requirements including limited
autonomous navigation capabilities.
internal structure to maintain the shape of its hull envelope. This paper is structured as follows. Section 2 discusses
Rather, its shape is maintained by a higher pressure of the gas. related works. The commercial blimp selected and other main
The only rigid components are the driving elements, the fins components are described in detail in Section 3. Section 4
and the gondola attached to the envelope. describes the design and implementation of the navigation
Unmanned blimp robots can be used in both indoor and system, which basically consists of a fuzzy logic obstacle
outdoor environments. The buoyancy force provides an avoidance controller and an altitude controller. Two different
energy-free form of lift, offering a non-traditional approach to control techniques were experimented with for the altitude
long-duration missions for which conventional aircrafts are not controller: a simple proportional–integral–derivative (PID)
well-suited. Miniaturization of sensors and actuators and the linear controller and a fuzzy logic controller. The experimental
development of long-duration batteries have also opened up results for these systems are also compared in this section.
opportunities for further progress in the development of these Finally, Section 5 summarizes our conclusions.
small-scale autonomous vehicles.
The first rigid airships, which were constructed in the early II. RELATED WORKS
20th century, consisted of a balloon with a metal frame Several researchers have recently developed autonomous
covered by fabric and filled with a gas (helium or hydrogen). robotic systems based on blimps and studied appropriate
These airships were mainly used in wars for military aerial control paradigms. Much of this research is devoted to large-
exploration and transportation. Nowadays, however, they are scale systems, with payloads of kilograms and therefore
mainly used for advertising and aerial filming. Nevertheless, capable of including a number of sensors (such as cameras)
they have great potential in terms of applications such as and remaining airborne for long periods of time. Elfes and
search and rescue missions, traffic monitoring, urban planning, colleagues [1] presented arguments that favor blimps over
airplanes and helicopters as ideal platforms for standard aerial
This work was partially funded by Spanish Ministerio de Ciencia y exploration missions. Kantor and colleagues [2] discussed the
Tecnología (DPI2005-06210) and the Xunta de Galicia (PGIDIT06PXIC use of solar energy as a renewable source of power for airships
303194PN).
1 using an outdoor blimp. Hygounenc and colleagues [3]
Dept. of Systems Engineering and Automation, University of Vigo,
focused on flight control and terrain mapping issues in
Spain.
2 cooperation between ground and aerial robots.
Dept. of Computer Science, University of Freiburg, Germany.
44 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

Other authors described the use of blimps in indoor Due to the limitation on payload, onboard hardware could
environments. Motoyama and colleagues [4] designed an not be equipped with sensors to measure absolute vehicle
autonomously controlled indoor blimp and an action-value position and this limited the autonomous navigation
function for motion planning based on a potential field capabilities of our vehicle (as described below).
method, evaluating its effectiveness in a simulated The main characteristics of the selected blimp components
environment [5]. Geoffrey and colleagues [6] used a and the design of other elements are described immediately
commercial indoor blimp, concluding that the vertical motor below.
was severely underpowered in tele-operated control. These
A. Blimp selection
researchers used a commercial wireless board to send the
sensor measurements to the ground computer and a separate Payload, which depends on balloon volume, is a critical
board to control the motors. A PC running Linux was used to aspect that constrains the choice of other onboard components
process sensor data and send control signals to the blimp and a such as the battery.
wireless communication unit (WCU) was used for sensor data The blimp that we selected was a hobby radio-controlled
communication and blimp-side servo control. Hydrogen and (RC) blimp from Plantraco, which comes with an RC
helium were used to increase the payload capacity. transmitter than can be connected via USB to a PC and
We used the same type of blimp as Geoffrey and colleagues controlled via a connection to a TCP socket. It has a 52” non-
[6], but with a different vertical motor, as a more powerful rigid hull made of a light material, achieving 200 grams of
motor and a big propeller were necessary to control altitude so estimated payload capacity (Figure 1) [9] [10]. This size is
as to have full altitude control in each indoor environment in very appropriate for indoor laboratory applications. It can fly
which the vehicle was tested. We also developed a lighter in a corridor with people and enter rooms through standard-
specific board to control the motors and sensors and thus had sized office doors. The payload capacity of this blimp was
more payload capacity for additional components. We only considered adequate for our purposes.
used helium, because hydrogen is flammable and so is very The blimp has a tri-turbofan gondola with three light DC
dangerous in indoor environments. micro-motors: a vertical motor allows altitude to be regulated
An important navigation problem is automatic control of and the other two motors control speed and rotation. Since the
altitude and of horizontal movement. If the blimp can be original vertical motor does not have enough power to
maintained at a specific altitude, it can be moved in a properly control altitude, we replaced it with a Futaba 3003
horizontal plane. Kadota and colleagues [7] used PID servo motor and a bigger propeller blade. With all these
controllers to control blimp altitude and horizontal movement, changes the gondola components weighed only 55 grams,
arguing that blimp trajectory could be unstable in the vertical leaving 74 grams free for other hardware components
direction. We also used PID and fuzzy logic controllers to including the battery (Table I).
control blimp altitude and evaluated the performance of the TABLE I
control systems developed using the controllers in two BLIMP COMPONENT WEIGHTS
different environments. Component Weight (g)
A second important navigation problem for blimps (and for Envelope (52"x37") 60.5
autonomous mobile robots in general) is obstacle detection and Gondola components 55.0
collision avoidance. Green and colleagues [8] used an infrared Fins and propellers 10.5
sensor to detect obstacles: when the collision avoidance system Onboard hardware 39.0
detects an obstacle, the blimp turned 180 degrees to avoid Battery 35.0
collision. The collision avoidance system does not take Total blimp weight 200.0
account of blimp dynamics, however; consequently, we used
an ultrasonic sensor to measure the distance to potential
obstacles and implemented a fuzzy logic controller to avoid
collisions.

III. BLIMP DESIGN


The design of an autonomous blimp has certain restrictions
arising from the assembled hardware. Various issues were
evaluated simultaneously in terms of making appropriate
choices. For a blimp system, the higher the volume of the
envelope—that is, the higher the ascending force—the higher
the possible payload. However, a blimp for indoor applications
has to be fairly small. For our blimp system, the goal was to
minimize both the size of the blimp and the weight of the
necessary onboard hardware.
Fig. 1. The commercial blimp selected for this research, with a tri-
GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 45

turbofan gondola at its base. Balloon size was 52”. The vertical motor and the onboard wireless communication link (see Figure 3).
controls altitude and the side motors control horizontal movements.
The main components are:
Power supply unit. It consists of a power regulator which
The choice of battery was another key aspect because of the provides two stabilized voltage levels: 8V and 5V and a
weight issue: a very light battery that still provided enough battery circuit which prevents total discharge of the battery.
current and autonomy was required. We selected a 350mA Microcontroller unit (MCU). It is based on a PIC
lithium polymer battery based on a new technology. Weighing microcontroller, which remains the best balance of cost,
only 35 grams, it supplies a maximum current of 3A and processing power, complexity, and power consumption. The
provides around 40 minutes of autonomy for the blimp. The PIC 16F873 microcontroller from Microchip appears to be a
nominal voltage level of 11V was too high for the electronic good choice. It has USART, analog ports, and I2C ports
components, and so a power management circuit to provide required to connect sensors and motor drivers. Some other
the required voltages (8V and 5V) was designed. A circuit to inputs and outputs are used to communicate with the wireless
control voltage level was also added to avoid damage to the transceiver.
battery when the voltage was under 9V.
B. Ultrasonic sensors
Two ultrasonic sensors provided the autonomous blimp with
information on the environment. An ultrasonic, lightweight
SRF05 sensor (with a resolution of 1 mm and a very narrow
beam) was mounted facing downwards at the bottom of the
gondola to measure the distance from the blimp to other
objects. Sensor measurements were integrated by means of a
Kalman filter which sequentially estimated blimp altitude. The
other ultrasonic sensor, located in the forward-facing part of
the balloon, was used by the obstacle avoidance controller. In
this case, we selected a lightweight SRF10 ultrasonic sensor
with an operating range of up to 6 meters that could be
connected to the microcontroller via a standard I2C bus
interface.
C. Electronic components
An electronic board was custom-designed for this
application because no commercial board met with our
requirements (Figure 2). An appropriate communication
system was also designed and implemented. The main reasons
motivating this approach were the following:
Fig. 2. Block diagrams of the gondola onboard and PC interface cards.
• A new motor speed control was necessary, firstly, because
the blimp could not be controlled using the original circuit, Motor drivers. They are necessary to control the speed of
which only worked at maximum speeds, and secondly, each motor. The drivers are based on a L293B circuit, which
because we could not control the new vertical motor with offers 1A per channel and can modulate the voltage and
the original board. control the motor speed. The pulse-width modulation (PWM)
• Good wireless communication coverage with a lightweight is directly controlled by the microcontroller.
circuit was desirable, so lighter components for Wireless communication unit (WCU). It is employed to
communications between the vehicle and the PC ground transmit data between the blimp and the PC ground station. It
station were selected, given that a wireless access point or consists of a serial wireless modem that works in the 400 MHz
a wireless router could not be used because of the limited band, and it is based on an ER400TRS transceiver. It also
payload capability. includes a buffer to protect the transceiver. Data are
• A bidirectional communication link to send and receive transferred in real time at a speed of 19,200 bps. The WCU
data from the blimp was required in order to be able to sends sensor data to the ground PC and receives servo
send data to and from the blimp and the PC ground station positions from the ground PC to control the blimp (Figure 2).
(in other words, we needed to close the control loop). PC interface card. The blimp was remotely controlled using
For the above reasons, two specific boards were designed: a a customized computer board that also contained a WCU. We
gondola onboard card and a PC interface card. also developed a software package module with generic
functions that enabled easy control of the blimp by the
Gondola onboard card. This board included all the programmer.
electronic components necessary to control the three motors
46 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

PID and fuzzy logic controllers. PID parameters can be


experimentally adjusted on the basis of well-known methods
and fuzzy logic controllers can be easily tuned on the basis of
trial and error.
Our control system was composed of two different
controllers that would ensure safe autonomous navigation in
indoor environments: an altitude controller and a collision
avoidance controller. Both controllers were implemented in
the PC ground station using the distance measurements sent by
the WCU, with computed control signals sent back to the
onboard microcontroller.
The reference blimp altitude was specified in the program
user interface, with the altitude control algorithm endeavoring
to automatically maintain the blimp at this altitude. Controlling
Fig. 3. Electronic components of the gondola onboard card. vertical motion reduced blimp movements by one degree of
freedom. Two different altitude controllers were implemented
The ground computer could both write and read from the and tested.
WCU using these functions. The design of the function The collision avoidance controller enabled control of the
interface was such that the programmer was not required to horizontal movements of the blimp so as to avoid frontal
deal with serial port characteristics and the communication collisions. Only a fuzzy controller was implemented for this
protocol between the blimp and the computer. The interface purpose because the results for PID controllers were poor.
had ten user-friendly functions so that the programmer could When the front sensor did not detect an obstacle in its path, the
develop software to move the blimp in few minutes. controller navigated the blimp along a straight line; when an
obstacle was detected, the controller kept the blimp at a certain
D. Onboard software distance from the obstacle.
The onboard software processed the sensor measurements
A. PID altitude controller
and sent them to the PC ground station. These data were
received by the PC ground station and used to compute new The current altitude of the blimp was measured by the
control signals for the gondola micromotors. The WCU in the SRF05 ultrasonic sensor and then sent to the PC. Sensor errors
PC interface card sent control data to the onboard WCU, were corrected with a Kalman filter, thus obtaining the
which were used by the microcontroller to set the estimated altitude. The filter parameters depended on sensor
corresponding commands for the micromotors. characteristics, the dynamic model of the blimp and the
previous measurements, but in our case they were tuned
IV. BLIMP NAVIGATION SOFTWARE experimentally. We did not make use of the Kalman filter
variance because of real time restrictions in controlling motor
The navigation software provided limited autonomous
action.
operation of our blimp due to:
The first altitude controller implemented was a PID-type
• The lack of odometry. controller. Control actions were calculated as:
• Misalignment caused by wind gusts and temperature t de(t )
changes. u PID = K p e(t ) + K i ∫ e(t )dt + K d (1)
• The non-linear nature of propeller action.
0 dt
• Blimp operation in three dimensions.
where Kp, Ki and Kd were parameters experimentally calculated
The control system for our autonomous blimp was designed
using the Zieger-Nichols method and uPID was the command
basically to keep the blimp moving in a straight line if its path
signal to the vertical propeller, responsible for up-and-down
was clear, maintaining the desired horizontal speed and
movements of the aerial vehicle.
reference altitude. When a head-on obstacle was detected the
control system attempted to maintain a certain distance from it.
Assuming a low speed, the control problem could be
decoupled in two sub-systems describing motions in both
vertical and horizontal planes [11].
Although some blimp control approaches are based on the
vehicle mathematical model (see [11] and [12], for example),
we preferred not to implement an analytical controller because
the blimp has complex dynamics due to its nonlinear
characteristics and environmental influences (air gusts,
temperature, altitude references, etc). We implemented both
GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 47

a)

Fig 5. Experimental environment 1: a corridor in Building 79, Freiburg


University.

The experiments have shown diverse results using the same


controller in these two environments. An overall good
performance can be observed during the experiments carried
out in environment 1. However, the same PID controller in
environment 2 has shown an oscillating behavior with an
important deviation from the mean. Although it can reach the
b)
altitude reference, there is a significant error as it is shown in
Fig 4. Behavior of the PID altitude controller when the altitude Figure 4b.
reference is set to 1 meter. a) Experiments in environment 1. b) Note that the controller response in the second environment
Experiments in environment 2.
could be improved by tuning PID parameters. In fact, these
Blimp altitude in Figure 4 is depicted in red, while average parameters had to be recalculated to take account of changes
values are represented in blue. The figure shows the blimp PID in environmental conditions while the blimp was navigating.
altitude controller behavior for two different scenarios. The
first was a corridor in Building 79 of Freiburg University
(Figure 5), while the second was a computer laboratory in the
same university (Figure 6). The computer laboratory had a
door that, when opened, could alter the draught inside. Eight
similar experiments were performed in both environments, all
commencing with the blimp lying on the ground and then
moving to its target position, one meter above the ground.

Fig 6. Experimental environment 2: a computer laboratory in Freiburg


University.

B. Fuzzy altitude controller


A non-linear altitude controller based on fuzzy logic was
also implemented. Fuzzy logic uses fuzzy sets to model
48 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

designer knowledge about the system to control, with


knowledge representation modeled using fuzzy rules. This The fuzzy logic controller is characterized by a set of
kind of controller has several advantages because it does not linguistic variables and fuzzy if-then rules. All input and
need to recalculate parameters when environmental conditions output linguistic variables have a finite number of linguistic
changes. values with membership functions that are empirically defined
A fuzzy controller is composed of a knowledge base and an after exhaustive simulation studies. The linguistic values
inference engine. The knowledge base contains rules and representing the linguistic variable altitude error and vehicle
linguistic variable descriptions, while the inference engine speed are very negative (VN), negative (N), zero (Z), positive
generates a control action as a function of state variable values (P) and very positive (VP). Linguistic values describing the
in a given time instant. output linguistic variable are very negative 3 (VN3), very
The altitude fuzzy logic controller in our blimp had two negative 2 (VN2), very negative (VN), negative (N), zero (Z),
inputs: altitude error and estimated current vertical speed. positive (P), very positive (VP), very positive 2 (VP2) and very
Altitude error was the difference between the desired altitude positive 3 (VP3). Their corresponding membership functions
and current altitude (Figure 7). A change in altitude error are shown in Figure 8.
indicated whether the aerial vehicle was approaching the Fuzzy rules describe the controller behavior in terms of
reference altitude or moving away from it. The controller relationships between input and control variables. A rule is
output was the vertical motor command. Note that the fuzzy usually of the type:
controller structure differed from the PID controller in that the If x1 is A1 and x2 is A2 then y is B,
latter has only one input, which ensures better and more where xi and y are, respectively, input and control linguistic
accurate altitude control. variables, and Ai and B are linguistic terms.
The altitude controller can be described with a small set of
rules, for example:
If altitude error is negative and vertical speed is very
positive then motor command is positive.
Figure 9 depicts, in tabular form, the fuzzy rule set used to
generate the motor command. These rules endeavor to
maintain the blimp at a specific height and represent human
expertise on how to control the system. Note that the rule
descriptions were developed on the basis of multiple
experiments.
Figure 10 depicts one of the situations described in the table
of Figure 9. In this situation, the blimp is descending and the
altitude error is significant. In the position shown in the figure,
the input variables have the linguistic labels VN for velocity
and VP for altitude error, respectively. So, the command signal
Fig. 7. Variable relations regarding altitude control of the blimp. Zref is the value must be VP3, as shown in the fuzzy rule table depicted
altitude reference; Z is the estimated height using a Kalman filter. Other
variables are shown on the left. in Figure 9.
Note that one set of rules describes when motor action must
be null. These rules have a dead band to avoid continuous
motor action.
Experimental results for the fuzzy altitude controller are
depicted in Figures 11a and 11b. The blimp showed only slight
oscillations in both environments and deviations from the
mean, shown in blue, were very small. A significant number of
tests carried out in different circumstances led to similar good
results in both environments.

Fig 8. Fuzzy linguistic variables for the fuzzy altitude controller.


GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 49

Fig 9. Fuzzy rules for altitude control.

b)
Fig. 11. Behavior of the fuzzy altitude controller when the altitude
reference is set to 1 meter. a) Experiments in environment 1. b)
Experiments in environment 2.

C. Collision avoidance controller


The collision avoidance system should cause the vehicle to
stop reliably when the frontal distance sensor detects an
obstacle in the vicinity. In such circumstances, the horizontal
navigation speed is changed by the collision avoidance
controller. Only the frontal sensor no longer detects an
obstacle close to the blimp, horizontal blimp speed is reset to a
certain value. Different approaches can be used to implement
the collision avoidance controller.
We again chose a fuzzy logic approach, based on a PID
controller which demonstrated oscillating behaviors in almost
Fig. 10. Behavior of the fuzzy altitude controller when the blimp all the experiments carried out in both environments. The goal
descends.
of this controller was to keep the blimp at a safe distance from
frontal obstacles. Controller inputs were frontal distance and
estimated speed and controller output was a speed index for
the horizontal motors.
Fuzzy rules for the collision avoidance controller are shown
in Figure 12. Figure 13 depicts a possible situation in which
the blimp is very close to an obstacle detected by the frontal
sensor. The blimp is heading straight for a vertical wall; speed
is V=VN and the distance error value is E=VP (near the wall).
In this situation, which is critical for the blimp as it is likely to
collide with the wall, the fuzzy collision avoidance controller
transmits the maximum control command (A=VP3) to the
motors, thereby transmitting the maximum opposite power to
ensure that the blimp avoids the collision. However, if this
speed is used for a long time the blimp develops inertia and it
a) becomes impossible to stop it in the desired reference. When
the speed is reduced so as to be comprised within the linguistic
variable V=N, the blimp does not need maximum power and
control action is reduced to A=VP2. These few rule examples
illustrate how the controller works.
50 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008

sensors were used for navigation; the small payload also did
not permit inclusion of an inertia sensor or miniature wireless
camera usually used for global position calculation [6].
We also implemented onboard and PC programs to control
the blimp with the PC ground station, with the communication
software module capable of receiving onboard sensor data and
sending commands to the blimp (as well as facilitating future
application developments).
Two different controllers were implemented. An altitude
Fig. 12. Fuzzy rules for the collision avoidance controller.
controller maintained the blimp at a certain distance from the
floor and a second controller avoided obstacles in the path of
the blimp. The altitude controller was implemented using two
different approaches: PID and fuzzy logic. A first experiment
was designed to compare the two controllers operating in the
same conditions, so as to determine which controller was
better suited to controlling blimp altitude.
In the first scenario, both controllers were tested on an
alternating basis in one environment (a corridor), an
experiment that was repeated eight times. Similarly, the blimp
Fig. 13. Behavior of the fuzzy collision avoidance controller when an was tested in a second environment (a computer room) with
obstacle is detected. the same perturbations on the blimp for both controllers.
Comparative results show that the fuzzy logic controller
produced balanced behavior in either of the two environments.
The PID altitude controller performed slightly better than the
fuzzy logic controller in environment 1 (see Figure 14a) but its
performance in environment 2 was significantly poorer than
that of the fuzzy logic controller. A better PID controller in
other environments would require online adjustment of PID
parameters. Using a fuzzy logic controller, on the other hand,
does not affect the behavior of the blimp too much and the
controller parameters do not require modification.
A second fuzzy logic controller, designed to avoid head-on
collision with obstacles while the vehicle was navigating in the
indoor environment, showed good experimental performance.

Fig. 14. Behavior of the blimp in a corridor. The blue line is the safety
distance from the floor.

Figure 14 shows the blimp distance to a vertical wall while


it is navigating in environment 1. Initially, it maintained a
constant horizontal speed. When it approached a wall at the
end of this corridor, the obstacle avoidance controller reduced
the speed accordingly. In the experiment, the vehicle
maintained a security distance of about 0.2 meters away from
the obstacle without crashing at any time.

V. CONCLUSIONS
a)
In this paper we have described the construction of a small
blimp based on a commercial kit, with the main goal of
designing and implementing different control navigation
techniques. It was necessary to change the vertical motor of
the commercial blimp and develop onboard control hardware,
including a wireless radio connection with a PC ground
station. Due to the vehicle’s small payload only two distance
GONZÁLEZ ET. AL.: DEVELOPING A LOW-COST AUTONOMOUS INDOOR BLIMP 51

[2] G. Kantor, D. Wettergreen, J. P. Ostrowski, and S. Singh, “Collection of


environmental data from an airship platform,” in Proceedings SPIE Vol
4571, Sensor Fusion and Decentralized Control in Robotic Systems,
2001.
[3] E. Hygounenc, I.-K. Jung, P. Soueres and S. Lacroix, “The autonomous
blimp project of LAAS-CNRS: Achievements in flight control and
terrain mapping,” The International Journal of Robotics Research, vol.
23, n. 4-5, pp. 473-511, 2004.
[4] K. Motoyama, H. Kawamura, M. Yamamoto, and A. Ohuchi,
“Development of autonomous blimp robot with intelligent control”, in
Entertainment Computing: Technologies and Applications, Eds. R.
Nakatsu and J. Hoshino, Kluwer Academic Pub. 2003, pp. 191-198.
[5] K. Motoyama, H. Kawamura, M. Yamamoto, and A. Ohuchi, “Design
of evaluation function in motion planning for autonomous balloon
b) robot”, in Proceedings of 2003 Asia Pacific Symposium on Intelligent
Fig. 13. Comparative results for the PID and fuzzy logic controllers. a) and Evolucionary Systems, 2003.
Experiments in environment 1. b) Experiments in environment 2. [6] G. A. Hollinger, Z. A. Pezzementi, A. D. Flurie, and B. A. Maxwell,
“Design and construction of an indoor robotic blimp for urban search
and rescue tasks”, Swarthmore College Senior Design Thesis, Spring
Note that we could not measure the absolute position of our 2005.
vehicle with the onboard sensor due to the low payload of the [7] H. Kadota, H. Kawamura, M. Yamamoto, T. Takaya, and A. Ohuchi,
“Vision-based positioning system for indoor blimp robot”, Proceedings
blimp. This limited blimp navigation in a desired direction of the 4th International Conference on Advanced Mechatronics-
without global path planning. For this reason we are Toward Evolutionary Fusion of IT and Mechatronics, 2004.
developing a new design with a bigger balloon than described [8] W. E. Green, K. W. Sevcik and P. Y. Oh, “A competition to identify key
in this paper that will enable new vehicle control functions to challenges for unmanned aerial robots in near-earth environments”, in
Proceedings. 12th International Conference on Advanced Robotics
be incorporated [13]. ICAR '05, July 2005.
[9] Balloon shop. http://www.heo.com.
ACKNOWLEDGMENTS [10] Plantraco shop. http://www.plantraco.com.
[11] S. van der Zwaan, M. Perrone, A. Bernardino, and J. Santos-Victor,
We would like to thank the people of the University of “Control of an aerial blimp based on visual input”, in Proceedings of
Freiburg (Germany) for their assistance during Pablo the 8th International Symposium on Intelligent Robotic Systems
SIRS’00, 2000.
González’s visit to the Autonomous Intelligent Systems [12] N. Rooz and E. N. Johnson, “Design and modeling of an airship station
Laboratory. holding controller for low cost satellite operations,” in Proceedings of
the AIAA Guidance, Navigation, and Control Conference and Exhibit,
REFERENCES August 2005.
[13] A. Rottmann, M. Sippel, T. Zitterell. W. Burgard, L. Reindl, and C.
[1] A. Elfes, S. Siquiera, M. Bergerman, and J. Guimaraes, “A semi- Scholl.” Towards an Experimental Autonomous Blimp Platform”, in
autonomous robotic airship for environmental monitoring missions,” in Proceedings of 3rd European Conference on Mobile Robots (ECMR
Proceedings of the 1998 IEEE. International Conference on Robotics '07), September 2007.
and Automation, 1998.
52 JOURNAL OF PHYSICAL AGENTS, VOL. 3, NO. 1, MONTH 2008
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/322308785

Design and Autonomous Control of a Solar-Power Blimp

Conference Paper · January 2018


DOI: 10.2514/6.2018-1588

CITATION READS
1 115

3 authors:

Changhuang Wan Nathaniel Kingry


The Ohio State University Collins Aerospace
7 PUBLICATIONS   4 CITATIONS    8 PUBLICATIONS   22 CITATIONS   

SEE PROFILE SEE PROFILE

Ran Dai
The Ohio State University
65 PUBLICATIONS   351 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Wearable Technology View project

All content following this page was uploaded by Changhuang Wan on 24 January 2019.

The user has requested enhancement of the downloaded file.


Design and Autonomous Control of
a Solar-Power Blimp

Changhuang Wan∗, Nathaniel Kingry† and Rai Dai‡


Iowa State University, Ames, IA, 50011, USA

In this paper, we present the design, modeling, and control law development of a solar-
powered blimp. Blimps offer flexibility and low cost to perform long endurance missions,
such as agricultural, surveillance and search and rescue missions. While their benefits
are apparent, blimps are subject to a variety of perturbations during flight. A six DOF,
nonlinear kinematic and dynamics model for a customized blimp is developed. These
nonlinear equations are then linearized to allow for developing real-time control laws. The
developed model and controller is simulated and then experimentally tested in an indoor
environment.

Nomenclature
A, B State matrix and input matrix of the linear model, respectively
CIB Rotation matrix from body frame to inertial frame
FΩ Rotation matrix for angular velocity
Jb Jacobian Matrix
F Force vector
T Torque vector
x Blimp state vector
u Blimp control vector
M General total mass matrices
m Mass, kg
B Buoyancy of the blimp
G Gravity of the blimp
φ roll angle
θ pitch angle
ψ yaw angle
U linear velocity vector expressed in body frame
Ω angular velocity expressed in body frame

I. Introduction
In the past decades, there has been extensive research in many areas of unmanned aerial vehicles (UAVs).
UAVs have played an indispensable role due to the development in a variety of engineering technology fields
and the rise of challenging demands.1 Blimps are a type of light-than-air UAVs which rely on their neutral
buoyancy, rather the aerodynamic lift, to stay afloat in the air. Therefore, light-than-air UAVs intrinsically
have better stability over other platforms in terms of maintaining the lofting status.2 Unlike rigid airships,
blimps maintain their shape through the high pressure acting on their envelope rather than the internal
∗ Graduate Student, Aerospace Engineering Department, Iowa State University, Ames, IA, 50011 Email: cwan@iastate.edu
† Graduate Student, Aerospace Engineering Department, Iowa State University, Ames, IA, 50011 Email: Nethank-
ing@iastate.edu
‡ Assistant Professor, Aerospace Engineering Department, Iowa State University, Ames, IA, 50011 Email: dairan@iastate.edu

1 of 12

American Institute of Aeronautics and Astronautics


structure.3 The main advantages of blimps compared with other flying robots are the low power consumption,
the safety for taking off and landing, and the long endurance capability. For scenarios where the vehicle is
required to remain aloft over a specific area for a long duration, blimps offer lower power consumption and
long endurance over airplanes and helicopters, which may require refueling to maintain the same endurance.2
Therefore, blimps have great potential for applications in many fields, such as advertising, search and rescue,
traffic monitoring, inspection, exploration, law enforcement, and telecommunication relay systems.3, 4
However, the challenges of applying blimps in real world missions lay on the load capacity, dynamic mod-
eling and the control design. The payload capacity of blimps is usually limited, especially for indoor blimps
featured by small size, low altitude, and slow flight velocity.5, 6 For the light-than-air UAVs, aerodynamic
virtual mass and inertia have significant effects on controlling blimps, which is different from conventional
fixed-wing aircraft. Due to the mass and inertia of the air, virtual mass and inertia effects are changing in the
dynamics of the blimps.1, 7 Furthermore, blimps are very sensitive to wind and temperature disturbances,
which makes them even more difficult to control.
Different from traditional fixed-wing aircraft, in the blimp control system, the trajectory control is the
external circle and attitude control is the inner circle. The basic requirement for a stable flight is effective
and smooth attitude control.5 Traditional control and advanced control methods are two existing strategies
developed for controlling blimps.8 Classical control strategies are characterized by easy implementation and
reliable control performance.9 Many advanced control methods, such as neural network,10 have been applied
to improve control performance when considering the complicated and unstable flight environment.
Recent studies of blimps focus on two parts, design and control law development. Geoffrey et al.11
constructed a lighter-than-air indoor blimp and realized blimp wandering motion without collision. Authors
in6 proposed a novel structure for an indoor quad-rotor blimp which will improve the balance performance
and increase the lift force of existing indoor blimps. Motoyama12, 13 built an autonomously controlled indoor
blimp and used action-value function for motion planning and evaluated its effectiveness using simulation.
Takanori Fukaot14 used back stepping techniques to design the image-based tracking controller of a blimp.
The literature15 examined the fuzzy logic approach for the designing of speed, heading, and height controllers.
Combination of Proportional, Integral, and Derivative (PID) and fuzzy logic controllers has been developed
to control the blimp altitude and evaluation of the control system performance has been conducted in two
different environments.3
In this work, a solar-powered blimp is developed. Solar energy is used to power motors and/or charge the
onboard battery, which greatly extends the endurance of a blimp.16 A PID controller is applied to control
the system states, including location and attitude of the blimp. In addition, results from experimental tests
are presented to verify the effectiveness of the blimp control system.
The paper is organized as follows: In §II, major components for building a solar-powered blimp are
described. The kinematics and dynamic equations are analyzed in §III. In §IV, the nonlinear dynamic model
is linearized and then PID controller for this linear model is developed. §V describes the experimental
environment and the results. We then conclude the paper with final remarks in §VI.

II. Blimp Configuration


In this section, the system architecture of the solar-powered blimp is presented. Several constraints exist
when designing an autonomous solar-power blimp. Generally, as the volume of the envelope increases, so
does the payload capacity. For a small-size blimp, which is typically restricted in size, the design strategies
focus on minimizing the weight of the on-board hardware to allow for a smaller envelope size. For a solar-
powered blimp, it must be capable of measuring the power consumption, power gain, and battery levels at
all times. The demonstration blimp, shown in Figure 1, was developed with several key aspects to allow for
simulation and experimental verification for the work presented in this paper.
As it’s shown in Figure 1, a flying toy ‘shark’, which is an inflatable balloon, was selected as the envelope.
The shape of the blimp is approximated by an ellipsoid with [Ra , Rb , Rc ] be the three semi-major axises.
The volume of helium filled into the blimp is about Vb = 190.5cm3 . According to Archimedes Principle, the
maximum weights of payload is calculated by

mload = Vb · (ρair − ρhelium ) = 202.1g (1)


3 3
where ρhelium = 169.3g/cm and ρair = 1226.0g/cm .
Figure 2 shows the system block diagram of the customized blimp. It is based on a balloon controlled by

2 of 12

American Institute of Aeronautics and Astronautics


an arduino mini pro board. The blimp is equipped with an 6W solar panel and a Maximum Power-Point
Tracker (MPPT) to convert the recovered energy to the vehicle’s operating voltage. System power flow is
monitored by two INA219B voltage/current (V/C) sensors. The two sensors are place within the system
so that the energy flow can be recorded in real time. One is placed between the MPPT and the battery,
which measures the total energy in and out of the battery. The second is placed between the MPPT and the
load, which is comprised of all of the system electronics, including Arduino mini pro, wireless synapse for
communication, two motor board driver boards, and three light DC micro-motors. A vertical motor installed
at the bottom allows altitude to be regulated and the other two motors placed at each side of the blimp can
control speed and yaw angle. The power consumption of three motors is about 2.7W in total. The layout of
the three motors are shown in Figure 3.
Due to the limitation of weight, the selection of battery is a key issue. First, the battery is required to
provided enough power while also being light. For the demonstrator blimp presented in this work, a single
cell, 1200 mAh Li-Po battery with 3.7 V output voltage was selected, weighing only 20 grams. Solar panels
is another issue. The solar panels are supposed to be flexible since the surface of blimp is not flat, and the
solar panels should supply enough power for the entire system. Two pieces of MPT 6-75 lightweight, thin,
flexible solar panels were selected17 for the customized blimp. The maximum wattage is about 6 W, which
is enough for the blimp system. The major components and their weights are listed in Table 1.

Figure 1: Autonomous Blimp Envelope

Figure 2: Autonomous blimp system block diagram

III. Blimp Dynamic Modeling


In this section, the formulations of the blimp kinematics and dynamics are presented. Since the dynamic
model of a blimp is nonlinear, the linearization is applied to simplify the model. Moreover, the shape of
a blimp is assumed to be an ellipsoid with major semi-axis Ra , minor semi-axis Rb and Rc . The blimp is
assumed to be non-deformable, which means it can be handled as a rigid body.

A. Coordinate Frames
In order to describe the states of the blimp, several reference frames shown in figure 4 are defined below
Body Coordinate Frame, FB : the origin locates at the center of the blimp, the XB axis is directed
along the axis of symmetry, YB axis to the left, and ZB axis to the up. It is used to analyze the linear
velocity and angular velocity during the flight.
Inertial Coordinate Frame, FI : also known as inertial reference system, is used to locate the blimp
relative to the earth. The origin locates at the surface of the Earth.

3 of 12

American Institute of Aeronautics and Astronautics


Figure 3: Layout of three motors and distribution of markers

Table 1: Blimp Components and Weights Distributions

Component Weights (g)


Envelope(140×78×60 cm3 ) 82.0
Motors and Propellers(×3) 7.0 × 3 = 21.0
Electronic Boards(Arduino,Wireless Synapse.etc) 42.6
Battery 20.0
Markers(×6) 2.0 × 6 = 12.0
Wires 12.1
Mounts 5.0
Solar panels(×4) 2.3 × 2 = 4.6
Total weight 200.3

Airflow Coordinate Frame, FA : also named speed coordinate system, is used to describe the relative
motion between the atmosphere and the blimp. It is used to analyze the aerodynamic force and torque. The
origin is located at center of the body (CB).
Local Horizon Frame, FH : This frame moves with the airship, but its axes remain parallel to the earth
fixed inertial frame axes and its origin CB is the center of volume of the vehicle.

B. Kinematics
Under the assumption of a rigid body, the blimp’s equations of motion is composed of translational and
rotational motion.

1. Translational Kinematics Equations


Based on the definitions of coordinates shown in the Figure 4, the translational motion of the center of a
blimp can be described by its position vector ~r.

~r˙ = V
~I (2)
~I denotes the inertial velocity vector in form of
where V
~I = V
V ~rel + V
~W , (3)

where V ~rel is the velocity vector of the blimp relative to the surrounding air. It can be expanded as V
~rel =
T
[u, v, w] = U ∈ < 3×1
in FB , where u, v, and w are the blimp’s velocity components in its body axis x-y-z.

4 of 12

American Institute of Aeronautics and Astronautics


Zh

Zb

Xh CV
CB

CG
ZI
r Position vector

Yh XI OI

YI

Figure 4: Definitions of Coordinate Frames

~W represents the wind velocity vector and it can be decomposed into V


V ~W = [Vx , Vy , Vz ]T = W ∈ <3×1
W W W
in FI . Then in the FI frame, Eq.(3) can be expressed as
     
ẋ u Vxw
ẏ  = CIB  v  + Vyw  (4)
     

ż w Vzw

where CIB is the rotation matrix from body frame to inertial frame.

2. Rotational Kinematic Equations


The Rotational motion of a blimp can be described by the orientation of frame FI relative to the local
horizon frame FI . It is defined in terms of Euler angles, roll (φ), pitch (θ), and yaw (ψ). Using the 3-2-1
convention of Euler angles, the rotational matrix CHB from FB to FH is written as
 
cθcψ cθsψ −sθ
CHB = sφsθcψ − cφsψ sφsθsψ + cφcψ sφcθ , (5)
 

cφsθcψ + sφsψ cφsθsψ − sφcψ cφcθ

where c ≡cos and s ≡ sin. Let Ω denotes the angular velocity expressed in FB . According to the definitions
of frames FI and FH , CIB = CHB .  
p
Ω = q  . (6)
 

r
Then, the rotational kinematic equation can be written as
    
φ̇ 0 sinψsecθ cosψsecθ p
 θ̇  = F · Ω = 0 cosψ −sinψ  q  . (7)
    

ψ̇ 1 sinψtanθ cosψtanθ r

5 of 12

American Institute of Aeronautics and Astronautics


Equations (4) and (6) can be organized in one matrix, expressed as
   
ẋ u
 ẏ  " # v 
   

  
  = CIB 03×3 w .
 ż   
(8)
 φ̇ 
  03×3 FΩ  p

 θ̇  q
   

ψ̇ r

C. Force Analysis
In this section, analytic expressions for the forces and torques acting on the Blimp are derived. There are
several types of forces acting on the blimp, shown in Figure 5, including the gravity force G, the buoyancy
force B, the aerodynamic forces, and the propulsion force generated by motors.

1. Gravity and Buoyancy


In the inertial frame FI , the gravity and buoyancy forces are expressed as
GI = [0 0 − mg]T
BI = [0 0 Vb · (ρair − ρhelium )]T . (9)

2. Propulsion Force
Propulsion force has a positive correlation with motor rotational speed and it is also related to the position
and direction of the propellers. It is assumed that the relationship between propulsive force by one propeller
Fp and rotational speed np is a polynomial function, written as:
Fp = η1 · n2 + η2 · n + η3 , (10)
where ηi , i = 1, 2, 3, are the coefficients and constants. The motor’s torque Tp is linear about Fp such that
Tf p = λ · Fp . (11)
The locations of three propellers in FB are
     
x p1 x p2 xp3
Lp1 =  yp1  Lp2 =  yp2  Lp3 =  yp3  (12)
     

zp1 zp2 zp3


. So the torque vector produced by propulsion forces can be expressed as
TXb = Tp1 + Tp2
TYb = Fp1 · zp1 + Fp2 · zp2 − Fp3 · xp3 (13)
TZb = −Fp1 · yp1 + Fp2 · yp2 + Tf p3
.

3. Aerodynamics
Since the blimp is light and operates at a low speed, the added mass or virtual mass can’t be ignored.
Referring to literature,2 the force produced by aerodynamics with virtual mass and inertia is formulated as
AA = A0 − A1 U̇ − A2 Ω̇, (14)
where A0 is the standard aerodynamic force in the body frame, expressed as
   
Fax 1/2ρair Vb2 Sref cN " #
2 U
A0 = Fay  = −  1/2ρair Vb Sref cT  − DU , (15)
   
2 Ω
Faz 1/2ρair Vb Sref cL

6 of 12

American Institute of Aeronautics and Astronautics


where cN , cT , and cL are the aerodynamic force coefficients in the body frame, Sref is considered as Θ2/3 ,
and DU is the translational portion of the Coriolis-centrifugal coupling matrix. Additionally, Vb is the blimp
velocity determined by p
Vb = u2 + v 2 + w2 . (16)
The aerodynamic force coefficients are function of angle of attack α and side-slip angle β. They are expressed
as

cL = c1 · cos(α) · sin(2β)
cN = c1 · cos2 (β) · sin(2α) (17)
cT = 0

Similar to the aerodynamic force, the aerodynamic torque Ti , i = 0, 1, 2 can be expanded as

TA = T0 − T1 Ω̇ − T2 U̇ , (18)

where T0 is the standard aerodynamic torque vector in body frame, expressed as


   
Tax 1/2ρair Vb2 Sref cn " #
2 U
T0 = Tay  = − 1/2ρair Vb Sref cm  − DΩ (19)
   

Taz 1/2ρair Vb2 Sref cl

where cn , cm , and cl are the aerodynamic torque coefficients in the body frame and DΩ is the rotational
portion of the Coriolis-centrifugal coupling matrix.

D. Dynamics
Based on the force analysis, applying Newton’s Laws, the dynamic model is expressed as

M~v˙ =
X
~ ex = F
F ~G +F ~B +F~P +F~A +F~C (20)

where M is the general mass and inertia matrix containing rigid-body inertia and added-mass terms, ~v is
the velocity vector including translational and rotational velocities, F~ G is the general gravity force, F
~ B is the
~
general buoyancy force containing buoyancy force and its tongue, FP is the force vector of the propulsion, F ~C
~
is the vector of Coriolis effects, and FA is the aerodynamic force vector. Since the blimp is in symmetric shape,
Eq.(20) is described in the body frame FB . Similar with kinematic analysis, translational and rotational
dynamic equations are also discussed separately in this section.

1. Translational Dynamic Equations


According to literature,2 using Newton’s second law and Euler’s equations, the three dimensional transla-
tional equations of the blimp are written as
X
× × 2
(mI3×3 + A1 )U̇ = m[SΩ (U − CBI Ẇ )] + CBI (G + B) + A0 + Fpi − m(SΩ ) ρCB − (mSρ×CB + A2 )Ω̇, (21)

where I3×3 is the identity matrix, G + B denotes the sum of gravity and buoyancy force vector expressed in
inertial frame, and ρCB = [rx , ry , rz ]T denotes the vector from the origin of the body frame to the location
×
of rigid body empty center of mass. SΩ is the skew-symmetric form of the vector Ω, expressed as
   
0 −r q 0 −rz ry
×  ×
SΩ = r 0 −p , SρCB =  rz 0 −rx  . (22)
  

−q p 0 −ry rx 0

7 of 12

American Institute of Aeronautics and Astronautics


B
L

Fp1 Tp1

Fp2 CB(CV)
Tp2

CG

Tp3
Fp3 G

Figure 5: Forces analysis for a blimp

2. Rotational Dynamic Equations


Using Newton’s second law and Euler’s equations, the rotational dynamic equations are formulated as
X
× ×
(Jb + T1 )Ω̇ = T0 + TG + Tpi SΩ Jb Ω + mSρ×CB (−SΩ U + CBI Ẇ ) + (mSρ×CB − T2 )U̇ , (23)

where Jb is the inertia matrix of the rigid (time-invariant mass) with respect to body frameFB , T0 represents
the aerodynamic torque vector in the body frame FB , which is a function Pof airspeed, and control surface
deflections, TG denotes the torque vector due to gravity in the FB , and Tpi denotes of the torque from
propulsion in FB .
Equations (8), (21, and (23) are the 6-DOF nonlinear model of the blimp system. By defining the state
vector x = [u, v, w, p, q, r, x, y, z, φ, θ, ψ]T and the control vector u = [n1 , n2 , n3 ]T , the nonlinear function can
be rewritten as
ẋ = F1 (x, u), (24)
where F1 is a set of functions about x, u.
For this customized blimp, roll angle control is inherent stable due to its configuration. Thus, the dynamic
model can be simplified by removing the roll and pitch controllers. Let X = [u, v, w, r, x, y, z, ψ]T , then the
simplified model is reformulated as
Ẋ = F2 (X, u). (25)

IV. Model Linearization and Controller Design


A. Linearization
The simplified model in Eq.(25) is a nonlinear model, which is difficult for real-time autonomous control
purposes. In this section, model in Eq.(25) is linearized. If the movement of the blimp is divided into desired
movement and disturbance movement, Taylor-series expansion is used to conduct the linearization of the
dynamic model.

8 of 12

American Institute of Aeronautics and Astronautics


Let X0 be the desired movement and δX be the disturbance, the state vector X = X0 + δX and
u = u0 + δu, then the linear model of the blimp system can be written in the form of
δX˙ = AδX + Bδu, (26)
where A and B are the Jacobian matrices, and they can be obtained by partial differential functions of F2 ,
expressed as
∂F2
A= (27)
∂X
∂F2
B= . (28)
∂u

B. PID controller
A PID controller is a classical closed-loop feedback control method. PID controller is well-known for its
simple structure and robust performance. In the section, PID method will be applied to the flight control of
the blimp. Define the observation model as
y = CX + Du (29)
where C = I3×3 and D = 03×3 .
Let Xd be the desired state variables, and Xp be the present state of blimp. The difference between Xd
and Xp is eX = Xd − Xp . So the PID controller is expressed as
Z
u = Kp · eX + Ki · eX dt + Kd · ėX (30)

where Kp , Ki , Kd are the proportional, integral, derivative gain matrices, respectively. The flowchart of
system controller is shown in Figure 6.

Figure 6: The Control Diagram of the Blimp System

V. Experimental Results
A. Experimental Environment
For the experiment tests, we constructed a time-invariant test area comprised of uniform terrain in an indoor
laboratory with 1.31×3.8 m2 area. More details of the environments can be found in the our previous work.18
In the indoor testing environment, the temperature of the experimental space is around 15.0 ◦ C with air
pressure of 1 atm. The helium density is 169.3 g/m3 at 15.0 ◦ C and the air density [g/m3 ] is 1226.0 at
15.0◦ C.
In Figure 7, the experiment flow chart of the solar-powered blimp system is presented. The test
includes three parts, the customized blimp, base station, and Vicon motion capture system.19 The Vicon
system is used to obtain the information about markers’ location put on the blimp and motion. Through
Vicon, we can obtain the location, orientation, and velocities of the blimp. All information, such as location
and power level of the blimp can be stored in the base station. Based on the information, the new control
vector is calculated using the proposed PID. Then the control commands are sent to blimp through wireless
communications. In this phase, the blimp is designed to operate in a straight path with a constant velocity
under wind disturbances.

9 of 12

American Institute of Aeronautics and Astronautics


Figure 7: Experiment Flow Chart of the Solar-Powered Blimp System

B. Results
This section presents results obtained from experimental testbed in the indoor testing environment. Given
the desired straight path through this area. The trajectory of the center of the blimp in three dimensional
space is shown in Figure 8. The red path is the desired straight path, and the blue line is the measured
trajectory. Figure 9 shows the time history of the control vector. Note that the maximum amplitude of
PWM for each motor was set as 200. Results with power consumption and management will be integrated
in the final version. We will also present outdoor test results in the final manuscript.

VI. Conclusion and Future work


In this paper, the design and control of an autonomous solar-powered blimp is presented. The key
components of the customized blimp are described. Furthermore, the six DOF nonlinear dynamic model of
the demonstration blimp is established and a PID controller is developed. Experimental results verify that
the developed blimp is capable of gathering energy and maintaining stability. Future research will explore
methods to optimize the blimp’s operation in outdoor environments.

References
1 Atmeh, G. and Subbarao, K., “Guidance, Navigation and Control of Unmanned Airships under Time-Varying Wind for
Extended Surveillance,” Aerospace, Vol. 3, No. 1, 2016, pp. 8.
2 Nakpiam, J., “Control of Airship Motion in the Presence of Wind,” 2012.
3 González, P., Burgard, W., Sanz Domı́nguez, R., and López Fernández, J., “Developing a low-cost autonomous indoor

blimp,” 2009.
4 Al-Rashedi, N. and Gerke, M., “3D Off-Line Path Planning for Autonomous Airships in Restricted Known Environments.”

Proceeding of the 18th International Conference on Process Control, 2011.


5 Miao, J., Zhou, J., Nie, Y., and Yang, X., “Yaw controller design of stratospheric airship based on phase plane method,”

Chinese Journal of Aeronautics, Vol. 29, No. 3, 2016, pp. 738–745.


6 Li, Y. and Wang, G., “Quad-Rotor Airship Modeling and Simulation Based on Backstepping Control,” International

Journal of Control and Automation, Vol. 6, No. 5, 2013, pp. 369–384.


7 Takaya, T., Kawamura, H., Minagawa, Y., Yamamoto, M., and Ouchi, A., “Motion Control in Three Dimensional Round

System of Blimp Robot,” SICE-ICASE, 2006. International Joint Conference, IEEE, 2006, pp. 1291–1294.
8 Liu, Y., Pan, Z., Stirling, D., and Naghdy, F., “Control of autonomous airship,” Robotics and Biomimetics (ROBIO),

2009 IEEE International Conference on, IEEE, 2009, pp. 2457–2462.


9 Kahale, E., Castillo Garcia, P., and Bestaoui, Y., “Autonomous path tracking of a kinematic airship in presence of

unknown gust,” Journal of Intelligent & Robotic Systems, 2013, pp. 1–16.
10 Park, C.-S., Lee, H., Tahk, M.-J., and Bang, H., “Airship control using neural network augmented model inversion,”

Control Applications, 2003. CCA 2003. Proceedings of 2003 IEEE Conference on, Vol. 1, IEEE, 2003, pp. 558–563.
11 Hollinger, G. A., Pezzementi, Z. A., Flurie, A. D., and Maxwell, B. A., “Design and construction of an indoor robotic

blimp for urban search and rescue tasks,” Swarthmore College Senior Design Thesis, 2005.
12 Motoyama, K., Kawamura, H., Yamamoto, M., and Ohuchi, A., “Development of autonomous blimp robot with intelligent

control,” Entertainment Computing, Springer, 2003, pp. 191–198.


13 Motoyama, K., Kawamura, H., Yamamoto, M., and Ohuchi, A., “Design of evaluation function in motion planning for

autonomous balloon robot,” Proceedings of 2003 Asia Pacific Symposium on Intelligent and Evolucionary Systems, 2003.
14 Fukao, T., Fujitani, K., and Kanade, T., “Image-based tracking control of a blimp,” Decision and Control, 2003. Pro-

ceedings. 42nd IEEE Conference on, Vol. 5, IEEE, 2003, pp. 5414–5419.
15 Avenant, G. C., Autonomous flight control system for an airship, Ph.D. thesis, Stellenbosch: University of Stellenbosch,

2010.
16 Kantor, G., Wettergreen, D., Ostrowski, J. P., and Singh, S., “Collection of environmental data from an airship platform,”

Intelligent Systems and Advanced Manufacturing, International Society for Optics and Photonics, 2001, pp. 76–83.

10 of 12

American Institute of Aeronautics and Astronautics


(a) The trajectory of blimp in 3-dimension space

(b) The trajectory of blimp shown in x-y plane

(c) The trajectory of blimp shown in y-z plane

Figure 8: The trajectory of the blimp

11 of 12

American Institute of Aeronautics and Astronautics


Figure 9: The PWM histories for the three motors installed on the blimp

17 “Lightweight,
Thin, Flexible Solar Panels,” 2014.
18 Kaplan,
A., Uhing, P., Kingry, N., and Dai Adam, R., “Integrated path planning and power management for solar-
powered unmanned ground vehicles,” Robotics and Automation (ICRA), 2015 IEEE International Conference on, IEEE, 2015,
pp. 982–987.
19 “Affordable Motion Capture for Any Application,” 2014.

12 of 12

American Institute of Aeronautics and Astronautics

View publication stats