You are on page 1of 24

Experimental Knot Tying with Aerial Drones

Dartmouth College Computer Science Technical Report


Michael S. Li

Advisor: Devin Balkcom


May 31, 2016

Experimental Knot Tying with Aerial Drones


Michael S. Li

Advisor: Devin Balkcom

Figure 1. Parrot Rolling Spider MiniDrone [20]


Abstract

As drone technology advances and drone control becomes more precise, aerial
drones are increasingly used in practical applications ranging from delivery to
construction. Here, we explore the application of drones to aerial construction, a field that
addresses construction of structures with the aid of flying machines, by tackling the
problem of knot-tying. We tackle the feasibility of a small drone to perform the task,
taking into consideration string tension, weight of string, and movement limitation due to
drone structure using manual open-loop control. We design a physical system using poles
to assist with knot-tying. Then, we design and implement a closed-loop system for
stabilization and simplified interface for drone pathing. We found that knot-tying was a
problem that was possible with a simple drone and that a closed-system could be
designed for drone stabilization and control. We anticipate our findings and drone control
framework to lay the groundwork for future related work in the robotics lab at Dartmouth
College.

1 INTRODUCTION

1.1

Introduction
Motivation

Unmanned Aerial Vehicles (UAVs), popularly known as drones, have been

around since the early 20th century with early usage limited mostly to the military [1].
However, as recent developments make the technology more advanced and cost of
production fall, UAVs are more commonly being applied to civilian day-to-day use [2].
UAVs are used for a multitude of purposes such as construction, delivery, monitoring,
search and rescue, and inspection [3]. In particular, our research focused on the
application of UAV technology in construction and more generally object manipulation.
We are also interested in developing generalized techniques for stabilization and
optimization of drone control. We tackle the problem of tying a simple overhand knot
with the drone and developing a simple closed-loop system for stabilizing the drone
pathing to accomplish the task. Knot-tying is a common procedure that can be applied to
many construction applicationsfrom securing structures together to forming secured
rope structures (i.e., bridges, nets) [4]. However, although knot-tying is simple for
humans and specialized robots, the task has many complications when tackled with a
drone. In this way, the problem we are tackling is novel and interesting. Because drone
technology is becoming more widespread in the field of construction, we believe the
research is very relevant as well.

1.2

The Problem

Our research tackles two problems both separately and then jointly. The first is

determining the feasibility of knot-tying with a simple aerial drone. We consider the

1 INTRODUCTION

general problem of tying a knot (i.e., the arrangement and steps required for knot
formation). Then, we consider the various obstacles that may complicate knot-tying in the
context of an aerial drone. These obstacles include the tension, weight of spool, blade and
string interference, and string and environment interaction.
The second problem is creating a closed-loop control system to stabilize and
autotomize drone pathing. This control system will make it easier to test our approaches
for knot-tying as well as other future endeavors in drone object manipulation and
construction.

1.3

Results

We were able to design a procedure to tie a simple overhand knot with an aerial

drone. We also discovered many limitations in tying the knot including tension, friction,
and string interactions. However, we were able to overcome these limitations by
engineering solutions to address them. In the end, we showed that at least for an overhand
knot, arranging the string into a knot form was possible with an aerial drone. In addition,
we were able to design and implement a closed-loop system for drone pathing and
stabilization. We believe our work provides a good foundation for future research.

1.4

Paper Structure

In this paper, we discuss the context of the research, the tools we used, the issues

we encountered, the solutions we created, and the results we found. Section 2 is focused
on related works in the field of drone control and robotic knot-tying. Section 3 gives
specifications of the drone used in our research. Section 4 discusses the physical set-up
and procedure designed for drone knot-tying. Section 5 discusses limitations to the knot-

2 RELATED WORKS

tying problem in context of aerial drone. Section 6 discusses the way we addressed the
limitations brought up in previous section. Section 7 discusses the design and
implementation of the closed-loop control system. Section 8 discusses conclusions and
future direction for related works.

Related Works

There has been much related research in recent years on drone use in aerial

manipulation. Much like in our knot-tying problem, aerial drones are being used to not
only spectate their environment (i.e., surveillance), but also to manipulate and interact
with it.
One area of research has been on equipping aerial drones with manipulators to
allow them greater dexterity in tackling aerial manipulation problems [5], [6]. In [6],
Huber, F., et. al. analyzed and experimented with aerial manipulation using robotic arms.
This area of research is relevant to our knot-tying problem, which also deals with
improving precision and control for aerial manipulation.
Another area of research has been in improving control methods to allow drones
to more safely be in contact with their environment and humans [7]-[12]. In [10], Tomic,
T., et. al. developed methods for collision detection and control reflexes of drones. In
[11], Hehn, M., et. al. developed an algorithm which allows for calculation of
quadrocopter trajectory and improved control. In [12], MITs CSAIL demonstrated
software which allows drones to make hairpin movements in a forest of distinct obstacles.
This area of research is relevant to our paper in that we also face and address droneenvironment interactions in the knot-tying problem.

3 THE DRONE

There has also been much research in aerial construction, a field that addresses
construction of structures with the aid of flying machines [4], [13]-[20]. In [4],
Augugliaro, F., et. al. were able to use aerial drones to create a stable rope bridge
structure. Though they did not tie knots, they manipulated rope into precise and secure
arrangements. In [13], Augugliaro and his team are able to construct a 6-meter-tall tower
using four quadrocopters. In [14], aerial construction of truss structures was explored. In
[19], Leach, D., et. al. explored the building of web-like structures using a thermoplastic
adhesive. These different construction problems, though seemingly different than our
knot-tying problem, share many of similar problems of precision and dexterity in
manipulation.

The Drone

The drone used is a Parrot Rolling Spider MiniDrone (Figure 1). The drone can

travel 18 kilometers/hour and has a range of 20 meters. It has a weight of 55 grams and is
140 mm in diameter with 4 propellers with diameter 55mm. It has capabilities to make
forwards, backwards, up, down, left, right, left turn, right turn, front flip, and back flip
movements. It contains an ultrasonic sensor, vertical camera, 3-axis gyroscope, 3-axis
accelerometer, and pressure sensor for drone stabilization and control. Signaling is via
Bluetooth V4.0 BLE.

4.1

Set-Up and Procedure


General Knot-Tying Problem

We note that the general knot tying problem can be broken down into two distinct

steps. The first step is the arrangement step. This requires the string to be put into a loose

4 SET-UP AND PROCEDURE

arrangement in which the strings relative position and overlap with itself is sufficient for
knot formation when we pull the ends of the rope apart. The simplest valid arrangement
is a loop structure in which the ends of the string goes through the loop to the opposite
side. This will lead to the basic overhand knot. Figure 2 shows this arrangement.

Figure 2. Arrangement for Overhand Knot [21]


The second step is the tightening step. In this step, the two ends of the string are
pulled together so that the arrangement from the previous step becomes as tight as
possible and the knot is secured. The focus of this project the arrangement step.

4.2

Physical Set-Up for Knot Tying

To assist with drone knot-tying of a basic overhand knot, we designed a physical set-

up using three pole structures for knot tying support. The set-up consists of three poles
one large target pole and two smaller helper poles. These are placed in a triangular
formation with the large target rod facing the drones initial position. The poles are one
meter away from each other and formed 60 degree angles with the other two poles (i.e.,
form equilateral triangle). The target pole is 1.5 meters tall and the helper poles are 1.2
meters tall each. The drone is initially placed in 2 meters in front of target pole. Figure 3
shows the set-up.

4 SET-UP AND PROCEDURE

Figure 3. Pole System Set-Up


In addition, the spool of string is attached to the initial fixed point on the ground
via a small rod (allowing free rotation). The string is attached to the end of a 10
centimeter stick pointing outward from the bottom front of the drone. Figure 4 shows this
set-up.

4.3

Figure 4. Drone Initialization Set-Up

Procedure for Knot Tying

To tie a simple overhand knot with drone, we developed the following procedure with

consideration of factors such as tension, friction, and string interactions (see Figure 5 for
depiction of the procedure):

4 SET-UP AND PROCEDURE

1) Fly drone upwards to half the height of the target pole.


2) Fly the drone around the three poles in a clockwise manner
going down each loop such that the rope spirals downward.
Pull outwards if string needs to be extended. Repeat this 3
times.
3) Fly the drone over the loop created and into the center of
the three poles.
4) Fly the drone down under the loop created and outwards so it
is outside of the loop.
5) Fly the drone to the side of the two helper poles.
6) Pull upwards so that the string slips off the helper poles.
7) Pull outwards so that the string is tightened around the
target pole.

It should be noted that steps 1-4 are arrangement steps and steps 5-7 are the tightening
steps. In our demonstration, we will only perform steps 1-4. This is because of a
limitation of the small drone to overcome tension of the string during steps 5-7.

Figure 5. Knot-Tying Procedure

5 LIMITATIONS

Limitations

Here we discuss the various physical limitations we encountered and took into

consideration when designing our knot-tying procedure.

5.1

Tension

Tension of the string became an issue during step #2 and the tightening step #6,

#7 of our procedure. The effects of tension on drone pathing was very large in the
tightening steps and prevents us from being able to carry out those steps. Though the
drone can still perform step #2, the effect of tension changed the angle of the drone such
that the front of the drone faced the direction of the force of tension.

5.2

Weight

Weight presented a limitation in our set-up design. Initially, we had two potential

spool set-ups (shown below in Figure 6).

Set-up A

Figure 6. Drone Spool Set-ups

Set-up B

Set-up A has the spool of string attached directly to the bottom of the drone. Set-up B has
the spool of string attached to a fixed rod on the ground. Set-up A was advantageous in
that it allows us greater control of string length as we move the drone. Specifically, it

5 LIMITATIONS

allowed us to avoid the issue of tension and friction with rods when we pulled the string
to extend its length. However, although Set-up A was advantageous in this regard, the
weight of the spool prevented us from choosing this design. Although the drone was able
to fly with spool attached (Supplemental Video 3), control was difficult as there was a lag
between input commands and actual execution by the drone. In addition, Supplemental
Video 4 demonstrates sporadic behavior in this set-up. It is likely this had to do with the
spool interfering with stabilization sensors located at the bottom of the drone.

5.3

String-Blade Interference

We note that a major limitation in the drones ability to move freely is the

interaction between the blade of the drone and the string. If the string hits any of the four
propeller blades of the drone, the drone will halt. This can happen in two different ways.
First, the drone can hit string already fixed to the set-up (i.e., the loop formed during
knot-tying). This occurs simply if the drone flies to close to the loop formed. This can be
seen in Supplemental Video 10. Second, the propeller blades can hit string just peripheral
of the attachment point on the drone. If the string is attached to the drone directly at the
bottom of the drone, this happens when the drones elevation is below the current fixed
point of the string (defined as the current point at which the string is being pulled from
this is either at the initialization point or the most recent point wrapped around on a pole)
and the drone makes a lateral move. Figure 7 shows this interaction.

5 LIMITATIONS

10

Figure 7. String-Blade Interference with Peripheral String

5.4

String-Pole Interaction

String-pole interaction can be broken down into two types of interactions

friction and fixation. Both of these interactions were important to consider when
designing our knot-tying procedure.
Frictional force between string and pole can present strain on drone movement as
the drone flies outwards to extend the length of the string in step #2. We had initially
anticipated this frictional force to be so great that it would present a problem for string
extension. However, after tests, we confirmed that the drone is capable of overcoming
this frictional force to extend the string (Supplemental Video 9). However, after a few
wraps around the pole, the frictional force becomes strong enough to prevent further
extension of the string. Frictional force is great in the tightening step and is a key force
that prevents our drone from carrying out the tightening step. Lastly, frictional force is
important to hold the loop suspended and allow for enough space below the loop for the

5 LIMITATIONS

11

drone to fly under it in step #3. Figure 8 shows that frictional force was sufficient to hold
the loop up.

Figure 8. Loop Formation on Poles


Fixation is the interaction that secures string in place. At any given arrangement, the
string can be divided into two sections a fixed section and a hanging section. The fixed
section is all the string that is bound between two fixed points. The hanging section is the
string that is between the most recent fixed point and the point of attachment on the
drone. Fixed points are either the initialization point (with spool) or wrap-around points
located on the pole. Figure 9 shows this idea.

Figure 9. Fixed and Hanging Section of String

7 CLOSED-LOOP DRONE CONTROL SYSTEM

12

Fixation is an important factor to consider with regards to string-blade interaction


since fixed and hanging string interact differently with the drone as it moves.

Solution and Performance

Here, we look at the solutions designed to address the limitations brought to

attention in Section 5. In Section 6.5, we look at the overall performance of our knottying procedure with manual control.

6.1

Tension

We did not address the issues tension had on the tightening steps as this was a

limitation of the power of the drone itself and could not be overcome with an engineered
solution using the tools we had. To address string tensions undesirable effect on drone
angle, we decided to readjust the angle of the drone after each undesirable angle change
event. This can be done manually or automatically using the closed-loop control system
we designed (see Section 7). The closed-loop control system we designed addresses this
problem by automatically detecting deviations from expected angle and correcting for
them.

6.2

Weight

Because weight of the spool presented a stability issue for drone movement in

Set-up A, we ended up using Set-up B (see Figure 6). Set-up B is very stable in take-off
and landing (supplemental video 8a) and forward, backward, left, right movements
(supplemental video 8b). Though there was a slight lag in movement due to weight and
tension of the string itself, this was very negligible compared to the lag present in tests
with Set-up A. In addition, we had initially anticipated friction and tension would inhibit

7 CLOSED-LOOP DRONE CONTROL SYSTEM

13

our ability to extend the string with Set-up B. However, after tests, we see that this is a
non-issue (see Supplemental Video 9).

6.3

String-Blade Interference

There were two types of string-blade interference to addressinterference with

fixed string and interference with peripheral hanging string. Interference with fixed string
was simply addressed by planning the drone path in such a way that avoided drone
contact with the loop.
Interference with peripheral hanging string however cannot be addressed by
simply altering the planned path. In the set-up where the string is attached at the bottom
of the drone, this type of interference occurs whenever the drone makes a downward
movement placing it below the most recent fixed point and then a lateral movement. In
order to avoid interference with peripheral hanging string, we would limit our pathing
options greatly. In fact, because our knot-tying procedure requires the drone to go below
the loop and make a lateral movement (steps #3), it is impossible to perform our knottying procedure with this limitation. To address this issue, we made a simple
modification to our drone set-up. Instead of attaching the string directly to the bottom of
the drone, we attached it to a point in front of the drone ahead of the front propellers via a
stick. All downwards movements can be made by simply rotating the drone such that the
front faces the location of recent fixed point before making the downwards movement.
Lateral movements after the downwards move can be made by first rotating drone so that
the front faces opposite the target direction (essentially positioning the drone to pull the
string). Figure 10 shows this method.

7 CLOSED-LOOP DRONE CONTROL SYSTEM

14

Figure 10. Method to Avoid Peripheral String-Blade Interference

6.4

String-Pole Interaction

We addressed the two string-pole interactions friction and fixation. The force of

friction on step #2 of our procedure increases with the number of times the string has
wrapped around the poles. To limit the force of friction, we simply limited the number of
times the drone wraps around the poles to three. At this number, the force of friction
presents a negligible effect on our drone movement. The force of friction is great in our
tightening steps. However, we did not address this problem as it is a limitation in the
power of our drone and cannot be overcome with an engineered solution using the tools
at our disposal.
Fixation does not itself present a problem. However, it is a factor to consider
when planning drone pathing in order to avoid blade interaction with fixed and hanging
string. Section 6.3 discusses how we addressed these issues.

7 CLOSED-LOOP DRONE CONTROL SYSTEM

6.5

15

Overall Performance

Using our general knot-tying procedure and solutions addressing the limitations

discussed, we attempted to manually tie an over-hand knot up to the arrangement step.


The knot-tying was successful (Supplemental Video 13, 14b) and demonstrated that knottying is possible with aerial drones.

Closed-Loop Drone Control System

We designed and implemented a closed-loop drone control system to assist with

automation and stabilization of drone pathing. We believe it will be applicable for our
knot-tying problem as well as future research. Here we discuss the design and
implementation of our closed-loop system.

7.1

Technologies

The design and implementation of our closed-loop drone system required many

hardware and software components with a variety of technologies.


We used Java as the primary language for our control system and Node.js as the
primary language for command signaling to the drone. We utilized JavaCV, a Java
interface to OpenCV, for webcamera image capturing and parsing.
We used the Logitech C920 HD 1080p Pro Computer Webcam as our primary
input to our control system. We used Plugable USB Micro Adapter which used Bluetooth
4.0 LE technology to signal to our aerial drone.

7 CLOSED-LOOP DRONE CONTROL SYSTEM

7.2

16

Physical System

The main objective in the physical design of our drone control system was to

obtain accurate and consistent input information of drone position and orientation. For
simplicity, we decided that using a single webcamera image as our only input was
sufficient for our drone control system. The webcamera was secured to the ceiling such
that it had an unobscured view of dark-colored ground below it. The drone flight space is
the space covered by vision of the webcamera. In our system, this space is 3.54 meters by
1.96 meters (corresponding to 1920 pixels by 1080 pixels in the captured image). This
space is clear of all objects aside from the those used in our experiments (i.e., the poles,
drones, etc). Figure 11a shows the physical set-up for our drone control system. Figure
11b shows the flight space as seen from above by the webcamera.

(b)
(a)

Figure 11. Physical Set-Up of Control System

7 CLOSED-LOOP DRONE CONTROL SYSTEM

17

In addition, the drone was modified with a red dot sticker placed on its top side
exactly at its centroid and another green dot sticker placed on its top side directly in front
of the red dot towards the front of the drone. These stickers were used to calculate
position and orientation of the drone. The center of the red dot is the position (x,y) of the
drone and the angle between the center of the red dot and the center of the green dot
represents the orientation () of the drone.

7.3

Objective

We noted during drone flight testing that although the aerial drone has internal

stabilization mechanisms, there were still some inconsistencies in drone pathingfor


example, the drone will make unanticipated lateral movements when rising or on standby.
In addition, in our knot-tying system, there are external forces such as tension which act
upon the drone and cause unplanned changes in its position and orientation. Thus, we aim
to create a control system which addresses these inconsistencies and corrects for
unplanned positional and orientation changes as they occur.
Thus, our control system should have as input a planned path (i.e., a series of
command moves to the drone). It should send these commands out to the drone so that
the drone moves along the intended path. However, the control system should also check
position and orientation of the drone frequently to detect deviation from expected
orientation and position. If the control system finds any deviations, it should send
commands to correct for them. If implemented correctly, the control system should
stabilize drone movements along the input path.

7 CLOSED-LOOP DRONE CONTROL SYSTEM

7.4

18

Algorithm Design

Our control system code is divided into two components: the controller and the

signaler. The controller, which is implemented in Java, is in charge of getting input (both
from initial input path and from webcamera) and making command decisions to send to
the signaler. The signaler, implemented in Node.js, is in charge of taking those command
decisions and sending them to the drone as Bluetooth signals. We outline high-level
pseudocode for both components. The flow of the control algorithm can be seen in Figure
12.

Controller:

While there are movement commands in input path file:


Get next input movement command

Output to the pipe for signaler if pipe ready; else wait


Wait a delay, then capture image frame from webcamera

Calculate current position, orientation of drone from image


frame and update drone model (i.e., internal model of where
drone is)
Update target position, orientation from latest move from
input path file
While current position, orientation and target position,
orientation are off:

Determine if forward/backward movement command can


correct for significant part of difference; if so, set
as correction move
If no correction move set:

Determine if sideways movement command can


correct for significant part of difference, if
so; set as correction move

If no correction move set:

7 CLOSED-LOOP DRONE CONTROL SYSTEM


Determine correct turn movement command to
correct for difference; set as correction move

Send correction to pipe for signaler if pipe ready;


else wait

Signaler:

Wait a delay, capture image frame from webcamera and


update drone model

Until we get exit command:

If pipe is ready to read:

Get next command from pipe


Send move to drone

Figure 12. Control System Algorithm Flow

19

8 CONCLUSIONS AND FUTURE DIRECTIONS

7.5

20

Current State and Difficulties

Currently, we have finished coding the entirety of the control system (see source

files). Conceptually, we have rigorously gone through the algorithms flow and believe it
works. However, we have yet to produce working demos of the system and this is our
next immediate goal. We are currently facing difficulties and bugs in controller-signaler
communication and timing (since Node.js is event based and Java is synchronous). After
we addresses these bugs, we believe we still have work in calibrating the system so that
movement commands move the drone the expected length across the flight space,
centroid colors are detected accurately by the webcamera, and delays in the system are
ideal for accurate and timely movements.

8.1

Conclusions and Future Directions


Conclusions

In the end, we were able to tie a knot with aerial drones. In addition, we were able

to design a closed-loop system for drone pathing stabilization and automation. We


conclude that the basic problem is not a difficult one. However, we believe that there is
much research to be done in the field. We hope that our research lays a stepping stone for
more complicated related research at the Robotics Lab at Dartmouth College.

8.2

Future Directions

There are many potential future directions to go with knot-tying. First, we were

unable to perform the tightening step with our drones. However, future work utilizing
more powerful drones could tackle the tightening problem to tie a complete knot. In
addition, we have gained some insight into which drone paths are valid in given our

8 CONCLUSIONS AND FUTURE DIRECTIONS

21

physical set-up (i.e., those that do not violate the limitations discussed). It may be
possible to model this in a computer simulation. This can help plan valid drone paths for
knot-tying which can be used with our closed-loop control system. Lastly, we can extend
this research into more complicated knots such as the half hitch, bowline, gnat hitch, or
half knot.

References:

[1] "Timeline of UAVs." PBS. PBS, Nov. 2002. Web. 31 May 2016.
<http://www.pbs.org/wgbh/nova/spiesfly/uavs.html >
[2] "Drones: A History of Flying Robots." Nesta. N.p., n.d. Web. 31 May 2016.
<http://www.nesta.org.uk/drones-history-flying-robots>.
[3] Tesla, Chuck. "The past and the Future of Drones." Tumotech. N.p., 17 June 2014. Web. 31 Mar. 2016.
<http://www.tumotech.com/2014/06/17/the-past-and-the-future-of-drones/>.
[4] A. Mirjan, F. Augugliaro, R. DAndrea, F. Gramazio and M. Kohler, Building a Bridge with Flying
Robots, Robotic Fabrication in Architecture, Art and Design 2016, in press
[5] M. Fumagalli, R. Naldi, A. Macchelli, F. Forte, A. Q. Keemink, S. Stramigioli, R. Carloni, and L.
Marconi, Developing an Aerial Manipulator Prototype: Physical Interaction with the Environment, IEEE
Robotics & Automation Magazine, vol. 21, no. 3, pp. 4150, 2014.
[6] F. Huber, K. Kondak, K. Krieger, D. Sommer, M. Schwarzbach, M. Laiacker, I. Kossyk, S. Parusel, S.
Haddadin, and A. AlbuSchaffer, First analysis and experiments in aerial manipulation using fully actuated
redundant robot arm, in 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems,
2013, pp. 34523457.
[7] L. Marconi and R. Naldi, Control of Aerial Robots: Hybrid Force and Position Feedback for a Ducted
Fan, IEEE Control Systems, vol. 32, no. 4, pp. 4365, 2012.
[8] S. Bellens, J. De Schutter, and H. Bruyninckx, A hybrid pose / wrench control framework for
quadrotor helicopters, in IEEE International Conference on Robotics and Automation, 2012, pp. 2269
2274.
[9] F. Augugliaro and R. DAndrea, Admittance control for physical human-quadrocopter interaction, in
European Control Conference, 2013, pp. 18051810.
[10] T. Tomic and S. Haddadin, A unified framework for external wrench estimation, interaction control
and collision reflexes for flying robots, in IEEE/RSJ International Conference on Intelligent Robots and
Systems, 2014, pp. 41974204.
[11] http://www.nt.ntnu.no/users/skoge/prost/proceedings/ifac11-proceedings/data/html/papers/3178.pdf
[12] http://news.mit.edu/2016/csail-drones-do-donuts-figure-eights-around-obstacles-0119
[13] F. Augugliaro, S. Lupashin, M. Hamer, C. Male, M. Hehn, M. W. Mueller, J. S. Willmann, F.
Gramazio, M. Kohler, and R. DAndrea, The Flight Assembled Architecture installation: Cooperative
construction with flying machines, IEEE Control Systems, vol. 34, no. 4, pp. 4664, Aug. 2014.
[14] Q. Lindsey and V. Kumar, Distributed Construction of Truss Structures, in Algorithmic Foundations
of Robotics X, ser. Springer Tracts in Advanced Robotics, E. Frazzoli, T. Lozano-Perez, N. Roy, and D.
Rus, Eds., 2013, vol. 86, pp. 209225.
[15] Integrated Project Aerial Robotics Cooperative Assembly System. [Online]. Available:
http://www.arcas-project.eu/
[16] G. Hunt, F. Mitzalis, T. Alhinai, P. A. Hooper, and M. Kovac, 3D printing with flying robots, in
IEEE International Conference on Robotics and Automation, 2014, pp. 44934499.
[17] F. Augugliaro, A. Mirjan, F. Gramazio, M. Kohler, and R. DAndrea, Building tensile structures with
flying machines, in IEEE/RSJ International Conference on Intelligent Robots and Systems, 2013, pp.
34873492.
[18] S. von Mammen, S. Tomforde, J. Hohner, P. Lehner, L. Forschner, A. Hiemer, M. Nicola, and P.
Blickling, OCbotics: An organic computing approach to collaborative robotic swarms, in IEEE
Symposium on Swarm Intelligence, 2014, pp. 18.
[19] D. Leach, L. Wang, D. Reusser, and F. Iida, Automatic building of a web-like structure based on
thermoplastic adhesive. Bioinspiration & biomimetics, vol. 9, no. 3, p. 036014, Sept. 2014. [20] A.
Mirjan, F. Gramazio, and M. Kohler, Building with Flying Robots, in Fabricate. gta-Verlag, 2014, pp.
266271.
[20] Figure 1: http://www.flyingtoys.com/image/cache/data/news/Parrot-Minidrone-C-600x500.jpg
[21] Figure 2: http://assets2.handipoints.com/worksheets/how-to-projects/knots/step-1.png

You might also like