Home

Open Access - Lund University Publications

1. 10 20 30 40 50 60 70 80 Time s Figure 4 5 Raw and low pass filtered data from the pressure sensor converted into meters when the quadrotor sits on the ground with motors off The natural variations in the atmospheric pressure can have a magnitude up to m Angle Sensor The two angle sensors that was to be attached to the LinkQuad bottom were two Hall effect sensors They measure each of the angles of the slung load a and p see Figure 4 4 and were to be combined to derive the slung load s position as it was done in the experiment of the gantry crane see Section 4 1 The design was an exact copy of the design of the head of the gantry crane developed by Rolf Braun The sensors were attached to the base of a fork which was attached to the wire of the slung load The base tilts and actuates each sensor as the fork swings along with the wire of the slung load Pressure Sensor Pressure sensors are not normally used indoors to measure the altitude since the pressure changes continously The lack of other distance sensors such as ultrasonic or infrared distance sensors implies that the pressure sensor is the only altitude sensor on the existing platform which could be used for measuring the altitude When the pressure sensor on the LinkQuad was used the first time it was observed that the raw sensor data was very noisy Even when the LinkQuad was completely still the pressure sensor s variance was about one meter see Figure
2. Raw PID output 915 510 Ol O ol PWM output aN ide z 490 485 480 475 100 100 5 101 101 5 102 102 5 103 103 5 104 104 5 105 Time s b The control signal to the step change of 1 5 m in Figure 5 3b The quadrotor was level and maintaining its altitude without any feedforward action until the step change occured When the quadrotor started to accelerate upwards it also started oscillate with small angles and the feed forward part compensates but does not dominate The effect of aerodynamic disturbances such as turbulence can also be seen on the control signal here The required thrust to hover and climb varies over time e g the thrust to hover at time 101 1 s is the same as the thrust to climb at 102 6 s Figure 5 4 The control signals to the step responses in Figure 5 3 85 Chapter 5 Experiments and Results Step change of ca 1 m I Altitude m sis 88 90 92 94 96 98 100 a The response in altitude to a step change of circa 1 m The slow integral action shows clearly as it approaches the new setpoint A small overshoot can be observed but is rejected within three seconds The step change is roughly set since it is induced by a user of the Computer Client The setpoint is red the raw altitude is grey and the filtered altitude is blue Control signals 1000 PWM output 400 b
3. g8 We 5 Hz Altitude m Altitude m Altitude m Altitude m i Oo 4 2 Sensors Raw data o 0 5 Hz l L L 19 19 5 20 21 5 22 22 5 23 23 5 24 Time s 20 5 21 b wc 0 5 Hz ou LAA Ni ij l Raw data o 1 Hz ii m S 1 i 19 19 5 20 20 5 21 21 5 22 22 5 23 23 5 24 Time s d wc 1 Hz I ARINA INN I Be f Ny j N Il AP i Raw data o 2 Hz all MING i l i AU iii i 1 1 1 1 1 L i 1 19 19 5 20 20 5 21 21 5 22 22 5 23 23 5 24 i s Time we 2 Hz Raw data g 5 Hz i 1 i i 1 1 1 i i 1 1 19 19 5 20 20 5 21 21 5 22 22 5 23 23 5 24 i s Time h we 5 Hz Figure 4 7 Low pass filtered pressure sensor with a first order Butterworth filter with different cut off frequencies The left plots show how a spike from a door is filtered and the right plots show a climb to 2 m altitude followed by a descent to the ground again 57 Chapter 4 Slung Load Control on a Quadrotor Raw data oO 1 Hz Raw data oO 1 Hz Altitude m Altitude m 2 L 1 1 L L 1 L 1 ul 1 ul L L L 26 5 27 27 5 28 28 5 29 29 5 19 19 5 20 205 21 215
4. stil or ree kaaa Ly ag t a A AN a La Wao as ahel r kad LJ Mpa ie La a T 2500 a G To iad oo z5 v7 2500 m a L 4500 5000 a Roll angle blue and 1 cos a green b Pitch angle blue and 1 cos q green Figure 4 16 Pitch and roll angles compared to the output from the approximation of 1 cos Q The experiment consisted of three parts First the quadrotor was tilted 30 in both directions along each axis Xg and Ypg Second it was tilted 30 in between the two axes to verify the behaviour in all directions Finally the quadrotor was shaken and tilted slightly in different directions to study how accelerations affected the approximation of 1 cos The final controller can be seen in Figure 4 17 It is a parallel implementation of a PID controller with anti windup tracking manual tracking and a feedforward solution for the tilting It is integrated into the existing control according to Figure 4 18 Discretization Since the PID controller was to be implemented on a computer it had to be discretized Each part s discretization will be presented and discussed separately The proportional part is static so it requires no approximation P kh K r kh y kh 4 30 where kh is the time instant t kh when the computer samples the values and A is the sample time Manual input Figure 4 17 The final controller Parallel implementation of a PID controller with an
5. Connection Settings interface that was design can be seen in Figure 3 3 Phe maa 13 09 Trim Terminate Left Trim 0 Turn speed 0 Forward speed 0 Side speed 0 Alltitude speed 0 Land or Lift Figure 3 4 The Steering interface In this view all the action regarding the flight of the LinkQuad is made While enabling the steering of the LinkQuad with the Android Client in a way that is reminiscent of how the RC is used several functions that the user should be able to perform were created The user should be able to change both the attitude and the altitude of the LinkQuad These functions would enable the Android Client to steer the LinkQuad Some more functions were added to enable termination of the Server and automated lift and land functions By using the mapping of controls see Table 3 5 a natural mapping is retrieved using 39 Chapter 3 User Control of a Quadrotor the IMU buttons and a virtual joystick The terminate button was decided to be red with white text because of using a standard solution of mapping dangerous actions gives a good impression of that pressing the terminate might not be a safe action The trim buttons were placed with the trim right button on the right side and the trim left button on the left side In order to use a natural mapping of the trimming of the yaw The green button that is used to activate the IMU readings of the pitch and roll values was placed on the right side to be easily
6. This was also a general idea when placing the buttons to have lot of free space to decrease the number of accidents A text log was placed in the middle of the screen to give the user direct feedback of the internal state and the actions that are performed This so that the user would be able to evaluate that what he or she did actually was what was intended The Connection Settings interface that was design can be seen in Figure 3 4 Implementation This section describes and gives an insight of how the Android Client was implemented and designed Because of how the flow of the code works in Android applications which work with short lived activities it was hard to implement a continuous loop that listens to the network while checking what the user does at the interface This was instead solved by a Steer activity creates a thread that takes care of the network communication and functions as the state machine while the steer activity takes care of the user interface This is also a general solution in the implementation to break loose the interface from the functionality by using the default functionality of the Android activities The network implementation for the Android Client was implemented using the structural design pattern Template Method Gam96 p 325 p 330 The Template Method pattern was used to create a abstract class Connection which supplies the network using classes with a common interface to a generic connection This to ensur
7. 4 5 This would be very bad since that could lead to the LinkQuad oscillating with a magnitude of one meter To counter this a low pass filter was designed to filter the noisy values and give a more accurate signal without introducing too much delay A second aspect to consider was occasional spikes in the pressure that could come from the opening and closing of doors in the surroundings If these spikes were allowed to enter to the control loop the altitude could be misinterpreted by several meters Pressure to altitude conversion A conversion from pressure to altitude was 39 Chapter 4 Slung Load Control on a Quadrotor required to use the pressure sensor to calculate the altitude of the LinkQuad The conversion formula was provided by the developers of the LinkQuad and can be seen in 4 9 101325 1000 p ltitude ____ 4 altitude 119 4 9 The pressure sensor provides the pressure p in kPa Since the altitude over sea level is sought after the pressure p is subtracted from the average pressure at sea level in Pa The conversion factor converts the pressure difference to altitude in decimeter l 119 Low pass Filter During the development of the low pass filter several different types of filters were tested in MATLAB by using the pressure data from real flights with known altitudes and spikes from door openings By testing the different filter types it was shown that a complex filter did not produce any better
8. E an 0000 0 1 0 Of a 0 0 ly 0000 0 0 o j 0 B 0000 0 0 0 1 B 0 0 0000 0 0 0 B 0 4 51 Chapter 4 Slung Load Control on a Quadrotor where p and py describe the position of the cart py and py are the velocities of the cart and p are the angles from Figure 4 4 is the length of the load arm and g is the acceleration due to gravity This model is mass independent which is attractive for the later application on the quadrotor since it can then be loaded with a variety of loads and the model would not have to be altered When the new approximation had been derived it was introduced to the existing Simulink files of the process and the control loop was to be modified The existing control consisted of a linear quadratic regulator LQR and since the system was easily converted from using a spherical pendulum to using the new model the LOR was kept with some modifications to the cost matrices to keep the load hanging straight down The final cost matrices can be found in 4 8 100 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 10 0 0 0 0 10 0 g o o o 0o 100 0 0 0 oh o a 0 0 0 0 0 1 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 1 The focus was instead shifted to making a reference generator that could be controlled from Android Client or Computer Client The main goal of the reference generator was to allow control of the position of the cart via the network protocol that had been developed in
9. Martin and Rahul Mangharam Cooperative flight guidance of autonomous unmanned aerial vehicles In Cooperating Objects Network of Excellence 2011 Erich Gamma Design Patterns Addison Wesley 1996 Google Android developer homepage http developer android com index html May 2011 Date of visit 2011 05 16 Tore H gglund Reglerteknik AK F rel sningar Department of Automatic Control Lund University 2009 Course literature for the basic course in Automatic Control at Lund University Gabriel M Hoffman Haomiao Huang Steven L Waslander and Claire J Tomlin Quadrotor helicopter flight dynamics and control Theory and experiment In AZAA Guidance Navigation and Control Conference and Exhibit 2007 Wayne Johnson Helicopter Theory Princeton University Press 1980 Michael Lichtenstern Michael Angermann and Martin Frassl Imu and gnss assisted single user control of a mav swarm for multiple perspective observation of outdoor activities In Proceedings of the ION International Technical Meeting 2011 January 2011 Per Ola Larsson and Rolf Braun Construction and control of an educational lab process the gantry crane In Reglermo te 2008 Lule June 2008 Lorenz Meier Gumstix datasheet gumstix overo fire http pixhawk ethz ch wiki electronics overo May 2011 Date of visit 2011 05 16 Donald A Norman Design of Everyday Things Basic Books 2002 QNX11 Sonl 1 WAAO09 Chapter 7 Bibliography Q
10. Swedish Master s thesis student 2 How easy can you see the func tionalities of the controls Is the joystick easy to use or does it contain any faults After getting an fast introduction to the application how would you connect to the LinkQuad After that how would you steer the LinkQuad Would you say that the applica tion is user friendly It seems pretty intuitive I think I could use it after a couple of sec onds In upwards downwards direction the joystick feels a litle too digital in comparison to an analog one Using Connection Settings enter the correct IP Adress then press Ok af ter that press Launch By tilting the phone and using the joystick Not really familiar with the UDP and TCP option The joystick control was a bit tricky if you compare to an analog one Connection settings enter IP number and choose protocol and after that press launch First press lift to take off and then press toogle button and tilt to steer Use the joystick to go up or down or turn Very easy Nice large icons Works nice Hard to really know how it works since you can t actu ally feel what you are doing Enter connection settings and en ter the IP address then choose the proctocol and then press launch You control the altitude and yaw with the joystick and the pitch and roll angles by tilting the phone while pressing the accelerometer button Yes I think so but haven t tried it
11. achieved by answering the seven questions for each interface They were also used to ensure that the user gets the necessary feedback to ensure the user that the action that was performed indeed did fulfilled the goal of the user DO Chapter 3 User Control of a Quadrotor How easily can one determine the function of the device How easily can one tell what ac tions are possible How easily can one determine the mapping from intention to physical movement How easily can one tell what state the system 1s in Table 3 2 Using the seven stages to ask design questions page I of 2 Menu Interface Connections Settings Interface Steering Interface This question was solved by adding Menu item names that were self ex plaining For an example Connec tion Settings handles the informa tion needed for the Steering activity to connect to the Server The names of the Menu items ex plain what can be done This is more an Android question By choosing understandable names as Menu items Android provides the solution of just touching the Menu item to travel through them The menu activity only has one State either the user is in the menu or the user already has clicked on a menu item and navigated to a new activity The item from the Menu which is the only road to travel to this activ ity has a clear and understandable name that describes what the task is in this activity By having describing labels above the fiel
12. an computer to ssh to the gumstix Gumstix development board e Development platform gcc g Eclipse Visual Studio 109
13. and receive functions Implementing this Template Method as the abstract class Connection enables the implementation of the same functionality for both TCP and UDP by creating two implementations TCPConnection and UDPConnection of the abstract class Connection The PhoneMessage is a class struct that describes what a message should look like and give functions to easily make a sendable package It is used to make it easy to handle messages and being able to make changes in message structure without having a lot of dependencies to the rest of the code Field Tests The design and the implementation were field tested by four different tests The first test was to connect the Android Client to the Gantry Crane Section 4 1 and steer it with the IMU as well as with the joystick The second test was to connect to the Server aboard the LinkQuad and control the panning of the external camera The Third test was done by studying the pulse width modulation PWM output while trying to steer the LinkQuad with the motors shut down The last test was flying the LinkQuad with altitude control It was noticed in these tests that steering with the IMU was much more easy than steering with the joystick The IMU gave a smooth control over the steering whilst the joystick gave a more rough steering experience It was also noticed that the pitch and roll inputs gave a proportional reaction in all of the tests Also the joystick behaved in a similar fashion but it
14. and a scared user gains access to the set point he she can pull down the set point to ground level to stop the quadrotor resulting in a crash A second condition has to be added and a good property to study would be the velocity If the quadrotor is close to the set point and has a low velocity the transition to Flying can be made All that is needed then is the derivative of the altitude and that has already been developed in the controller Since the derivative action of the PID controller calculates the derivative of the altitude it can be used in the second condition As the PID controller is on parallel form it is easy to access the derivative action and no modifications are required The control itself is the same throughout the states in the state machine and no parameter changes are made anywhere However the PID controller does support bumpless parameter changes and it would be interesting to study if other parameters would improve the performance at lift off and landing situations TI Chapter 4 Slung Load Control on a Quadrotor For further details about the software design of the state machine see Section 2 3 4 4 Control of the Slung Load The original plan was that the altitude control was to be sufficiently good to make the control of the slung load on the quadrotor very similar to the one developed for the gantry crane However multiple factors such as lead time for hardware and underestimated complexity of subproble
15. configuration file General settings welcome message Welcome message that is printed on start up of the system Altering this property is a good way to find out whether you are using the proper configuration file Values Arbitrary Default Mandatory protocol Type of protocol that should be used for connecting to the other part Values tcp or udp Default Mandatory port Port of the network communication lt ip address gt lt port gt Values 49152 65535 Default 50000 Client specific settings ip IP addresss to the Server Values The IP address of the Server in IPv4 format Default 192 168 0 1 send frequency This is the frequency of how often the Computer Client will read the connected human interface device HID and send the input to the Server Unit Hz Values 10 50 Default 20 yaw trim You often need to do minor adjustments to the setpoint of the yaw control to get the yaw rotation at an equilibrium It is called trimming and this property is the initial trim on start up Values Rig dependent It can be both positive and negative Default 0 listen2PS3 Choice of HID can be either a PlayStation 3 3 gamepad PS3 gamepad or a keyboard If this property is set to true Computer Client will use a PS3 gamepad If this property is set to false it will use a keyboard 103 Appendix B Manuals Values true or false Default Mandatory PS3 sensitivity The sensitivity of the axises on the PS3 game
16. connect to the Server directly through a ad hoc network network An android smartphone today version 2 2 of android can not connect to ad hoc network network by default To accomplish this we need to have a rooted android smartphone and change the smartphone behavior We solved this by using a rebuild on wpa_supplicant that we configure to accept ad hoc network connections by the guide partly made by bla10 Using the Android Client In our implementation we are always using the Android smartphone in landscape mode The first thing we will see when we start the application is a menu containing the links connection steering and exit as can be seen in Figure B 1 Tie Connection Settings Launch Exit Figure B 1 Menu Activity In the connection activity Figure B 2 we will be able to change the ip address of the device to connect to and also see our own address This address is not set 1f you not are connected to any network We will not direct connect to the device when pressing OK it will connect when we want to start steering the LinkQuad When we have set the address we can go to the link steering in the menu This will bring up a view containing 5 buttons a text field and a joystick as can be seen in 101 Appendix B Manuals MAS 16 45 Ip Address Local adress 0 0 0 0 Using TCP protocol Figure B 2 Connection Activity Figure B 3 In the status field messages such as if we successfully connected to
17. gamepad Figure 3 9 A PS3 gamepad seen from the top 46 3 4 Computer Client Figure 3 10 The original RC controller 47 4 Slung Load Control on a Quadrotor This chapter will discuss how the slung load control was investigated and which different problems that were found and which problems were solved The slung load control problem originated from the intention of flying with a load attached to the quadrotor with a string This would result in an oscillative behaviour that would perhaps affect the quadrotor as the load would act as a pendulum The LinkQuad has a lifting capability of 300 g according to the manufacturer UAS Technologies Sweden AB and this was the target mass of the load to be controlled The quadrotor was to be able to carry this load and to cancel out its oscillative behaviour and disturbances in hover The existing control of the LinkQuad consisted of attitude control and an open loop control of the thrust To reduce the complexity of the slung load problem to a two dimensional problem it was decided to close the loop of the thrust as an altitude control The quadrotor would then be able to cancel out the oscillations by moving in the horizontal plane which is a problem that could initially be investigated on the gantry crane an existing lab rig consisting of crane with a slung load that can move in the horizontal plane On this lab rig a model of the load could also be developed and evaluated The su
18. had made some changes which led to a more challenging development than was expected The challenges and their solutions are listed below An Android application consists mainly of activities which are executed as the user interact with the screen This is not suffice since the application will have to maintain a constant network connection and read continously from the IMU It should thereby not be possible to have the application running when the smartphone is sleeping or the application is not in focus This could be solved by extending the application with either a Java service or a Java thread to handle these tasks in the background It was decided to use the latter since it was considered to be easier to implement The challenge of closing the thread and exiting the application on lost focus was solved by killing the application as soon as the actitivy which the thread belonged to was paused or exited Together with a non blocking read operation in the network connection it could be said that the Android Client is a single threaded application with activities attached to it Computer Client The Computer Client application will reside on computer systems and since C was chosen as one of the programming languages for the thesis see below the problem of machine and operative system dependencies came forth For example a Windows machine does not support Posix threads pthreads per default and nix systems does not support Windows thre
19. keyboard must therefore only be used to test communication and never to fly the LinkQuad itself 44 3 4 Computer Client PS3Listener se Key StrokeListener Figure 3 8 The class diagram of the Computer Client PS3 gamepad The PS3 gamepad uses the Bluetooth HID protocol and is thus registered as a joystick on computers which support the HID protocol Since it 1s registered as a joystick one can use any simple gaming library to access the input from the PS3 gamepad In Computer Client the choice fell upon Simple DirectMedia Layer SDL since it could be used on several platforms and it was easy to apply QNX11 Computer Client uses SDL to collect the available inputs at an instant as events and Stores their values and states When all available events have been collected all values and states are interpreted to produce a correct message which is put it in the outgoing mailbox see Section 2 4 for details on mailboxes After that the procedure is periodically repeated The controls of the PS3 gamepad are mapped to match the existing RC controller with some adjustments for the nature of the sticks of the PS3 gamepad and the accessibility of the software The RC s mapping can be found in Table 3 6 an image of the controller can be found in Figure 3 10 the PS3 gamepad s mapping can be found in Table 3 7 and an image of the gamepad can be found in Figure 3 9 As can be seen in Figure 3 10 the left stick is not f
20. low pass filter of the pressure sensor is a second order Butterworth filter whose purpose is to reduce noise and the impact of spikes induced by events in the surroundings Classification system and or index terms if any Supplementary bibliographical information ISSN and key title ISBN 0280 5316 Language Number of pages Recipient s notes English 1 114 Security classification http www control Ith se publications Abstract The objective of this thesis is to investigate the use of commercial devices as user interfaces on a quadrotor and to investigate solutions to the problem of slung load control A slung load is a uniform mass attached with a wire which 1s allowed to swing freely to the bottom of the quadrotor The purpose of substituting the radio control RC controller with a commercial smartphone is that they are more easy to grasp and might therefore be easier to use for a novice than an RC controller The existing radio control link does not exist on a smartphone so it was complemented by a wireless network connection via transmission control protocol TCP or user datagram protocol UDP The smartphone does not have the same interface as the RC controller either so the same functionalities were implemented by using the touch screen and the inertial measurement unit IMU of the smartphone However this requires altitude control since the lack of multitouch in Android does not allow several inputs at the same time
21. no time to test its functionality on the quadrotor 6 4 Altitude Control The altitude control consists of a PID controller with a feedforward gain for the tilt angle It acts upon a low pass filtered signal from a pressure sensor as a measurement of the altitude and receives its setpoints from the Computer Client or the Android Client via a wireless network connection The controller is a parallel implementation of a PID controller with limited derivative gain anti windup bumpless transfer from manual to automatic and a feedforward gain for the tilt angle see Figure 6 1 for the continous version This was discretized with a sample time of 0 01 s and parameter values K 20 7 10 Tg 0 27 and N 10 The anti windup and bumpless transfer is implemented by having the integral action track either the saturation or the manual input by a feedback loop with gain 7 10 The feedforward action s purpose is to aid the PID controller with a fast counteraction of the loss of thrust when the quadrotor tilts and divides its pure vertical thrust into one vertical and one horizontal effective component This is done by applying a gain of 1 cos a where is the tilt angle to the control signal of the PID controller This results in a gain of 1 when the angle is O and an increased gain as the angle grows However large gains during short time intervals were observed during an experiment and were found to originate from unknown acceleration spikes a
22. off Flight Trim Computer 11 Lift Direction Trim Right Lift off Flight Trim Computer 12 Ping Ping Table 2 6 The test set of the computer client Input is the action that the user intro duces Expected output is the log of the client For the Computer Client it was also important that the tests were done for the Unix Mac and Windows platforms to ensure portability The tests on the Android Client were done by using the same test techniques and test set Table 2 6 as the Computer Client In comparison with the personal computer environment it was more challenging to monitor the execution of the server program since the Server is executed on the 29 Chapter 2 System Overview Angstr m distribution on the Gumstix board Because of this the black box technique equivalence classes partitioning was used By letting each message being an equivalence class and creating a single test for each equivalence class the test set can be seen in the Table 2 7 Bur03 p 67 LQServer 1 Direction Server parameters LQServer 2 Lift Beginning to send values hover LQServer 3 Terminate Emergency land end sending values LQServer 4 Close Emergency land close existing connections LQServer 5 Trim Right Adding value to the yaw LQServer 6 Trim Left Adding value to the yaw LQServer 7 Ping Ping in logger and ping back Table 2 7 The test set of the LinkQuadServer Input is the message sent from client Expected o
23. on quadrotor it was not so extensive Two videos were found from a project at the Aerospace Controls Laboratory ACL at Massachusetts Institue of Technology MIT in which a quadrotor successfully rejected disturbances on a slung load However no official papers or web page could be found 1 8 Outline of the report The report is divided into the following chapters Each chapter contains a number of sections that are described below e Introduction This chapter presents background information problem description context of the problem methodology and related work e System Overview This chapter describes the system as a whole platforms the main software of the server network communication serial communication logging and how tests and verification was executed e User Control of a Quadrotor This chapter describes the solution of the user control of the LinkQuad e Slung Load Control on a Quadrotor This chapter describes the gantry experiment sensors the altitude control and the progress of the control of the slung load e Conclusions This chapter describes the conclusions of the thesis and recommendations for future work e Appendices Documents that describe how to use the applications that has been developed and other aspects that help to set up the system 16 2 System Overview This chapter will discuss how the system as a whole is composed and how the different components interact The chapter first presen
24. parallel This way the clients of the system and the protocol could be tested on a real process to seek out missing parts and test its performance The reference generator borrowed the necessary functionality of Server and it was adjusted to fit into a MATLAB S function A S function is a way to implement the functionality of a Simulink block in C or C and by doing so introducing 1 e network communication The S function became a simple version of Server which could interpret direction messages into set points of the cart s position This was introduced into the modified Simulink files and run successfully with both Android Client and Computer Client steering the rig 4 2 Sensors A solution to the problem of altitude control and control of the slung load requires some sensors The LinkQuad has a set of sensors e g an accelerometer and a pressure sensor that can be reached through a serial bus to the Sensor MCU SMCU The control of the slung load requires another sensor as well This was to be solved by attaching an angle sensor to the LinkQuad and anchor the slung load to ie This section will discuss first the angle sensor that was built originally for the gantry crane experiment and which was ported to the LinkQuad Second the pressure sensor will be introduced and how its noisy behaviour was filtered 52 4 2 Sensors Raw data o 10 Whi rN i fi i wifi i il PN IR i ih Altitude m
25. problem for Android is the hardware dependecy and the open source customability Manufacturers may develop different implementations of the Android which may require porting of software Since the different manufactures are using their own hardware it may result in that a Android smartphone does not fullfil the hardware requirements Iphone smartphones are on the other hand restricted and they have all the similar hardware and are therefore more reliable The Iphone compared to the Android smartphone also supported ad hoc networks which is a crucial feature since the Gumstix boards on the LinkQuad had no DHCP server configured in its present version These problems limits the possibilities of which platforms can be used for an Android client Android was still considered to be the best choice since the ad hoc network problem could be solved by changing the wpa_supplicant the network software client and since lack of multitouch was not a major drawback The problem of different platform specifications still remains and this system was tested on a LG P500 Optimus One from here on referenced to as the test phone or the test platform Multitouch or Singletouch When the development first started it was planned to use the inertial measurement unit IMU to set pitch and roll and to use a virtual joystick to set thrust or altitude and yaw setpoints simultaneously The virtual joystick and a button to activate the capturing of IMU values were fi
26. result than using an simple filter e g a first or second order filter Filters that were considered were the first and second order Butterworth filters Filters of higher order e g Chebyshev filters were also considered in the beginning but since the simpler Butterworth filters were proven to be sufficient the focus remained on these Each filter type was evaluated with several cut off frequencies on a generic flight sequence from the ground to 2 m altitude and back to the ground At the end a spike was generated from a door being slammed shut The properties of the different filter outputs that were studied were how much delay the filter produced and how much it dampened the noise The first order filter has almost no delay at all cut off frequencies and a damping of the spike to less than 1 5 m but it lets noise with a magnitude of 0 3 m or more through see Figure 4 7 except 0 5 Hz where instead too much delay was introduced see Figure 4 7b The second order filter had a delay at lower cut off frequencies than 2 Hz but it had also an improved damping of the noise at this frequency see Figure 4 8b The higher cut off frequencies had much less delay but also had less dampening of the noise see Figure 4 8f and Figure 4 8h The cut off frequency of 2 Hz was chosen to be implemented since it dampened the noise to less than 0 2 m and the spike of 2 5 m to m without introducing too much delay see Figure 4 8d and Figure 4 8c If the firs
27. that they will use it The graduate will use the Android Client in his own research and would want to have an application that is easy to understand and use but most important easy to modify The quadrotor experienced pilot will use the Android Client to do test flights or to get a new flight experience The ordinary person is probably a friend of the graduate or to the quadrotor experienced pilot who wants to try the LinkQuad From this set of users it can easily be seen that the most probable user will have some but not an extensive knowledge of flying quadrotors This implies that the steering and the graphical interface should be as natural as possible so that a user with little knowledge of quadrotors can understand what the different controls are used for It should also be in focus that the steering should be reminiscent of the steering interface of the RC Since if one of the users has used a RC controller before the introduction would be easier The application is not meant to be accessible for the public Its main purpose is to be used for development and research Therefore the focus will also lie on making the software easy to modify and maintain 32 3 3 Design of the Interface 3 3 Design of the Interface Intention to act Evaluation of interpretations Sequence of actions Interpreting the perception Perceiving the state of the world THE WORLD Execution of the action sequence Figure 3 1 Th
28. the following 128 bits as the payload In the beginning of the thesis a version with 8 bits ID and 32 bits of payload was used but this resulted in a constraint on the payload The constraint was that a direction message had only a range of 127 to 128 which was sufficient for the current problem but was an unnecessary restraint for future usage The Gumstix board has support for the IEEE 802 1 1b g standard for wireless networks Meil1 implying a possible connection speed of 54 Mbit s this would result in that a 40 bit message sent over a TCP 23 Chapter 2 System Overview Message Type Message ID Payload Acknowledge Terminate Message Acknowledge Acknowledge Table 2 3 Different network messages with their ID number and how the paylod is interpreted Close Message connection 52 bytes head will result in a 57 bytes package Dyk03 This package would be sent in 8 44 10 s in a ideal situation with direct line of sight and no interference Since the control loop has a sample time of 10 ms this implies that 1184 different messages can be received during one period For the 160 bit message sent over a TCP connection 52 bytes head would result in a 72 bytes package and the package would be sent in 1 07 107 s This would enable 937 different messages to be received during one period of control Since receiving only one message in a loop of the control is sufficient the limitation of using eight bits is unnecessary and woul
29. the low pass filter positive and negative step responses the feedforward action and how well the clients were integrated into the system These aspects will now be analyzed in the given order Bumpless Transfers The first occurence of bumpless transfer from manual to automatic control was under ideal conditions see Figure 5 1a The quadrotor had no velocity in any direction and was hovering at an altitude of 1 5 m The result was a bumpless transfer During the second bumpless transfer from manual to automatic control the quadrotor had a velocity along the Z axis see Figure 5 1b The result was a small bump in altitude after the transfer but it is rejected within ten seconds and never reaches a magnitude larger than 0 8 m Low pass Filter The low pass filter was studied during a manual take off and flight and it had already converged on the ground before the logging started The filter s output was 79 Chapter 5 Experiments and Results compared to the raw data in Figure 5 2a and it suppresses the noise to a magnitude of 0 2 m or less without any phase delay The filter was then compared to the simulated ideal version of the same filter and both real and simulation data are consistent see Figure 5 2b Step Responses The two positive step responses had settling time of two seconds which exceeded the expectations of response time see Figure 5 3a and Figure 5 3b There was some stationary error but this is believed to origi
30. time would be larger it could have severe effects on the process see Figure 5 6b were observed that In retrospect corrupt values from the approximation of This bad behaviour originated from an unknown acceleration along the Zg axis see Figure 5 7b which could be a bad value from the sensor since the quadrotor had no sudden gain of altitude following the peak of acceleration Clients Both clients were successfully integrated and supplied the user with the same behaviour for similar tasks The Android Client was used to produce the bumpless transfer in Figure 5 1a and the positive step response in Figure 5 4a The Computer Client was used to produce the bumpless transfer in Figure 5 1b the positive step response in Figure 5 4b and the negative step response in Figure 5 5b 81 Chapter 5 Experiments and Results a The transfer from manual to automatic occurs at the dashed line at an altitude of 1 5 m The quadrotor was in an almost perfect hover with no velocity in any direction Altitude raw and filtered b The transfer from manual to automatic occurs at the dashed line at an altitude of 2 m The quadrotor had a velocity directed upwards at the transfer but the control cancelled out its effect within ten seconds and the bump did not exceed 0 8 m 82 Figure 5 1 The altitude at the bumpless transfers 5 2 Review Altitude raw and filtered I Raw data Filte
31. touch screen and the inertial measurement unit IMU of the smartphone However this requires altitude control since the lack of multitouch in Android does not allow several inputs at the same time thus making it impossible to adjust the thrust as the pitch and roll is adjusted Another commercial device that was investigated was a PlayStation 3R gamepad PS3 gamepad whose joysticks were very similar to the RC controller s but its shape was smaller and more ergonomic It also communicated via TCP or UDP over a wireless network connection The purpose of slung load control is to reduce the oscillations of the slung load and its effect on the quadrotors flight performance thus enabling it to handle heavier loads The slung load control consisted of several subproblems model of a slung load altitude control filtering of noisy sensors and the slung load control itself These were all solved but the last one where the slung load control was not implemented on the quadrotor due to lack of time The model of the slung load was done by letting the dynamics of two traversed simple pendulums approximate the slung load s angular movement The altitude control consists of a PID controller extended with anti windup and bumpless transfer which is combined with a feedforward control of the tilt angle The controller acts upon a low pass filtered pressure sensor as a measurement signal and receives its setpoint from one of the commercial devices mentioned above The
32. velocity orientation and gravitational forces linear quadratic regulator A state feedback controller which optimizes its poles according to a quadratic cost function least significant bit In computing the least significant bit is the bit position with the least significant value most significant bit In computing the most significant bit is the bit position with the most significant value proportional integral derivative controller A generic and simple feedback controller PlayStation 3Q gamepad Gamepad originally configured for Sony PlayStation 3 console It is possible to connect the gamepad to a computer via Bluetooth pulse width modulation A technique for controlling power to a motor Simple DirectMedia Layer A cross platform hardware API widely used in Linux games Sensor MCU The MCU that is responsible for sensor algorithms on the LinkBoard ad hoc network A network connection without Dynamic Host Configuration Protocol DHCP between two devices Android Client The client application for connecting and controlling the LinkQuad from a Android smartphone Android SDK The software development kit for android smartphones Computer Client The client application for connecting and controlling the LinkQuad from a computer LinkBoard The LinkQuad hardware LinkGS Graphical User Interface used to configure develop and operate the the LinkBoard LinkQuad A specific quadrotor used in the thesis as a test platform Server T
33. was easy to loose the grip of the joystick which made it hard to use it without looking on it When using the Android Client with altitude control in the final test it was noticed that the Android Client behaved equally with the Computer Client More information about the result of the flight can be found in Chapter 5 For these reasons the Android Client should be used with altitude control and not with thrust control 3 4 Computer Client The Computer Client is an alternative to the Android Client It supplies the user with two options of inputs a keyboard or a PS3 gamepad but lacks graphical feedback on the screen This is due to the need of keeping the pilot focused on the LinkQuad and to its main purpose which was to be used as a tool in tests The advantage of using the Computer Client instead of the Android Client is the higher accuracy of the PS3 gamepad and the fixed inputs of the keyboard However the disadvantage of using the keyboard is the loss of variable inputs Also both the HIDs need to have an extra computer to run the Computer Client on 43 Chapter 3 User Control of a Quadrotor Figure 3 7 The state machine diagram of the Computer Client The Computer Client can connect to the Server with either a TCP connection or an UDP connection via an ad hoc network The user inputs are then interpreted into different commands or setpoints and are sent to the Server which will use them in different parts of the control sys
34. was not required so it was Stopping has been successful and the disabled to avoid crashes disabled to avoid crashes software is shut down 76 4 3 Altitude Control Raw data o 10 Maa N d Mi i Altitude m 10 20 30 40 50 60 70 80 Time s Figure 4 22 Raw and low pass filtered data from the pressure sensor converted into meters when the quadrotor sits on the ground with motors off The natural variations in the atmospheric pressure can have a magnitude up to I m effect turbulence and battery depletion causing the quadrotor to require a variable thrust to hover continously Another solution that was discussed was to set the setpoint at 1 m above the ground and after the quadrotor had stabilized around the setpoint it would do a slow descent according to the first intuitive solution This was an interesting solution that was implemented to some extent in Emergency Landing but was never tested due to bad weather conditions during the final week of the thesis The problem with the atmospheric pressure variations was not such an issue for the Lifting state since it could be implemented to do an aggressive step response to a secure height e g 2 m The problem of this state was to determine when to switch to flying and allow the user to affect the set point It is not sufficient to just say that the switch should be made when the quadrotor is close to the set point If the quadrotor overshoots the set point
35. with a P Pole zero map of the closed loop with a Pl T T T T T T T T x x 1 5 4 1H R 0 5 4 2 E H i 4 T O igs wi 4 fan wc 05 J 4 L 4 1 5 4 x 1 1 1 2 1 1 1 1 1 1 1 x 0 5 0 0 5 0 1 0 08 0 06 0 04 0 02 0 0 02 0 04 0 06 Real Axis Real Axis a Pole zero map of the closed loop system with b Pole zero map of the closed loop system with a P controller Imaginary Axis a PI controller Polelzero map of the closed loop with a Polelzero map of the closed loop with a PD T T T T T X 4 15 4 4 1 4 4 0 4 2 E gt m H amp Uo Ei E B w 4 05 J 4 1 a 4 15 4 L 1 x ee 1 fi fi 4 0 5 0 0 5 20 15 10 5 0 Real Axis Real Axis c Pole zero map of the closed loop system with d Pole zero map of the closed loop system with a I controller a PD controller Pole Zero Map x Imaginary Axis o amp x L 1 1 L 1 L L 1 L 1 0 9 0 8 0 7 0 6 0 5 0 4 0 3 0 2 0 1 0 Real Axis e Pole zero map of the closed loop system with a PID controller Figure 4 11 Pole zero maps of the closed loop system with the considered controllers 66 4 3 Altitude Control Actuator model Actuator HA Figure 4 12 Parallel PID implementation with anti windup feedback The anti windup is active when saturation in the motors occurs and e 0 It is otherwise equal to zero and does not affect the control Courtesy of Kar
36. without losing any safety and in the same time fulfilling the seven principles The simplification of the tasks is that each task has one unique mapping short of the Land or Lift map Since of space issues in the graphical interface led to that they had to share the same button Tie Connection Settings Launch Exit Figure 3 2 The menu interface The menu view of the Android Client When the user uses the Android Client he or she always has a goal and the menu is just the road to that goal Therefore it 1s important that the menu is effective to use This is the same problem as the seven principles breaking a complex task into simple ones To be able to connect to the LinkQuad some connection details are needed such as the ip address and what network protocol is going to be used user datagram protocol UDP or transmission control protocol TCP It could always be solved by prompting when the Launch button is pressed but that would be unnecessary because usually that information is the same all the time This was solved by adding the menu item Connection Settings which stores the information so it can be reused when connecting The two other menu items are the Launch which connect to the LinkQuad and start the steering and the last are quite self explanatory The evaluation Table 3 4 and the seven questions Table 3 2 shows that keeping the menu small was a good choice because it is only the road to the action that the user
37. would be familiar The mapping of the controls was also natural 90 6 5 Android Client Manual input Figure 6 1 The final controller as a continous system Parallel implementation of a PID controller with anti windup tracking manual tracking and a feedforward solu tion for the tilting Figure 6 2 The modified control The setpoints for the attitude control and the es timated angles from AHRS are still sent from the SCMU to the CMCU The thrust input from the RC receiver has been overridden by the altitude control loop from Fig ure 6 l It receives its setpoint sp from the wireless network and the sensor values from the SMCU The PID output and the feedforward gain is calculated on the Gum stix Board and forwarded to the old thrust controller at the CMCU At the CMCU the controllers outputs are calculated saturated mixed into individual PWM outputs and sent to the motors mapped towards the process in the sense of if the user tilted the smartphone forward then the LinkQuad would also tilt forward The Android Client mainly uses the IMU and the virtual joystick to steer the LinkQuad During the evaluation of the Android Client it came forth that the virtual joystick 91 Chapter 6 Conclusions was hard to use This can be best understood if it is compared to an analog joystick An analog joystick provides the user with a physical feedback of the position of the joystick which gives a feeling of the steering The virtua
38. 22 225 23 235 24 Time s Time s a wc 1 Hz b we 1 Hz Raw data Raw data o 2 Hz J o 2 Hz Altitude m Altitude m 1 55 6 5 27 275 28 28 5 29 29 5 lig 19 5 20 20 5 21 21 5 22 22 5 23 23 5 24 Time s Time s c we 2 Hz d we 2 Hz 3 T T 3 T T Raw data Raw data 25 o 5Hz J o 5Hz 2 5F s i IAN A TA W j P INI Gi i 3 3 1p al l ii y M g J 1 eat af lj 0 54 p f M I 4 sl v 1 Mi Hy uy Ul ETN ee i HW Mid I Ih Bi of Wh Ne ibi i 7 itt i j l ali 36 5 27 275 28 28 5 29 29 5 1 9 19 5 20 20 5 21 21 5 22 22 5 23 23 5 24 Time s Time s e wc 5 Az we 5 Hz 3 T T 3 T T Raw data Raw data 10 Hz 10 Hz Altitude m Altitude m 36 5 27 275 28 28 5 29 29 5 a 195 20 20 5 21 2 5 22 22 5 23 23 5 24 Time s Time s g we 10 Hz h we 10 Hz Figure 4 8 Low pass filtered pressure sensor with a second order Butterworth filter with different cut off frequencies The left plots show how a spike from a door is filtered and the right plots show a climb to 2 m altitude followed by a descent to the ground again 58 4 3 Altitude Control 4 3 Altitude Control The existing control consisted o
39. Bur03 The client sends a message the server unpacks the message then repacks it again and sends it back to the client At the client the received message could then be verified to be the same as the sent message It was important that every message type was tested with both normal values and values that are outside of the limits of the payload and that messages with incorrect identification numbers are tested so it is proven that they are rejected without problems See Section 2 4 for more information about the payload It can then be proven that the network communication can handle all of the different types of values and messages The test set 1s required to be executed for both the TCP and the UDP protocols The test set is presented in Table 2 5 Since the development environment on the computer was equipped with debugger tools it enabled the use of the white box technique coverage analysis It implies that the execution of the code is studied and how inputs affects its flow The goal 1s to verify that all code statements executes in a proper way It is called coverage analysis since a complete coverage requires that all code statements are tested If not a code inspection must be done to design test cases which will maximize the coverage Together with a black box technique boundary value analysis a good set of tests could be created The input for the tests was different messages sent from the network connection see 2 4 The output was wri
40. Combining these two conditions the sample time should be within 0 0068 s and 0 0188 s so the actual sample time of 0 01 s is properly chosen in the middle of the interval The cut off frequency of the open loop system determines the speed of control and is dependent of the low pass filter s cut off frequency The second order Butterworth filter has a phase shift of 90 at the cut off frequency and it increases for higher frequencies see Figure 4 19 The choice of the open loop system s cut off frequency should be done so that the filter s characteristics the phase shift does not affect the open loop system s phase margin so that is unstable To do this the controller must be tuned such that a sufficient phase margin for stability is achieved and the system still has a fast response to inputs A theoretical PID controller has a maximum phase shift of 60 and since the derivative gain is limited the actual controller has a maximum phase shift of 50 see Figure 4 20 As the phase shift has to be positioned so that it creates a positive phase margin for stability the positioning of the cut off frequency of the open loop system 1s limited by the low pass filter s phase shift The parameters where thereby chosen to give a phase margin m 15 1 and a gain margin Am 4 73 According to rules of thumb H4g09 p 54 the gain margin is valid but the phase margin is small However this is a design choice since increasing the phase
41. ISSN 0280 5316 ISRN LUTFD2 TFRT 5910 SE Control of a Quadrotor Niklas Hansson Mikael Rudner Lund University Department of Automatic Control May 2011 Document name MASTER THESIS Date of issue May 2011 Lund University Department of Automatic Control Box 118 SE 221 00 Lund Sweden Document Number ISRN LUTFD2 TFRT 5910 SE Supervisor Anders Robertsson Dept of Automatic Control Lund University Sweden Author s Niklas Hansson Mikael Rudner eee Karl Erik Arz n Dept of Automatic Control Lund University Sweden examiner Sponsoring organization Title and subtitle Control of a Quadrotor Abstract The objective of this thesis is to investigate the use of commercial devices as user interfaces on a quadrotor and to investigate solutions to the problem of slung load control A slung load is a uniform mass attached with a wire which is allowed to swing freely to the bottom of the quadrotor The purpose of substituting the radio control RC controller with a commercial smartphone is that they are more easy to grasp and might therefore be easier to use for a novice than an RC controller The existing radio control link does not exist on a smartphone so it was complemented by a wireless network connection via transmission control protocol TCP or user datagram protocol UDP The smartphone does not have the same interface as the RC controller either so the same functionalities were implemented by using the
42. NX Simple directmedia layer webpage http www 1libsdl org May 2011 Date of visit 2011 05 30 Dag Sonntag A study of quadrotor modelling 2011 Bj rn Wittenmark Karl Johan str m and Karl Erik Arz n Computer Control An Overview Educational Version 2009 Department of Automatic Control Lund University 2009 97 A Photos Figure A 1 The LinkQuad in its original state with a PS3 gamepad Figure A 2 Top view of the repaired LinkQuad after a crash from 5 m above the ground 98 Appendix A Photos Figure A 3 Side view of the repaired LinkQuad after a crash from 5 m above the ground Figure A 4 The final version of the fork sensor for measuring the angle of the load 99 Appendix A Photos Figure A 5 The green in front of the building of the department of Automatic Control at LTH Figure A 6 The ball room of the student union at LTH 100 B Manuals B 1 Android Client The Android Client was created for the purpose to be able to control the LinkQuad by manual steering with the tilt functionality of an smartphone This was implemented by using java with Android SDK with the target version 2 2 of Android Connecting the Android Client to a network Connections to the Server can be made in several ways One is that we connect the Server and the Android Client to an existing network and use that network to communicate to each others A little more difficult way is that if we want to
43. Suggestions for measuring the altitude of the quadrotor ee Figure 4 10 lustration of how the controller closes the loop around the process Design of the Controller The second challenge was to find the simplest controller that could achieve asymptotic stability A way to do this was to find the closed loop transfer function and study its poles and zeros The process itself can be approximated as simple double integrator i y G 4 22 y PE u pu where the input u is acceleration the measured output y is the altitude and m is the mass of the system Closing the loop as in Figure 4 10 gives the closed loop system G Gp Gea 4 23 i 1 G G 4 23 as a transfer function from the reference value to the output The initial investigation of possible controllers considered five types of controllers P PI PID PD and I The controllers poles and zeros could now be studied using the transfer function in 4 23 and the transfer function for each controller A P controller G K increases the system gain but does not asymptotically stabilize the system When the gain K is altered the poles to the origin is altered but they remain on the imaginary axis see Figure 4 1 1a Since the P controller is only stable and since it does not have any integral action it will not suffice A PI controller G K 1 T introduces two unstable complex poles or one unstable real pole depending on the values of K and 7 see F
44. The control signal to the step change of 1 m The quadrotor was level and maintaining its altitude with only a minor feedforward action until the step change occured When the quadrotor overshot the set point and came close a table the pilot started to tilt the quadrotor to move it away from the table The feedforward action behaves properly in the time interval 96 5 s 96 85 s but after 96 85 s it saturates the control signal in both limits due to a unknown value from the accelerometer See Figure 5 6a for details on this subject Figure 5 5 A downwards step change and the corresponding control signal The fil tering of the altitude is done in retrospect with MATLAB 86 5 2 Review The approximation of 1 cos 4 96 96 2 96 4 96 6 96 8 97 97 2 97 4 97 6 97 8 98 Time s 1 a The approximation of a 1500 1000 4 1000 7 1500 g 2000 96 96 2 96 4 96 6 96 8 97 97 2 97 4 97 6 97 8 98 b The unsaturated control signal The saturation limits are O and 1000 and are here removed to show how the bad behaviour of the approximation dominates the control signal spect It saturated the control signal to both its maximum and minimum values It originates from an unknown acceleration along the Zg axis which could be a bad value from the sensor since the quadrotor has no sudden gain of altitude following the peak of acceleration The co
45. Thesis extends the existing system in two ways 1 User control via a commercial device 1 e a smartphone or a PlayStation 3 gamepad PS3 gamepad will be introduced the user should be able to steer the LinkQuad via this device without using the existing RC 2 The LinkQuad should be able to function as a mobile crane with a slung load A slung load is a uniform mass attached with a string to the bottom of the quadrotor and it is allowed to swing freely 11 Chapter 1 Introduction GPS i Magnetometer i Differentiell pressure sensos i a aa a i Accelerometer Sensor MCU i Absolute pressure sensor i eee eee EAT RS232 ee cm Left Motor Control MCU i Right Motor i PWM Back Motor i Front Motor Figure 1 2 This sketch shows the different components of the LinkBoard and the known protocols that are used between the components Motor 1 Motor 3 T aj N Motor 4 Figure 1 3 The existing control Setpoints are received from the RC receiver at the SMCU The setpoints and the estimated angles from AHRS is sent to the CMCU At the CMCU the controllers outputs are calculated saturated mixed into individual PWM outputs and sent to the motors The first problem introduces the following subproblems e Choice of smartphone platform e Communication between the smartphone and the LinkQuad e A user friendly interface on the smartphone which enables most of the functionalities of the RC Since
46. a problem Even though different areas of focus existed both students worked to keep an understanding in each other s areas Testing was done together since the LinkQuad requires both a pilot and an operator of LinkGS The tests were designed in different ways depending on what sort of functionality was to be tested If it was a pure software function that required no actuation of the process the functionality could be tested on the real platforms with the motors turned off However if the function required flying or some other live interaction with the LinkQuad it was always simulated first using models or scripts in MATLAB and then tested on the real quadrotor 14 1 6 Method Pop Scientific article Master Thesis report Manuals Clients developed LQ developed and Network and Server Computer client Seral development and Control and Choice of smartphone Altitude control Pressure filtering Figure 1 5 Workflow chart Documentation have been made continously to ensure that no knowledge was lost during the development and that the information was as precise as possible Distribution of work Mikael Rudner Niklas Hansson Keyboard Controller PlayStation 3 gamepad Serial Communication Network Protocol Control Server Android Client Gantry Crane Computer Client Testing Simulations Low pass Filter Table 1 1 The distribution of work between the two Master s Thesis stude
47. ads The thesis students had access to computers with Windows Unix and OS X as Operative systems so it was decided that the applications should be portable to all three operative systems This could be achieved by using the Boost C libraries that are portable to all of the operative systems in question Boost supports everything from threads and math to network communication and file handling which makes it ideal to use in a multiplatform application BD11 20 2 3 Server Server Since the Server is supposed to be executed exclusively on the ngstr m based Linux distribution it was decided to use the standard C libraries Development Tools The main issue was the decision of which language to implement Server and Computer Client in There was a probability in the beginning that the Computer Client would be integrated into the LinkGS Thereby the same programming language C was chosen for Computer Client Since Server was to be put on the Gumstix board without any dependices the list of possible languages was longer However since the Gumstix board uses the Angstrom operative system which supports gcc and the Computer Client was to be implemented in C the choice fell upon C for the Server as well Since the smartphone was based on Android it was to be developed in Java MATLAB was used together with Simulink to derive and simulate different control and filter algorithms 2 3 Server In this section the gener
48. al implementation of the Server and design patterns will be discussed Detailed implementations of specific parts of the Server are discussed in the following sections Serial Communication Section 2 5 Network Communication Section 2 4 Sensors Section 4 2 Control Section 4 3 and Logging Section 2 6 The Server consists of three major parts Master Connection and SerialCommunicator Master and its implementations contain a state machine and a control loop Connection and its implementations handle the network communication SerialCommunicator consists of two parts an interface of functions for Master to use for the serial communication with the Control MCU CMCU and the second part is a thread which handles the serial communication with Sensor MCU SMCU Several patterns were used to structure and implement the Server both to lower the code complexity and to make it easier to modify in the future Master Connection and SerialCommunicator are separated with either a monitor or a mailbox to ensure mutual exclusion on shared data The heart of the Server is the abstract class Master and its two implementations LoadedMaster and FreeMaster that are two different state machines whose behaviour depend on whether a load is attached or not The state machine functionality exists in Master where the current state is periodically invoked and the specific state s behaviour is defined and executed in the implementing class The current st
49. andles the incoming data from SMCU and a set of functions for sending data to CMCU This enables the control loop to send data direct to CMCU and to not waste execution time to read from SMCU The low pass filter is executed in SerialCommunicator due to the need of a different sample time than the control loop The SMCU can send data at a maximum 25 Chapter 2 System Overview frequency of 500 Hz and is currently sending at 250Hz to SerialCommunicator This is implied by the need of a high sample rate of the low pass filter of the pressure sensor see Section 4 2 SerialCommunicator puts the sensor values in SensorMonitor which is shared with the control loop Control Loop Figure 2 4 Illustration of serial communication between the control loop and the external microcontrollers SM stands for SensorMonitor and SerialComm stands for SerialCommunicator Serial Protocol to Control Microcontroller Unit The existing high level protocol supports arbitrary data to be sent to the CMCU packed in an array with eight floats This can be used in the CMCU by configuring the PID loops in LinkGS to apply receivedParamsX where X stands for the index in the float array as a target or an input The array is also logged and available for plotting during runtime in LinkGS In Table 2 4 the present use of the parameters is presented Pitch Roll Yaw Yaw trim Set point for the altitude control Control signal Current state or PID o
50. ate is kept in ReferenceMonitor and state transitions can be made from either Master or Connection This enables a transition to be triggered by commands from the network communication and thereby the user The state machines and the control are discussed further in Section 4 3 The Master classes communicate with Connection through two mediators MailBox and ReferenceMonitor which were created using the Mediator 2i Chapter 2 System Overview Figure 2 2 The class diagram of the Server The Connection are implemented ac cording to the Template Method pattern The MailBoxz ReferenceMonitor and SensorMonitor are implemented according to the Mediator pattern pattern Gam96 p 273 p 282 Mailbox can contain a message that should be sent to the connected client and ReferenceMonitor holds the current state and the setpoints of the control loop The Mediator pattern implies that the Master classes need only to know that the message is only required to be put in MailBox to be sent In the same way it implies that the Master classes are only required to use ReferenceMonitor to get current setpoints Both MailBox and ReferenceMonitor act as mediators limiting the interaction between the two classes thus making the two parties implementation independent of each other and easier to modify Both MailBox and ReferenceMonitor have mutual exclusive entry to all functions to ensure thread safety The network implementation for the Server were impl
51. ban dba c 90 6 4 Altitude Control ae eee he koe Sw EGE Be ES eS 90 6 5 Android Client oe ees oh oe ae eRe eG ee Kw eS 90 6 6 Computer Client s lt 524 4464 ee2e 64424 4 2B OEE ss 92 O N e eaa ereke nsere n 93 6 8 Future Work aoaaa D EER ER EEHRE S 93 7 Bibliography oaaae aaa eee ee 96 A PHOUOS 2 ca nee ee RRR EEE EEE a 98 Acronyms Manuals Gg Gao ea ee eee ee Re Re REE ee SY 101 B 1 Android Client 226 we om be Ae oo eo oe a ES SS 101 B 2 Configuration Pile 2 45 42 4588856456646 44246 o gt 103 B 3 Ad Hoc Network Configuration on a Gumstix Board 105 B 4 BOM Bill of Materials 2 109 Acronyms 1 Introduction 1 1 Quadrotors A quadrotor is a vertical take off and landing aircraft that is propelled by four rotors instead of using two rotors as a helicopter The advantages of a quadrotor in comparison to a helicopter is that quadrotors do not require to alter the angle of attack of the rotors to tilt the aircraft It increases the thrust on one rotor and decreases the thrust on the opposite rotor to affect an angle Mechanical linkages to vary the rotor blades angles are therefore omitted and thereby simplifying the design and reducing maintenance time and cost The usage of four rotors allows each individual rotor to have a smaller diameter than an equivalent helicopter rotor to produce the same thrust which lessens the kinetic energy that is produced This reduces the damage the rotors w
52. bproblems of the slung load control were to develop an altitude control to develop a model of the slung load develop a slung load control on the gantry crane and integrate all these solutions into a final slung load control on the quadrotor The chapter will first present the gantry crane experiment and the deriving of a model of a slung load After that the sensors that were investigated and the low pass filter of the pressure sensor that was developed to decrease noise will be introduced Third the altitude control will be presented and finally a short motivation to why the full integration was not successful will end the chapter 4 1 Gantry Crane Control The gantry crane is an existing lab process originally developed by Per Ola Larsson and Rolf Braun and used in courses at the Department of Automatic Control Lund University The lab process consists of a cart on a rail with a movable arm which has a slung load attached at the end see Figure 4 1 This process can approximate the slung load problem on the quadcopter if altitude control is assumed and the tilting of the load s pivot point is disregarded It was therefore considered as a good exercise to introduce position control of the cart with set points sent from Computer Client and Android Client while the slung load is kept perpendicular to the floor The original control on the rig makes the slung load follow a circular trajectory while it keeps the cart centered on the rail The
53. chine itself For an example if the user wants the LinkQuad to lift from the ground and become airborne he or she presses the lift land button and the state machine changes the current state into lifting sending the lift command to the Server and then later by itself changes to the steering state This to enable the steering of the LinkQuad It is important that the Android Client and the Server state machines are in sync because if one of the state machines thinks that the LinkQuad is airborne but the other thinks it is on the ground serious problems will happen This was counteracted by having the basic states of the Android Client structured in the same way as the Server and doing much testing But the Android Client does vary much from the Server because the user has more freedom of choosing the next state by his or her actions on the interface 42 3 4 Computer Client Connection To communicate the commands and data that we create or fetch in the steering activity a customized message protocol with TCP or UDP is used This protocol is described in Section 2 4 Because the different low level implementations of the UDP and TCP the code complexity would increase much if one would implement them both directly into the code every time a network operation should be done So by creating a Template Method the problem is abstracted away and the Template Method acts as the network connection to the rest of the application by providing its own send
54. control has to reside on a platform and the development tools and design choices rely on these platforms hardware design In this section the different platforms will be presented and the main focus will be on the smartphone since the hardware options are diverse but might introduce restrictions such as programming language and input possibilities Smartphone Client The Master s Thesis specification contained the assignment to develop a smartphone application that could be used to steer the LinkQuad Some questions existed from the beginning such as if the target platform should be Android or Iphone The Choice between Android and Iphone The first challenge to complete was the choice between Android smartphones and Iphone with 10S Both had their advantages and disadvantages The initial study of Android and 10S are shown in Table 2 1 Tan iPhone wm p To e l o mese FAccustomedness e Opensowe e Table 2 1 Initial study of advantages and disadvantages of Android smartphones and Iphones Plus sign denotes a positive feature Minus sign denotes a negative feature Android uses the Java programming language with the Android SDK and to develop an Android application no licenses are needed Android itself is also open source which in the authors experience gives some assurance that it has some degree of code quality Go011 To develop for the Iphone it is required to have a development license and a Mac computer
55. d and in the toggle button it is quite simple to see the three actions that can be made Write in the ip address toggle the communication protocol and also checking the lo cal address of the network that the smartphone is connected to This is more an Android question By choosing understandable names the Android provides the solution of just touching the field or button to interact with them and the labels gives a good hint of what task the item solves The are two states that are important to know First if the smartphone is connected to the LinkQuad net work Second the different states of the connection settings The con nection settings can be seen in the ip adress field and the toggle button that is used to change the connec tion settings The item from the Menu which is the only road to travel to this ac tivity have a good explaining name that describes what the task is in this activity The task contains many normal controls that are quite de scribing for this application giving a good hint of what it is used for Every button has text bound to it that describes its functionality but it is not as easy to determine what the green button or the joystick is used for No labels could be added due to lack of space This was solved by describing it s functionality in the manual It 1s easy to understand what the buttons do due to the labels but again the green button and joystick are the hardest to map This i
56. d and still maintain the same thrust Joh80 All of these properties affect the process in such a way that the models can not be accurate enough The first model has therefore been used to get pointers to how the controller should be tuned but the final parameters of the controller have been found empirically Altitude Control The gantry crane experiment was so successful that the solution was preferably to be reused To remove the aspects of vertical motion the altitude control became a subproblem to the slung load problem This section will discuss the following subjects e Coordinate systems e Choice of sensors e Design of the controller e Discretization of the controller Tuning of the controller Coordinate Systems Two different coordinate systems will be used in this section and will be introduced here The first coordinate system is called world coordinates X Y Z where X is horizontal parallel to the equator and positive in a westbound direction Y is horizontal perpendicular to the equator and positive in a northbound direction Z is vertical and positive in a direction from the center of the Earth It can be used to define the absolute position of a quadrotor The second coordinate system is called body coordinates Xg Yg Zg where Xg is parallel to the axis of the front and back rotors and positive towards the front rotor Yp is parallel to the axis of the left and right rotors and positive towards the left roto
57. d perhaps constrain the accuracy for future extensions The data of the payload is representing different properties depending of the message type For a command message type 0 and 2 through 5 the first 32 bits of the payload is either 1 or 0 to determine whether it is a command or an acknowledgment to a previous command message For a direction message the payload consists of four integers representing setpoints for pitch roll yaw and difference of thrust The setpoints of pitch roll and yaw are to be sent to the inner loops of attitude control and the setpoint difference of thrust is to interpreted either as how much the thrust setpoint should change manual flight or how much the altitude setpoint should change altitude control TCP and UDP The main advantage of using TCP is that the system gains a reliability of communication but also a disadvantage of slower communication During the testing it was noticed that the disadvantage was so small that it was not noteworthy However this reliability is not always a requirement When a client 1s sending a direction message the Server requires only the latest input and it does not matter if occasional messages are lost during the flight The main advantage of using UDP for realizing the network protocol is a fast communication but at a loss of reliability since packages may be lost During testing it was found that the loss of packages happens more often the further apart the reciever and t
58. d the angles are assumed to be stabilized with PID controllers This model has its disadvantages since it does not account for inertia and takes thrust as input It was therefore altered to take the existing input range from the radio controller 0 1000 and to convert that input into thrust inside the model The conversion between input and thrust was initially investigated in two experiments where input was compared to the resulting acceleration The first experiment was to let the quadrotor hover and the second experiment was to maximize the input to find the maximum acceleration Since the quadrotor hovers the acceleration from the propellers equals the acceleration from gravity thus hover 2 The maximum acceleration was found to be close to 2g Assuming that the relationship between input and acceleration is linear see 4 14 the constant c were derived with the mass m 1 4kg and accelerations from the experiments see 4 15 ma c u Lo ma gt gt c 4 14 u 39 Chapter 4 Slung Load Control on a Quadrotor 1 4 8 Chover 5900 z 0 02326 1 4 2 Cmax OO 0 0303 4 15 The relationship is nonlinear but it was decided to be linearized around the hover position since it is the area it will operate in most of the time The initial investigations of the modified model showed that its behaviour looked like a quadrotor s but to confirm this real input signals were fed to the model and the altitude outpu
59. de and yaw value An effort to implement the functionality to be able to press the green button while using the joystick was also made it succeeded but it was noticed that the Android multitouch was badly implemented The current implementation are not using multitouch for this reason The design problem with the joystick added with the implementation problem of not being able to use the multitouch disabling the use of the joystick and the IMU simultaneously made it near to impossible to fly the LinkQuad with thrust control because the pilot needs to be able to use both at the same time A normal case is when the LinkQuad tilts the previous vertical thrust is split into two effective components one vertical and one horizontal This means that the vertical thrust is reduced and that the pilot needs to counteract this reduction as he is setting setpoints for the attitude control This counteraction is handled by the altitude control Figure 3 6 The state machine diagram of the Android Client By doing an action on the steering interface the user changes the current state in the state machine Figure 3 6 If this basic functionality had not been implemented the code complexity of programming the behavior of the Android Client would be massive So instead of having this massive untraceable code a state machine was implemented in the BackgroundThread This to be able to change state depending on what action the user had done and by the state ma
60. e action cycle starts by the user having a goal The user will then execute a set of actions needed to perfom the goal Then the user will perform an evaluation that the actions that has been done indeed fulfill the goal A human action has two aspects execution and evaluation The execution aspect involves the doing of something and the evaluation is the comparison of what happened with what was expected to happen the goal The user starts by setting a goal The goal is then translated into an intention to do some action The action is translated into a set of internal commands an action sequence that is executed upon the world After the execution aspect has been performed the evaluation aspect starts It starts with the human that performed the action using his or hers perception on the world This perception must then be interpreted to his or hers expectations Then compared evaluated against the intentions and the goal Figure 3 1 describes the cycle of execution and evaluation of a users actions the so called seven stages Nor02 p 45 p 51 The seven stages can be a valuable design aid becuase it provides a basic checklist of questions to ask to ensure that the gulfs of evaluation and execution are bridged The checklist and the answers can be seen in Table 3 2 Nor02 p 52 p 53 The question were used to design the user interface by ensuring that the user can see what action he or she needs to perfom to reach the their goal This was
61. e that future users do not need to bother about the specific implementations of the network communication For further details see Section 2 4 40 3 3 Design of the Interface Figure 3 5 The class diagram of the Android Client More information about the implementations will follow in the sections below In Figure 3 5 a simple class diagram shows the basic structure of the application The activities act as the interface to the user while the Connection act as the network link to the Server Every calculation and operation that are executed due to interaction with the interface is carried out in the BackgroundThread The following sections will describe where the patterns are used and some of the structure and implementation of the application State The program has a static class State that holds the current properties of the process e g if the LinkQuad is airborne or which IP address the client should connect to MenuActivity The menu view is nothing but that a table representation with three normal buttons Each button except one links to a new activity with a new view ConnectionSettingsActivity The field where the ip adress is inserted is a simple EditText field The problem with a EditText field is that it accepts all types of inputs This has been taken care of by having a parser that checks if the inserted data is a valid ip address if valid it will set the connection address in the State to the given address The second featu
62. e to lie on an external computer However if local sensors could be used it would be possible to have the solution on the Gumstix Since a sensor solution from a present laboration rig is easier to use and it is available it will be the first choice To apply the solutions to the slung load control and communication device altitude control of the LinkQuad is required This could be provided by the global positioning system GPS but only outdoors Due to this and to that the GPS was not fully implemented at the time of the thesis it was disregarded but for future solutions it could be used to provide absolute positioning Measurements of the altitude could still be derived from a pressure sensor 1 4 Context and Purpose The objective of this report is to investigate the use of commercial devices as user interfaces on a quadrotor and to investigate solutions to the problem of slung load 13 Chapter 1 Introduction control The purpose of substituting the RC controller with a commercial smartphone is that they are more easy to grasp and might therefore be easier to use for a novice than an RC controller The purpose of slung load control is to reduce the oscillations of the slung load and its effect on the quadrotor s flight performance thus enabling it to handle heavier loads ELLIIT is a collaboration between multiple universities within common fields 1 e realtime systems and artificial intelligence and the development of t
63. eds to perform five actions mini tions one write action and one in mum one action to be able to steer spect action are needed to perform the LinkQuad the whole task How easily can one tell if the The menu itself has no purpose be The are two states that are important This is easily verified because of the system is in the desired state sides being a hub for navigation to know First if the smartphone feedback log that always print the Therefore the menu only has one is connected to the LinkQuad net values of the controls and what the state that can be seen by checking work Second the different states Android Client does if the menu is shown of the connection settings The con nection settings can be seen in the ip adress field and the toggle button that is used to change the connec tion settings How easily can one determine By clicking on a menu item the user Easiest checked by saving some set This is easily verified because of the mapping from system state to in travels directly to that menu items tings and test to connect feedback log that always print the terpretation corresponding activity values of the controls and what the Android Client does aovfuaquy ays fo usisaq EE Chapter 3 User Control of a Quadrotor Table 3 4 The evaluation performed by asking three students at the Lunds University five simple questions The German exchange student Swedish Master s thesis student 1
64. eface This has been a fun and challenging thesis and we would like to thank Karl Erik Arz n and Patrick Doherty for offering it to us We would also like to give thanks to Piotr Rudol Mariusz Wzorek and Jerker Nordh for being excellent mentors and to Fredrik Heintz for guiding us in the end of the thesis Rolf Braun and Stefan Skoog have been very kind to provide experience rubber bands and support throughout the thesis We would like to thank Tore Hagglund Karl Johan str m and Per Ola Larsson for advice within and figures from their fields of expertise Edward Linderoth Olson should also be recognized for helping us with spelling and grammar checks We would like to thank Anders Robertsson for interesting discussions and for tipping us about the thesis We also like to give thanks to our girlfriends for their unwavering support and belief in us Glossary attitude and heading reference system The attitude and heading reference system of the LinkQuad bill of material A bill of materials is a list of the parts and the quantities of each needed to manufacture an end product unmanned aerial vehicle A powered aerial vehicle that does not carry a human operator Control MCU The microcontroller unit MCU that is responsible for control inputs on the LinkBoard human interface device A device that provides direct interaction with humans inertial measurement unit An electronic device that measures and reports on a craft s
65. emented using the behavioral design pattern Template Method The Template Method pattern allows a superclass Connection set a template of execution and deferring some steps to subclasses to implement Gam96 This enables future users to extend with their specific implementations of network communication For further details see Section 2 4 SerialCommunicator consists of two parts an interface for sending data to the CMCU and a thread which handles all communication with the CMCU This is to enable fast access for the Master classes to send data and to reduce the computational delay for the control loop by removing the responsibility of receiving data The thread of SerialCommunicator is separated from Master with SensorMonitor SensorMonitor is also implemented according to the Mediator pattern and is protected by mutual exclusion Its main purpose is to supply the Master classes with the latest available sensor data SerialCommunicator is thread safe since all shared data is protected in SensorMonitor and the interface to CMCU does not share any data with the thread in SerialCommunicator 22 2 4 Network Communication 2 4 Network Communication The clients need a uniform way to communicate with Server as they are implemented on different platforms Their inputs need to be forwarded to the Server and interpreted without consideration of which client the inputs originate from A protocol was therefore developed for the communication between t
66. ess mode ad hoc wireless essid YourOwnAdhocNetwork address 192 168 0 1 netmask 255 255 255 0 HHEHHHHEHHHHHHEHHHHHHEHHHHHHEHHEHHAHHEHE Wired interfaces auto eth0O iface ethO inet static address 192 168 1 202 netmask 255 255 255 0 network 192 168 1 0 gateway 192 168 1 201 The file may contain a lot of garbage as well but what we want to focus on is the wireless interfaces An entry always start with auto lt interfacename gt and after that follows it settings If your interface has any existing settings you can remove them now and replace with the following settings auto wlanO iface wlanO inet static wireless mode ad hoc wireless essid YourOwnAdhocNetwork 107 Appendix B Manuals address 192 168 0 1 netmask 255 255 255 0 The line starting with iface wlan0O is mandatory and we just enter it as is The next line is where we set the wireless mode to ad hoc instead of the mode where it is monitored by a DHCP Then we come to wireless essid which is the ESSID of the network When your laptop or smartphone lists the available networks it will display this as the name of the network The address is the IP address of this interface and the one you will connect to via SSH or using Computer Client or Android Client Finally netmask is set to 255 255 255 0 to allow an address pool of 255 addresses and this should be entered on the unit you use to connect with too Note Remember to set a static IP e g 192 168 0 5 o
67. f attitude control and an open loop control of the thrust of the rig The thesis problem of controlling a slung load required both a control of the slung load s angles and an altitude control for closing the loop of the thrust A state machine was added to allow different modes of operation such as flying lift off or ground where different settings are to be used This section will discuss the following subjects in given order e Model of the LinkQuad e Altitude Control e State Machine e Control of the Slung Load Model of the LinkQuad A model of the LinkQuad was needed to simulate its behaviour in and evaluate controllers without the risk of damaging the rig It was to be as simple as possible and still accurate enough to give confidence in the controllers performance The model was to be derived from existing work since the authors lack enough knowledge in system identification and mechanics to be able to derive the model from scratch Etter Martin and Mangharam EMM 11 derived a simple model of a quadrotor s mechanics T X cosy siny 0 sin 0 Y sinw cosy 0 sin 2 4 13 Z 0 0 l cos 0 g na Rotation matrix where is the pitch angle is the roll angle w is the yaw angle m is the mass g is the acceleration due to gravity X Y and Z are the world coordinates and Tor is the thrust from all four propellers These equations are complemented with integrations to get velocities and position an
68. first ten seconds no longer sessions were used Neither of these models handles the different aerodynamic effects that can occur on the real process and this contributes to the difference between the real process and the models Examples of such effects are listed below e Blade flapping e Turbulence e Ground effect When the quadrotor is moving in the horizontal plane the advancing blade of a rotor has a higher velocity relative to the air than the retreating blade Since the lift depends on the velocity of the airflow over the blade a difference between the two blades is induced This difference causes the blade to flap up and down once every revolution making the rotor plane tilt away from the direction of motion This has numerous effects on the dynamics of the quadrotor 1 e stability in attitude HHWT07 Turbulence increases or decreases the effect of the rotors since the rotors lift depends on the airflow velocity This effect has been visible when flying indoors and 62 4 3 Altitude Control the air started to circulate within the room the quadrotor could suddenly lose or gain height This can also bee observed when wind affects the propeller Joh80 Ground effect is a phenomena that occurs when a rotor aircraft is close to the ground i e an altitude of half the length of rotorblade or less The rotors constrain the rotor wake and creates a pillow of air This pillow implies that the rotors can rotate at a slower spee
69. fore be approximated as two traversed simple pendulums that were defined along these axes The final model as a LTI system 0 0 j 0100 0 0 0 0 spy a fa 0000 0 0 0 Off _ by 0001 0 0 0 Off 0000 0 0 0 0 0O 1l Py Py 6 1 ao 9 000 0 1 0 O lg 0 OF i g l i i 0000 F 0 0 Ojja 7 0 B 0000 0 0 0 14ff B 0 0 B 0000 0 0 5 o B o J i where py and py describe the position of the cart p and py are the velocities of the cart and B are the angles of the angle sensors is the length of the load arm and g is the acceleration due to gravity This model is mass independent which is attractive for the later application on the quadrotor since it can then be loaded with a variety of loads and the model would not have to be altered A problem that can occur when using this model on the quadrotor is that the attachment point of the load will tilt when the quadrotor tilts This still needs to be solved by introducing the quadrotor s equations of dynamics in the model see Section 4 3 6 2 Pressure Sensor The low pass filter is a second order Butterworth filter with a cut off frequency at 2 Hz This filter dampens the noise and random spikes while it does not introduce too much phase delay so that the pressure sensor can be used reliably as an input to the PID controller 89 Chapter 6 Conclusions 6 3 Angle sensor The fork and attachment point were finalized during the last week of the thesis and there was
70. he LinkQuad quadrotor is one of its projects This Master s thesis is a part of the ELLIIT collaboration 1 5 Delimitations The thesis considers only two types of smartphones those with Android OS or Apple 10S It was not relevant to do an evaluation of all smartphone operative systems since it was more important that the smartphone fulfilled the requirements of the thesis and that the development could start early The load is considered to be attached or disattached and that no release or attachment mechanism exists This implies that no effects of releasing or attaching the load will be controlled 1 6 Method The thesis started with an investigation of what possibilities and challenges it contained This was done by using a Divide and Conquer approach where the thesis was split into smaller subproblems and then a workflow was created from these subproblems see Figure 1 5 When the thesis had been broken down into workable challenges the development started The thesis was done in a more dynamic way by focusing on the most present and doable problem instead of having different phases designated for 1 e development or testing The planning was therefore done by using a bottom up traversal of the workflow so that all dependencies would be met before a new task was initiated It was also decided that the two authors should focus on their own area of expertise even if that was occasionally disregarded and both cooperated to solve
71. he clients and the control loop in the Server see Section 2 4 for a detailed description of the protocol Messages are packed according to this protocol and then sent to Server via a transmission control protocol TCP or user datagram protocol UDP connection The choice of underlying protocols is also motivated in this section As can be seen in Figure 2 3 the clients handle the communication with TCP UDP directly while Server separates the network responsibility from the control loop with Connection to reduce the computational delay in the control loop Connection is a network handler which is in charge of the network communication of a current connection It interprets all received messages into setpoints or into commands and feed them to the control loop via a ReferenceMonitor All outgoing messages from the control loop are put in a mailbox which Connection pulls from and sends to the other party For details see page 24 Client TCP UDP TCP UDP Figure 2 3 Illustration of active network communication between Computer Client and the control loop in Server MB stands for MailBox and RM for ReferenceMonitor Network Protocol First the design of the internal protocol is described and then the implementations of the TCP and UDP versions will be introduced A summary of the different messages is given in Table 2 3 When a message is to be sent it is packed into a 160 bits large package with the first 32 bits as the ID and
72. he transmitter where Both protocols are implemented in two versions of Connection TCPConnection and UDPConnection Network Handlers Server uses a dedicated network handler Connection to separate the main functionality from the network communication The purpose of this is to reduce the 24 2 5 Serial Communication computational delay of the main functionality 1 e the control loop in Server and increase the stability of the system A second benefit of this is the abstraction level between the network implementation and the main implementation which implies understandability and usability An instance of the network handler is created when a client tries to connect to Server and there is a limit of one active instance to ensure that only one client can steer the LinkQuad simultaneously The network handler executes a simple loop which reads from the connection if data is available and after that it writes to the connection if there is a message to send available The incoming messages are interpreted into setpoints and commands which are forwarded to the ReferenceMonitor which supplies the control loop with setpoints and its current state and the outgoing messages are fetched from MailBox Mailbox A mailbox stores a single message that has been received or that should be sent It can only store one message at a time but it prioritizes what messages that should be stored by looking at the id of the message The greater the id n
73. his is the server program that was made to take care of the outer loop control in the gumstix on the LinkQuad simple pendulum A simplification of a pendulum which consists of a mass moving without friction on the circumference of a circle spherical pendulum A simplification of a pendulum which consists of a mass moving without friction on a sphere Glossary Acronyms AHRS attitude and heading reference system BOM bill of material CMCU Control MCU GPS global positioning system HID human interface device IMU Inertial Measurement Unit LQR Linear Quadratic Regulator LSB least significant bit LTI linear time invariant MCU microcontroller unit MSB most significant bit PID controller proportional integral derivative controller PS3 gamepad PlayStation 3Q gamepad PWM pulse width modulation RC radio control SDL Simple DirectMedia Layer SMCU Sensor MCU TCP transmission control protocol UAV unmanned aerial vehicle UDP user datagram protocol Contents ADSIEAC 2462 ed sehen ee eo heed zene bedsbeetee sees 1 Preface sso eee ee Geee taut taue nape kaaa aeea 2 eT a 464 4 E E E E E E E E 3 ACronyMS oaa 5 1 Introduction 46 6 lt e 4684 SSR tRER REBATE Bw eS 9 1 1 Quadrotors 2u4cuaedeeagwht ee bee bebe eae aens 9 lee The LinkQuad ake we cdo the wee deo ea hee eR ESS 10 1 3 Problem Description gt so saccra s ee ae ee eee 11 1 4 Context and Purpose ooo aaa 0 00 2 eee ee ee 13 L3 Delm
74. igation of the target smartphone platform and then the implementation design and evaluation of the smartphone client Finally the Computer Client will be presented 3 1 Android Client The Android Client is the main application for steering the LinkQuad The advantage of using the Android Client is that it provides an interface that is naturally mapped to the process by using the IMU of the smartphone to steer and a graphical interface that is naturally mapped to cultural standards and the RC controller Nor02 p 23 A disadvantage is that the implementation of the virtual joystick which set the setpoint for the altitude or the thrust does not give the same feeling of control as the PS3 gamepad or the RC controller This is due to the lack of physical feedback Another disadvantage is that Android smartphones usually have different hardware configurations which may lead to bad performance see Section 2 2 The Android Client requires a rooted and properly configured Android smartphone to connect to an ad hoc network It does not support multitouch even if the functionality exists in the code This is purely in case of a better implementation of multitouch being developed for the Android smartphones see Section 2 2 For more information about how to use the Android Client see Appendix B 1 This section will first discuss what kind of users that might use this application and what ramifications this will have for the design and implemen
75. igure 4 11b This controller was therefore disregarded 64 4 3 Altitude Control A I controller G F also introduce two unstable complex poles see Figure 4 11c and was thereby disregarded A PD controller G K 1 sT moves the poles into the asymptotically stable area but it does not improve the performance Their placement can be altered by adjusting K and 74 but since the pressure sensor is noisy an increase in the derivative action would amplify the noise Second the controller does not have any integral action All this factors implied that this controller was disregarded Finally a PID controller G K 1 T sT4 was investigated It implies a closed loop transfer function K Ely Ts 1 K 622 eae Kn 4 24 i S Kin Tas Kms F which implies a possibility to freely place the poles by altering the values of K T and T4 see Figure 4 1 le Therefore this was the controller to be used However for asymptotic stability the poles must be situated in the left half plane and for a transfer function with a third order denominator s ays a2S8 a3 this holds only if all coefficients are positive and 4142 gt a3 Hag09 p 46 For this system these equations imply that K gt JQ Ti gt 0 Ti gt OQ V 8 for the system to be asymptotically stable 4 25 65 Chapter 4 Imaginary Axis Slung Load Control on a Quadrotor Polelzero map of the closed loop
76. ilter can be used where the pressure 93 Chapter 6 Conclusions sensor s value is used together with the accelerometer This could reduce the effect from the natural variations in atmospheric pressure and increase the dampening of spikes from doors and other disturbances For example a Kalman filter with a model of the process which takes the pressure and accelerations as inputs could be used The pressure should then be used as the major input and the accelerometer should be used determine whether to disregard spikes in the pressure signal or not Altitude Control The different modes of the state machine needs to be tested further to the extent that it can be relied upon to work both indoors and outdoors The altitude control consisting of a PID controller could also be extended to an adaptive solution since the characteristics of the process varies with several factors e g wind turbulence and atmospheric pressure The computing power of a Gumstix computer on module board is more than enough to run the current software so it could be interesting to introduce a full Kalman filter and investigate if more accurate values and estimates of the position and attitude of the quadrotor can be achieved than the original estimation Control of a Slung Load on a Quadrotor This problem was not finished within the time of the thesis work but the remaining steps are clearly defined as follows e Integrate the derived model of a slung load w
77. ime of the pressure sensor All these properties affect the stability of the closed loop system and how fast responses that are possible The pressure signal can be noisy even after low pass filtering so the derivative gain must be chosen such that the noise is not amplified and corrupts the control signal This can be done by choosing T4 sufficiently small so the noise does not pass through However this can collide with the condition from 4 25 KmTa gt 1 T where a too small 74 might introduce instability Hag09 This also matches the theory of rule based empirical tuning The sample time of the controller can be chosen using two rules of thumb The first rule of thumb h 0 05 to 0 14 4 36 where is the crossover frequency in radians per seconds of the continous time system gives a Nyquist frequency is roughly 23 to 70 times higher than the crossover frequency This implies a good approximation of the continous system to a discrete system WAAO9 p 45 The crossover frequency for the closed loop system is 7 45rads_ which implies 0 0067 lt h lt 0 0188 12 4 3 Altitude Control The second rule of thumb ensures that the sampling period is so short that the phase lead is not affected and does not affect the derivative action too much It is formulated as 4 37 and the final parameters of the PID controller T4 0 27 and N 8 implies 0 0068 lt h lt 0 0203 hN 0 2 to 0 6 4 37 Tq Arz09
78. implemented in the same way as the reference monitor but holds sensor values instead of state and position changes It has mutual exclusive entry to all functions to ensure thread safety 2 6 Logging All of the developed applications have support of logging The Android Client uses adb logcat for logging and debugging The Computer Client and the Server use custom logging procedures Printer and FileLogger The logging procedures can be configured with the configuration file for the Computer Client and the Server Printer Printer prints the log messages to the screen This enables the user to follow the execution live on the screen FileLogger FileLogger prints the log messages to a file whose name is set in the configuration file 2 7 Testing and Verification This section will give an introduction to how the different functions of the system were tested and their functionalities verified The foundation of the system is the network communication so it was to be verified first The protocol of network communication and the applications that use the protocol had been implemented in different languages so it was chosen to use a 24 Chapter 2 System Overview Computer Client Keyboard Network Communication Android Client LinkQuad LinkQuad Server Serial Communication Sensor Meu i Control MEU Figure 2 5 The different components of the system that was tested black box technique called black box random testing
79. ing Control The existing control previously mentioned as the inner control loops consists of stabilization of attitude and an open loop control of the thrust The nature of the dynamics allows pitch roll yaw and thrust to be controlled individually Each angle is stabilized with a PD controller and the thrust is not controlled but is forwarded directly to the motors through the mixer see Figure 1 3 Each PD controller use the corresponding input from the RC controller as a setpoint and the angle to be controlled and its corresponding angular velocity as measurement values The output is the reduction or increase in thrust that should be put on each rotor which is aligned with the angle Each controller is then connected to a mixer which calculates each motor s control input This is done by adding the thrust input to the output from the PD controllers of yaw and the angle which 1s parallel to the current motor As an example the back motor s input is the sum of the output from the thrust control the output from the yaw control and the negated output from the pitch control The output from the pitch control is negated to have the inverse effect on the back rotor compared to the front rotor Each angle and its angular velocity are derived from an attitude and heading reference system AHRS 68 5 cm 45 8 cm 20 cm a Top View b Front View Figure 1 1 Dimensions of the LinkQuad AB1I1 1 3 Problem Description
80. itatons ss reassess snare TEREFE REER 14 LO Mothod si se sraya ea a a 14 Lay Related WOK se s setaa BOSSES ERR ew eee eee 16 LS Outline OL the report s sss ss oe aoe eee be ee EES 16 2 System Overview 20 0002 eee ee ee ee 17 2L OVCINICOW sa uke hee eee eee eee eR eRe 17 22 VIOUS 2a4aaeea ee oe ee eRe eRe eee 18 22 DOW a gen eee eee eee wR Eee RS Ee ee ewe eee 2i 2 4 Network Communication a oa a oa a a 23 2 5 Serial Communication 244 ka ee be a Se ee 25 20 DOOD 4 e aeae a e eee wee ee ee ee ee ee 27 2 7 Testing and Verification ee een oe oa eee eR SE ae 21 3 User Control of a Quadrotor aaa aa aa 31 3 1 Android Client oa aaa a 31 3 2 Users and Environment 5 a kk deed REESE Re rarr 32 3 3 Design of the Interface wo ee Gao ewe ewe ee ae we Gs 33 3 4 Computer Clent 4aue eee eaen ea SESS eR RE SE aS 43 4 Slung Load Control ona Quadrotor 48 4 1 Gantry Crane Control s4 de bebe ee eK the dawee es 48 A DENSOS a pee ee e ea eee ee e 32 4 3 Altitude Control aoao 0 002 000 0000 08 59 4 4 Control of the Slung Load oaa 78 5 Experiments and Results anaa aaa 79 JL Expres esere eee ween ted eee the Puwe edn 79 D2 RV gine ee eR RES eR Gee ee Eee Ee ee eS 79 6 Conclusions 0 000 eee ee ee 89 Opl Ganty Lane ata and Roe eee Eee EERE EER GASH 89 6 2 Pressure Sensor yee ee ee mY RAS PE SEE GS SHEDS EG 89 6 3 Angle sensor ndndeu do eGdheh babes tw hb
81. ith the dynamics of a quadrotor e Verify that the angle sensors with the fork solution can be used on a quadrotor e Introduce control of the slung load using movement in the horizontal plane as an actuator with either linear quadratic regulator LQR control or two PID controllers e Implement a modified version of the state machine to account for the slung load at lift off and landing A solution for the ideal case with control for absolute positioning can be seen in Figure 6 3 Android Introducing a better implementation of the yaw and altitude steering would increase the application s usability There exists functionality to interpret the yaw of the smartphone but it is very inaccurate in its original state Some smartphones have a magnetometer or a gyroscope and these could be fused with the accelerometer to a more accurate measurement of yaw of the smartphone The implementation of multitouch is also an area that could be reinvestigated if newer versions of Android introduce a proper multitouch functionality which is accurate and stable 94 6 8 Future Work s fee Alcon X a r as ees APS Mixer io im Z Figure 6 3 This is a suggestion of how the slung load control could be solved in the ideal case with absolute positioning system APS The cascaded solution of absolute positioning is patched with the slung load controllers output to be added to the setpoints of the X and Y control This allo
82. itude can be controlled by changing the rotation rate of each rotor and thus the induced thrust of each rotor A quadrotor is said to be hovering when there is no horizontal movement each rotor induces the same thrust and their combined thrust equals the force induced by gravity Each rotor s thrust at hovering is called hover thrust Ty where m is the mass The combined thrust of the rotors will control the acceleration in the Z axis if the quadrotor is horizontal To change the thrust of all rotors the rotation rate of all Chapter I Introduction rotors shall be changed by an equal amount This combined thrust can be seen as the throttle of quadrotor if its attitude is stabilized Attitude control is achieved by controlling each angle pitch roll yaw individually Pitch can be controlled by creating a difference in thrust between the front and back rotor If the front rotor s thrust 1s increased it will raise the front rotor The back rotor will be decreased with the same amount to maintain the total thrust By increasing the back rotor s thrust and reducing the front rotor s the front rotor will be lowered The behaviour for pitch and roll are analogous except that for roll the left and right rotors are used The torque of a rotor is related to its rate of rotation and the torque can be reduced or increased by reducing or increasing the rate of rotation Yaw is affected by the rotation created by the difference in
83. ixed in an upright position Since it is used almost as a throttle it does not have any springs and has instead notches along the axis to keep the stick in the given position The PS3 gamepad does not have this behavior on its left stick and therefore the thrust is implemented as an adding subtracting function on the left stick in the same way as the Android Client For an example to increase thrust the stick is moved upwards and downwards to decrease the thrust Other differences are the lack of control of the camera since it is out of scope of this thesis and the lack of mode control The missing mode control is due to that the interface to the existing software does not allow access to this property 45 Chapter 3 User Control of a Quadrotor Stick Switch Orientation Set point Left stick North South Thrust Left stick East West Yaw Right stick North South Pitch Right stick East West Roll Top left knob CTRL 7 Camera pan Top right slider CTRL 5 Camera tilt Top right toggle switch SW 6 7 Mode operator Table 3 6 The mapping of controls on the RC controller Stick Switch Orientation Set point Left stick North South Add subtract thrust Left stick East West Yaw Right stick North South Pitch Right stick East West Roll Cross Land Lift Circle Emergency land Top left front button L1 Yaw trim left Top right front button R1 Yaw trim right Table 3 7 The mapping of controls on the PS3
84. k communication but during the development of the Android Client a new feature was added This feature was the human interface device HID PlayStation 3 gamepad PS3 gamepad that could be used instead of the keyboard which was used during the testing of the network communication The PS3 gamepad can supply the software with multiple simultaneous inputs with a higher resolution than a smartphone thus enabling the pilot to accurately control several aspects of the quadrotor at the same time Because of this the Computer Client was successfully used to do the first test flight with manual thrust control indoors The Computer Client however lacks a proper graphical interface since the Computer Client was originally designed to be used for debugging and testing Which leaves much to improve for the user handiness The mapping of the controls are listed in Table 6 2 92 6 7 Server Stick Switch Orientation Setpoint Left stick North South Add subtract thrust Left stick East West Yaw Right stick North South Pitch Right stick East West Roll Cross Land Lift Circle Emergency land Top left front button L1 Yaw trim left Top right front button R1 Yaw trim right Table 6 2 The mapping of controls on the PS3 gamepad 6 7 Server The Server where designed to allow communication to the LinkQuad through a wireless connection to enabling steering of the LinkQuad This was done by creating a network protocol that the clie
85. l Johan str m and Tore H gglund This PID controller was extended to handle limited derivative gain integrator windup and bumpless transfer from manual operation to automatic control The derivative action can result in a very large amplification of measurement noise and must therefore limited This was done by the following approximation of the derivative part sT d Tai S mee 1 sT4 N 4 26 This approximation works well at low frequencies while at high frequencies the gain is limited to N Furthermore it is not good to let the derivative act on the setpoint signal The PID controller becomes then U s K Rs _ s R s s ale 4 27 where U s R s and Y s are the Laplace transforms of the control signal u t the setpoint signal r t and the measurement signal y t WAAO9 p 50 The motors of the LinkQuad can easily be saturated since their PWM inputs are limited to the range 0 1000 If the controller saturates the integral action may then integrate up to a very large value Its value may become so large that when the error is later reduced it may take a while until the integral action returns to a normal value This behaviour is called integrator windup and can be counteracted with so called anti windup This can be done by modeling the motors as a saturation and adding an extra feedback loop to the integrator The feedback loop takes the difference between the input to and the outpu
86. l joystick however is implemented in a touch screen that can not provide this physical feedback Therefore the user instead wants to look on the control to make sure that it provides the right value This makes it hard to steer the LinkQuad at the same time The joystick could not be used simultaneously as the IMU was used which resulted in that the pilot could not counteract the loss of vertical thrust when he or she sets the setpoints of the attitude for the LinkQuad This is due to the lack of a well implemented multitouch on the Android smartphone These two reasons that the virtual joystick was hard to use and that multitouch was not available resulted in that the Android Client should not be used with manual thrust control The Android Client should instead be used with altitude control which implies that the quadrotor does not require the simultaneous inputs from the IMU and the joystick The mappings of the controls are listed in Table 6 1 Input Orientation Set point Green button tilt North South Pitch Green button tilt East West Roll Joystick North South Add subtract altitude Joystick East West Yaw Trim Right Yaw trim right Trim Left Yaw trim left Land or Lift Land Lift Terminate 7 Terminate Table 6 1 Mapping of controls on the Android Client 6 6 Computer Client The initial purpose of the Computer Client was for debugging and testing First it was used to verify the functionality of the networ
87. lan NAME wlan0O net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 19 88 31 fa 88 auto rowbreak ATTR dev_id 0x0 ATTR type 1 KERNEL wlan NAME wlani net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 15 c9 28 d0 18 auto rowbreak ATTR dev_id 0x0 ATTR type 1 KERNEL eth NAME etho net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 19 88 20 fa b6 auto rowbreak ATTR dev_id 0x0 ATTR type 1 KERNEL wlan NAME wlan2 In this example file the wlanO interface is missing in iwconfig output and it lists instead wlan1 and wlan2 which are neither any ad hoc networks Assuming that the etc network interfaces file is properly configured for an ad hoc network we will only need to replace the MAC address of wlanO with the MAC address of wlan2 and do a proper reboot Do not forget to remove the entry of wlan2 root overo etc udev rules d cat 70 persistent net rules This file was automatically generated by the lib udev write_net_rules program run by the persistent net generator rules rules file H H H You can modify it as long as you keep each rule on a single line and change only the value of the NAME key net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 19 88 3e 86 d8 auto ro
88. long the Zg axis These acceleration spikes could not be traced in the measurements of the altitude and are assumed to be measurement errors in the accelerometer Since this is not a required part of the system and this behaviour could cause a crash if these unknown accelerations occur over a longer period of time it should not be used until the hardware can be guaranteed to be flawless or that the time interval of these spikes is observed to be constrained to a short period of time An alternative solution would be to introduce a low pass filter to dampen the undesired spikes but this might impair the performance of the feedforward action The controller exists in a state machine to allow different modes of operation such as lifting landing and flying but only the flying mode was properly tested due to poor weather conditions hindered further test sessions The entire solution of the modified control is shown in Figure 6 2 6 5 Android Client The Android Client is the client application for the Android smartphone The Android Client uses its inertial measurement unit IMU to set setpoints to the attitude control of the LinkQuad and it uses a custom implemented virtual joystick to set setpoints for the LinkQuad altitude and yaw angle The interface and behavior was designed to mimic the radio control RC controller This was done by using a natural mapping of the controls so that if the user had used the RC controller before then the controls
89. margin would decrease the speed of the system The saturation limits were originally set to O and 1000 as these were the limits of the PWM inputs to the motors However after a flight were the controller saturated it was observed that it was impossible for the inner loops to maintain the attitude This is due to the fact that the inner loops control pitch and roll by increasing the thrust on one rotor and decreasing the trust on the opposite rotor on the same axis The same behaviour would occur if the motors were saturated in the bottom thus making the quadcopter drop and perhaps turn itself upside down The saturation limits were therefore set to tow 200 and Uypper 900 to avoid constraining the inner loops stabilization of the attitude The final parameters of the controller can be found in Table 4 3 and an evaluation of the final altitude control can be found in Section 5 State Machine The system requires different behaviour in different situations i e the motors should be inactive on the ground A state machine was introduced to handle the different states of operation and the following states were identified Ground Lifting Flying Landing Emergency Landing Stopping and Stop see Figure 4 21 for a sketch of the transitions and the states The original plan was to have different behaviours if a load was attached to the quadcopter or not so two different state machines were planned initially The state 73 Chapter 4 Sl
90. ms contributed to delaying this task so much that it was not executed due to lack of time 78 5 Experiments and Results This chapter presents the experimental results of the experiments that were carried out on the LinkQuad Unfortunately due to a malfunction of the gantry crane data is not available The rig could not be repaired during the course of the thesis The chapter begins with a presentation of the experiments which is followed by a review of the logged experiment 5 1 Experiments During the thesis six experimental sessions were carried out on the integrated system Each session is presented in Table 5 1 where a short summary is given together with the locale and the atmospheric pressure of the day The two environments for flying were the green in front of the building where the department of Automatic Control resides and the ball room of the student union at LTH see Figure A 5 and Figure A 6 in Appendix A Before the experiments begun the LinkQuad exterior had the appearance as in Figure A 1 in Appendix A After a crash in Experiment 2 the exterior and the chassis of the LinkQuad was broken in half and the repaired quadrotor can be seen in Figure A 2 and Figure A 3 in Appendix A Experiment 3 was the only experiment that was logged and will be reviewed in the next section 5 2 Review During Experiment 3 the following properties were studied bumpless transfers between manual and automatic altitude control
91. n logger s tendency to backtrack and overwrite parts of the log The same occured for the control input for this sequence The setpoint is red the raw altitude is grey and the filtered altitude is blue Step change ca 1 5m 36 Raw altitude Filtered altitude Setpoint from client 35 5 4 34 5 Altitude m wo EN 33 5 100 100 5 101 101 5 102 102 5 103 103 5 104 104 5 105 Time s b The response in altitude to a step change of circa 1 5 m The step change is roughly set since it is induced by a user of the Computer Client The setpoint is red the raw altitude is grey and the filtered altitude is blue Figure 5 3 Two upwards step responses The filtering of the altitude is done in retro spect with MATLAB 84 5 2 Review Control signals 660 Control signal with feed forward 640 Raw PID output 620 PWM output a The control signal to the step change of 0 8 m in Figure 5 3a The effect of the feedforward gain is visible throughout the experiment since the quadrotor was oscillating with a very small angle The data for the control with feedforward gain blue in the time interval 80 85 s 83 60 s is corrupt due to the built in logger s tendency to backtrack and overwrite parts of the log The same occured for the raw PID output red in the time interval 80 85 s 81 25 s Control signals 525 Control signal with feed forward 520
92. n the connecting unit When the configuration is done do a proper reboot The Infamous Proper Reboot A proper reboot is done by doing either a halt followed by a power reset or a shutdown with the correct flags for rebooting see below When doing a halt you will lose your SSH connection and can therefore not see when you should power down A good praxis is to wait at least 30 seconds to let the system store the settings properly root overo halt Broadcast message from root pts 0 Mon Apr 11 12 19 36 2011 The system is going down for system halt NOW root overo Connection to 192 168 0 1 closed by remote host Connection to 192 168 0 1 closed A reboot by using shutdown is done by the following command line root overo shutdown r now lt Message to all other processes gt where r is the flag for reboot now is the time for the reboot to occur and these are followed by an optional message to other processes running When doing a reboot with shutdown you will also lose the SSH connection but you will not need to power down the LinkBoard Just reconnect with the SSH connection when possible This is how it looks in action root overo shutdown r now Reboot incoming Broadcast message from root pts 0 Mon Apr 11 12 41 37 2011 Reboot incoming The system is going down for reboot NOW root overo Connection to 192 168 0 1 closed by remote host Connection to 192 168 0 1 closed According to the G
93. nate from varying turbulence and since the slow integral action was not fast enough to suppress this it remained Both step responses have a small overshoot and do not oscillate too much and both this properties are advantageous as the quadrotor flies indoors The control signals in Figure 5 4a and 5 4b were hard to interpret since the thrust to hover varied over time which is considered to be induced by turbulence However the control signal is not very aggressive in any changes as its full range is O 1000 and it is kept within 475 and 540 at both steps The feedforward action might have been aiding the PID control in both step responses and it might be the reason to why the PID output was never changed to any great extent Its effect was visible throughout the step response in Figure 5 4a and becomes visible at the step change in Figure 5 4b As the feedforward action was designed to aid with fast increases of the thrust at tilting and not at reductions a worse response time should occur at the negative step response The step change of 1 m can be seen in Figure 5 5a and the altitude had settled 1 Altitude control was achieved with a The green Unknown bumpless transfer from a manual flight at an altitude of 5 m It was filmed but not logged 2 Altitude control was achieved with a The green Unknown bumpless transfer from a manual flight at an altitude of 2 m and a step change to 5 m was successful However the log was lost d
94. nts In the beginning it was planned that Niklas Hansson would focus on the control problems of the LinkQuad while Mikael Rudner would focus on the programming problems However the dependencies of the subproblems implied that some areas of focus was intermixed The distribution of the work can be seen in Table 1 1 15 Chapter I Introduction 1 7 Related work The interest in quadrotors has expanded over the latest few years since its simple design and control makes it an attractive solution for UAVs and if properly configured as radio controlled toys The research within steering quadrotors with smartphone is not so extensive but there are two significant projects where the first is the commercial AR drone which is steered with either an 1Phone or an Android smartphone It can also be extended by creating a custom client for a computer and connecting a human interface device HID to the computer for inputs An second solution was investigated by Lichtenstern Angermann and Frass at German Aerospace Center DLR where an Android smartphone is used to control a swarm of quadrotors to film outdoor events LAF11 The area of slung loads is host to a wide array of research projects for both large and small single rotor helicopters An example of such a project is the work of Bisgaard Bendtsen and la Cour Harbo whom has investigated modelling of a slung load system on a small scale helicopter BBICHO06 As for control of a slung load
95. nts use to communicate with the Server The Server then controls the LinkQuad by sending data through a serial connection to the Control MCU CMCU The serial connection together with some configuration of the inner control loops enables the Server to send control signals to the attitude and thrust control of the LinkQuad The Server consists of the following parts e Serial communication e Network communication e Control loop in a state machine The serial and network communication was placed in a thread each and separated from the control loop to reduce the computational delay of the control loop The low pass filter was also put in the thread of serial communication to be filtered at a higher sample frequency than the control loop 6 8 Future Work There are many parts that can be improved or added to this project Network Communication In the present version a single connection exists between the client and the Server The stability of the communication could be improved by extending to two connections One connection should then be a transmission control protocol TCP connection for command messages e g terminate and one connection should be a user datagram protocol UDP connection for direction messages This way the system would benefit from the speed of UDP and the stability of TCP Sensors The accuracy of the estimation of altitude can be improved significantly if sensor fusion with a complementary filter or a Kalman f
96. ollers of the manual mode are separated on the Control MCU CMCU whose source code is closed and it is therefore not possible to implement the tracking in this manual input The tracking of the controller in the manual input was disregarded in the available software as well since the majority of manual takeovers are made with a main mode switch One issue that the current controller can not handle fast enough is the loss of vertical thrust when the quadrotor tilts When the quadrotor tilts to move the previously vertical thrust becomes divided into two effective components where one is vertical and one is horizontal see Figure 4 14 It is therefore necessary to do a feedforward part of the tilting action that can compensate for this loss of thrust If the LinkQuad is seen as a plane with a normal 7 the tilt angle can be found as the angle between the normal vector and the gravity acceleration g see Figure 4 15 In the body coordinate system the normal is defined as 7 0 0 1 and gt amp gx By g and by using the geometric interpretation of the scalar product g Figure 4 13 Parallel PID implementation with bumpless transfer by manual tracking As for anti windup the goal for bumpless transfer is to make the integral action track so the same feedback loop can be used with some modification Courtesy of Karl Johan str m and Tore H gglund 68 4 3 Altitude Control Figure 4 14 The upward thrust of the quad
97. otors By using the seven design questions the seven principles natural mapping principles and the evaluation the following design decisions were made Input Orientation Set point Green button tilt North South Pitch Green button tilt East West Roll Joystick North South Add subtract altitude Joystick East West Yaw Trim Right Yaw trim right Trim Left Yaw trim left Land or Lift Land Lift Terminate Terminate Table 3 5 Mapping of controls on the Android Client When designing the mappings of the controls Table 3 5 it was important that it would be an easy task to execute the controls as well to understand what they do This is the reason why the seven principles were used to choose how the mapping should be done Because of the criteria that the controls should be naturally mapped which in general is a very good idea to do anyway the focus was on designing these natural maps by the principle of design for errors The reason of focusing on the designing for errors principle can be seen in the tilting task If the user did not have 37 Chapter 3 User Control of a Quadrotor to press down a button while using the IMU to steer the LinkQuad it could result in disaster because of tilting that 1s not meant for the LinkQuad such as failing to close the application and then putting the smartphone into the pocket By using error design and natural mapping the different tasks could be made as simple as possible
98. ould do if they hit an object and thus it is safer to use in close proximity to humans or delicate equipment These advantages make the quadrotor an excellent air vehicle to be used both indoors and outdoors This is one common reason why quadrotors are used as both radio control RC quadrotor models and as unmanned aerial vehicles UAVs Flight Dynamics This section will give a general introduction to the dynamics of a quadrotor Definitions The first coordinate system is called world coordinates X Y Z where X is horizontal parallel to the equator and positive in a westbound direction Y is horizontal perpendicular to the equator and positive in a northbound direction Z is vertical and positive towards the center of the Earth It can be used to define the absolute position of a quadrotor The second coordinate system is called body coordinates Xg Yg Zg where Xg is parallel to the axis of the front and back rotors and positive towards the front rotor Ypg is parallel to the axis of the left and right rotors and positive towards the right rotor Zg is parallel to the normal of the plane spanned by Xg and Yg and positive in a downwards direction Euler angles of the body axes are 0 y with respect to the world axes and are referred to as pitch roll and yaw in the given order The front and back rotors rotate clockwise and the left and right rotors rotate counter clockwise Dynamics The quadrotor s acceleration and att
99. pad can be adjusted to the user s preference A higher value results in more dampening and a lower value results in a more sensitive control Values 300 600 Default 328 PS3 deadzone Axes on a PS3 gamepad have a range of 32768 to 32767 which implies that even a sneeze can give a reading Therefore a deadzone is needed to remove a lot of glitching Values 800 1500 Default 1000 Server specific settings serial CMCU Path to the serial port to the Control MCU CMCU Values dev ttySX where X is either 0 1 or 2 Default dev ttyS2 serial SMCU Path to the serial port to the Sensor MCU SMCU Values dev ttySX where X is either O 1 or 2 Default dev ttyS2 load_attached You should enable this if a slung load is attached to the LinkQuad and if it should be controlled Values true or false Default Mandatory There is two types of loggers FileLogger and Printer for two different uses The names state the functionality well as FileLogger logs to a file and Printer prints to the screen To use Printer as a logger you type printer case sensitive as a name of the log All other strings are interpreted as the name of the file Filelogger will write to main and network log The initiation part of the program shares a logger with the network communication This is due to they share the same thread as well Values Arbitrary Default printer control loop log This is the log of the control loop Value
100. pressed by the thumb while the rest of the hand holds the smartphone It was chosen to have the button in a green color because the color is used in traffic lighting when the users should start walking This indicates to the user that pressing the button is probably going to start something like in this case the reading of the pitch and roll values from the IMU Since the green button is positioned to the right the joystick is put on the opposite side to allow use with the other thumb The evaluation Table 3 4 shows however that the choice of using a joystick has its disadvantages The seven questions can be used to solve these problems see Table 3 2 All of the three students thought it was hard to get the feeling of the position of the joystick when they used it Since that a virtual joystick does not give any physical feedback of its state as an analog joystick does This concern was shared by the developer and it is stated that the Android Client is a good substitution for the RC controller but should not be used when controlling the thrust power This because controlling the thrust is an delicate operation and without feeling or maybe having control it could result in disaster Altitude control however does not need as much delicacy So it is recommended to use the Android Client with the altitude control The land lift button was placed at the bottom of the screen to decrease the odds that the user would accidentally press the wrong button
101. r Zp is parallel to the normal of the plane spanned by Xg and Yg and positive in an upwards direction Choice of Sensors The LinkQuad was supplied with an existing inactive solution to an absolute positioning control which altitude control is a part of It was supposed to use a global positioning system GPS unit in combination with a pressure sensor to get an absolute position of the quadrotor but this control was however untested and inactive on the thesis version of LinkBoard The first challenge of the altitude control was to find a good measurement of the altitude and there were different ways to solve this problem see Table 4 2 The choice fell upon using the low pass filtered pressure sensor for its simplicity and its ability to be used both indoors and outdoors 63 Chapter 4 Slung Load Control on a Quadrotor Pressure sensor filtered with a low Simplest solution It is however re pass filter lying on the environment see sec tion 4 2 Integration of accelerometers and Drift and bias errors makes it unre gyroscopes liable Sensor fusion of the pressure sen Most accurate solution however sor the accelerometers and gyro complex and time consuming to de scopes in a Kalman filter which ac velop counts for drift and bias 4 GPS combined with the pressure The GPS loses its connection in sensor doors 5 Camera positioning system from a Limited area of operation related thesis Table 4 2
102. r with o 2 Hz a The actual second order Butterworth low pass filtering with cut off frequency e 2 Hz compared to the raw output The raw output is grey and the low pass filtered signal is blue Low pass filtering I Actual filter MATLAB implementation 30 5 4 b The actual second order Butterworth low pass filtering with cut off frequency 2 Hz compared to an ideal implementation of the same filter in MATLAB The actual low pass filtering is blue and the MATLAB filtering is red Figure 5 2 A study of the low pass filter during a manual take off and flight indoors The altitude is defined from the pressure at sea level 1 atm 101 325 kPa and due to a high pressure on the day of the experiment the altitude for the ground became 31 5 m above sea level instead of the actual 80 m The dip below 31 5 m around the time 20 s is due to an increase in pressure from the motors 83 Chapter 5 Experiments and Results Step change ca 0 8 m 36 Raw altitude 35 5 Filtered altitude Setpoint from client 35 34 5 ow D Altitude m Oo Oo z 33 32 5 32 31 5 31 18 19 80 81 82 83 84 85 86 87 88 Time s a The response in altitude to a step change of circa 0 8 m The step change is roughly set since it is induced by a user of the Android Client The data for the data in the time interval 80 85 s 81 25 s is corrupt due to the built i
103. re 4 21 The state machine for the control pressure sensor s accuracy when the ground was to be determined Even if the quadrotor was to land from a flat surface where a reference value for the ground had been taken on lift off it could not be guaranteed that the pressure at the ground is the same at the time of landing This is due to the continous natural variations of the atmospheric pressure see Figure 4 22 The first intuitive solution would be to set the thrust to just less than required to hover and slowly descend to the ground This is sadly not possible due to ground 75 Chapter 4 Slung Load Control on a Quadrotor Table 4 4 Desired behaviour of each state in the different state machines State Manual flight with clients Altitude control without au Altitude control with autonomous tonomous liftoff and landing liftoff and landing Ground The quadrotor should be on the ground The quadrotor should be able to fly The quadrotor should be on the ground with motors turned off with the RC controller and the con with motors turned off troller should track the thrust input from the user Lifting This state was not required so the state This state was not required so the state The quadrotor should lift off to a set machine goes straight to Flying machine goes straight to Flying Dur point of two meters above the ground ing the mandatory loop through the lift When it has stabilized around the set ing the behavio
104. re of this is that it fetches the ip address for the connection to the wifi that the smartphone has if no connection it will return 0 0 0 0 which means no address was found and thus no connection This feature is implemented by using WifiManager that is a Android SDK class SteerActivity The steering view has five buttons The Terminate and Land Lift button are normal buttons that directly links to an onTouch function that is declared in the SteerActivity class The green button is a little different since it is a ImageButton and it implements a onTouchListener that it used to sense if the button is pressed or not that so the user can use the button together with tilting using the IMU of the smartphone The left and right trim buttons work in the same way In the begining of the thesis it was thought of using the yaw of the smartphone to control the yaw of the LinkQuad But because of the badly implemented IMU no good values were received and it was agreed to use a virtual joystick instead The virtual joystick was a little more complex to implement it was implemented using two Image Views and one FrameView One of the ImageViews were the 4 Chapter 3 User Control of a Quadrotor background and the second was the joystick knob These two ImageViews was merged into the Frame View that is used as the boundaries of the joystick By using an onTouchListener the position of the knob ImageView can be captured and used to calculate the altitu
105. rotor is split into two effective components one vertical and one horizontal Po Quadrotor Figure 4 15 The tilt angle between the quadrotor and the horizontal plane is the same as the angle between ni and amp since n is perpendicular to the quadrotor see 4 28 can be approximated according to 4 29 i g n g cosa 4 28 cos UA a ae 4 29 lale lgl Instead of performing a time consuming lookup for the arccos function the value cos can be used advantageously as a feedforward of the tilting If the output from the PID controller is scaled with 1 cos it will be unaffected if 0 and increase if increases This implementation is also independent of advanced trigonometry to define amp from the existing pitch and roll angles 69 Chapter 4 Slung Load Control on a Quadrotor The key for using this approximation is to find the gravity acceleration vector and this can be done if the origin of the coordinate system is placed at the accelerometer The output vector from the accelerometer can then be used as an approximation of g in 4 29 and its performance can be seen in Figure 4 16 ni li all Ii j aah unl le AE z oo ar NAA AAA AM A VA g oF g PE aa E al ne WA WA WA ya MV wn NV H l IER Jae dp EEE i I A SIRE ES SiG lity wl li ly Hl LL Lal WW IN Vf AAN Yai
106. round can receive a videofeed from the analog camera through a framegrabber The SMCU handles all the communication with the sensors For other components to use any sensor data a request query to the SMCU must be sent This request containing identifiers for the data that is requested is sent via a serial bus to the SMCU The SMCU will respond by continously pushing the latest sensor values of the requested data In the same way if another component should set setpoints of the existing attitude control instead of steering with the RC a serial connection has to be maintained and inputs sent through it to the CMCU which handles the inner control loops and the pulse width modulation PWM outputs to the motors Through LinkGS the inputs and the parameters of the inner control loops can be configured to assign the user specified inputs from the serial bus as setpoints instead The Gumstix boards have a WiFi circuit which supports the 802 11 b g protocol and allows a wireless connection from computers or other WiFi supporting units to the LinkQuad The Gumstixs can be used as the components that can request sensor data and send inputs to the motors via the CMCU and they are the hosts for all the 10 1 3 Problem Description software developed for the LinkQuad in this thesis A sketch of the different components and the connections in between them can be seen in Figure 1 2 The dimensions of the LinkQuad AB11 can be seen in Figure 1 1 Exist
107. rrupt value affected the control signal during a short period of time and could thereby never cause a great impact on the process The plots continue in Figure 5 7 Figure 5 6 Corrupt values from the approximation of was observed in retro 87 Chapter 5 Experiments and Results Altitude raw and filtered 34 I I Raw altitude Filtered altitude 33 5 i 33 E 3 32 51 4 32 2 31 5 96 96 2 96 4 96 6 96 8 97 97 2 97 4 97 6 97 8 Time s a The raw and filtered altitude The output from the accelerometer in z 30 T T T T T 20 107 4 2 S a Or oO N o oO oO lt 1 0 l aa 20 F 4 30 li i 96 96 2 96 4 96 6 96 8 97 97 2 97 4 97 6 97 8 98 Time s b The acceleration along the Zp axis Figure 5 7 Plots continued from Figure 5 6 88 98 6 Conclusions The conclusions of the thesis are presented here along with suggestions of possible future work 6 1 Gantry Crane The gantry crane experiment was successful and a simple model of a slung load was derived and tested The slung load was originally thought of as a spherical pendulum but that model could not be used since it introduced a singularity in the downright position The model was approximated instead by using the fact that the two angle sensors of the crane were parallel to one axis of motion of the cart each The slung load could there
108. rst implemented using the multitouch ability of the test phone The multitouch would enable the simultaneous use of the IMU and the virtual joystick but its performance was poor The specific implementation of multitouch mixes up the finger inputs and sometimes even loses track of a finger This was unacceptable since the application could then send wrong control values resulting in a crash Multitouch was therefore 19 Chapter 2 System Overview considered to be inadequate and a singletouch approach was used instead implying that thrust and yaw could not be controlled in parallel with pitch and roll Ad Hoc Network One of the bigger challenges was that the Android smartphone did not support connections to ad hoc network An initial assumption was that it could be avoided by configuring the Gumstix board to act as an DHCP server for the network it created but its operative system did not have support for it The solution to the problem was found by using a rooted Android smartphone and a rebuilt wpa_supplicant file bla10 The guide contained a prebuild on the wpa_supplicant which changes the network filter s behavior to show and accept ad hoc networks An ad hoc network is shown in the same place as other networks but has a prefix of an asterisk Threads After the initial study of smartphones one of the arguments for using Android was the Java programming language which the students has experienced with However the Android SDK
109. s Arbitrary Default cll 104 B 3 Ad Hoc Network Configuration on a Gumstix Board serial com log This is the log of the serial communication Values Arbitrary Default scl B 3 Ad Hoc Network Configuration on a Gumstix Board The configuration of the ad hoc network may occasionally be lost so this manual will help you to reconfigure the setup into a working state again Perhaps you just want to change the name of the network and this guide will show that as well First things first write iwconfig to display the current configuration of the wireless devices It should look something like this root overo iwconfig lo no wireless extensions wlanO IEEE 802 11b g ESSID LU LinkQuad_0_1 Mode Ad Hoc Frequency 2 412 GHz Cell 02 2F AF A1 3E 9F Bit Rate 11 Mb s Tx Power 13 dBm Retry short limit 8 RTS thr 2347 B Fragment thr 2346 B Encryption key off Power Management off Link Quality 0 100 Signal level 94 dBm Noise level 94 dBm Rx invalid nwid 0 Rx invalid crypt 0 Rx invalid frag 0 Tx excessive retries 80 Invalid misc 18625 Missed beacon 0 Some details may vary 1 e ESSID and Cell which are independent details depending on your current settings The Ce11 is the card s MAC address and it will be needed later so please note it down for future reference If it does not look anything like that do not worry That is why you are reading this However if you see any wlan listings please note down the MAC addresses for them as
110. s eas iest solved by testing against the LinkGS This is easily verified because of the feedback log that always print the values of the controls and what the Android Client does 34 SE Table 3 3 Using the seven stages to ask design questions page 2 of 2 Menu Interface Connections Settings Interface Steering Interface How easily can one perform the If a user wants to change the con When the user has navigated from First the right connection settings action nection settings he or she only the menu to the Connection Set need to be set this can be as much needs to touch the menu item Con tings he or she needs to touch the as four different actions or zero nection Settings and he or she will ip address field and fill in the ip and actions if it is already configured navigate to the Connection Settings then choose which protocol he or correctly Then in the menu navi activity Touch screen and under she wants to use This and check gate through the launch button to standable names provide quick and ing that the smartphone is con the steering activity Then the user easy navigation nected Then confirming by touch is free to steer the LinkQuad di ing the ok button is the only things rectly by natural mapped controls he or she needs to do to perform and an automatic altitude control this task Consequently maximum This gives that the user at maximum of four touch actions two touch ac ne
111. slung load could be modeled as a spherical pendulum in the original control with the following equations of motion 48 4 1 Gantry Crane Control Figure 4 1 The gantry crane process Py Cart position Load Control Y Pz Px directions Traverse Figure 4 2 Crane layout and coordinates The cart position i e the pivot point of the crane load can be moved in the px py plane Courtesy of Per Ola Larsson 210 wWceos 8 1 Wsin Ux SiN Y uy cos Y Q gsin 0 awe sin26 ux cos 0 cos YW uy cos 0 sin y 0 4 1 where u t and w t are accelerations in the corresponding rail directions and 0 t and y t are the angles of the load see Figure 4 2 The new control could not use 49 Chapter 4 Slung Load Control on a Quadrotor this model because of a singularity at the downright position This singularity 1s visible when the equations are linearized and written as a linear time invariant LTT system The downright position is chosen as linearization point see 4 2 and through a simple reformulation of Equation 4 1 the model can be expressed as 4 3 4 2 Sco eco Go ec oo Uyo w t a u t sin w t u cos w t 216 1 W t cos 1 4 3 Problem LB08 When 4 3 is linearized with respect to uy and uy and the linearization point a singularity is introduced at the downright position 0 0 This problem implied tha
112. smartphones do not support the same radio links as the RC another method of communication needs to be found The Gumstix board on the LinkBoard has access to a Wifi connection which could be used instead This implies that the communication is required to go through the Gumstixs instead of passing through the existing radio link that the RC use This can be solved by creating a server application on the Gumstix which uses the serial communication to forward inputs and commands The second problem introduces oscillations from the slung load and will impair the 12 1 4 Context and Purpose N soso cS w Figure 1 4 Illustration of the system flight capabilities Therefore control of the slung load must be introduced to reduce the pendulum effect of the load The quadrotor should then be able to lift fly hover and land with a slung load and during the hover be able to reject disturbances see Figure 1 4 for a sketch of the system in action The second task could not be completed fully but the following subproblems were found and completed e Control of a slung load in the horizontal plane on an experimental rig e Investigation of the existing sensors to get an estimate of the altitude e Altitude control The control of the slung load could either lie on an external personal computer or in the server application on the Gumstix board If the angles of the slung load would be derived by computer vision algorithms it would hav
113. t another approximation of the load had to be found and the first attempt was to use two traversed simple pendulums This proved successful so no further research was done The model s derivation follows below The motion of a simple pendulum is described by d mia mg sin 0 4 4 where is the angle m is the mass is the length and g is the acceleration due to gravity see Figure 4 3 By assuming that the amplitude of oscillation is sufficiently small and that sin as an linearization it can be simplified to 4 5 as a new equation of motion BBO5 d wet o79 This LTI system can be expressed on state space form see 4 6 so 3 6 The two simple pendulums had definitions of load angles that differed from the spherical pendulum in the original model However the original angles were derived from two angular sensors values and p see Figure 4 4 The simple pendulums were defined so that they were aligned along each axis in Figure 4 4 and this implies 50 4 1 Gantry Crane Control J Figure 4 3 A simple pendulum pivot point Figure 4 4 Definitions of angles a and B measured by angle sensors Courtesy of Per Ola Larsson that one simple pendulum could use amp as angle in Equation 4 5 and one could use P The modified model can be described as a LTI system Dx 0100 0 0 0 O p 0 0 Ds 0000 0 0 0 Off p 1 0 Dy 0001 0 0 0 o0ojlip 0 0 j 0000 0 0 0 0 0 1 uy P Py
114. t from the saturation as the error e and feeds it through the gain 1 7 see Figure 4 12 The error e is zero when the motor is not saturated and when a saturation occurs the feedback loop tries to reduce e back to zero again The time constant 7 is called the tracking time constant WAAO9 p 52 67 Chapter 4 Slung Load Control on a Quadrotor The anti windup functionality can be used to introduce bumpless transfer from manual operation to automatic control The integral action will track the manual input if the manual input is connected to the anti windup according to Figure 4 13 The error e will have a value only if the output from the PID differs from the manual input thus the feedback loop makes the integral action follow However it will not make the manual input track the PID loop This can be done by applying an integrator to the manual input and connect an equal tracking loop to that integrator AH06 The tracking of automatic input was disregarded in this application since the switch from automatic to manual often occurs when the LinkQuad is about to crash or is to be brought down manually The standard routine for this action is to make a main mode switch via the radio control RC on the LinkQuad from semi automatic to manual mode The manual mode is set up in LinkGS to only contain tested settings and take inputs from the RC controller while the semi automatic mode allows other inputs combined with the manual inputs The contr
115. t order Butterworth filter with cut off frequency at 1 Hz is compared to the chosen filter 1t has a similar performance It was not chosen since it implies more noise at a lower cut off frequency Higher cut off frequency implies that faster control is possible and this is discussed further on page 73 in Section 4 3 The second order continous time Butterworth filter s transfer function in the is 2 w Go S u 4 10 p s s2 2 s w2 l and it needs to be discretized into a digital filter so it can be implemented in Server The Server receives the pressure data from the SMCU at a sample time of 4 ms so this is chosen as the sample time A for the discrete low pass filter The discretization can be done by zero order hold sampling to the pulse transfer function H z 2 Wo bi z b2 S OOO gt s 26 as of G s s z a z a H z 4 11 54 4 2 Sensors where b 1 a 2 y MO W 1 6 a a 227p O eS Ooh aj 2a0B B cos h a amp y sin ah and Wp is given in radian per seconds rad s The pulse transfer function can be written on backward shift form as bitb 1 a z a2z7 _ Big z a z a H z H z z This gives the filter s equation ye kh H z u kh yp kh z 1 a z a277 u kh b 22 yr kh 1 ary kh ary kh 1 byu kh bzu kh 1 yp kh 1 ayy kh avy kh 1 byu kh bzu kh 1 It where y kh is the fil
116. t was compared to the real process altitude The constants from the experiment were found to imply a behaviour in the model which did not coincide with the real behaviour The conversion constant c was therefore altered to c 0 0234 It gave a matching behaviour between the model and the real data and by interpreting the error in later parts of simulation as an error from the double integration a behaviour in the model that imitates the real process was found see Figure 4 9 The small dip below zero after two seconds in the real process altitude is due to the pressure change from starting the propellers Altitude behaviour Real process Model Alftudetn Time s Figure 4 9 Comparison between the real process and the model with c 0 0234 Another model which was developed in another thesis Son11 as a part of the LinkQuad project was also investigated to see if it was more accurate This model was also linearized around the hover equilibrium but it considered more dynamics than the other model The system was written on state space form as 4 16 i t Ax t B 60 4 3 Altitude Control where u is the pulse width modulation PWM input to motor i and the states are 4 17 where is the pitch angle is the roll angle w is the yaw angle m is the mass g is the acceleration due to gravity X Y and Z are the world coordinates and Tor is the thrust from all four propellers These equations are completed with in
117. tation Finally the design of the different interfaces will be discussed 31 Chapter 3 User Control of a Quadrotor 3 2 Users and Environment Flying the LinkQuad is not an easy task for an novice pilot and it would be even harder if the steering interface was poorly designed or implemented Thus it is important to know in which situations the interface will be used which possible users exists and what their needs are for the application are The following assumptions are made from the Department of Automatic Control at LTH The most likely user is a graduate student that uses the work of this thesis for his or hers thesis He or she will probably have a strong knowledge base in embedded systems computers and control but will not have used a quadrotor before It is also possible that an experienced quadrotor pilot will try to use the Android Client when he or she acts as a test pilot The least probable user is an ordinary person with nearly no experience of using the LinkQuad In the Table 3 1 the three different types of users and theirs different sets of knowledge are displayed the graduate the experienced pilot and the ordinary person Type Name Important knowledge and skills Probability Graduate Embedded systems and control High Quadrotor Experienced Pilot Quadrotors and RC Low Ordinary Person None Low Table 3 1 The possible users of the user control of the quadrotor their possible knowledge and the probability of
118. tegrations to get position and angles All other variables are specified in Table 4 1 The state matrix 000400000 0 0 0 0 0000 0000 0 0 0 0 000000000 4 5 1 _i 000000000 0 O FF 0 00000000 gt 0 gt 0 4 18 000000000 B6 B B B 000 0 00000 An 0 0 0 0000 00000 0 A 0 0 0000 00000 0 0 A 0 000000000 0 0 O Am and the input matrix 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 B 4 19 y y Y By2 aTo 0 0 0 0 Bm2 vaT 0 0 0 0 Bm2 aTo 0 0 0 0 B2 aT where d Amlprop _ a 4 20 p al 2aQol l 61 Chapter 4 Slung Load Control on a Quadrotor and 4 21 e Acceleration induced by graviy 982m 987m a thst content Nat measured 1 59 10 7 NG Table 4 1 Variables of the model in equations 4 16 through 4 21 This model differs from the one in equation 4 13 since it considers inertia of the rig and the propellers as well as properties in the motors Also the input is the PWM input which is the same unit as the inputs to the real motors However when the model was simulated with the values given in the original model it behaved different from the real LinkQuad e g the thrust input to hover was much smaller It would thereby be necessary to do some experiments to do a proper system identification of the model s variables The system identification was considered to be time consuming and the first model was used to evaluate controllers Since it could only be said to be reliable during the
119. tem Since the client was originally considered as testing tool the design of the client differs slightly from the rest of the application The other applications Android Client and Server have a dedicated class for network communication The network communication for Computer Client became instead the main functionality WifiClient since its main purpose is to periodically read inputs from the user and send them to the Server WifiClient fetches user inputs from the abstract class UIListener and its implementations KeyStrokeListener and PS3Listener This design pattern is called Template Method which allows a template to specify a skeleton of operation and deferring some steps to subclasses Gam96 The state machine from Android Client was also introduced in PS3Listener see Figure 3 7 It was put in PS3Listener to ensure the same behaviour as the Android Client but not to hinder the simple nature of KeyStrokeListener The protocol specific implementations of the network communication TCPClient and UDPClient were also designed according to the Template Method pattern Gam96 A class diagram of the Computer Client can be seen in Figure 3 8 Keyboard The keyboard interface was developed to test the communication between Server and Computer Client Any type of keyboard which is connected to the computer running Computer Client can be used as a HID The values that are interpreted into messages are static values which are binary The
120. tered signal u kh is the input sample of the signal to be filtered and kh is the time instant t kh when the computer samples the values This is advantageous since the output can be calculated a sample in advance and there will be minimal computational delay when the output is to used 1 the final filter can be seen 4 12 nA y kh 1 1 929y kh 0 9314y kh 1 0 001234u kh 0 001205u kh 1 4 12 With 2 6 2831853 rad s and The Bode plot of the filter can be seen in Figure 4 6 33 Chapter 4 Slung Load Control on a Quadrotor Magnitude dB Phase deg Bode Diagram 0 180 _270 o ee ee ee gt o s o oa baiiia 107 10 10 10 10 Frequency Hz Figure 4 6 Bode plot of the discrete second order Butterworth filter with cut off fre quency 2 Hz The dashed line marks the filter s the cut off frequency and the dotted line the open loop system s cut off frequency 56 Altitude m Altitude m Altitude m Altitude m Raw data o 0 5 Hz 27 27 5 28 28 5 29 29 5 Time s a we 0 5 Hz Raw data o 15 Hz 27 27 5 28 28 5 29 29 5 Raw data o 2 Hz 2 1 1 1 1 L 26 5 27 27 5 28 28 5 29 29 5 Time s e wc 2 Hz 3 i Raw data 254 5Hz 2 L Time s
121. that much 36 3 3 Design of the Interface The Seven Principles were used to break down complex tasks into more simple ones and to design an interface where the user can execute tasks and actions intuitively These principles were used to decide what actions that the user should be able to do and how the actions should be mapped to the controls Nor02 p 188 The Seven Principles for transforming difficult tasks into simple ones e Use both knowledge in the world and knowledge in the head Simplify the structure of tasks e Make things visible bridge the Gulfs of Execution and Evaluation e Get the mappings right e Exploit the poser of constraints both natural and artificial e Design for error e When all else fails standardize From the investigation of the users it was found that 1t was important to have a naturally mapped application whose functionalities are reminiscent of those of the RC controller The design and positioning of the controls were therefore taken forth by using natural mapping principles Nor02 p 23 p 27 The questions and principles were complemented by an evaluation that was performed by asking three students at Lund University that had not used the application before five simple questions about the design The questions and the answers can be seen in Table 3 4 All of the three students were graduate students with knowledge in embedded systems and control but with no prior experience with quadr
122. the device will be printed to give us a good feedback over what is happening in the system The button Terminate should only be used when we want to do a emergency shutdown of the engines The Land Lift button should be used to make the LinkQuad to land or lift from the ground The joystick is used to change the altitude and turn the LinkQuad To drive the LinkQuad you need to press down the green button and hold it pressed Then the smartphone will register the pitch and roll of the smartphone and translate it to forward velocity and left strafe velocity Phe maa 13 09 Terminate Trim Right Trim 0 Turn speed 0 Forward speed 0 Side speed 0 Alltitude speed 0 Land or Lift Figure B 3 Steering Activity As last note of using the Android Client is that if you somehow exit the steering view of the application the Android Client sends a close message to the LinkQuad so be sure that the LinkQuad has landed safely before closing the application 102 B 2 Configuration File B 2 Configuration File This manual will give you a complete introduction to the configuration file The settings are listed below with an explanatory text typical values and a default value If a setting 1s missing in the file the default value will be used Some settings are listed with a default value mandatory and these settings are required for the application to start If they are missing the application will terminate and print a read error from the
123. thus making it impossible to adjust the thrust as the pitch and roll is adjusted Another commercial device that was investigated was a PlayStation 3 R gamepad PS3 gamepad whose joysticks were very similar to the RC controller s but its shape was smaller and more ergonomic It also communicated via TCP or UDP over a wireless network connection The purpose of slung load control is to reduce the oscillations of the slung load and its effect on the quadrotors flight performance thus enabling it to handle heavier loads The slung load control consisted of several subproblems model of a slung load altitude control filtering of noisy sensors and the slung load control itself These were all solved but the last one where the slung load control was not implemented on the quadrotor due to lack of time The model of the slung load was done by letting the dynamics of two traversed simple pendulums approximate the slung load s angular movement The altitude control consists of a PID controller extended with anti windup and bumpless transfer which is combined with a feedforward control of the tilt angle The controller acts upon a low pass filtered pressure sensor as a measurement signal and receives its setpoint from one of the commercial devices mentioned above The low pass filter of the pressure sensor is a second order Butterworth filter whose purpose is to reduce noise and the impact of spikes induced by events in the surroundings Pr
124. ti windup tracking manual tracking and a feedforward solution for the tilting 70 4 3 Altitude Control Morr moos Figure 4 18 The modified control The setpoints for the attitude control and the es timated angles from AHRS are still sent from the SCMU to the CMCU The thrust input from the RC receiver has been overridden by the altitude control loop from Fig ure 6 l It receives its setpoint sp from the wireless network and the sensor values from the SMCU The PID output and the feedforward gain is calculated on the Gum stix Board and forwarded to the old thrust controller at the CMCU At the CMCU the controllers outputs are calculated saturated mixed into individual PWM outputs and sent to the motors The integral action given by 4 31 is approximated with a forward approximation resulting in 4 32 However the tracking for anti windup and bumpless transfer needs to be included into the approximation see 4 33 The integral part depends only on present values to calculate the future value It can therefore be updated after the control signal has been put to the process thereby reducing the computational delay K t F J E 4 31 I kh h I kh Kl eCkh 4 32 l Kh h i t kh 4 33 The derivative action can not be discretized with a forward approximation since it might become unstable then Therefore it is instead approximated by backward 71 Chapter 4 Slung Load Control on a Quadrotor appro
125. total torque in between the rotors This torque difference is usually cancelled out by having a pair of rotors rotating clockwise and another rotating counter clockwise By increasing a pair of rotors rotation rate and decreasing the other s a torque difference can be created and be used to control the yaw As before the increase of rotation rate on the clockwise rotating rotor pair requires a decrease on the other pair to maintain the total thrust 1 2 The LinkQuad The LinkQuad is the quadrotor that has been used as a test platform in this thesis It has the capability to act as a UAV but in the normal case it is controlled from a RC that communicates to the LinkQuad through a radio link An application for monitoring and configuration LinkGS is delivered together with the LinkQuad This application can be used to change how the LinkQuad should be controlled and change how the LinkQuad internal control algorithms should behave The LinkQuad contains the high performance circuit board LinkBoard III that 1s built up by two Gumstix computer on module boards two microcontrollers referred to as Sensor MCU SMCU and Control MCU CMCU sensors camera and some other components The LinkBoard contains the following sensors One 3 axis accelerometer three gyrosensors a magnetometer a GPS and a pressure sensor The LinkQuad also has an external analog camera whose video output is not accessible through the LinkQuad However a computer on the g
126. ts an overview over the system and then continues to the platforms of the system and the related development tools Third the server design and its implementation will be introduced Fourth the network communication will be described The fifth section contains the implementation of the serial communication between the different microcontroller units After that the logging is presented and finally the testing and verification of the system will be described 2 1 Overview The system can be distributed on three platforms an arbitrary computer with support for wireless network and Bluetooth a Android smartphone with root access rooted which is configured to support ad hoc networks and the LinkQuad itself The computer and the smartphone make up the client side of the system where either enables the user to steer the LinkQuad The LinkQuad hosts a server for a client on a Gumstix board and a client can connect to it via the ad hoc network The user will need a PlayStation 3 gamepad PS3 gamepad to fly with the application on the computer which is connected via Bluetooth and registered as a joystick on the computer An illustration of the system can be seen in Figure 2 1 From here on the clients will be referred to as Computer Client and Android Client and the server as Server N rrenan Figure 2 1 Illustration of the system 17 Chapter 2 System Overview 2 2 Platforms Each part of the system e g clients for user
127. tten to the logger and studied Bur03 p 72 p 101 As for the computer client s functionality the focus was put upon the human interface devices HIDs The test set was first executed by using the keyboard to 28 2 7 Testing and Verification Network 1 Direction with invalid pitch value Network 2 Direction with invalid roll value Network 3 Direction with invalid altitude value Network 4 Direction with invalid yaw value Network 5 Direction with normal payload Network 6 Lift Network 7 Terminate Network 8 Close Network 9 Trim Right Network 10 Trim Left Network 11 Ping Network 12 Incorrect Negative Message ID Network 13 Incorrect Positive Message ID Table 2 5 The test set of the network communication Input is messages sent from the client computer Expected output is the message that is sent back from the server create messages and then repeated for the PS3 gamepad The test set can be seen in the Table 2 6 Computer 1 Direction No output No lift off yet Computer 2 Lift Direction Lift off Flight Computer 3 Lift Direction Land Lift off Flight Landing Computer 4 Terminate Termination message Computer 5 Close Closing message Computer 6 Lift Direction Terminate Lift off Flight Emergency landing Computer 7 Lift Direction Close Lift off Flight Emergency landing Computer 8 Trim Right Trim Computer 9 Trim Left Trim Computer 10 Lift Direction Trim Left Lift
128. umber the higher priority e g if a Ping Message id 0 is stored in the mailbox a Land Lift Message id 4 will overwrite it However if the new message is of the same type as the stored message the new message will always overwrite the stored one This is to ensure that the latest data 1s used at all times Reference Monitor The reference monitor stores the latest set points and the current state that the control loop should use An example of such setpoints are the desired attitude of the LinkQuad 2 5 Serial Communication The LinkQuad has an existing interface for serial communication to and from the SMCU and the CMCU which enables applications on the Gumstix boards to access sensor data and send data to be used in the existing control loops The interface supplies the user with both an existing protocol and functions to import and export data to the protocol This section will therefore only introduce the reader to this high level protocol and its uses but will omit how it was implemented on the serial bus From an example given by the developers of the LinkBoard a simple serial communication layer was implemented which is used to send data to the CMCU The same example was also used to implement a listener to the serial communication and thereby separating the control loop from the serial communication with a monitor Both these parts exist in SerialCommunicator see Figure 2 4 SerialCommunicator is both a thread which h
129. umstix user community the reboot of Gumstix might hang itself due to an issue with an audio driver If you experience this you might try to switch snd soc gumstix to snd pxa2xx ac97 Note This is an untested procedure and you do it on your own risk We take no responsibility for any effects or damage to hardware performance or software by this procedure 108 B 4 BOM Bill of Materials B 4 BOM Bill of Materials These section contains the bill of materials BOMs of the applications that have been developed during the Master Thesis The BOMs describes how to configure the environment to continue development on the applications Makefiles for Unix and Mac OS development exists for all the applications Android Client Windows or Unix Android smartphone with Android version 2 2 or newer and with a inertial measurement unit Mini USB between the smartphone and the developer computer Android SDK Development platform with Android SDK capabilities Netbeans or Eclipse Computer Client Windows Unix or Mac OS X Boost C libs and include files version 1 44 only for Windows Posix C libs and include files only for Mac OS X and Unix HIDs as PS3 gamepad and keyboard Wifi supporting connections to ad hoc network Development platform gcc g Eclipse Visual Studio Server Fully equipped LinkQuad Gumstix with Unix installation Gumstix per default uses ngstr m Posix C libs and include files HID such
130. ung Load Control on a Quadrotor Bode Diagram Magnitude dB Phase deg 10 10 10 10 10 Frequency Hz Figure 4 19 Bode plot of the discrete second order Butterworth filter with cut off frequency 2 Hz The dashed line marks the filter s the cut off frequency and the dotted line the open loop system s cut off frequency Parameter Value 2 10 Value 10 027 10 00r 200 20 10 Table 4 3 The final parameters of the controller machine for handling the load was however not implemented see Control of a Slung Load Instead the state machine for unloaded behaviour has three different implementations One for manual flight with either Android Client or Computer Client one for doing takeovers from manual flights with the RC controller and one with autonomous lift off and flying capabilities All three versions rely on the user to control the quadcopter in the horizontal plane Their desired behaviours are listed in Table 4 4 The two first versions were successfully implemented but the autonomous altitude control s Landing state was not implemented This was due to the lack of faith in the 74 4 3 Altitude Control Bode Diagram of the PID controller Magnitude dB oo D gt o1 oO o1 oO oO ou 45 Phase deg oO 45 90 10 10 10 10 10 10 Frequency Hz Figure 4 20 Bode plot of the PID controller with K 20 T 10 Ty 0 27 and N 10 Figu
131. ur is the same as for point it should hand over the responsi ground bility of setting the set point to the user by going to Flying Flying The quadrotor should take the set The quadrotor should take over the al The quadrotor should take the set point points sent from the clients and for titude control and it should take the to the altitude control from the user via ward them to the inner loops on the set point from either of the clients It Android Client or Computer Client and LinkBoard should also be able to fly manually should be able to fly manually in the in the horizontal plane with set points horizontal plane with set points from from the client the client Landing This state was not required so it was This state was not required so it was The quadrotor should land the quadro disabled to avoid crashes disabled to avoid crashes tor and go to Ground on a successful landing Emergency Landing This state was not needed so it was dis This state was not required so it was This state is the same as Landing ex abled to avoid crashes disabled to avoid crashes cept it goes straight to Stopping after a successful landing Stopping This state was not required so it was This state was not required so it was All motors should be turned off and all disabled to avoid crashes disabled to avoid crashes values zeroed out thereby disabling the quadrotor Stop This state was not required so it was This state
132. uring a crash 3 Altitude control was achieved twice The ball room 100 9275 kPa with bumpless transfers from a man ual flight at altitudes of 1 5 m and 2 m Two step changes of 0 8 m and 1 5 m upwards and one step change of 1 m downwards were successful Both clients were used to set the setpoint for the altitude control and send com mands to the state machine Table 5 1 The experiments that have been carried out on the integrated system throughout the thesis The two locales for flying were the green in front of the build ing where the department of Automatic Control resides and the ball room of the student union at LTH 80 5 2 Review within 8 seconds with a stationary error with a mean of 0 1 m The proportional part of the controller can be seen in the edge at the step change at time 92 6 s in Figure 5 5b The initial slow response and the overshoot was probably due to the integral action This behaviour is desirable if the pressure sensor would react to any pressure spikes e g from the slamming of doors In Figure 5 5b two saturations and aggressive behaviour caused by the feedforward action can be observed This will be further discussed in the next section Feedforward Action l cos had not affected the flight in any visible way to the observers see Figure 5 6a and Figure 5 7a However it saturated the control signal to both its maximum and minimum values during a short period of time and if this period of
133. utput if manual tracking is live Altitude 0 1 2 3 4 5 6 7 Table 2 4 Received parameters and their contents ordered after index The parame ters are sent from the Server to the Control MCU through a serial connection 26 2 6 Logging Serial Protocol to and from Sensor Microcontroller Unit The SMCU can supply the other microcontroller units MCUs with data on the serial bus To gain access of this data a request must is sent of what sensor values that are wanted and how often these values should be sent The request is an array which contains the identification numbers of the sensor values that are wanted The identification numbers can be found in AB11 The SMCU can send at a maximum frequency of 500 Hz and at present time the Server uses a frequency of 250 Hz to run filters on the data at a high frequency There is no need to push the SMCU after the request the sending of data starts at once The data is unpacked into a struct and there easily accessed for further use Serial Listener The serial listener handles the responsibility of the high frequency communication from the SMCU applying a low pass filter to the noisy pressure signal and supplying the sensor monitor with the latest data The low pass filter is put here to decrease the computational delay of the control and to allow it to have a higher sample frequency than the control loop for more details see Section 4 2 SensorMonitor The SensorMonitor is
134. utput is written to the server log After each component or group of components had been tested an integration test was designed to ensure reliability when combining the different components This was done by first executing all the the different tests again but on the whole system and complementing by doing several field tests Chapter 5 30 3 User Control of a Quadrotor This chapter will discuss the investigation and implementation of the clients which allow the user to control the LinkQuad The user control problem originates from the need of controlling the LinkQuad without radio control RC Smartphones are not supplied with the same radio link capabilities as the RC so other channels of communication were required This was solved by having a server on the Gumstix which was connected to the Control MCU CMCU through a serial connection and to the clients through a wireless connection A client on a computer was first developed to be able to debug the network communication and the server with a keyboard A second client on a Android smartphone was developed to steer the LinkQuad The Android Client uses its inertial measurement unit IMU and a virtual joystick to control the attitude and the altitude of the LinkQuad During the development of the Android Client another human interface device HID was integrated into the Computer Client namely a PlayStation 3 gamepad PS3 gamepad This chapter will first present the invest
135. wants to perform The menu interface that was designed can be seen in Figure 3 2 Regarding the design of the menu interface the Connection Settings interface has one task with two important settings that it needs to store the ip address and the communication protocol The easiest way to break down this task is to have one field with an artificial constraint of only accepting valid ip addresses to store The other setting is what network protocol that should be used This is solved by using a toggle button to toggle between the TCP and UDP options having TCP as default because it is the safest protocol This and by adding utility labels that describe the button and the ip address field makes it easy to understand what the task is and where to do it Another feature is 38 3 3 Design of the Interface Ell 16 45 Ip Address Local adress 0 0 0 0 Using TCP protocol Figure 3 3 The Connection Settings interface This view takes care of the connection settings that the Connection Settings interface should give the user feedback if the smartphone is connected to the network and what local address it obtained Having no local address is the same as the smartphone not being connected to the LinkQuad This so that the gulf of execution and evaluation are bridged The answers in Tables 3 2 and 3 4 shows that this was indeed the case By storing the data between sessions it is also ensured that this task is not always necessary The
136. wbreak ATTR dev_id 0x0 ATTR type 1 KERNEL wlan NAME wlan0 net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 19 88 20 fa b6 auto rowbreak ATTR dev_id 0x0 ATTR type 1 KERNEL wlan NAME wlani net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 15 c9 28 d0 18 auto rowbreak ATTR dev_id 0x0 ATTR type 1 KERNEL eth NAME etho 106 B 3 Ad Hoc Network Configuration on a Gumstix Board If your chosen interface is not configured into an ad hoc network yet please continue to the next section before you reboot Verify that the interfaces are configured correctly with iwconfig after the reboot is done Configuration of an ad hoc network The setup of an ad hoc network is pretty straight forward once you have gotten the naming of the interfaces working All the settings are set in etc network interfaces and it is here you can change the name of the network as well In this example wlanO will be configured to an ad hoc network and to use wlan1 just switch wlanO with wlan First let us take a look on the file root overo cat etc network interfaces etc network interfaces configuration file for ifup 8 ifdown 8 The loopback interface Ht auto lo iface lo inet loopback Wireless interfaces HHH WORKING AD HOC 4 HHHHEHHH auto wlanO iface wlanO inet static wirel
137. well It might be one of those you want to use The settings for the wireless network interfaces are specified in the file etc network interfaces and the naming of this interfaces are made on start up from the file etc udev rules d 70 persistent net rules Most problems originate from differences in naming between these files and almost all settings are set in them as well so the manual will focus on these files Existing network interfaces without ad hoc configuration The Gumstix have two network interfaces usually numbered wlanO and wlan1 so if you have some other interface i e wlan3 it is probably one of these but it was auto defined in 70 persistent net rules The interface is auto defined when its MAC address is not found in the file This can happen if you edit this file and enter a non existing MAC address as wlanO or wlanl You can solve this in two simple steps 70 persistent net rules looks like this root overo etc udev rules d cat 70 persistent net rules 105 Appendix B Manuals This file was automatically generated by the lib udev write_net_rules program run by the persistent net generator rules rules file H You can modify it as long as you keep each rule on a single line and change only the value of the NAME key net device SUBSYSTEM net ACTION add DRIVERS auto rowbreak ATTR address 00 19 88 32 86 FE auto rowbreak ATTR dev_id 0x0 ATTR type 1 KERNEL w
138. where the development license costs 99 per year The programming language for Iphone 10S is Objective C which is a modified version of C and C When an application has been developed it has to be added to Apple s App Store to able distribution to more Iphones then the one it was developed on This is required since Apple does not tolerate unmonitored third party applications and requires all applications are reviewed by them This related to public distribution and details regarding any other options that might be available was not investigated App11 Since Android smartphones were cheaper and that neither of the thesis students had any experience with Objective C it was quickly decided that Android was the preferred choice After the development had started it was noticed that the initial study missed some important parts 1 e multitouch A second study took place and is presented in Table 2 2 18 2 2 Platforms Android Java Open Source Accustomedness Multitouch Ad Hoc Objective C App Store Development License gt i e ra aa Reliability SS i Different Platforms Table 2 2 Secondary Study of pros and cons Plus sign denotes a positive feature Minus sign denotes a negative feature Iphone has a more accurate multitouch than Android smartphones which has less accurate multitouch This was noticed during the development of the joystick for the Android application A second
139. ws the slung load controller to position the quadrotor to reduce the angles of the load a and B 95 7 Bibliography AB11 AH06 App11 Arz09 BB05 BBICH06 BD11 bla10 Bur03 Dyk03 EMMI11 Gam96 Gool1 Hag09 HHWT07 Joh80 LAF11 LB08 Meil 1 Nor02 96 UAS Technologies Sweden AB LinkBoard III User Manual Preliminary March 2011 Karl Johan Astr m and Tore H gglund Advanced PID Control ISA Instrumentation Systems and Automation Society 2006 Apple Iphone developer homepage http developer apple com devcenter ios index action May 2011 Date of visit 2011 05 16 Karl Erik Arz n Real Time Control Systems Department of Automatic Control Lund University 2009 Gregory L Baker and James A Blackburn The Pendulum a case study in physics Oxford University Press 2005 Morten Bisgaard Jan Dimon Bendtsen and Anders la Cour Harbo Modelling of generic slung load system 2006 David Abrahams Beman Dawes Boost c libraries webpage http www boost org May 2011 Date of visit 2011 05 27 blackplatypus Android ad hoc support hack wpa_supplicant http forum xda developers com showthread php t 754961 August 2010 Date of visit 2011 05 15 Ilene Burnstein Practical Software Testing Springer 2003 Phil Dykstra Protocol overhead http sd wareonearth com phil net overhead April 2003 Date of visit 2011 05 30 William Etter Paul
140. ximation since the result is always mapped to be stable KT4N Ty Nh D kh 4 D kh h Ty Nh y kh y kh h 4 34 WAA09 The control signal becomes v kh P kh I kh D kh for the PID controller and when the model for the motors saturation and the feedforward signal which are both static are added the final control signal becomes 4 35 ioe sat v Kh 4 35 Tuning The major part of the tuning was done by rule based empirical tuning on the model described in 4 13 The rules are e Increasing the proportional gain decreases stability e Error decays more rapidly if integration time is decreased e Decreasing integration time decreases stability e Increasing derivative time improves stability AH06 The methods to study the performance and stability has been simulating step responses and bumpless transfers in Simulink and plotting the Bode plots of the closed and open loop systems and the transfer function for disturbances However after two crashes with parameters designed for the simulated model the entire tuning was focused on Bode plots and by studying the real performance of the quadrotor Besides rule based empirical tuning there are five other aspects to consider while tuning the cut off frequency of the open loop system the cut off frequency of the low pass filter the noisy signal s effect on the derivative part the sample time of the controller and the sample t

Open Access - Lund University Publications

Contents

Download Pdf Manuals

Related Search

Related Contents