Drone Project

Physics 118 - Final Report

Krish Kabra,
Stefan Orosco

Physics & Astronomy Department

University of California, Los Angeles
June 2018
Contents
1 Introduction 1
1.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Design 2
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Battery Control Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Motors & Electronic Speed Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Transmitter & Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5 Internal Measurement Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.6 Flight Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Parts List 10

4 Timeline 11

5 Project Evaluation Metric 11

6 Conclusion 12

7 Acknowledgements 12

1 Introduction

1.1 Objective
The project objective was to build a quadcopter drone that uses an Arduino as its main flight controller. This
build was largely inspired by electronic hobbyist Joop Brokking, whose YouTube videos we often referred to
[1].
We set both primary and secondary objectives, with the secondary objectives existing simply to serve as
an interesting bonus outgrowth of our original project plan. Our primary objectives were as follows:

1. Build a remote-controlled quadcopter

2. Stabilize the flight of the quadcopter using a feedback control loop

Although the primary objectives appear concise, there are many intricate aspects to the building such
an “Auto-leveling” quadcopter. Firstly, the remote-control aspect of the quadcopter requires seamless, lag-
free and stable communication between the transmitter and quadcopter motors for the quadcopter to fly
smoothly. Secondly, due to the inherent unstable nature of quadcopter flight, there is a necessity for some
stabilization mechanism in-built to the quadcopter flight controller. For this, we plan on using a proportional-
integral-derivative (PID) feedback control that uses input for an onboard internal measurement unit (IMU)
to communicate with the motors in such a way the quadcopter flight is less unstable, if not completely stable.

1
 If these primary objectives were completed in time, we had modular secondary objectives as add-on features
for the quadcopter, which included:

1. Sonar-based collision avoidance system

2. Voice control system

These objectives are currently based upon ongoing developments in industry standard quadcopters, and
were to be bonus features we could add onto our main quadcopter build to enhance the barebone drone.

1.2 Applications
Autonomous drones technology has an extremely promising future in the evolving transportation, retail and
entertainment markets. The rapid development of deep learning and the relative affordability of microcon-
trollers and other single board computing technology has made the development of the consumer facing
drone industry economically feasible. Hardware companies like DJI have been steadily growing their own
consumer/hobbyist product line. However, more importantly is the growth of the business facing drone in-
dustry for both hardware and software development. The promising advancements in neural net processing
and open source AI projects has pushed the boundaries on the concept of possible. For example, companies
like Skydio R1 are pushing forward drones with powerful facial recognition software for video tracking and
photography. While facial recognition and autonomy is a game changer to the entertainment industry with
smart cinema cameras and selfie drones, the true utility for this technology lies in the agriculture and ship-
ping industries. Autonomous drones present an extremely convenient method for mapping and irrigation for
large crop fields. Any sizeable farm will employ more than a dozen people for the sole purpose of watering
and mapping the crop fields. This work could be given to a drone (or fleet of drones) and save time/energy.
The commercial shipping industry is also investing in drone technology for quicker package delivery (Amazon
being the prominent company). Large tech conglomerates have introduced large capital investments , for
example Intel recently invested 60 million dollars into Yuneec an electrical aviation company specializing in
drones.

2 Design

2.1 Overview
The overall project can be split up into 5 main components: developing a battery control circuit that powers
our entire quadcopter, controlling the motors and electronic speed controllers (ESCs), retrieving pulse width
signals from a radio frequency transmitter and receiver, obtaining data from an internal measurement unit
(IMU), and, finally, building a flight controller on an MCU, specifically an ATmega328p (which we work
using an Arduino Uno development board).
The design of the primary objective part of the project is very modular. The secondary objective features
can easily be added to this ‘barebone’ quadcopter; for example, the voice control kit or sonar-based collision
system can be set up with an external MCU, such as a Raspberry Pi or another Arduino, and be synced
with the main Arduino flight controller via the I2C communication bus (corresponding to the SDA and SCL
pins), which the MPU-6050 is currently connected to.

2
 Figure 1: Full schematic of the electronics for the quadcopter.

2.2 Battery Control Circuit

Figure 2: Schematic of battery control circuit

The battery control circuit is the heart of the quadcopter. You can reference figure 2 with the Vin, A0
and A2 pins representing Arduino connections pins. The four resistors used are identical precision resistors
which are all within two ohms of each other. This was done to provide the relative voltage outputs to the
Arduino’s analog input pins, A0/A2. The battery circuit code is quite simple. The raw voltage output of
the LiPo battery is roughly 13.1V (much higher than its rated 11.2V). This is too high for the on board
voltage regulator on the Arduino UNO (the voltage regulator on the Arduino can handle 12.0V max). To
accommodate for this we use the 3 pin MOLEX header that is used to charge each cell of the battery to draw
only 7.4V from 2 cells of the 3 cell battery pack. The voltage divider is read the Arduino using analog pins

3
A0/A2 and below a given threshold the Arduino will flash an on board LED to indicate that the battery
is low. This same effect could be achieved with just one pair of resistors, however, the second pair was
added as a precaution in case there is a hardware malfunction mid flight. This second pair also allows for a
more accurate measurement from the Arduino, because the values read at the analog inputs can be averaged
to obtain a more consistent value. Finally, the diode is connected directly to the battery so as to prevent
backflow current from entering the battery.

2.3 Motors & Electronic Speed Controllers

The electronic speed controllers (ESC) and motors are the muscles and limbs of the quadcopter respectively.
We used LHI’s a 2212 920KV Brushless DC Motor (BLDC) and SimonK’s 30A ESC [2, 3].

Figure 3: Right: ESC - middle top MOSFET chip has a visible short on the bottom three pins.
Left: Brushless Motor - Internal magnet rotated by the induced alternating magnetic field from wire loops.
ESC used to output A-B-C pulses at frequency which matches desired speed[4].

BLDC motors work in a primarily different manner to commutator (brushed) DC motors. Instead of hav-
ing a rotor shaft rotate by having an alternating current passing through a wire loop in between a permanent
magnet, BLDC motors rely on switching the direction of current passing through electromagnets that sur-
round a central permanent magnet which acts as the rotor. For the simple case of having 1 electromagnet,
the current through the coil windings of the electromagnet is only alternated with every 180◦ . Typical
BLDC motors nowadays, however, have 3 or more electromagnets, and so the current is alternated between
the different coil windings (electromagnets) at different phases depending on the number of electromagnets
and desired rotor RPM (refer to Figure 3, left). Due to the lack of friction in them, BLDC motors are more
efficient, longer lasting, and faster in comparison to the older commutator DC motors.
The way we correctly output the necessary multiple phased pulses to the BLDC motor is with an ESC. The
ESC interprets pulse-width-modulation (PWM) signals and converts them into the correct pulse currents to
be sent to the electromagnets within the BLDC. This is done using multiple transistors, which are typically
power MOSFET modules. Therefore, we are able to control the motors of our quadcopter using an Arduino
by sending the calibrated PWM signal to the ESC, which corresponds to the RPM we would like the motor
to spin at.

4
 One issue that we have encountered during our project was a faulty ESC shipped from the manufacturer.
There was a short through one of the attached MOSFETs on the ESC’s circuit board. Unfortunately, when
connecting this to a motor for testing, the motor shorted out and began smoking. We naively believed that
the issue occured with the motor and not the ESC and so we tested the ESC on another motor and the
same issue occurred. At this point we disassembled the ESC and the motors to troubleshoot the issues. The
motors were tested using inductance meters because of the large quantity of copper wire windings associated
with brushless motors. A working motor should read similar inductances around each of the leads, however
if there has been a short then the inductance will vary extremely. After doing this process on each of the
motors, we found that two of the motors had a short circuit. To test the remaining ESCs we checked the
connection between the MOSFET source and drain pins on the circuit board and found that two of our
ESC’s contained shorts (refer to Figure 3, right). We promptly ordered a new set of ESCs and brushless
motors and checked the ESCs and motors before testing.
Combined with knowledge of the specifications of our battery (11.1V, 2800mAh, 35C)[5], we can make an
estimate on the minimum flight time of our quadcopter by assuming it is running at full-speed for the entire
flight:
Battery Capacity × Battery Discharge
T =
Max. Total Amp Draw
2800mAh × 35C
T = = 49 minutes
30A × 4 motors

2.4 Transmitter & Receiver

(a) Multiple pulses of constant (b) Pulse length for 0% throttle (c) Pulse length for 100% throttle
time and frequency being received corresponds to approximately corresponds to approximately
from the radio transmitter. 1000us. 2000us.

Figure 4: Example output from receiver channel 3, which corresponds to the transmitter’s throttle.

The radio transmitter and receiver are the nervous system of the quadcopter. We use a 6-channel 2.4Ghz
radio controller transmitter and receiver from Flysky [6], although we only need to utilize 4 sticks (i.e 4
channels) which correspond to the throttle, pitch, yaw and roll movements. The Flysky FS-T6 controller
communicates with the FS-R6B receiver via radio frequency signals. Although the physics behind this
communication is interesting, we will not choose to explain how the transmitter sends these signals in more
detail.
Instead, what the receiver outputs as we move the sticks on the transmitter is of utmost importance. As
can be seen in figure 4, the receiver channels output a constant frequency of pulses whose width is modulated
according to the stick movements - this is exactly the pulse width modulation needed by the ESCs to control
the motors.

5
 This raises the question of what PWM signal we should send to the ESCs - the PWM generated by the
Arduino, or the PWM generated by the receiver? The correct answer is the PWM from the receiver. This is
for a few reasons: firstly, the Arduino PWM function is very slow, and not very stable. This makes it hard
to control a quadcopter whose motor speeds need to be precisely controlled and be responsive to a change
in user control, otherwise it will crash very easily. Secondly, using the receiver PWM signals allows for a
quicker interaction between the pilot and motors.

Figure 5: Complete interrupt sequence code contained in Arduino flight controller sketch.

In order to store the receiver pulse lengths, we use an interrupt sequence, as seen in figure 5. This interrupt
sequence triggers on a state change. For example, let’s look at how the pulse time will be obtained for channel
1, which is connected to the digital input pin 8 on the Arduino. When the channel goes from low to high,
the interrupt sequence is triggered and we go through our control flow statements (note that for the first
trigger, we initialize the last channel variable to be 0). Since the last state of the channel was a low, we
collect the current time. After the interrupt is triggered again when the state falls from high to low, we go
through the second control flow statement, which collects the current time and subtracts it off the previous
time we had collected when our state had risen from low to high. This lets us calculate how long our pulse
high state was on for in microseconds.
We can now use these pulse times to output the corresponding PWM to the ESCs, as seen in figure 6. In
this code, we simply turn each ESC to a high state for a period of time that corresponds to the pulse time
recorded from the receiver by the interrupt sequence using a loop function that exits when the corresponding
pulse time has elapsed.

6
Figure 6: ESC output function contained in Arduino flight controller sketch. The timer values are
obtained from the pulse signals from the receiver recorded via the interrupt sequence.

2.5 Internal Measurement Unit

Figure 7: Quadcopter axes corresponding to the roll, pitch and yaw movements [7].

The internal measurement unit (IMU) is the eyes and ears of the quadcopter. The IMU is composed of two
key hardware parts the accelerometer and gyroscope. The IMU essentially measures the current orientation
of the quadcopter as angles from the yaw, roll and pitch motions (see figure 7. This angle calculation is
done from the raw data output of the gyroscope and accelerometer readings. The gyroscope measures the
angles as a rate and represents that rate as a raw integer value (depending on the sensitivity setting of
the gyro the max rate in degrees per second is 2000◦ /s. With the full scale gyro range set to ±500◦ /s the
base rate of one degree per second rotation will yield a value of 65.5 for each axis. Thus to scale the raw
gyro output reading to be in degrees per second the raw output is divided by 65.5. This angle reading
is still insufficient for a quadcopter IMU unit however, this is due to the initial noise and offset present
in the first gyroscope readings. To correct for this the axis outputs are averaged over 2000 readings and
these averages are subtracted from the gyroscope measurements. To obtain a traveled angle reading from a
rate measurement, the rates must be summed consecutively (analogous to taking an integral of velocity to
obtain position). The rates are summed at a sample rate of 250Hz (one sample every four milliseconds) and
multiplied by a factor to give a travelled angle measurement. This factor is 1.66 × 10−5 s−1 which accounts
for the gyros full scale measurement (65.5) and the sample rate of 250Hz. The last requirement is that the
axis of the gyro be coupled together. This means that rotation involving more than one axis can be measured

7
by the quadcopter. The equation for the coupled axis are:

Angle Roll Coupled = Angle Roll − Angle Pitch × sin Angle Yaw
Angle Pitch Coupled = Angle Pitch + Angle Roll × sin Angle Yaw

The coupled angle readings for the travelled angle are now usable for a short period of time. Because this
is a real world system there is an drift factor present in the angle measurement from just the gyroscope, thus
for use on a quadcopter the a gyroscope IMU is not useful.
The accelerometer is used to correct for this drift term and provide a starting angle position for the IMU.
The accelerometer works very similarly to the gyroscope, however it measures the acceleration present rather
than the velocity. The measurement comes out as a raw integer value and is converted to a quantity in g’s
(1g being the base force on the surface of the Earth). The vector direction is used to obtain a measured
angle on the pitch and roll axis. The following equations show how the angle is calculated:
q
atotal = a2x + a2y + a2z
ay
anglepitch = α × Angle Roll × arcsin
atotal
ax
anglepitch = −α × Angle Roll × arcsin
atotal
α = 57.3◦

Note that is a correction term dependent on the sensitivity full scale range of the accelerometer, in our
case we utilized a sensitivity range of +/-8g. While the accelerometer angles work in theory the vibrational
acceleration due to the motors provides too much noise for practical use. The final IMU utilizes a combination
of the gyroscope and accelerometer angle calculations.

2.6 Flight Controller

The flight controller is the brain of the quadcopter. We designed out flight controller on the ATmega328p
MCU using an Arduino development board [8]. The purpose of the flight controller is connecting the pilot’s
movements on the transmitter with the desired movement on the quadcopter. Since the quadcopter is
inherently an unstable system due to vibrations from the motors, wind and other factors, it is necessary to
have some feedback control system that regulates the flight control output to the motors with the desired
user input from the transmitter. For this reason, we utilze a proportional-integral-derivative (PID) controller
that is integrated into our flight controller.

Figure 8: Block diagram of PID feedback control within flight controller

8
 The PID loop control mechanism works by considering a proportional, integral and derivative aspect of the
error between the user inputted output set-point and actual output [9]. . Consider the following equation:
Z t
de(t)
u(t) = Kp e(t) + Ki e(τ )dτ + Kd
0 dt

Here, e(t) denotes the error, u(t) denotes the output, and Kp , Ki , Kd denote the proportional, integral
and derivative gain constants respectively. It is clear to see why we would need a proportional gain to reduce
the error and reach the user set-point - the larger the error, the larger the output response to counteract this
error. We can see two problems with this kind of error though: firstly, there is a possibility for the output
response to overshoot the set-point, and cause oscillations about this point. We can solve this problem by
considering a time derivative term that reacts to how the error is changing with time. Secondly, for there to
be any output response from the system there must always be an error, causing the final steady-state output
to be above or below the user set-point. To solve this problem we use an integral component which sums
the error up over time, adding up to fix this steady-state error.

Figure 9: Complete PID calculation function contained within Arduino flight controller sketch

For our quadcopter, we required three PID controllers each for 3 movements: pitch, roll and yaw. The
full code method that implements equation 2.6 for these movements is seen in figure 9. You can also refer to
figure 8 to see a block diagram version of how the PID controller is used in our project. We tuned our gain
parameters manually, through flight and testing. Unfortunately, we do not quite have the best gain settings
as of yet.

Figure 10: Snippet of Arduino flight controller sketch that calculates the necessary ESC output

9
 The last aspect of the flight controller is to communicate these PID outputs to the ESCs. This can be seen
using the code snippet in figure 10. Here, the outputs of the pitch, yaw and roll PID calculations are either
subtracted or added to the throttle value. To determine whether the value should be added or subtracted,
you simply have to consider what the motor response should be for a given movement. To pitch the nose
up, you need the front two motors to be spinning faster than the rear two motors. To roll right, you need
the left two motors spinning faster than the right two motors. Finally, for yawing right, you need the front
right motor and left rear motor spinning faster than the other two motors. For the opposite movement, the
vice versa is true. This completes the primary objective flight controller.

3 Parts List

PARTS VENDOR UNITS COST/UNIT TOTAL

Drone Frame
YoungRC F450 Quadcopter MultiCopter Frame Amazon 1 $18.99 $18.99
Kit
Propellers
Hausbell 2x CW+CCW 9450 Self-tightening Pro- Amazon 2 $9.99 $19.98
pellers
RAYCorp R 1045 10x4.5 Propellers. 8 Pieces(4 Amazon 1 $9.99 $9.99
CW, 4 CCW)
Motors+ESC
4x 920KV Brushless Motor (CW/CCW) + 4x Amazon 2 $63.00 $126.00
30A ESC
Battery
3S 11.1V 2800mAh 35C Li-Po Battery Pack Amazon 1 $20.99 $20.99
Li-PO/Li-Fe Balance Charger Amazon 1 $19.99 $19.99
Internal Measurement Unit (IMU)
GY-521 MPU-6050 6DOF Amazon 1 $5.98 $5.98
Radio Transmitter & Receiver
Flysky FS-T6 6-CH TX Transmitter + Radio Amazon 1 $48.00 $48.00
Control System
Miscellaneous Items
Glarks 635Pcs Connector, Headers & Wire Cable Amazon 1 $13.97 $13.97
Assortment Kit
Secondary Objective Parts
HC-SR04 Ultrasonic Sensor Distance Module Amazon 1 $9.97 $9.97
(5pcs)
Google AIY Voice Kit Amazon 1 $15.94 $15.94
CanaKit Raspberry Pi 3 Complete Starter Kit - Amazon 1 $74.99 $74.99
32 GB Edition
Elegoo For Arduino Nano V3.0 (X3) Amazon 1 $13.86 $13.86
$398.65

Table 1: Finalized parts list for the project. *Please note that we had ordered some DigiKey parts, which
never arrived. We have not included these parts as they were replaced by similar parts from Amazon.

Table 1 shows our finalized parts list for the project. Unfortunately, we had gone slightly above budget.
This is largely due to the fact that we had damaged 2 motors and 2 ESCs during the motor testing stage.
Furthermore, we had preemptively ordered our secondary objective items thinking that our flight controller

10
would work on first attempt smoothly, and that the parts would take about 2 weeks to arrive.

4 Timeline

(a) Initial timeline (b) Final timeline

Figure 11: The original and final timelines

The initially proposed timeline (figure 11a) was definitely an ambitious estimate of the projects progress.
Originally, we had planned to have a week to stress test the quadcopter and a final week to implement the
ultrasonic guidance control system. These bonus features had to be cut for the sake of time. The PID flight
stabilization took longer to realize than expected and due to hardware ordering issues we had to postpone
our first flight until the end of tenth week. This was namely due to the confusion surrounding the propeller
fitting that is connected to the motor. We found that every company sells a different propeller fitting for use
with their proprietary motors, this set us back about a week during which we had finished the PID feedback
but we couldn’t test the flight capabilities. Another hardware mishap occurred in the first batch of ESC’s
that were delivered, in which one of the ESC’s had a mosfet that was shorted. When this ESC was connected
to a motor, it immediately blew the motor. The shorted ESC destroyed two motors in total which pushed
us back by roughly a week in the hardware goals for the project. Finally, the extra two weeks alloted for
stress testing and bonus features were very useful for extending the original flight demonstration timeline.

5 Project Evaluation Metric

(a) Initial project evaluation metric (b) Final project evaluation metric

Figure 12: The two project evaluation metrics are show as pie charts in the figure above.

The original proposed project evaluation metric shown in ,the figure above definitely, had a few flaws.
Firstly, the physical drone build took up too much space on the metric this was corrected in the final metric.
Assembling the drone was much easier than anticipated and should not have constituted nearly half of the

11
evaluation metric. Also the PID flight stabilization was increased in the final metric, as this was the core of
the project. Finally, when considering the difficulty present in implementing the RF control into the flight
controller this category was chosen to be increased.

6 Conclusion
As stated in the timeline section above, we were overly ambitious for this project and should have focused
solely on our primary objectives rather than considering the secondary objectives. Nevertheless, we learned a
lot about programming the ATmega328p specifically rather than relying on Arduino based functions, coding
in C/C++, and understanding how ESC and BLDC motors, I2C communication, PID feedback loop control,
and, obviously, quadcopters work. The final setback of ordering the wrong propellers was tough and cost us
a lot of time and testing.
The final result of our project was a quadcopter that could lift off and turn, but not to a full enough range
where it could be considered flyable. Over the first week of summer, we plan to tweak our quadcopter to
make it ‘flyable’, and perhaps begin adding our secondary objective features.

7 Acknowledgements
We would like to thank Professor Christian Schneider for all his guidance and teaching done for this class as
it was greatly appreciated. Lastly, we would also like to thank Anthony Ransford and Jacob Saret for their
help, support and motivation throughout the building of our project.

References
[1] J.M. Brokking. Project YMFC-AL - the arduino auto-level quadcopter, Apr 2017.
[2] LHI 4x 2212 920KV Brushless Motor (CW / CCW) + 4x SIMONK 30A ESC For DJI Phantom. https:
//www.amazon.com/LHI-920KV-Brushless-SIMONK-Phantom/dp/B00XQYTZQ2/.
[3] Lynxmotion. SimonK ESC User Guide. Revision 1.0.
[4] Brushless DC Motors Control - How it Works (Part 1 of 2). https://www.youtube.com/watch?v=
ZAY5JInyHXY, June 2012.

[5] FLOUREON 3S 11.1V 2800mAh 35C Lipo Battery Pack with XT60 Plug for RC Air-
plane RC Helicopter RC Car RC Truck RC Boat, RC Hobby. https://www.amazon.com/
FLOUREON-2800mAh-Battery-Airplane-Helicopter/dp/B00RXA02QW.
[6] Flysky. FS-T6 Digital proportional radio control system, Instruction Manual, 2012.
[7] Faisal Zaman. Nothing Beats a Clean Signal (especially for Drones/UAVs). http://www.qualtre.com/
2015/10/nothing-beats-a-clean-signal/, October 2015.
[8] Atmel. ATmega328/P Datasheet Complete, November 2016.
[9] John Bechhoefer. Feedback for physicists: A tutorial essay on control. Rev. Mod. Phys., 77:783–836, Aug
2005.

12