paper2012draft.tex

\documentclass[12pt,letterpaper]{article}
\usepackage[margin=1in]{geometry}
\usepackage{fancyhdr}
\usepackage[utf8]{inputenc}
\usepackage{palatino}
\usepackage{microtype}
\usepackage{hyperref}
\usepackage{graphicx}
\usepackage{lastpage}
\usepackage[hang,small]{caption}
\usepackage{titlesec}

\renewcommand{\headrulewidth}{0pt}
\fancyfoot{}
\fancyfoot[C]{\sffamily Page \thepage\ of \pageref{LastPage}} \pagestyle{fancy}

\titleformat{\section}{\bfseries\MakeUppercase}{\arabic{\thesection}}{1em}{}
\titleformat{\subsection}{\bfseries}{\arabic{\thesection}.\arabic{\thesubsection}}{1em}{}
\titleformat{\subsubsection}{\itshape}{\arabic{\thesection}.\arabic{\thesubsection}.\arabic{\thesubsubsection}}{1em}{}

\setlength{\parindent}{0cm}
\setlength{\parskip}{1em}

\captionsetup[figure]{labelfont=it,font=it}
\captionsetup[table]{labelfont={it,sc},font={it,sc}}

\hypersetup{colorlinks,
    linkcolor = black,
    citecolor = black,
    urlcolor  = black}
\urlstyle{same}



\begin{document}

\begin{center}
{
    \bfseries\huge
    Oregon State University \\
    Autonomous Aerial Robotics Team \\ [1em]
}
{
    \small
    Daniel Miller, Team Lead, millerd4@engr.oregonstate.edu, 541-728-8090\\
	Kyle Dillon,  Tim Niedermeyer, Ryan Skeele, Michael Williams, Soo-Hyun Yoo \\ [0.5em]
    \emph{1148 Kelley Enginnering Oregon State University, Corvallis, OR 97330}
}
\end{center}

\begin{center}
\begin{minipage}{5.5in}

\section*{Abstract}

The Oregon State University Autonomous Aerial Robotics Team has developed an
indoor autonomous quadrotor with custom hardware and software to compete in the
International Aerial Robotics Competition (IARC). Onboard, an ATXmega128a3
microcontroller runs a 200 Hz PD orientation controller. The quadrotor is
capable of sending live video, LIDAR scans, and altitude measurements to the
base station which passes navigational commands back to the quadrotor. The
quadrotor possesses a passively compliant robotic hand that will be used to
pick up the USB flash drive in competition.

\end{minipage}
\end{center}


\section*{Introduction}

Our strategy for indoor automous flight involves specific functional goals for
each part of the system. Flight stability is handled by the flight platform,
however, all navigational data gathered from distance sensors and cameras
mounted on the flight platform is transmitted to the base station for
processing. The base station then transmits navigational commands to the flight
platform.

Stabilization sensory is handled using a three axis gyroscope, accellerometer
and magnetometer. Navigational data comes from a number of distance sensors
mounted on servos that act as an improvised LIDAR. This solution was chosen
because of the low electrical and computing power required to process distance
sensor data. A seperate wireless camera is implemented to provide object
recognition using the OpenCV libraries.


\section*{Hardware}

\subsection*{Chassis}

Ultimately, we aim to construct a chassis that is lightweight, durable,
modular, and safe.

For initial tests, however, a simple ``X'' configuration quadrotor was
assembled from laser-cut plywood. The symmetric construction of the ``X''
allowed us to make convenient assumptions about the vehicle's flight
characteristics.

The final frame is constructed in an ``I'' configuration, which simplifies the
task of mounting cameras and distance sensors on the periphery of the
quadrotor. The carbon fiber booms are strong yet lightweight. The booms are
also separable from each other, facilitating repairs and modification. The
motors are mounted to the frame with aluminum clamps. Similar clamps are used
to mount the sensors elsewhere on the booms. Finally, carbon fiber shrouds
prevent any of the four propellers from contacting objects or people.


\subsection*{End Effector}

One of the primary objectives of the competition is to grasp a USB flash drive.
Our grasping mechanism will consist of an under-actuated, passively compliant
four-fingered hand. Each of the four fingers will contain two flexure joints. A
single cable runs through the center of each finger to its tip. These four
cables are actuated by a single linear actuator. The use of only a single
actuator in this under-actuated system helps keep down weight and cost and
allows the hand to automatically adapt to the shape of the object being grasped
without the need for a separate control system.


\subsection*{Electronics}

A single 3-cell lithium polymer battery powers the motors and motor controllers
on the quadrotor. The battery is also regulated down to 5 volts to power an
Atmel Xmega microcontroller, a 2.4 GHz XBee radio, and various distance
and inertial sensors. The wireless camera is powered by a seperate 9 volt
source.

The organization of the electrical system is illustrated in
Figure~\ref{fig:el_blockdia}.

\begin{figure}[h!]
    \centering
    \includegraphics[width=6in]{controlhardware.png}
    \caption{Control Hardware Block Diagram}
    \label{fig:el_blockdia}
\end{figure}


\section*{Software}

\subsection*{Aerial Stabilization and Flight Control}

% \section*{Flight Control} The on board flight control system was developed to
% provide basic stability for the flight platform. It does not have
% contingencies to prevent drift. Integration of gyroscope samples is used to
% create an estimate of the orientation of the flight platform. This combined
% with the raw data from the gyroscopes is as input to a position differential
% (PD) controller. To prevent the gyroscope positional estimates from drifting
% they are averaged with magnetometer and accelerometer samples.

\subsubsection*{Orientation Kinematics}

An accurate measurement of the quadrotor's orientation is key to autonomous
stabilization. The orientation can be represented with a direction cosine
matrix (DCM), which is a 3x3 matrix containing the cosines between each of the
9 possible pairs of axes of two separate Cartesian coordinate systems.

In the context of inertial measurement in robotics, a 3D vector could be
represented in either the global (earth) or the local (body) frames of
reference. For example, the location of an end effector may be represented as
$\langle1, 0, 0\rangle$ in the local frame but have different (and changing) X,
Y, and Z components in the global frame, and vice versa.

If the controller loop frequency is high enough (on the order of 100 Hz), an
axis of the DCM, its rotational vector, and its linear velocity are
approximately orthogonal to each other. Thus, the magnitude of the angular
velocity of a unit vector approximately equals its linear velocity, which means
that the DCM can be calculated by integrating the gyroscope readings.

Unfortunately, since the gyroscope only measures the change in the orientation,
the DCM will drift over time. A 3-axis accelerometer can be used to correct the
roll and pitch drift. Similarly, a 3-axis magnetometer can be used to correct
for yaw drift. With these corrections, an accurate DCM can be maintained.

%In order to achieve this, we wish to keep the calculated global Z vector
%codirectional with the negative gravity vector. That is, the cross product of
%the two vectors should equal zero, and we can use whatever the cross product
%actually is as our correctional rotation vector. It is important to keep in
%mind, however, that since the accelerometer reports acceleration in the body
%frame of reference, we must express Z in the same frame of reference as the
%acceleration vector.

%This body frame Z vector is simply found by taking the third row of the
%transpose of the DCM. We cross multiply Z with the gravity vector, add the
%resulting correction vector to the angular displacement vector, and take a
%weighted average.

%Small imperfections in the mounting of the accelerometer can be offset by a
%rotation matrix.


\subsubsection*{Cascading PID Controllers}

The D controller in a simple angular position PD controller can hinder the
responsiveness of the quadrotor, which is crucial to stability. A better
alternative is an angular position P controller that feeds a desired velocity
to an angular velocity PD controller.

At first thought, it might seem that this combination of a position P
controller and a velocity PD controller is no different than a position PD
controller, since both the position D and the velocity P are based on velocity
measurements. However, the two controllers do different things with the
velocity measurements.

The position D controller has a damping effect on motion in that it always
resists a change in position. This means that if the quadrotor experiences a
perturbation away from some desired position, the D controller will help resist
the motion, as desired. However, as the quadrotor tries to recover from this
error state, the D controller blindly hinders the movement back towards the
desired position, which is not at all optimal.

The velocity P controller, on the other hand, pushes the velocity to whatever
it should be, whether that means slowing it down or speeding it up. The
velocity D controller ensures that the velocity change does not happen too
abruptly, helping reduce the chance of overshoot. The controller is able to do
this by keeping track of a desired velocity in addition to the current
velocity, which provides a context upon which the controller can ``decide''
whether it should help a positional movement or hinder it, instead of blindly
hindering all movement as is the case with a position D controller.

This means that the P/PD controller as a whole can take a desired position
input and accelerate the body towards and maintain a target angular velocity
until the current position nears the target position. Only then will the
controller actively slow down the movement. This makes for a controller that
can respond much more quickly while maintaining stability, making flight
possible (Figure~\ref{fig:flight}).

\begin{figure}[h!]
    \centering
    \includegraphics[width=6in]{posterbackground_cropped.png}
    \caption{Prototype Flight Platform Operating Outdoors}
    \label{fig:flight}
\end{figure}


\subsection*{Navigation}

\subsubsection*{Navigation Hierarchy}

Each component of our system has a specific responsibility with little overlap
with those of others. The onboard microcontroller is solely responsible for
maintaining angular orientation. The microcontroller streams data from distance
sensors through the XBee link to the base station. A video stream is sent to
the base station from a wireless camera. The base station is responsible for
processing this data and send navigational commands back to the quadrotor in
order to complete the mission.


\subsubsection*{Navigation Strategy}

In order to navigate indoor environments we implement an array of simple
distance sensors, with some mounted on pan or tilt mechanisms, to serve as an
improvised LIDAR system. This approach was chosen because of its accessibility
and low power and bandwidth requirements.

Distance sensor data gathered on the flight platform is sent to the base
station. A navigation system at the base station processes both distance sensor
data and images sent from the flight platform to make navigational decisions.
These decisions are then relayed back to the quadrotor as movement commands.


\subsubsection*{Robot Operating System}

Robot Operating System (ROS) is a collection of libraries and tools that allow
developers from around the world to contribute software packages such as device
drivers, messaging libraries, and visualizers in a consistent format for others
to use. We have developed our navigation system to operate within this software
architecture. This allows us to use third-party joystick drivers, point cloud
libraries, and SLAM algorithms, which frees us from having to reimplement
solved problems.

\end{document}