Winter End-of-Term Report.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

457 lines
21 KiB

\documentclass[10pt, draftclsnofoot,onecolumn, compsoc]{IEEEtran}
\def \CapstoneTeamName{Automated Fenceless Grazing}
\def \CapstoneTeamNumber{CS3}
\def \GroupMemberOne{Ryan Alder}
\def \GroupMemberTwo{Danila Fedorin}
\def \GroupMemberThree{Matthew Sessions}
\def \CapstoneProjectName{Automated Fenceless Grazing}
\def \CapstoneSponsorCompany{Oregon State University}
\def \CapstoneSponsorPerson{Bechir Hamdaoui}
\def \DocType{End Of Term Report}
\makebox[2.75in][r]{#1} \hfil \makebox[3.25in]{\makebox[2.25in]{\hrulefill} \hfill \makebox[.75in]{\hrulefill}}
\par\vspace{-12pt} \textit{\tiny\noindent
\makebox[2.75in]{} \hfil \makebox[3.25in]{\makebox[2.25in][r]{Signature} \hfill \makebox[.75in][r]{Date}}}}
% 4. If you have a logo, use this includegraphics command to put it on the coversheet.
\huge CS Capstone \DocType \par
{\large Prepared for}\par
\Huge \CapstoneSponsorCompany\par
{\large Prepared by }\par
% 5. comment out the line below this one if you do not wish to name your team
% \begin{abstract}
% The Fenceless Grazing Collar system aims to reduce the amount of work
% needed by farmers to keep herds of grazing animals. The project
% will be implemented using the LoRa wireless communication protocol to allow
% for long-range interaction between animal-worn collars and a gateway device.
% The gateway device will also provide an HTTP-based JSON API to apply configuration
% changes to collars through an application built for Android mobile devices.
% The MariaDB SQL database management system will be used to store the data
% received from the collar for viewing and analysis.
% \end{abstract}
\section{Project Recap}
The Fenceless Grazing System (FGS) project aims to reduce the need for manual labor
for farmers and ranchers caring for large numbers of grazing animals. The
system does so by automating the common task of herding through the use
of GPS-equipped collars that emit negative auditory or electrical stimuli.
Animals that leave a rancher-defined grazing area are met with increasingly
potent uses of the aforementioned stimuli, intended to discourage
such behavior. In addition to the automated herding capabilities,
the FGS also provides data logging capabilities. The data from the
system can be analyzed both by ranchers, in order to make decisions
about the livestock, as well as researchers seeking to measure
the efficacy of the system.
The Fenceless Grazing System is split into three major components:
\item \emph{GPS Collar Prototype:} This part of the system
is a prototype collar, which will eventually be placed
on a grazing animal. The collar uses a GPS module to
determine its location, and a single-channel LoRa (Long Range)
transmitter to communicate its position and other data
to the \emph{Gateway Server}.
\item \emph{Gateway Server:} This component of the system
is used to manage connections from large numbers of
collars, as well as to store the data transmitted by the
collars. This component is currently implemented
using a Raspberry Pi, a low power, miniature single-board
computer running Raspbian Linux. The gateway server also provides
a JSON API to external clients, allowing them to view
and modify collar settings. The current reference
client is the \emph{Android Application}.
\item \emph{Android Application:} The Android application
uses the JSON API provided by the gateway server to view
the locations of the grazing animals, as well as adjust
the grazing boundaries. To protect the data, the application
uses JSON Web Token authentication.
\section{Current Project State}
During this term, we were able to purchase the majority of the
required hardware, and created a working prototype of the system.
Below are the descriptions of the current state of each
of the project components.
\subsection{GPS Collar}
The collar controller (implemented as an Arduino Nano with an ATmega328p microprocessor)
is currenly able to interface with the on-board GPS module and
determine its own location, within about a foot. This was tested
in a somewhat populated area, meaning that such precision is in
spite of various obstacles such as buildings and trees
that would not commonly be present in a grazing field. The collar
can also use a single-channel LoRa transceiver to communicate its
location to the gateway server.
Currently, the collar is assembled using prototyping materials
--- jumper wires and breadboards -- and is not soldered or placed
into a case. This is to allow for quick replacement of broken
or damaged hardware, which occurs somewhat frequently during
development. Despite its current construction, the GPS
collar is quite study, and we anticipate no further need
for soldering or otherwise hardwiring the components. We
do, however, intend to 3D print or laser cut a case.
The current version of the collar can be seen in Figure \ref{fig:gps_collar}.
To standardize the communication protocol between the collar and the gateway,
the project uses a data encoding/decoding tool named Protobuf (short for Protocol Buffers).
The tool provides a separate language for describing data structures to be encoded,
and has the advantage of tightly packing data, reducing the number of bytes needed
to store the information. This is especially useful for our purposes, since it
minimizes the amount of data to be transmitted over radio, and therefore minimizes
the potential for error as well as increasing scalability.
Currently, the communication protocol fits in two Protobuf data structures. The
code is as follows:
The above snippet declares two data structures: a point, which represents
a pair of longitude and latitude, and a ``collar response'' data structure,
which is sent by a collar every time it is prompted by the server.
In the future, the ``collar response'' will also include information
such as the battery level, whether or not the collar is outside of
the prescribed grazing area, and the collar identifier. It is this
information that will be logged by the gateway server. A curious
fact about this message format is that it deliberately uses 32-bit floating
point numbers, rather than their 64-bit version. This is due to the fact
that the ATmega328p microcontroller used by the collar does not
support 64-bit floating point operations. It is unclear at present whether
or not this is an issue. From initial testing, the 16 digits of precision
provided by 32-bit floating point numbers seem sufficient for our purposes.
\caption{GPS Collar Prototype}
\subsection{Gateway Server}
The gateway server can correctly serve as the intermediary between
GPS collars in the field and the Android application. It does so by providing
two separate services: a REST API, written in Python, and a LoRa client,
written in C.
The LoRa client is used to interface with the Raspberry Pi LoRa shield,
and provides all the communication-related functionality. Currently,
this entails waiting for location broadcasts from the GPS collars,
decoding them into C data structures from Protobuf, and storing
this information into an SQLite database using SQLite's C API.
Because the C Protobuf API is much more low-level than the
Arduino / ATmega328 API, it requires the users (us) to provide
it with implementation details, such as the method for memory
allocation. We use C's standard \texttt{malloc} function
to do so:
The REST API uses the Flask microframework to provide an HTTP-based
interface to the Android application. Currently, it receives
login information from the client, and verifies it against its
database of user accounts. If the account information is correct,
it creates a JSON Web Token which can then be used by the Android
application in further requests.
The JSON Web Token is an encoded piece of data in JSON format.
It can contain any arbitrary information, which is inaccessible
to the end-user (the Android application or other client) without
access to the secret key. Thus, while the REST API server can easily
decode JWTs to determine their contents (such as the user ID who
this token belongs to), this information is inaccessible to
external parties, preventing tampering and unauthorized access.
Currently, the REST API provides two operations to the Android
application: a way to get the list of currently active collars,
and an endpoint to list the recent history of a single collar.
The REST API and the LoRa client share the same SQLite database.
However, whereas the SQLite C API (used by the LoRa client) is
very low-level, the Flask-based API server is capable of using
a high-level interface with the database, called an Object Relational Model (ORM).
Through this interface, the Python code can treat SQL data like Python
objects. This makes it significantly easier to access the modify the data
in the database in a more idiomatic way. However, the ORM interface
is incapable of expressing certain SQL queries. This makes
it necessary for the Flask API to use several SQL queries
to access data, rather than using a single, complex query.
The below code snippet demonstrates our use of the
ORM interface.
In the above snippet, note, in particular, the various uses
of \texttt{query}. This occurs once in \texttt{Collar.query},
which retrieves the list of active collars from the database.
Then, for every active collar, the \texttt{query} is used
again, selecting the most recent data point. This means
that the server will have to make one additional SQL
query to the database for every active collar. This
may not scale well in the future, and would necessitate
the use of raw SQL, without the ORM wrapper.
Additionally, note the use of the \texttt{@jwt\_required}
decorator. Using this decorator, we express that the
endpoint described by the above code (which returns
the list of active collars, together with their
current locations) is only accessible to clients
that have already signed in and were issued a JWT token.
Adding this decorator is sufficient in our code to
protect a route.
The current version of the collar can be seen in Figure \ref{fig:gateway_server}.
\caption{GPS Collar Prototype}
\subsection{Android Application}
The Android application is capable of retrieving login
information from the user, using it to retrieve a JSON Web
Token, and using that token to retrieve and display
the current collar positions on a map.
When the user initially opens the application, they
are presented with a login screen. The current
appearance of this screen is shown in Figure \ref{fig:login_screen}.
When the ``Log In'' button is pressed, the application sends
a request to the REST API server with the user name and password
(using secure HTTPS), attempting to authenticate. If the authentication
succeeds, the application stores the generated JWT and uses
it for further interaction with the gateway server. It also
uses the JWT to retrieve a current list of active collars
(triggering the code snippet in the previous section).
The application is capable of using the OpenStreetMaps API to display
the locations of the active collars on an interactive map. This
map supports zooming in and out and panning. Each individual
collar is presented on the map as a single marker. In
the future, tapping a marker will take the user to a screen
with more information about the selected collar, such
as its recent location history, battery level, and "out-of-bounds"
status. A screenshot of the collar list screen is shown
in Figure \ref{fig:collar_list_screen}.
\subfloat[Login Screen]{
\subfloat[Collar List Screen]{
\caption{Screenshots from the current version of the Android application.}
Currently, the application does not request updates from the server
(this technique is called "polling"). Rather, the user has to manually
press the "back" button and press "log in" again to refresh the
data on the map. Naturally, we intend to change this, so that the map
updates in near real time as the animals move around.
Though the functionality for regular polling isn't implemented,
the code for handling updated data is present. When receiving a
list of collars, the application must avoid placing duplicate
markers on the map. The application must also remove now-inactive
markers from the map, so that collars that were turned off
are no longer displayed. Additionally, collar names can
change, which must also be reflected in the updated data.
This functionality is implemented by the following Kotlin snippet:
In the above code, \texttt{currentSet} is a set of collar IDs that were marked active
in the latest data update. Regardless of whether or not a marker for a collar
with that ID exists (\texttt{collarOverlays[]}) or a new one needs to be
created (\texttt{Marker(map)}), the new collar's ID is added to the set. After
all the updated data is processed, any collars not in the "current"
list are removed from the map.
\section{To Be Done}
Although it is far along, the project is not yet complete. We believe that the following
functionality / features still need to be completed prior to the code freeze:
\subsection{Client $\rightarrow$ Collar Communication:}
One of the goals for the Android application is to serve as the main communication channel between
the users of the system (ranchers or farmers) and the collars in the field.
To this end, it must be able to not only view the current locations
of the collars and track their history, but also to modify
the settings on these collars.
First and foremost, this means that the Android application must be able
to communicate with the collars in some way. The main difficulty in
implementing this is the bridge between the Python-based server application,
which receives API calls from the Android application, and the C-based
LoRa client application, which can actually send and receive data to collars.
We anticipate establishing communication between the two server-based programs
through a message queue in the shared SQL database: whenever the Python server
receives a commands it must relay to the collars, it will place an entry
into the SQL database. The C client will continuously poll the database
for these messages, and send them to collars as they arrive.
\subsection{Grazing Boundary Tracking and Response}
Currently, while the collars are able to keep track of their location,
they are not able to determine whether or not they are in a "valid" location,
or whether they should emit the auditory or electrical stimulus. This
must be implemented, along with the ability to configure the grazing
boundary through the LoRa protocol.
We will likely define the grazing area to be a 16-point polygon.
The coordinate for each vertex of the polygon will be delivered
through the Protobuf-encoded LoRa communication to the collar,
and an existing algorithm will be used to determine if the current
coordinate of the collar is within the polygon. This is likely
sufficient, since grazing fields are unlikely to become large enough
for the Earth's spherical nature to significantly affect the calculations.
The client had previously stated that it is sufficient for our
collar to shine an LED light to indicate whether or not it is out
of bounds. That is, the actual stimuli are not our main priority,
and thus, we do not have concrete plans to complete that part of
the project.
\subsection{Multi-channel LoRa Communication and LoRaWAN}
A large subset of the communication needs of our project
has already been implemented by the LoRaWAN protocol, which
is built on top of LoRa. This protocol provides signal scheduling,
allowing for a large number of collar devices to be communicating
without sending data concurrently (and, thus, interfering with
each other). We plan to use LoRaWAN in the final version
of the project, though we are currently using LoRa.
A major barrier to the use of LoRaWAN for our team
was the need for specialized hardware. While a single-channel
(i.e., capable of only listening to one input at a time)
device is sufficient for LoRa, LoRaWAN requires a
multi-channel (i.e., capable of concurrent communication
with multiple devices) transceiver. We have recently
been able to purchase this component, but have as
yet been unable to integrate it into our system.
Switching to LoRaWAN would also require changes
to the current code of the LoRa client on the
gateway server and the collar.
\subsection{Data Analysis and Presentation}
Although the current application is capable
of displaying the present location of various
collars, it does not display the grazing
boundary (which is as of yet nonexistent
in the code). The application also does not
provide any significant ability to analyze the
gathered data. The client has requested that
we address this, especially in preparation
for the engineering expo.
In particular, the client has asked that
we allow the user to view the history
of an animal's movements on a map,
and be able to determine how many times
the animal has left the prescribed boundary.
This was part of our original design document,
and we plan on including this functionality
in the final product. Other, more visually
interesting diagrams --- such as graphs
of the number of "escapes" per day --- will
be considered.
\subsection{Physical Casing}
The team has yet to create a physical casing for the collar
component, which would allow for safer transportation and
more robust testing. This will likely be done with
the help of the university's laser cutter, since
the casing will likely take too long to 3D print
using the university's infrastructure. Additionally,
longer 3D prints have a higher probability of failure,
and may require several attempts. We believe that
using a laser cutter, in combination with standard
CAD software, will be sufficient for us to design
and built a box to protect the collar hardware.
The only problem encountered by the team during
this term is the recent coronavirus outbreak. Under normal
circumstances, our team is organized in a very productive way -
every team member is responsible for a component, and multiple
members can rendezvous to work on integrating the components
with each other.
Due to the social distancing measures now strongly recommended for
US citizens, we are having a much harder time with integrating
the components, and testing them with each other. Of particular
concern is the integration of the gateway server and the collar
prototype, which need to be in range of each other to be properly
tested. This means that at most one team member can work on these
components at any given time, and switching between team members
becomes especially unproductive. Because the components are
largely physical, it is nearly impossible to test them without
having access to the hardware. This is especially true because
the majority of the bugs that we have encountered were caused
by problems with the physical assembly, and not with the software.
This issue does not \emph{stop} progress, as much as it \emph{impedes}
it. We are therefore likely going to continue working on the project
with the social distancing measures, having a single team member
work on the hardware components at a time. We have already began
coordinating with Dr. Winters regarding the situation, and we
are currently on track to complete the project despite the delays
caused by the epidemic.
Other than the recent outbreak, the team has not encountered
any issues. All is well.