Ausarbeitung/implementation.tex

   1 \chapter{Implementation}
   2 \hyphenation{im-ple-men-ta-tion im-ple-men-ta-tion-spe-ci-fic}
   3 \todo{more?}
   4 This chapter describes the implementation that was used for the aforementioned
   5 experiments. It consists of the measuring programs themselves, which use the
   6 Wiselib Roomba Control, and are written in C++. Additionally, there are several
   7 Bash and Perl scripts to help with the analysis of the measured data.
   8
   9 All code used for this thesis resides in the Wiselib source tree\footnote{see
  10 \url{https://github.com/ibr-alg/wiselib/}, or the \acs{CDROM} attached at the
  11 end of this book} in the directory \filepath{apps/pc\_apps/roomba\_tests}.
  12
  13 \section{Measuring}
  14 \label{sec:impl:measuring}
  15
  16 The measuring implementation is a set of programs running on a x86-compatible
  17 Linux system which uses the Wiselib to control a Roomba attached to the system
  18 over a serial interface. It features a graphical user interface using the
  19 Qt application framework. Basically, it carries out a movement with specified
  20 parameters (velocity, and destination or angle, depending on the experiment),
  21 then prompts the user to measure and input the actual destination or angle the
  22 Roomba has moved after it has finished the movement. In this context, the user
  23 can choose between two operation modes. In the manual mode, the user can control
  24 the target values for the Roomba movement himself, in automatic mode, the
  25 application uses a list of pre-programmed values for the target values, so the
  26 user only has to input the measured data.
  27
  28 \paragraph{Logging}
  29 All measured data is written to a log file in the current directory whose name
  30 is based on the type of experiment performed. Every line in that log file
  31 describes one measurement made by the user, and consists of pairs of the form
  32 \code{key=value}, separated of by whitespace. In particular, the data on each
  33 line are (with key name, in the order of appearance on the line):
  34 \begin{enumerate}[noitemsep]
  35   \item\texttt{svn:} the \ac{SVN} or Git revision the program was compiled from
  36     (statically compiled into the program)
  37   \item\texttt{roomba\_id:} the ID of the Roomba the measurement was performed
  38     with (given by the user)
  39   \item\texttt{ground\_type:} the ground type used for measurement:
  40     \magicvalue{iz250flur} for laminated floor, \magicvalue{seminarraum}
  41     for carpet floor (given by the user)
  42   \item\texttt{diff\_ticks\_left:} the difference of encoder counts between the
  43     beginning and the end of the movement, on the Roomba's left wheel
  44     (packet~ID \magicvalue{0x2b} in the \ac{ROI} Specification)
  45   \item\texttt{diff\_ticks\_right:} the difference of encoder counts between the
  46     beginning and the end of the movement, on the Roomba's right wheel
  47     (packet~ID \magicvalue{0x2c} in the \ac{ROI} Specification)
  48   \item\texttt{raw\_ticks\_left:} the absolute value of the encoder count of the
  49     Roomba's left wheel (packet~ID \magicvalue{0x2b})
  50   \item\texttt{raw\_ticks\_right:} the absolute value of the encoder count of
  51     the Roomba's right wheel (packet~ID \magicvalue{0x2c})
  52   \item\texttt{batt\_charge:} the charge of the Roomba's battery in mAh
  53     (packet~ID~\magicvalue{0x19})
  54   \item\texttt{batt\_capacity:} the capacity of the Roomba's battery in mAh
  55     (packet~ID~\magicvalue{0x1a})
  56   \item\texttt{batt\_voltage:} the voltage of the Roomba's battery in mV
  57     (packet~ID~\magicvalue{0x16})
  58   \item\texttt{batt\_voltage:} the current of the Roomba's battery in mA
  59     (packet~ID~\magicvalue{0x17})
  60   \item\texttt{move:} movement type; \magicvalue{straight} for straight moves,
  61     \magicvalue{turn} for turn moves (given by the user while selecting the
  62     experiment type)
  63   \newcounter{logitems}\setcounter{logitems}{\value{enumi}+1}
  64 \end{enumerate}
  65 For straight moves (\magicvalue{move=straight}) follow:
  66 \begin{enumerate}[noitemsep,start=\value{logitems}]
  67   \item\texttt{input\_distance}: the target distance in mm sent to the Roomba
  68     via the \ac{ROI} \cmd{Drive} command (given by the user in manual mode,
  69     or determined by the program in automatic mode)
  70   \item\texttt{velocity:} (the velocity in mm/s sent to the Roomba via the
  71     \ac{ROI} \cmd{Drive} command (given by the user in manual mode, or
  72     determined by the program in automatic mode)
  73   \item\texttt{measured\_x:} the actual covered distance in mm in the Roomba's
  74     original viewing direction (measured by the user)
  75   \item\texttt{measured\_y:} originally, the distance shift perpendicular to the
  76     viewing direction of the Roomba. Not used anymore.
  77   \item\texttt{deviation\_orientation:} originally, the shift in orientation of
  78     the Roomba. Not used anymore.
  79 \end{enumerate}
  80 For turn moves (\magicvalue{move=turn}) follow:
  81 \begin{enumerate}[noitemsep,start=\value{logitems}]
  82   \item\texttt{turn\_angle:} the turn angle in degree sent to the Roomba via the
  83     \ac{ROI} \cmd{Drive} command (given by the user in manual mode, or
  84     determined by the program in automatic mode)
  85   \item\texttt{measured\_angle:} the actual turned angle of the Roomba in degree
  86     (measured by the user)
  87   \item\texttt{velocity:} (the velocity in mm/s sent to the Roomba via the
  88     \ac{ROI} \cmd{Drive} command (given by the user in manual mode, or
  89     determined by the program in automatic mode)
  90 \end{enumerate}
  91
  92 \begin{figure}[htbp]
  93   \centering
  94   \includegraphics[width=0.9\textwidth]{images/Implementation-Diagram.pdf}
  95   \caption[File structure of the measuring implementation]{Simplified file
  96     structure of the measuring implementation\label{fig:impl:struct}\\
  97     yellow: common files used for all three experiments; green: files for
  98     Experiment~1; blue: files for Experiment~2; red: files for Experiment~3.\\
  99     An arrow from node $x$ to node $y$ means ``$y$ depends from $x$''.}
 100 \end{figure}
 101
 102 Figure~\ref{fig:impl:struct} shows the file layout of the measuring
 103 implementation. It reuses components whereever possible, and therefore
 104 consists of three parts (green, blue, red) specific for the three performed
 105 experiments (see Sections~\ref{sec:exp1}, \ref{sec:exp2} and~\ref{sec:exp3}), as
 106 well as a common part (yellow), which contains the elements of the
 107 implementation shared by the other three parts.
 108
 109 \subsection{Common part}
 110 % no \file{} here please, we don't want "stuff.{cc,h}" in index...
 111 The files \filefont{stuff.\{cc,h\}} and
 112 \filefont{target\_value\_input\_dialog.\{cc,h\}} build up the general part of
 113 the implementation (shown with a yellow background in
 114 Figure~\ref{fig:impl:struct}). These are included in all other three parts of
 115 the implementation.
 116
 117 \paragraph{Global declarations}
 118 These files \file{stuff.cc} and \file{stuff.h} contain global helper functions
 119 and variables, data structures specific to the implementation, and all kinds of
 120 convenient typedefs. In particular, there are functions for formatting and
 121 logging messages to a log file, according to the format described above. While
 122 \file{stuff.h} contains the declarations and typedefs, \file{stuff.cc} contains
 123 the appropriate definitions. Besides, since \file{stuff.h} is included
 124 indirectly in most other source files, it also includes the Wiselib headers
 125 needed for Roomba control.
 126
 127 \paragraph{Input dialog for measured orientation values}
 128 \hyphenation{Tar-get-Va-lue-In-put-Dia-log}
 129 The class \class{TargetValueInputDialog}, which is an implementation of a simple
 130 dialog box using the Qt widget framework, resides in the files
 131 \file{target\_value\_input\_dialog.cc} and
 132 \file{target\_value\_input\_dialog.h}. This dialog allows the user to input a
 133 measured angle (although the name mentions a ``target value'', which is probably
 134 named this way only for the sake of confusion\ldots). It contains an input box
 135 with adjacent spin box for entering the measured value, OK and Cancel buttons,
 136 and also information about the target orientation and the current real
 137 orientation as well as the current capacity of the Roomba's battery.
 138
 139 When the dialog is displayed, it obtains the orientation as it was before the
 140 last movement, and the target angle, and sets the input box to the specified
 141 target value. When the input box is changed, the dialog calculates the new
 142 orientation from the last orientation and the value in the input box, and
 143 displays this new orientation. This way, the user can simply enter the measured
 144 angle by increasing or decreasing the value by operating the spin box, and
 145 adjusting the value until the displayed orientation reflects the real
 146 orientation of the Roomba.
 147
 148 To achieve this, the class uses the Qt signal-slot
 149 mechanism~\cite{qt-signalslots}, and connects the \fnfont{intValueChanged(int)}
 150 signal of the input box to the method
 151 \fnfont{turn\_dialog\_value\_changed(int)},
 152 which performs the calculation and display of the new value.
 153
 154 \subsection{Implementation for Experiment~1: Application \prog{roomba\_test}}
 155 The implementation for Experiment~1 is used to build the application
 156 \prog{roomba\_test}. Its main part is constituted by the file \file{main.cc},
 157 which uses the global definitions of \file{stuff.h} and the dialog box from
 158 \file{target\_value\_input\_dialog.h}.
 159
 160 \paragraph{Initial setup}
 161 The program first checks for command line arguments, where it expects the type
 162 of test performed and if automatic or manual mode, the ground type, and the
 163 Roomba's ID, or the \code{-{}-help} switch to display a usage message.
 164
 165 After that, it instanciates and initializes a \class{RoombaModel} instance,
 166 while also providing for the needed instances of the \concept{Timer} concept
 167 and the \concept{SerialCommunication} concept, which is needed for the serial
 168 communication with the Roomba. Furthermore, an instance of
 169 \class{ControlledMotion} is initialized by passing the \class{RoombaModel}
 170 instance. This instance is later used to control the Roomba's movements.
 171
 172 Due to an implementation detail inside the \class{RoombaModel} class, the sensor
 173 data returned by the class itself may be corrupt if it is read while new sensor
 174 data packets arriving from the Roomba are processed. To prevent this corruption,
 175 we register a callback function which gets called everytime new sensor data is
 176 available (i.~e. the sensor data packets have been fully processed, which is
 177 every 15~ms as we are using the \cmd{Stream} command), and let the callback
 178 function perform a deep copy of the sensor values we are interested in.
 179
 180 If the user has specified automated measurement on the command line, an array of
 181 velocities and target values is set up, as described in
 182 Section~\ref{sec:exp1:setup}, and slightly scrambled to ensure independence of
 183 the measurements. Furthermore, the file name for the log file to record the
 184 measured values is determined from the operation mode, the type of experiment
 185 performed, and the Unix timestamp\footnote{The Unix timestamp is a signed
 186 integer which is defined as the seconds elapsed since January 1st, 1970, 0:00
 187 \ac{UTC}}. To ease measurement of angles, if the type of experiments are turn
 188 movements, the user is prompted once to input the current orientation of the
 189 Roomba.
 190
 191 \paragraph{Input/measurement loop}
 192 The rest of the program consists of a loop which does the following things:
 193 \begin{enumerate}[noitemsep]
 194   \item \textbf{In manual mode:} Prompt the user for a new pair of velocity and
 195     input value \\
 196     \textbf{In automatic mode:} For each pair of velocity and input value
 197     specified in the previously definied arrays, do the following:
 198   \item Carry out the specified movement using the \class{ControlledMotion}
 199     instance
 200   \item Prompt the user for the measured value using the
 201     \class{TargetValueInputDialog} for angles, or a simple Qt
 202     \class{QInputDialog} for distances
 203   \item Write the values to the log file, as specified at the beginning of this
 204     section, using the log function from \file{stuff.h}
 205   \item Repeat until the user cancels.
 206 \end{enumerate}
 207
 208 \subsection{Implementation for Experiment~2:
 209 Application \prog{mean\_correction\_test}}
 210 Additionally, the program \prog{mean\_correction\_test} for Experiment~2 uses
 211 the class \class{CorrectedMeanMotion} from \file{corrected\_mean\_motion.h}.
 212 This class adapts the input value according to the fit function determined from
 213 the data in Experiment~1 (see Chapter~\ref{sec:exp2}), and exposes the exact
 214 interface as \class{ControlledMotion} does. Thus, both classes are easily
 215 interchangeable, and \class{CorrectedMeanMotion} serves as an example of the
 216 stackability principle mentioned in Section~\ref{sec:wiselib:arch}.
 217
 218 The essential parts of \class{CorrectedMeanMotion} are the functions
 219 \fnfont{set\_profile} to specify the fit function (i.~e. the ground type),
 220 and \fnfont{move\_distance} and \fnfont{turn\_to} which use that specified fit
 221 function to adapt the input value (angle or distance) and then move the Roomba
 222 around accordingly. As in the implementation of \class{ControlledMotion}, the
 223 encoding counts of the Roomba's wheels are monitored to determine when the
 224 Roomba has to be stopped.
 225
 226 The main function in \file{mean\_correction.cc} is mostly the same as in
 227 \file{main.cc}, except that it does not use a \class{ControlledMotion}
 228 instance but a \class{CorrectedMeanMotion} instance to control the Roomba. The
 229 profile that \class{CorrectedMeanMotion} uses is determined by the floor type
 230 given on the command line.
 231
 232 \subsection{Implementation for Experiment~3: Application
 233 \prog{soft\_start\_test}}
 234 As the implementation for Experiment~2, the application \prog{soft\_start\_test}
 235 which is built from the file \file{soft\_start.cc} has the same basic layout as
 236 \file{main.cc} from Experiment~1. As previously, it defines a custom movement in
 237 the class \class{SoftStartMotion} (file \file{soft\_start\_motion.h}), which
 238 is also compatible with the interface defined by \class{ControlledMotion}. This
 239 class causes the Roomba to constantly accelerate at the start of the movement,
 240 and also constantly decelerate at the end of its movement, until the target
 241 distance or angle is reached.
 242
 243 Internally, the class defines a time interval for the duration of the start and
 244 stop processes, and breaks down the process into multiple acceleration steps
 245 which each increase the velocity by a constant value, so for sufficient steps,
 246 the acceleration remains nearly constant. Therefore, a timer is used, which
 247 increases resp. decreases the velocity over time (depending if starting or
 248 stopping), until the target velocity or zero is reached, and sends \cmd{Drive}
 249 commands to the Roomba over the \ac{ROI}. The methods \fnfont{move\_distance}
 250 and \fnfont{turn\_to} on the other hand calculate the distance at which
 251 the deceleration process begins, and fire the initial timer to increase the
 252 velocity.
 253
 254 \section{Evaluation}
 255 \label{sec:impl:eval}
 256 For evaluation of the log files, a few Bash and Perl scripts were used. They
 257 use Gnuplot to plot graphs and \ac{GNU} R for statistical analysis, and reside
 258 in the Wiselib source tree under \filepath{apps/pc\_apps/roomba\_tests/logs}.
 259
 260 \paragraph{\prog{graph.sh}}
 261 This wrapper script uses Gnuplot to create interactive 3-dimensional plots of
 262 the original behaviour from Experiment~1, and also includes the fit function as
 263 determined by \ac{GNU} through linear regression. For each pair of ground type
 264 (carpet floor, laminate floor) and experiment type (straight movement, turn on
 265 spot), it opens a Gnuplot window containing a graph with the target value on the
 266 $x$ axis, the velocity on the $y$ axis and the measured value on the $z$ axis,
 267 and the fit function. The user can rotate and tilt the
 268 plot with the mouse.
 269
 270 \paragraph{\prog{graph-mean.sh}} This script does the same as \prog{graph.sh},
 271 except it visualizes the measured data from Experiment~2 (Mean Correction), and
 272 a fit function for that data.
 273
 274 \paragraph{\prog{graph-soft.sh}} Also, like both scripts mentioned above, this
 275 script uses the measured data from Experiment~3 (Constant Start/Stop
 276 Acceleration), and fit function for that data.
 277
 278 \paragraph{\prog{graph-mean-soft.sh}} This wrapper script plots both results
 279 from Experiment~2 and 3 into one coordinate system, for better comparison. It
 280 does not include the fit functions.
 281
 282 \paragraph{\prog{graph-errorlines.sh}} In contrast to the other scripts,
 283 \prog{graph-errorlines.sh} plots the 2D graphs seen in
 284 Chapters~\ref{sec:exp1},~\ref{sec:exp2} and~\ref{sec:exp3}, with target value on
 285 the $x$ axis, measured value on the $y$ axis, and velocities as single
 286 functions each.