Ausarbeitung/experiment1.tex

   1 \chapter{Experiment 1: Original Movement Behavior}
   2 \label{sec:exp1}
   3
   4 The concepts described in the previous chapter are now used to set up a test
   5 series with the purpose to measure the error in the Roomba's movement and
   6 to improve its accuracy.
   7
   8 In the first experiment, the Roomba's original movement behavior was
   9 measured to get an overview of the errors that occur while moving, and to
  10 establish a pool of data for correction approaches to work on later.
  11 There was no error correction involved (apart from any possible error
  12 correction the Roomba itself implements in its underlying logic, and which is
  13 not known to anyone except iRobot). To achieve this, the Wiselib Roomba Control
  14 is used to access the Roomba over the \ac{ROI}\index{Roomba Open Interface}.
  15
  16 \section{Setup}
  17 \label{sec:exp1:setup}
  18 \begin{figure}[htbp]
  19  \centering
  20  \includegraphics[width=0.45\textwidth]{./images/IMAGE_00079.jpg}
  21  \caption{Roomba with netbook\label{fig:roombasetup}}
  22 \end{figure}
  23 \begin{figure}[htbp]
  24  \centering
  25  \includegraphics[width=0.6\textwidth]{./images/IMAGE_00148.jpg}
  26  \caption{Measuring turn angles with laser pointer\label{fig:laserpointer}}
  27 \end{figure}
  28
  29 The test equipment consisted of a small x86 netbook\index{netbook} which was
  30 mounted on an iRobot Roomba~530\index{Roomba} robot, as seen in
  31 Figure~\ref{fig:roombasetup}. The netbook controlled the Roomba over a
  32 \acs{USB}-to-serial converter plugged into the \ac{ROI}\index{Roomba
  33 Open Interface} port on the Roomba, and hosted as the environment for
  34 executing the Wiselib \index{Wiselib} code.
  35
  36 In this experiment, the Roomba started and stopped with the full velocity the
  37 movement was executed with; so there was (ideally speaking) an infinite
  38 acceleration and deceleration at the start and the end of the movement. As
  39 mentioned before, due to limitations in the \ac{ROI}\index{Roomba Open
  40 Interface} it is only possible to explicitly start and stop the Roomba's
  41 movements at different times, so the Wiselib's implementation of the Roomba
  42 control code first starts the Roomba's movement, keeps track of the turned angle
  43 and covered distance, and then stops the Roomba if these values exceed the
  44 target values.
  45
  46 The tests were done in two atomic drive modes: letting the Roomba walk a
  47 specific straight distance with a specific velocity in its viewing direction and
  48 letting it turn on the spot with a specific velocity about a specific angle.
  49 Each of the two modes was carried out on two different floor types\index{floor
  50 type}, a laminated floor and a carpet floor, to see if the movement behavior
  51 significantly depended on the floor type. The side brush was removed, since the
  52 Roomba tends to turn slightly towards the left when driving straight on a
  53 carpet floor. Without the side brush, this was not the case.
  54
  55 The actual traveled distance of the straight drive tests were determined using
  56 a measuring tape with an accuracy of 1~mm. Only the distance in the Roomba's
  57 original viewing direction was considered, as it turned out that the offset
  58 perpendicular to the viewing direction and a possible shift in orientation were
  59 too small to be measured precisely.
  60
  61 The actual turn angles of the turn tests were determined using a
  62 \acs{ISO}/\acs{DIN}~A0 sheet of paper with a printed polar coordinate system in
  63 which a circular hole was cut in the center to let the Roomba's wheels touch the
  64 floor. The sheet was fixed on the floor, and the Roomba was aligned in the
  65 center of the paper. A laser pointer\index{laser pointer} attached to the Roomba
  66 pointed to the current orientation on the paper, as shown in
  67 Fig.~\ref{fig:laserpointer}. The accuracy for these tests was 1~degree.
  68
  69 After the initial setup, the application \prog{roomba\_test} (see
  70 section~\ref{sec:impl:measuring}) was started on the netbook for half-automatic
  71 testing. It used a predefined array of nominal distances, angles and velocities
  72 and for each pair of distance (for straight drive tests) or angle (for turn
  73 tests) and velocity it used the Wiselib implementation to move the Roomba. Then
  74 it asked the user to input the measured distance or angle the Roomba had moved
  75 and repeated the process until each combination of distance/angle and velocity
  76 was used. The nominal and input values were written to a log file, as well as
  77 the floor type, the Roomba ID, the \ac{SVN} revision the program was built
  78 from, the battery status\index{battery status}, and other internal
  79 implementation-specific values.
  80
  81 For the straight drive tests, the arrays with predefined values were:\\
  82 \begin{tabular}{@{}ll@{}}
  83 Distances: & 20, 50, 100, 200, 500, 1000, 2000 and 4000~mm \\
  84 Velocities: & 20, 50, 70, 100, 150, 200, 300 and 400~mm/s
  85 \end{tabular}
  86
  87 For the turn tests, the arrays with predefined values were: \\
  88 \begin{tabular}{@{}ll@{}}
  89 Turn angles: & \phantom{0}5, 15, 30, 45, 90, 120, 180, 360, 530 and 720~degree\\
  90 Velocities: & 20, 50, 70, 100, 150, 200, 300 and 400~mm/s
  91 \end{tabular}
  92
  93 According to the implementation of the Wiselib Roomba control, the velocities
  94 were given in mm/s and referred to the distance the wheels traveled when the
  95 Roomba turned on the spot, which was a circle of 230~mm in diameter.
  96
  97 \section{Results}
  98 \label{exp1:results}
  99 The following graphs show the difference from the measured value to the input
 100 value for driving or turning with different velocities. Positive values mean
 101 that the Roomba had turned too much or traveled more than the target value,
 102 negative values mean that the Roomba had turned or traveled less. The
 103 plots show multiple test runs; given are the minimum, the maximum and the
 104 arithmetic mean\index{arithmetic mean} of all results for a data point.
 105
 106 \todo{clearpage, cleardoublepage?}
 107 \begin{figure}[p!]
 108  \centering
 109  \includegraphics[width=\textwidth]{images/iz250flur_drive_data.pdf}
 110  \caption{Original behavior on laminated floor, straight drive movements
 111   \label{fig:orig:lam:drive}}
 112 \end{figure}
 113 \begin{figure}[p!]
 114  \centering
 115  \includegraphics[width=\textwidth]{images/iz250flur_turn_data.pdf}
 116  \caption{Original behavior on laminated floor, turn movements
 117   \label{fig:orig:lam:turn}}
 118 \end{figure}
 119 \begin{figure}[p!]
 120  \centering
 121  \includegraphics[width=\textwidth]{images/seminarraum_drive_data.pdf}
 122  \caption{Original behavior on carpet floor, straight drive movements
 123   \label{fig:orig:carpet:drive}}
 124 \end{figure}
 125 \begin{figure}[p!]
 126  \centering
 127  \includegraphics[width=\textwidth]{images/seminarraum_turn_data.pdf}
 128  \caption{Original behavior on carpet floor, turn movements
 129   \label{fig:orig:carpet:turn}}
 130 \end{figure}
 131
 132 Figure~\ref{fig:orig:lam:drive} shows that the error becomes greater with
 133 increasing input distance when driving straight on the laminated floor, however,
 134 in Figure~\ref{fig:orig:carpet:drive} we see the opposite effect on the carpet
 135 floor, the error decreases with greater input distance. This could happen due to
 136 imprecise measurement of distances in either the Roomba's sensors or the Wiselib
 137 implementation that controls the Roomba, or both, adding up over the time the
 138 movement continues. Also slippage of the wheels on the laminated floor could be
 139 possible, as well as slowdown through the carpet floor, explaining why the error
 140 increases on the laminated floor, but decreases into negative values on the
 141 carpet floor when the distance grows.
 142
 143 On the other hand, rising the velocity always seems to cause the
 144 error to increase. This illustrates the effect of accumulating errors which
 145 represents the crucial drawback of dead reckoning\index{dead reckoning}.
 146
 147 The same effects also apply for turn tests. Additionally, there could be errors
 148 resulting from false assumptions about the Roomba's wheel base\index{wheel
 149 base}, resulting in bogus calculations of the circle the Roomba's wheels
 150 describe while turning, and therefore leading to false results.
 151
 152 In the following, two approaches to correct errors and improve the accuracy of
 153 the movement are presented.