The benefit of having such a parser is that it allows much more freedom in the specification of the format for the raw data. As long as the raw data has a suitable version number identifier, the parser can hide changes and even abstract to different robots. And, with the data collection process saving as complete a description of the world as is practical, it is trivial to extract and use different representations during learning.
machine% parse datafile configfile > parsedfileA sample configuration file is provided, as is a lisp readable frame generated.
The specification language is:
SONAR Output recorded sonar readings.
LASER Output recorded laser readings. RECTANGLE name x1 y1 x2 y2 xres yres Convert vision data bounded by the rectangle {x1,y1}{x2,y2} into rectangles of size {xres, yres} in row/coloumn order. A slot titled "name" will be created in the parsed output. (See sample files for examples)
machine% CTREPLAY -t datafileThis will open 2 windows on the X window display entitled 'Map' and 'Last image' respectively.
The Camera View is toggled using the verbose flag ('v' at the CTREPLAY command
line), and display the current visual perception - an intensity display or
actually image on a color monitor. Use 't
The map window depicts the position of the robot relative to the start of a
run (located at (0,0)). 24 rays protrude from a central location - these
signify the sonar reading at any point in time. In addition, by entering 'e',
toggling erase mode, old lines are not erased, which leads to a map of free space (save
for noise and sensor error), being built up. If you just wish to see the
objects detected by the laser and the sonar, 'l' toggles line/point mode. In
point mode, the laser is also visible as a horizontal dot (the sonar being a
vertical dot). As the robot moves, if 'e' is off, walls and obstacles will be
traced out. In addition, the window can zoom in and center on portions of the
world. On a color screen, the robot path will appear in green, the laser will
always be present as a yellow trace, with the sonar being increasingly white,
depending on the likelihood of an object actually having being detected. Full
online help is provided by entering '?'.