**WHAT IS THIS
Mapping is a suite for studying sonar and laser based mapping robotic mapping algorithms.
The name is only an attempt, and it will FOR SURE change in the next releases.
All of the code has been developed by Giorgio Grisetti <grisetti@dis.uniroma1.it>, and is
released under the GPL license.
The software is able to process log files in the Carmen format.
In particular in the suite the following packages are provided:
1)a scanmatcher based on gradient descent
2)a (simple) implementation for landmark based FastSLAM
3)a (very cool) grid_fast_slam algorithm
4)a map builder for sonar data, based on forward model.
**REQUIREMENTS:
an X86 linux machine with
gcc>3.2.x
qt, libraries and development > 3.x
gsl, gnu scientific libraries > 1.3
**COMPILING AND INSTALLING:
What to do for compiling the software:
1) uncompress the archive
tar -xzvf mapping-xx.tgz
2) change into the source directory
cd mapping
3) set the LD_LIBRARY_PATH variable so that it includes the mapping/lib directory
. setlibpath
4) generate the configuration files
./configure
5) compile the software
make
At rhis point the following binaries should have been generated in the mapping/bin directory
autoptr_test
converter
fastslam_test
fmp
gfs2log
gfs_gui
gfs_logplayer
gfs_test
hscanmatch_test
log_test
particlefilter_test
range_bearing
scanmatch_test
weakness
while the mapping/lib directory should contain the following shared objects
libconverter.so
libforwardmodel.so
libgui.so
libsensor_base.so
libsensor_range.so
libFastSLAM.so
libgridfastslam.so
liblog.so
libsensor_odometry.so
libutils.so
Each library is generated from the corresponding directory, in the source tree.
Not all of the binaries in mapping/bin are useful, since most of them are test programs for
the libraries.
**USING
In the following we focus on the following tools:
-the scanmatcher
-the gridfastslam algorithm
and the related conversion tools
SCANMATCHER:
the command line for the scan matcher is
hscanmatch_test [options] -filename <carmen_log>
The rest of the options of the scan matcher are illustrated at the end of the document
The scanmatcher generates a file named scanmatched-<carmen_log>, that contains the laser readings and the robot poses corrected.
It is possible, suddenly, to process such a file with a tool for generating occupancy grids (like carmen), to obtain the map
from the log file.
COMMAND LINE OPTIONS [scanmatcher]
Here are the options that can be passed to the command line: <i> integer, <d> double <s> string
-filename <s> : the input log filename in carmen format
-xmax <d> -xmin <d>,
-ymax <d> -ymin <d> : the map size, in meters
-delta <d> : the patchsize of the hierarchical map used (default 1m)
-patchDelta <d> : the map discretization (default 0.1m)
-maxrange <d> : maximum laser range
-maxUrange <d> : maximum laser range usable by the algorithm
-regscore <d> : the registration score for the scan, if over regscore the scan will be not registered in the map
-critscore <d> : the critical score for the scan, if below cs the scan will be registered in the map using teh odometry.
must be lower than regscore.
-kernelSize <i> : the (half) of the size of the convolution kernel (number of grid cells) minus 1 (default 0)
-sigma <d> : the sigma of the convolution gaussian
-iterations <i> : the number of recursive iteration used by the scan matcher
-lstep <d> : the scanmatcher linear search step, in meters
-astep <d> : the scanmatcher angular search step, in radians
-maxMove : the maximal move allowed in the log among adiacent scans
GRIDFASTSLAM:
The grid fast slam (gfs) algorithm is the coolest algorithm in the suite.
It allows to build maps with a fast slam based algorithm, by adopting a smart choice of the prior distribution.
There are three gfs tools here:
1) the gui based tool (gfs_gui), which allows to track what is appening while building a map
2) the off line tool (gfs_test), which allows to process a carmen log and to produce a *.gfs file (which is the gfs output format),
containing all of the informations produced by the algorithm while running
3) the gfs2log utility, which converts the gfs outoput file in a log file, tracking the best particle.
It is possible to suddenly process the log file in order to obtain the occupancy grid as specified in the scanmatcher.
The options of gfs_gui and gfs_test are the same, and moreover, since they are based on the scanmatcher, they accepts all of the options for the scanmatcher.
COMMAND LINE OPTIONS [gridfastslam]
-outfilename <s> : the output filename in gfs format;
-srr <d>, -srt<d>,
-str <d>, -stt<d> : the motion model parameters (sigma). see gridfastslam/MotionModel.{h,cpp}
-particles <i> : the number of the particles to be used
-angularUpdate <d> : the angular update distance, the algorithm executes an update step each time the traveled distance from the last
updates exceed this value;
-linearUpdate <d> : the linear update distance, the algorithm executes an update step each time the traveled distance from the last
updates exceed this value;
-lsigma <d> : sigma of the gaussian for evaluating the likelihood in the correlation based model;
-lobsGain : gain to be used while evaluating the likelihood, for smoothing the resampling effects
-lskip : the laser readings to skip in the likelihood evaluation;
parseInt("-mapUpdate", mapUpdateTime);
parseFlag("-drawFromObservation", m_drawFromObservation);
GFS2LOG:
During the operation gfs_gui shows only the binary state of the best particle map cells. This results in a less accurate and less nice map
representation that the one allowed by the algorithm (which internally keeps track of the mean of the laser points falling in each cell).
In order to recover such a precision it is possible to recover a log from a gfsfile, through the command:
gfs2log <gfsfile> <logfile>.
and suddenly to recover the occupancy grid from the logfile.
