Using RT-SLAM

Configuring the demo

There are two configuration files :

• --config-setup=data/setup.cfg that contains everything related to the hardware setup (camera, IMU, calibrations, hardware noise specs, ...). If none specified, the default file is data/setup.cfg in live mode, and <data-path>/setup.cfg in replay mode.

• --config-estimation=data/estimation.cfg that contains everything related to estimation and Slam (size of map, number of landmarks, strategies, ...). If none specified, the default file is data/estimation.cfg.

You can use the '@' character to refer to <data-path> when specifying the location of the config file to use : --config-estimation=@/estimation.cfg

Running the demo

   1 cd $JAFAR_DIR/build/modules/rtslam
   2 demo_suite/<platform>/demo_slam <options>

Control options:

• --disp-2d=0/1 use 2D display

• --disp-3d=0/1 use 3D display

• --render-all=0/1 (needs --replay 1)

• --replay=0/1 (needs --data-path and having dumped here an online demo first)

• --dump=0/1 dump the images (--replay=0) or the rendered views (--replay=1) (needs --data-path). If --replay=0, --data-path should be in a ram disk.

• --log=0/1 log result in text file

• --rand-seed=0/1/n. 0=generate new one, 1=in replay use the saved one, n=use seed n

• --pause=0/n. 0=don't, n=pause for frames>n (needs --replay 1)

• --data-path=/mnt/ram/rtslam

• --config-setup=data/setup.cfg

• --config-estimation=data/estimation.cfg

• --verbose=0/1/2/3/4/5. Off/Trace/Warning/Debug/VerboseDebug/VeryVerboseDebug (only if compiled in debug)

• --help

• --usage

Slam options:

• --robot=0/1. 0=constant velocity, 1=inertial

• --trigger=0/1/2. 0=internal, 1=external with MTI control, 2=external without control

• --simu=0/1

• --freq=<double,Hz>. camera frequency (only with trigger==0/1)

• --shutter=<double,seconds>. shutter time (only with trigger==1)

Controlling the demo

When doing a replay, you can control how slam is running by interacting with the 2D viewer:

• Press "[space]" to switch between play and pause mode.

• Press "N" to process one image when in pause mode.

• Press "F" to switch between render-all mode and fast mode.

• Press "Q" to quit.

• Click on an observation to get textual and visual information about the observation and associated landmark.

• Click on an empty part of the image to get information about the sensor and the associated robot.

Interpreting the demo

Color code:

• magenta: InitialParametrization landmarks that were observed this frame

• dark red: InitialParametrization landmarks that were not observed this frame

• cyan: ConvergedParametrization landmarks that were observed this frame

• dark blue: ConvergedParametrization landmarks that were not observed this frame

Another way to remember the color code:

• bright colors represent observed landmarks

• dark colors represent non observed landmarks

• reddish hues represent InitialParametrization landmarks (like inverse depth for points)

• bluish hues represent ConvergedParametrization landmarks (like euclidean for points)

2D landmark representation:

• a colored + cross: the predicted position of the landmark

• a colored ellipse: the 3-sigma prediction uncertainty

• an orange x cross: the measured position of the landmark

• a colored label: the id of the landmark, followed by the matching score (/100)

Reasons for a landmark not being observed:

• outside of the sensor frame (you can still see them in the 3D view)
• matching failed (you can see the matching score that is below the threshold)
• the observation was inconsistent (the measure is out of the uncertainty ellipse)
• too many landmarks are in the sensor frame and we didn't try to observe some of them (no matching score is present). In that case the landmark is regardless drawn with the "observed" color, because it wasn't due to a problem with the landmark.