Operation: Localization#

  • Default available maps:

    • ETH_large_loop/TTIC_large_loop

_images/eth_large_loop.png

Fig. 56 ETH_large_loop#

Operation instructions#

In order to operate your autolab you need to set it up according to the following instructions:

  1. Install docker-compose for the desktop machine: instructions

  2. Make sure you are on the ente version of the shell commands by running:

    dts --set-version ente

  3. Update the commands:

    dts update

  4. On the local machine, for ALL robots (Autobot, Watchtower, Duckietown):

    dts duckiebot update [HOSTNAME]

  5. On robots, based on the robot type, create the following configs:

    • For ALL robots (Autobot, Watchtower, Duckietown): Replace [MAP_NAME] with the map name for this Autolab, e.g. ETH_large_loop or TTIC_large_loop:

      echo [MAP_NAME] > /data/config/autolab/map_name

    Attention

    Autobot only!

    • Replace [TAG_ID] with the AprilTag ID (just the number) on top of that Autobot:

      echo [TAG_ID] > /data/config/autolab/tag_id

  6. On Watchtowers only, make sure that the dt-core container is running. You can do so by running this on your desktop/laptop:

    dts stack up -H [HOSTNAME] core -d

    TODO

    add apriltag detector topic name

    Tip

    Make sure the AprilTag detector is spinning and you get detection messages in rostopic hz.

  7. On the local machine, for ALL robots (Autobot, Watchtower, Duckietown), we need the Autolab layer running as well. You can do so by running this on your desktop/laptop:

    dts stack up -H [HOSTNAME] autolab -d

    Note

    This guide was written for the map ETH_large_loop. Replace it with your map name every time you find it in a command.

  8. Check if you get data by cloning the autolab localization repo and running the following command from its root:

    dts devel build -f
dts devel run -L test-tf -- --hostname ETHlargeloop

    Attention

    The separator -- is not a typo; the hostname is the map name without underscores, so ETHlargeloop instead of ETH_large_loop. The map name is case sensitive.

    You should see an output like this:

    watchtower03/camera_optical_frame tag/326
watchtower03/camera_optical_frame tag/307
watchtower03/camera_optical_frame tag/326
watchtower02/camera_optical_frame tag/308
watchtower02/camera_optical_frame tag/322

    Make sure you get detections from all the watchtowers. Every 10s, the autobots will publish the transform between their footprint and the AprilTag on top of them:

    watchtower04/camera_optical_frame tag/314
autobot01/footprint tag/403
watchtower05/camera_optical_frame tag/301

    The Duckietown robot will also publish TF between the map frame and each ground tag (every 10 seconds):

    watchtower02/camera_optical_frame tag/343
world map
map tag/301
map tag/303
map tag/309
map tag/310
map tag/311
map tag/312
map tag/314
map tag/315
map tag/318
map tag/317
map tag/321
map tag/322
map tag/307
map tag/326
map tag/343
watchtower05/camera_optical_frame tag/301
watchtower05/camera_optical_frame tag/307

    Make sure you get all these types of observations.

Another way of looking at this data is by running the following command (from the root of the same repository):

dts devel run -f -X -L single-experiment -- --hostname ETHlargeloop

This should wait 12 seconds for the observations to come in and put everything in a pose-graph that will be optimized and rendered. You should see something like this appear:

image-20201207-232938.png

Online localization#

To execute localization in realtime (i.e. online) you can run the following command, again from the root of the repository:

dts devel run -f -X -L single-experiment-online -- --hostname ETHlargeloop

Integrate Odometry data (wheel encoders)#

In order to integrate encoder data into the localization system, you need an extra container that takes the odometry data from the robot and feeds it into the localization system.

Note

You need to run this on every Autobot.

You can run the odometry integration with the command:

dts stack up -H [HOSTNAME] encoder -d

You can test that the data is correctly received by the localization system by running the test-tf application as done above. You should see something like the following, indicating that a transformation between the reference frame footprint of the robot at different timestamps is received:

autobot01/footprint autobot01/footprint
autobot01/footprint autobot01/footprint
autobot01/footprint autobot01/footprint