Announcing argument change in Autoware Launch for switching localization modes

[kminoda]

TLDR

• We plan to modify the argument for selecting which localization methods to launch in autoware_launch
• Let us know if there is any issues or counter-proposal, otherwise this modification will be merged in about a week

Backgrounds

Thanks to the supportive contributors in Autoware community, the number of localization methods supported in Autoware Universe has increased from one to three, currently supporting ndt_scan_matcher, yabloc, and eagleye. I believe the diversity of the supported localization methods is crucial for Autoware to expand towards various ODDs with flexible sensor configurations.

However, current approach for selecting among these localization methods are somewhat complicated, making it difficult for users to switch the methods, and for developers to add any additional localization methods. For example, if you want to use eagleye instead of ndt_scan_matcher, you first have to modify the following part in autoware.launch.xml

...
  <!-- Localization -->
  <group if="$(var launch_localization)">
    <include file="$(find-pkg-share autoware_launch)/launch/components/tier4_localization_component.launch.xml"/>
    <!-- <include file="$(find-pkg-share autoware_launch)/launch/components/map4_localization_component.launch.xml"/> -->
  </group>
...

to

...
  <!-- Localization -->
  <group if="$(var launch_localization)">
    <!-- <include file="$(find-pkg-share autoware_launch)/launch/components/tier4_localization_component.launch.xml"/> -->
    <include file="$(find-pkg-share autoware_launch)/launch/components/map4_localization_component.launch.xml"/>
  </group>
...

and then use pose_estimator_mode to select which mode to use for eagleye. This is rather complicated while all of the methods mostly obey the Recommended Architecture in Localization Design Docs.

We should instead have a comprehensive interface to select among several localization methods available. In other words, we should have an alternative way that can take the advantage of the current modular implementation of localization methods based on the recommended architecture.

Proposal

We introduce two new arguments instead of localization_mode

• pose_source
• twist_source

which is to select pose_estimator and twist_estimator implementation respectively (see Recommended Architecture in Localization Design Docs). Using these two arguments, you are now able to select localization methods as follows

NDT as pose_estimator, Gyro Odometer as twist_estimator

$ ros2 launch autoware_launch autoware.launch.xml ... pose_source:=ndt twist_source:=gyro_odom

NOTE: This is going to be the default argument, and thus no need for specifying them explicitly in this case.

YabLoc as pose_estimator, Eagleye as twist_estimator

$ ros2 launch autoware_launch autoware.launch.xml ... pose_source:=yabloc twist_source:=eagleye

Eagleye as pose_estimator and twist_estimator

$ ros2 launch autoware_launch autoware.launch.xml ... pose_source:=eagleye twist_source:=eagleye

Related Links:

• Autoware Universe Implementation: feat: use pose_source and twist_source for selecting localization methods autoware.universe#4257
• Autoware Launch: feat: use pose_source and twist_source for selecting localization methods autoware_launch#442
• Autoware Documentation: feat: use pose_source and twist_source for selecting localization methods autoware-documentation#418

[kminoda]

We are going to merge above PRs in about 24 hours.