2024-08-01 Racing Working Group Meeting

[pojenwang]

Administrative

Previous Meeting Minutes

Attendees

• Po-Jen Wang (AWF)
• Mitsudome-san (Tier.IV)
• David Walmroth (Open AD Kit / Pix-Moving)
• Atanasko

Minutes: Po-Jen Wang

Topics

Autoware on F1tenth Update

• Previously the F1tenth branch was updated to the Autoware’s release/2024.06 version. However, it was found that this version has new build issues (because of a recent build-dependent package update?).
• The F1tenth branch is now updated to the Autoware release/2024.07 version to resolve build issues. This latest version also has a lot of refactoring done and our F1tenth packages were modified to adapt to the changes:
  -- Package reorganized and planning modules were added a “autoware” prefix in the package name
  -- simple_trajectory_follower and controller codes were updated. The autoware_auto msgs were replaced by autoware_msgs
• Issues adapting the recent changes:
  -- The recordreplay_planner packages that our demo is based on still use the autoware_auto_planning_msgs that include action messages(for record and replay) that were not yet added to the autoware_planning_msgs
  -- The utility functions(part of the Autoware core packages) called by the simple_trajectory_follower now requires autoware_planning_msgs
  -- Need to interface the recordreplay_planner(autoware_auto_planning_msgs) and trajectory_follower(autoware_planning_msgs)
  -- To resolve this, our trajectory_follower_f1tenth node is modified to subscribe to the trajectory in the old autoware_auto_planning_msgs format. The trajectory is then converted to the autoware_planning_msgs format before further processing

Autoware on F1tenth Future Plan

• The current recordreplay demo runs reduced Autoware stacks (Planning: recordreplay and Control: trajectory_follow). It also uses the F1tenth RViz UI so it's hard for others to tell it is running Autoware.
• Goal: Expand the usage of Autoware packages (especially in planning) and run the Planning simulation demo on the F1tenth
• Introduce mission planning and scenario planning so users can run more complicated scenarios (ex. Parking scenario)
• Utilizes the Autoware UI instead of the F1tenth UI
  
• Develop for two platforms: F1tenth with 2D Lidar and F1fifth(autoware training vehicle) with 3D Lidar
• High-level system architecture
  -- Sensing: utilizes the f1tenth_stack for sensor interface
  -- Localization: F1tenth - particle_filter for 2D Lidar localization, F1fifth - ndt_scan_matcher for 3D Lidar localization
  -- Perception: removed and not in the current scope
  -- Planning: mission planning(lane driving, parking), obstacle avoidance
  -- Control: trajectory_follower
• Because Autoware requires pointcloud and lanelet2 map to run and F1tenth only has a 2D Lidar, a tool will be developed to convert an F1tenth 2D occupancy map into a pointcloud map (offline), which is then used to generate the lanelet2 map
  -- For F1tenth the pointcloud map is mainly for vector(lanelet2) map generation and obstacle visualization in Rviz, not for localization
  -- Use simple image processing techniques on the occupancy map to get the contours of the obstacle blocks. Then, pointcloud can be generated by extruding the contours in the Z direction at a specific height and density

Other topics:

• AWF Work Group overview: https://autoware.org/join-a-work-group/