Autonomy Software WG Meeting 2022/08/30

[BonoloAWF]

Administrative

• Previous Meeting Minutes

Common Resources

• Bus ODD software architecture proposal
• Bus ODD architecture design discussions
• Bus ODD candidate simulator comparisons
• Bus ODD milestone development plan
• Bus ODD Project Board
• Autoware contributors list

Announcements

Phillip will be presenting on TUM planning modules at the meeting next week.

Autoware.Universe

• Bus ODD July-Aug Milestone
• PRs waiting for review
• Bus ODD high priority issues not in the current milestone
• Autoware.Universe reported bugs
• Autoware design discussions and demos

Discussion topics

Enabling the code-owners feature in AWF repositories

• See @kenji-miyake comments below
• Proposal to assign a code owner for each package (which tracks the maintainers). When a pull request is created against that package, the correct reviewer can be assigned.
• Package.xml file of each package should have a valid username and email address.
• The CODEOWNERS files should not be edited directly (update package.xml and CI workflow will automatically update the CODEOWNERS file).

Bus ODD progress updates:

ITRI bus integration update (In meetings with 7am UTC time slot)

• See slides in comments for details
• On-vehicle integration is 90% complete and on-track for completion by end of next week
• DriveNet data is working, LidarNet input is ongoing work
• Ouster sensors are fully integrated, TF conversion node for lidar is ready
• Trimble hardware is fully integrated, localisation node will be tested when the vehicle is on the road
• Bus stop functionality was fixed

LeoDrive bus integration update

• no update (public holiday in Turkey)

Action Items

• @BonoloAWF to set up a call between AWF and Arm to finalise TVM support in Autoware discussion Run Lidar Centerpoint with TVM autoware.universe#908 and Add TVM inference capability autoware.universe#628

Review of Issues and Discussions

The following Discussions and Issues were highlighted (see comments for details):
Issues:

• Autoware keeps following old local path after behavior_velocity_planner is dead autoware.universe#1253 - ongoing
• Image messages have latency or dropped autoware.universe#1228 - bag files converted, Fatih will test tomorrow
• Run Lidar Centerpoint with TVM autoware.universe#908 - moved to next milestone
• Create speed_bump behavior module autoware.universe#611 - working on draft PR
• Motion_velocity_smoother should depend the steering_rate_limit autoware.universe#1023 - ongoing PRs
• Evaluate performance of object detection pipeline autoware.universe#565 - waiting for reviewer comments
• Finalize the field type for pointcloud autoware.universe#563 - ongoing documentation additions
• Implement sensor drivers for ROS 2 autoware.universe#551 - ongoing work to provide drivers
• Sometimes the sensing module doesn't publish concatenated pointcloud autoware.universe#1009 - multi-thread nodes referring to the same library could be the cause
• Ground segmentation fails for points behind the low objects autoware.universe#669 - no updates
• Positioning error in wrong zone when NDT starts autoware.universe#1421 - ready to close
• StepPolygon of obstacle_stop_planner is instable when changing the params of avoidance_module autoware.universe#1499 - PR awaiting review
• Checking only 4 corners in the drivable area autoware.universe#1716 - no updates
• planner node dies inside RCLCPP_COMPONENTS_REGISTER_NODE macro autoware.universe#1389 - check if there is a public avenue for testing
• Adding NDT reinitalization statement and reinitialization while moving autoware.universe#1375 - ongoing
• planning mode doesn't change if lane change path can't be executed in behavior path planner autoware.universe#1328 - this was reproduced, data needs to be shared
• Can not give a new goal behind in the same lane autoware.universe#1715 - known issue
• Drivable area from behavior_path_planner is not expanded in this case. autoware.universe#1702 - bug is fixed
• tensorrt_yolo sample yolov5 model throw error on inference autoware.universe#1647 - possible fix outlined

Discussions:

• Lanelet map convention for static obstacles #2826 - extracting physically non traversal borders for planning
• [proposal] A new pointcloud map interface for dynamic map loading #2812 - proposal for supporting dynamic map loading
• Avoid obstacles near the center line in behavior_path_planner #2799 - more discussion needed
• Performance Analysis Report with CARET #2822 - there will be an upcoming presentation to present these results
• CARLA Simulator + Autoware Universe Tutorials #2828 - integrating Carla with Autoware

[kenji-miyake]

@BonoloAWF @xmfcx @mitsudome-r
Hi, I'd like to enable the code-owners feature in AWF repositories.
https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/customizing-your-repository/about-code-owners

The feature can automatically assign appropriate reviewers based on the settings in the CODEOWNERS file. I believe it can make reviews more efficient.
For more detailed behaviors, please refer to this comment: autowarefoundation/autoware.universe#1619 (comment)

The CODEOWNERS file is automatically updated using the maintainer tags information in package.xml like this.
https://github.com/autowarefoundation/autoware.universe/pull/1671/files

[kenji-miyake]

Based on this proposal, I have some requests for the maintainers.

• Please keep the maintainer tags of your package up to date.
• Please link your email address that is used in package.xml files to your GitHub account.
  • If it's not linked, the CODEOWNERS file raises errors.
    
  • You can link multiple email addresses, so you don't have to remove your personal email.
    

[hueirutseng]

ITRI Bus Project Update: Presentation

[mitsudome-r]

I have discussed with @ambroise-arm and AWF software architects regarding TVM.
Here's the slides with the minutes: TVM Status.pdf

Here's the summary:

• Although we have decided to use TVM as inference engine in Autoware in the past discussion, we will still make it as a "recommended" inference engine until we have enough performance evaluated in Autoware and also have enough developers who can write/review TVM packages.
• Also, we would have to train more developers who can use TVM. As a starting point, @angry-crab will try to learn it. @mitsudome-r will also check CoE if there are anyone who is interested in learning the library as well.