Autonomy Software WG Meeting 2022/05/31

[BonoloAWF]

Administrative

• Previous Meeting Minutes

Attendees

Chaired by Fatih

• Ahmet
• Ambroise
• Berkay K
• Beyza
• Bonolo
• David
• Esteve
• Fatih
• Hatem
• Haoru
• JianKang
• Goktug
• Kaan
• Maxime
• Mehmet
• Mert
• Phillip
• Piotr
• Rama
• Ryohsuke
• Sena
• Sugwan
• Shark
• Xinyu
• Yoshi

Common Resources

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

Announcements

Autoware.Universe

• Bus ODD issues
• Autoware.Universe reported bugs
• Autoware design discussions and demos

Discussion topics

Proposed by Hatem:

• I will present the current status of Open Planner architecture

Summary of Open Planner presentation (see slides in comments for full details):

• Architecture is a self contained functionality where the library implements all the functions. ROS nodes simply connect to the library and convert message types required.
• General map semantics like traffic lights are supported.
• Internal road network is the same as in Autoware.AI.
• Works with AISAN vector maps (.csv files), KML (.kml files), Lanelet2 (.osm files) and OpenDrive (.xodr file) formats
• Features global planning using distance in the cost function
• Human machine interface (HMI) bridge supports voice commands
• Obstacle avoidance, curb avoidance, yielding to others, stop sign and traffic light behavior is implemented
• Forward & backward simulation is used to navigate 90 degree turns
• Lane change is supported
• Probabilities are assigned to trajectories and behaviours
• Future development roadmap includes using Markov Decision Process instead of deterministic trajectory evaluation, overtaking behaviour, occlusion awareness, ROS 2 implementation of Open Planner 2.5+ to integrate with Autoware.Universe
• The formal structure of Open Planner development within AWF needs further discussion (as noted in the slides)

Fatih - for developing directly in Autoware.Universe start with a PR since the other packages do not depend on Open Planner. Compilation time of Autoware is the only concern. The maintainers will need to discuss this. The first goal is to get it running and we can improve iteratively. Github Discussion section can be used and the documentation will be updated.

Ryohsuke - adding Open Planner to Autoware calendar requires the TSC to first create a formal WG. If the WG is approved then meetings can be added to the calendar. We don't want to create a new WG for each algorithm, We should discuss further at the TSC meeting, it might be organised differently. David - Discord could be an option and running Zoom meetings from there. Hatem - there is already a Discord channel which we are using.

Ryohsuke - what is the difference between the current planning pipeline in Autoware and Open Planner? Hatem - this is outlined in the presentation slides but we need people to run bench-marking tests and provide feedback.

Sugwan - are you going to keep compatibility with Autoware.AI post migration? Hatem - Yes I will continue to maintain this in my own repository. Sugwan - there is a list of organisations using Open Planner, will they migrate to Autoware.Universe as well? Hatem - I will recommend moving to Universe once everything is working but this will take some time.

Proposed by Ryohsuke:

• I would like to get some feedback on the following issues/discussions:
  Migrate launch XML files to Python scripts autoware_launch#46
  Runtime Manager Sensing tab, update filter cmd for Synchronization bu… #371

Ryohsuke - converting to Python is not high priority but please raise any concerns. Fatih - all new launch files should be written in Python which offers greater flexibility but replacing existing launch files is not a priority. Esteve - I was working on reducing duplication in launch files (eg. vehicle characteristics appear in several files). To solve this issue in Autoware.Auto we wrote Python launch files so we can just apply the same solution. Ryohsuke - I don't think Python is required to use globally loaded params. Esteve - correct but the same issue was solved in Autoware.Auto using Python launch files. Ryohsuke - there are more features in Python but I'm not sure the timing is right to do the migration.

Fatih - launch files are not large so we should use Python. Each package could have its own launch file and these can be combined in the autoware launch repo. Ryohsuke - the sub-level launch system is because companies add different launch files and they only modify top level launch and can easily switch to different components provided by the args. Please add your comments to the issue for further discussion.

Esteve - is the consensus to write new launchfiles in Python? Ryohsuke - Yes.

Xinyu - we can keep another copy of launch and config files under launch repo. Fatih - if someone changes params of the behaviour path planner they need to update it in another repo. The top level launch in Universe calls single files and details are in the Universe launch files.

Action Items

1. @xmfcx to create an issue to test the existing lidar-only perception pipeline with the data provided Collect raw sensor data to test sensor drivers autoware.universe#562 (comment)
2. @hatem-darweesh to prepare presentation slide about OpenPlanner for the next ASWG call
3. @mitsudome-r to propose to ODD WG about adding parking scenario for Bus use cases if it is included in ITRI's course.

Review of Issues and Discussions

The following Discussions and Issues were highlighted:

• Runtime Manager, fix camera_id selection at Synchronization ON #369 - analysis tool for Autoware. Tools are useful to get info on execution time of each node but can't automatically match which topic corresponds to which node (this can be inferred from the graph manually)
• Runtime Manager, fix func overwrite at dialog close #214 - differential drive with Autoware will be added to the documentation with various use cases discussed.
• Runtime Manager Simulation tab, fix showing of rosbag pos remains on … #368 - The Board asked Ryohsuke to merge various repositories and organisations of Autoware. Please read the thread and post feedback by the end of this week.
• Sometimes the sensing module doesn't publish concatenated pointcloud autoware.universe#1009 - could be ROS related
• Vehicle stops ~1m further away than goal point or stop line autoware.universe#993 - steps to reproduce the bug are required

[mitsudome-r]

I would like to get some feedback on the following issues/discussions:

• Migrate launch XML files to Python scripts autoware_launch#46
• Runtime Manager Sensing tab, update filter cmd for Synchronization bu… #371

[xmfcx]

Finally created the issue to test the data collected by AutoCore.ai : autowarefoundation/autoware.universe#1008

[hatem-darweesh]

Today's Presentation introducing OpenPlanner and migration plans:
OpenPlanner_3.0_Presentation.pdf