Autoware TSC meeting minutes for May 15, 2019

Autoware
1

May 15, 2019, 22:00 UTC

Attendees

  • Geoffrey Biggs (Tier IV)
  • Adrian Bedford (Parkopedia)
  • Joseph Buckner (AutonomousStuff)
  • Esteve Fernandez (Apex.AI)
  • Kenji Funaoka (Tier IV)
  • Servando German (Linaro)
  • Masaki Gondo (eSOL)
  • Brian Holt (Parkopedia)
  • Shinpei Kato (Tier IV)
  • Seonmaan Kim (LG)
  • Angelo Mastroberardino (Parkopedia)
  • Nikos Michalakis (TRI-AD)
  • Ryohsuke Mitsudome (Tier IV)
  • Dejan Pangercic (Apex.AI)
  • Antonis Skardasis (StreetDrone)
  • Daisuke Tanaka (Tier IV)
  • Simon Thompson (Tier IV)
  • Akihiko Tsukuda (eSOL)
  • Yang Zhang (Linaro)

Minutes Geoffrey Biggs (Tier IV)

Agenda

  1. Opening remarks and new member introductions (Board)
  2. Admin Confirmation of previous minutes (All)
  3. Notice Moving TSC home from Slack to Discourse (Geoff)
  4. Follow-up Action items from previous meeting (All)
  5. Project update Autoware.AI/Autoware.Auto (Apex.AI/Tier IV)
  6. Project update Map format (Parkopedia/Tier IV/Apex.AI)
  7. Project update Vehicle interface design (StreetDrone)
  8. Follow-up CI infrastructure (Apex.AI)
  9. Follow-up Quality control of Autoware.AI and problems with 1.11 (Apex.AI)
  10. Follow-up Silicon Valley meet-up (Apex.AI)
  11. Proposal Threat model for Autoware.Auto platform and use case (Apex.AI)
  12. Proposal Automating map and vehicle model imports with Parallel Domain (Apex.AI)
  13. Planning Milestones for development of Autoware.Auto (All)
  14. Presentation Arm and the ELISA foundation (Arm)
  15. Presentation Perception process unit hardware and software design (Autocore.AI)
  16. Presentation Apex.Autonomy vs Autoware.Auto (Apex.AI)

Action items

  • Close the Slack channels
    • Geoff
  • List planned contributions that go towards Foundation work and can be used to meet the milestone -demos. Place planned contributions in this topic.
    • LG
    • Kalray
    • AutonomousStuff
    • Velodyne
  • Move the Autoware.AI code to GitLab
    • Autoware maintainers team
  • Set up a meeting to talk about how to contribute to Autoware.Auto
    • Tier IV (Geoff)
  • Set up a meeting to discuss map format work
    • Tier IV (Geoff)
  • Set up a meeting to discuss vehicle interface work
    • StreetDrone
  • (Contingent on AWF board approving budget) Coordinate with TRI-AD to get CI going in AWS
    • Autoware maintainers team
  • Begin organising an integration verification team by discussing team make-up and specific tasks on Discourse and/or GitLab
    • Tier IV, AutonomousStuff, StreetDrone
  • Coordinate the threat modelling work of NDIAS and Alias Robotics
    • Tier IV, Apex.AI

Detailed minutes

Opening remarks and new member introductions (Board)

  • eSOL is a new premium member, and joining the TSC from this meeting. Akihiko Tsukuda will be their main representative on the TSC.

Confirmation of previous minutes (All)

  • Minutes approved

Moving TSC home from Slack to Discourse (Geoff)

  • As agreed in the previous TSC meeting, we will shift from Slack to Discourse for Foundation-related communication.
  • Permissions for the Autoware TSC category on Discourse have been mostly completed.
  • The Autoware TSC category on Discourse is accessible to the following people:
    • TSC representatives and their alternates
    • FTEs assigned by premium members
  • Some TSC members and FTEs have not yet sent their Discourse user names to Geoff.
    • Please do so immediately because otherwise you will not have access to the TSC materials from today!
  • The following categories on the Autoware Slack will be closed today:
    • #team-operation
    • #team-tsc
  • Please do not use the Autoware Slack for Foundation-related discussion as much as possible. Prefer to use Discourse, including both the Autoware TSC category and the Discourse private message function. Records are kept, notifications are sent by email and desktop, etc.

Follow-up: Action items from previous meeting (All)

  • Move TSC discussion place and AWF operations discussion place from Slack to Discourse
    • Tier IV (Geoff); will do in stages by the next meeting
    • Action taken: Category organised and ready for use; Slack channels will be closed when this meeting finishes.
  • Remove the public sub-categories of the public Autoware category at Discourse
    • Tier IV (Geoff)
    • Action taken: Sub-categories removed.
  • Update the support guidelines and bug/issue templates for Autoware.AI to use ROS Answers instead of Discourse for support discussion
    • Tier IV (Geoff)
    • Action taken: Guidelines updated and users beginning to use ROS Answers.
  • Arrange review of licenses for Unreal, Unity, CARLA and LGSVL
    • Apex.AI/Board (Yang)
    • Action taken: Will hear the results in a future meeting.
  • Provide differences between Apex.Autonomy and Autoware.Auto, and between Apex.OS and ROS 2, in next meeting
    • Apex.AI
    • Action taken: To be discussed in this meeting.
  • Review, discuss and improve milestones for Autoware.Auto development
    • All
    • Action taken: None
  • Coordinate with Tier IV and Kalray to bring their work on algorithm acceleration into the computing platform effort
    • Linaro (Yang)
    • Action taken: Tier IV is reaching out to eSOL and Xilinx (not a premium member), and trying to identify which algorithms can be offloaded to hardware-based implementations. A test ASIC chip that implements NDT, YOLO and Kalman filters is being targeted. A project is being started with montly meetings.
  • List planned contributions that go towards Foundation work and can be used to meet the milestone -demos. Place planned contributions in this topic.
    • Tier IV
    • LG
    • Kalray
    • Arm
    • Linaro
    • AutonomousStuff
    • TRI-AD
    • Autocore.ai
    • Huawei
    • Velodyne
    • Action taken: Autocore.AI, ARM, TRI-AD, Linaro and Tier IV have listed items.
  • Coordinate on prototyping work for Lanelets2 use in Autoware
    • Parkopedia, Tier IV
    • Action taken: Tier IV has done prototyping work (to be presented below). Parkopedia has modelled the AS carpark using OSM, and loaded up the XML map using Lanelets.
  • Talk to OSRF about using one of their GSoC slots for Autoware work
    • Apex.AI
    • Action taken: Unfortunately the OSRF does not have any spare slots. We will need to be more proactive next year, and in the mean time we can try applying for Outreachy in September. Revisit this topic in July/August.
  • Set up a teleconference to talk about the vehicle interface work for Autoware (especially Autoware.Auto)
    • StreetDrone
    • Action taken: Not yet held due to absence of necessary members. A new date will be arranged.

Project update: Autoware.AI/Autoware.Auto (Apex.AI/Tier IV)

  • Autoware.AI:
    • Work on splitting repositories continues. Should be completed by the end of next week.
    • Methods of adding the ROS 1 to ROS 2 bridge as a dependency are being explored. This is expected to be completed well in time for 1.12.
    • Feature freeze for 1.12 is on track to happen as scheduled on the 20th of May, giving us 6 weeks to test before release.
    • Tier IV is organising several days of test track time to test 1.12 before release.
  • Autoware.Auto:
    • Apex.AI has integrated the new architecture for pure pursuit.
    • Much cleaning up of ports from Apex.Autonomy to Autoware.Auto.
    • Apex.AI has implemented NDT matching for ROS 2 and this has been added to Autoware.Auto, but it needs cleaning up and re-design work.
    • Tier IV is reorganising the architecture of Autoware.AI, and wants to reflect that work in the architecture of Autoware.Auto.
  • Tool chain:
    • GitLab is mostly up and running for Autoware.Auto.
    • We would like to move Autoware.AI to GitLab as well to take advantage of the advanced tools that we have got access to for free. However there has been some discussion about whether this has properly been agreed to by the TSC.

Discussion

  • Some TSC members are not sure how to contribute to Autoware.Auto
    • Tier IV to set up a meeting to talk about contributing to Autoware.Auto for other TSC members.
  • Tool chain:
    • TRI-AD: Microsoft and GitHub have a roadmap for catching up to GitLab in features within the next year. We should try for a solution that allows to move back to GitHub when the features become available. TRI-AD proposes that we tell Microsoft why we are shifting to GitLab so that we might get access to these features sooner.
    • Tier IV: We cannot sit around and wait for a year to get tools we need.
  • Decision: TSC agrees to the plan to move Autoware.AI to GitLab and leave a pointer repository with the .repos file at GitHub.
    • Use a GitLab group called AutowareFoundation .
    • Use a GitLab sub-group called AutowareAuto that holds all the Autoware.Auto repositories.
    • Use a GitLab sub-group called AutowareAI that holds the Autoware.AI code repositories (shifted from GitHub).
    • Leave a repository at GitHub to hold the stars, hold the autoware_ai.repos file, and point new users at GitLab.

Project update: Map format (Parkopedia/Tier IV/Apex.AI)

  • Angelo from Parkopedia has been trying out the Lanelets2 library, and found it to be very versatile for things like adding traffic rules to junctions, etc.
    • Parkopedia has made a few pull requests to Lanelets2 as a way to get to know the Lanelet2 developers and introduce ourselves to them.
    • One of the PRs they have done is for projection from local coordinates of the lanelet to the coordinate system of the map.
    • The Lanelets2 developers did not accept the PRs but they copied the proposed code into the source anyway. They justified this as because they are currently doing a massive refactoring of the Lanelets2 code, so it is more likely that they were not sure how to manage the PRs properly in this situation than that they are closed to outside contributions.
  • Angelo will spend the next week looking at connecting Lanelets2 and Autoware.
  • Apex.AI wants Parkopedia to make it clear to the Lanelets2 developers that they are representing Autoware, a very large project, not just a single company or small project.
  • Parkopedia has modelled the AutonomousStuff carpark using the OpenStreetMap tools (JOSM).
    • They have found that this is probably something no one has done before, because there have been lots of warnings and errors in trying to use the tools.
    • They have an OSM file they can distribute now, but they cannot upload it to OSM because of all the errors and warnings.
    • This work is being done based on imagery from a proprietary image source, not on point cloud data from LG. LG has not yet provided the point cloud data to Parkopedia.
  • Parkopedia has been working on a converter from OSM to Lanelets2. It works on the carpark map. They want to compare hand-modelled maps done in JOSM and maps generated by their software.
  • Parkopedia wants the map format to be a topic for the upcoming Autoware Hackathon in the UK.
  • LG is working on the map of the AutonomousStuff carpark for the LGSVL simulator and creating a 3D environment for it.
    • They will deliver the map in time for the IV2019 workshop.
  • Tier IV has done a detailed investigation of the Lanelets2 format and a comparison with the current format used by Autoware.
  • Tier IV is interested in the organisation behind Lanelets2 becoming a member of the Autoware Foundation. At the least, the AWF should have someone who is an expert in or developer for Lanelets2.
    • Parkopedia thinks they are becoming this.

Project update: Vehicle interface design (StreetDrone)

  • A meeting was supposed to happen, but has been postponed to next week.

CI infrastructure (Apex.AI)

  • We need to decide what infrastructure we should use for CI for Autoware.
  • Cost for cloud CI infrastructure will be $300 to $400 per month because we need CUDA-compatible hardware.
  • We need to have the same CI infrastructure for Autoware.AI and Autoware.Auto to ease the moving of code.
  • Options are Amazon Web Services and Microsoft Azure.
    • TRI-AD votes for Amazon. All the work they are doing is for AWS, and they is transitioning to AWS because they had issues at scale when using Azure. TRI-AD also does not want to investigate a second cloud.
    • AWS is proactive about providing new types of runners, e.g. new GPUs and new operating systems.
    • TRI-AD has a good relationship with AWS.
  • TRI-AD has worked on GitLab CI so that they can work for Autoware.
    • If TRI-AD is provided with access to an Autoware AWS account then they can set up all the stuff we need.
  • Decision: The TSC agrees to use AWS. Board will need to vote to provide the money (next board meeting is next week).

Quality control of Autoware.AI and problems with 1.11 (Apex.AI)

  • Apex.AI tried to do a demo using Autoware.AI 1.11 following the documentation for the quick start, and it did not work.
    • They had problems compiling, with launch files that did not work and were not kept up to date, and NDT code crashing constantly.
  • Ultimately they had to give up using 1.11 and reverted to 1.10. But even that didn’t work out of the box (missing imports in Python code, broken configuration files, etc.) and required changes to get it going.
  • Other problems they encountered:
  • We need to ensure that Autoware.AI has at least one demo that always works out of the box, without needing to modify launch files, etc.
  • This situation of things being generally broken is unchanged from a year ago. This situation should not happen if we have a good product.

Discussion

  • We must have a rosbag-based demo that always works out of the box.
  • We should also have a demo that works on actual (specified) hardware out of the box.
  • We need to decide where to put the resources: adding tests and other quality control measures to Autoware.AI to improve its quality, or developing Autoware.Auto as a high-quality product from scratch.
  • Linaro: We do need to have a certain level of resources provided to maintain the quality of Autoware.AI during the transition phase to Autoware.Auto.
  • Tier IV: Even the PRs being provided for 1.12 mostly don’t have tests (and are too large to review properly).
  • Apex.AI: Currently Autoware.AI is so hard to use that only experts from Tier IV can use it properly.
  • Parkopedia: Some notes that it is actually hard to use Autoware might improve the situation.
    • Tier IV: It should still not be hard to at least run a demo with a given bag file or defined hardware setup.
  • Apex.AI: Should we have someone who’s job is to make sure the quick start and demo always work?
    • Tier IV: Perhaps Tier IV and AutonomousStuff and StreetDrone should work together on this?
    • AutonomousStuff: This is a good approach, the challenge is finding the resources to do this.
    • Tier IV proposes an integration team. Apex.AI wants the primary goal to make sure the rosbag demo works out of the box with less than 30 minutes’ work. Only then should the integration team move on to hardware.
  • Decision: As a start, Tier IV, AutonomousStuff and StreetDrone will set up an integration team. Tier IV to start a conversation about this on Discourse.

Proposal: Threat model for Autoware.Auto platform and use case (Apex.AI/Tier IV)

  • Recently the ROS 2 TSC’s security working group has done quite a bit of good work in developing a threat model for two robots:
    • The Turtlebot 3
    • Acutronic Robotics’ Mara modular robot. This work was done by a company called Alias Robotics from Spain. They are interested in doing a threat model for autonomous driving.
  • Tier IV is working with a company called NDIAS, which is a spin-off from Denso specialising in security for the automotive domain.
  • Proposal: A member company should undertake to develop a threat model for Autoware.Auto and its use case.
  • Proposal: Tier IV will work with NDIAS to produce a threat model for an autonomous car that the foundation can provide to ROS 2.
  • Decision: Tier IV and Apex.AI will do initial coordination to bring NDIAS and Alias together so they can work together to produce a single threat model.

Proposal: Automating map and vehicle model imports with Parallel Domain (Apex.AI)

  • A company called Parallel Domain has capabilities for automating the creation of simulation worlds and vehicle models.
  • Should we get in touch with them and see if they want to work with us?
  • Decision: Tier IV and the board are getting in touch with them.

Silicon Valley meet-up (Apex.AI)

  • Held at AS. Results posted on Discourse.
  • 40 to 45 people. Everyone was happy.
  • Apex.AI and AutonomousStuff intend to continue doing this.
  • Others are encouraged to have meet-ups in their area.

Milestones for development of Autoware.Auto (All)

  • We must choose milestones to direct our development efforts for the next generation Autoware.
  • See milestones topic.
  • Milestones have seen no discussion or input except from Tier IV and Apex.AI.
  • We must agree on at least the first two or three milestones today.
  • Decision: First two milestones accepted. Parkopedia in particular will contribute requirements to the milestones, but all TSC membners must do so.

Apex.Autonomy vs Autoware.Auto (Apex.AI)

https://gitlab.com/AutowareAuto/AutowareAuto/issues/82

  • Everyone read and we will discuss in the next meeting only if there are comments or questions that cannot be handled on Discourse.

Arm and the ELISA foundation (Arm)

  • Dropped due to time limits.

Perception process unit hardware and software design (Autocore.AI)

  • Dropped due to time limits.
1 Like