Open AD Kit WG Meeting 2022/04/07

[ghost]

Administrative

Previous meeting minutes

Attendees

Chaired by Lalith

• Armağan Arslan (LeoDrive.ai)
• Christophe Le Dorze (SUSE)
• Daisuke Hashimoto (T4)
• David W (A3VG)
• Eiji Sekiya (T4)
• Fumihito Ito (T4)
• Hiroshi Igata (T4)
• Kei Kato (T4)
• Lalith Vipulananthan (T4)
• Naoyori Tanzawa (T4)
• Ryohsuke Mitsudome (T4)
• Yoshiya Fujii (T4)

Agenda

• Introduction of new members
• Review of previous meeting's topics
• Announcements
• Review of backlog actions
• Review of open discussions
  • Relationships between artifacts in AWF working groups
  • Open AD Kit v2.0 scope
    • Open AD Kit draft schedule
    • Microservices based architecture design for Autoware
• Discussion topics
• Q&A follow-up
  • SUSE
  • eSync

Introduction of new members

• No new members

Review of previous meeting's topics

• Skipped, since the same topics were planned for discussion in this meeting

Announcements

• None

Review of backlog actions

• None

Review of open discussions

Relationships between artifacts in AWF working groups

• Armağan noted that this proposal is an excellent one for setting high-level requirements for projects
• A closing date for comments on this proposal has been set for April 14th. If no comments or concerns have been received by this time, then the proposal will be formally accepted by the working group.

Open AD Kit 2.0 scope

• Scope is currently being based on the proposal from Hashimoto as the co-chairs think that is an excellent basis for 2.0 scope
  • Deadline for comments is set for April 13th
  • If there are no concerns or questions from the working group by that time, then the proposal will be accepted and we will start considering use cases
• Armağan said that we need to determine how to achieve the goals based on the ConOps proposed by Hashimoto-san
  • "Playback field data" seems to be the easiest one to achieve, and also fundamental to AD system development
  • For example, if there is a localization failure with no recording, then the test must be re-run to try to recreate the behaviour -> we need an automatic recording system
• We need to set objectives for this use case
  • How can the necessary sensor data be recorded?
  • What is the necessary sensor data for the ITRI bus project?
  • Rather, what is the minimum data set required to test the autonomous data stack. Processed image data or combined pointcloud data may not be necessary (since processed data can be processed again during playback), but raw LiDAR data should be available etc
  • data could be the last hour (if data size is too big, then remove camera data)
  • Tesla demonstration videos show that users can tap a button on the UI to upload log data when an error occurs
• Recorded data is also fundamental to the other OpsCons proposed for Open AD Kit 2.0
  • Enhance the perception under MLOPs -> how can we do this without recorded data?
  • OTA updates -> how can this be achieved without two-way communication?
• Any comments/objections from T4?
  • Does T4 already have a solution for this can be made available for OSS?

Questions/concerns:

• David: Is there is any existing functionality available to record/playback data?
  • Mitsudome: For Universe, there is an ongoing discussion around porting the "record/replay planner" package from Autoware.Auto (similar to waypoint recorded in Autoware.AI) that can record and replay trajectory.
  • Armağan: We can port it but it does not satisfy the condition to record driving data.
  • Mitsudome: If we think about Open AD Kit as interacting with different systems, we need something that can record and replay from the planner as a starting point.
  • Armağan: Perhaps we can create an issue to discuss and decide which data should be recorded at a minimum (some data is processed and so doesn't need to be recorded, but raw data must be recorded) and decide on the recording architecture. Open AD Kit can then consider recording data and uploading online
• David: Does T4 concur with the idea of starting with data recording, or if anything else should be considered?
  • Mitsudome: What is the purpose of recording the data?
  • Armagan: First of all, ConOps was reviewed by the chairs and we think it is a great idea, but data collection needs to be first. In his experience, wasted a lot of time recreating tests that failed due to lack of recorded data (which is not always possible) -> needs to be recorded and uploaded to test the complete AD stack.
  • Mitsudome: I have some concerns (rosbags are not deterministic, so same error may not occur when replaying the data) but will discuss more in the issue.
• Lalith: Any other questions or concerns about the scope described by the ConOps?
  • Armağan: In the short term, we have a camera pipeline but there are GDPR concerns about privacy so we need a way to anonymise recorded camera data (automatic blurring of faces and license plates via ML). One approach is to find a partner who can do this and invite them to the AWF. To start, the "Enhance the perception under MLOPs" OpsCon can be achieved with just LiDAR data but what about camera data?
  • Even for LiDAR data, we need a way of labelling the data (so will need a partner for that as well).
  • All of the goals on the ConOps are great, but need to consider and solve the related issues.
• Action item: Set up a short sync meeting with the WG co-chairs plus Mitsudome and Fatih to discuss requirements for data recording
• Action item: Create a Design discussion with a list of requirements from the Open AD Kit WG and loop in the software team (Mitsudome and Fatih) so that they can voice any concerns (eg: rosbags are not deterministic, so same error may not occur when replaying the data)

Open AD Kit 2.0 draft schedule

• Shared draft schedule for Open AD Kit 2.0, noting that it is very rough and subject to change
• Regarding map data, ITRI have provided map data for their public bus route, but if the bus demo will be done within the private ITRI campus, should ITRI provide that map data as well?
• Mitsudome agreed that we should ask ITRI to provide the private campus map data, and that we can ask directly in the next ODD WG meeting. Armağan noted that in the worst case, we can recreate the map from sensor if we have to, but we can cross that bridge later.
• Action item: Ask ITRI in the next ODD meeting if they are able to provide their private test campus map data

Microservices based architecture design for Autoware

• T4 proposes that the WG should first agree on the scope of Open AD Kit 2.0 (currently being discussed), create detailed use cases based on that scope and then determine which technology is best suited to achieving those use cases.

Discussion topics

Q&A follow-up

SUSE

• David: Going forward, thoughts on a demo leveraging SUSE Linux for Open AD Kit 2.0? How do we fold SUSE in at some point?
• Christophe: Andreas wrote some detail in the discussion. Where should effort be focused? Micro-services and Docker container could be remade in a new container image (but not interesting, since if it already works in one container then why remake it?), but could investigate orchestration via K3s rather than Docker Swarm.

eSync

• Armağan: What is the concept of the alliance? Will they make eSync compatible with Autoware? Will they provide FTEs for Autoware? If so, which WG?
• Lalith: Based on Excelfore's presentation, they are working on a PoC of eSync that will be compatible with Autoware.
• Armağan: eSync is not OSS so what license are they providing? If we are working with a bus manufacturer who is using Autoware, but not an AWF member, can we still use eSync?
• Action item: Ask Excelfore to clarify what "license for use in development and demonstration" actually means.

Action Items

• Arrange a short sync meeting to get the ball rolling on Open AD Kit requirements for data recording @LalithVipulananthan - due 04/08
• Create a discussion for the Open AD Kit WG to discuss data recording requirements with the Software WG (eg: which topics are required for a minimum data set) @LalithVipulananthan - Pending sync meeting scheduled for 04/11
• Ask ITRI if they are willing to provide map data for their private campus @Igata-ctrl - due 04/15
• Ask Excelfore to clarify the license that will be provided with the eSync Autoware PoC @LalithVipulananthan - due 04/08