Open AD Kit WG Meeting 2022/06/30

[ghost]

Administrative

• Previous meeting minutes
• Open AD Kit project board
• Open AD Kit 2.0 schedule

Attendees

Chaired by Armağan

• Armağan (Leodrive)
• Bonolo M (AWF)
• CJ (AWF)
• Chase Cao (PIX Moving)
• Daisuke Hashimoto (T4)
• David W (A3VG)
• Eiji Sekiya (T4)
• Fatih (AWF)
• Fumihito Ito (T4)
• Hiroshi Igata (T4)
• Lalith Vipulananthan (T4)
• Mark Lambe (KLAS)
• Matt Spencer (ARM)
• Nancy Lee (PIX Moving)
• Naoyori Tanzawa (T4)
• Ryohsuke Mitsudome (T4)
• Stephen Li (Autocore)
• Tsigigenet Dessalgn (Marvell)
• Yigit Cagatay Kuyu (Karsan)

Agenda

• Introduction of new members
• Announcements
• Review of backlog actions
• Discussion topics
  • PIX Moving presentation
  • Foxglove Studio
  • MIH demo plan
  • Data Collection (GDPR)

Introduction of new attendees/members

• Matt Spencer (ARM), returning to the WG to cover for Kasper
• Nancy Lee (PIX Moving)
• Yigit (Karsan), working as an engineer at Karsan (global path planning) and a researcher

Announcements

• No announcements.

Review of backlog actions / open issues

• Integrate Autoware Universe with Open AD Kit and review the results
  • Verify that Autoware Universe can be run with Open AD Kit 1.0 autoware-projects#32
• Document Autoware Universe ROS nodes
  • https://github.com/autowarefoundation/autoware.universe/issues/827
  • Autoware documentation PR submitted: feat(how-to-guides): add dependencies how to autoware-documentation#162
  • If anyone has any comments or suggestions for improvement, then please do comment on the PR!
• Create an autoware.org meeting invitation group for the Open AD Kit WG and add it to each meeting as an invitee
  • New group created: https://groups.google.com/a/autoware.org/g/open-ad-kit-working-group-meeting-invitations/
  • Invitation group is not quite configured correctly -> currently "debugging"
• Create separate containers for testing Open AD Kit #2454
  • Questions from Fatih were passed on to ARM and answered
  • Fatih noted that he is still working on this and will need another couple of weeks to finish

Discussion topics

PIX Moving presentation

PIX Moving Intro for AWF 20220701.pdf

Questions

• Armağan: Any issues updating multiple vehicles?
  • Nancy: All platforms use Autoware (80% of vehicles), but very basic functionality (entry-level). Customers then add their own algorithms and hardware on top.
  • Armağan: This group is aiming to develop better solutions in the future so that PIX could potential offer robots as a service instead of just a basic platform
  • Nancy: Trying to determine which scenario has the most potential out of an autonomous bus or robot delivery service.
• Lalith: Have you used any other versions of Autoware beyond Autoware.AI?
  • Nancy: Only Autoware.AI at this time. Have developed ROS 2 drivers for use with Autoware.Auto
• Fatih: What computer do you have in the chassis?
  • Nancy: Depends on the platform size. For the Hassenburg (half a metre in length), using the NVIDIA Xavier. For the Hoop or larger platforms, using a Neosys platform.
  • Fatih:
• Armağan: PIX Moving has been added to the Autoware supported hardware list, so it would be good if PIX can check and verify the addition.

MIH demo plan

• Armağan: Any questions/comments from Fatih?
  • Fatih noted that he will work with Daniel for the demo plan, and is currently working on testing Autoware in two containers on Ubuntu. He is not focusing on the EWAOL side of things since he does have the necessary hardware (AVA or Autocore PCU), and wants to focus on his existing software WG tasks
  • David noted that a Jetson AGX Orin should also work (as this is something Leodrive have on order), and was confirmed by Matt Spencer who said that any GPU should work with EWAOL as long as there is an ARM driver available (details available from Kasper, or Ben in the UK in Kasper's absence).
  • In theory, testing on EWAOL supported machines will be carried out by Daniel at National Taiwan University and Kasper on his return
• Armağan: Any issues from the Open AD Kit 1.0 and Universe?
  • Ito-san (T4) is waiting for an ARM Autoware image and an x86 image for the Scenario Simulator (the latter needs to come from T4)
  • Fatih added a link to the latest ARM image on the issue Verify that Autoware Universe can be run with Open AD Kit 1.0 autoware-projects#32
  • ACTION: Lalith to confirm who will be providing the x86 image and if there are any blockages

Foxglove Studio

• Open source visualization and debugging tool for robotics
• David noted that Foxglove Studio looks really interesting and shows the value of a nice HMI/GUI. Does it have utility, is there anyone who is interested in using this? Should we invite them?
• Fatih briefly described the tool:
  • Foxglove Studio has two versions, a desktop app and a web app running off the same core code
  • The desktop app would be used by Autoware mostly, since the web app cannot be used due to DDS limitations
  • It can be connected directly to ROS 2 (one-way communication, can publish topics), but rosbridge can be used to communicate with outside things
  • If we can improve the communication to enable bi-directional communication, then this can be a potential replacement for RViz
    • Can add various panels, buttons to run ROS commands (like publish etc)
  • Their support is good, as he created a small issue and it was fixed quickly.
• Armağan asked if anyone in the WG is in contact with Foxglove (developers or managers) but no one responded. He then asked CJ about how affiliated organizations are added to the AWF
  • Write a high level MOU (Memorandum Of Understanding) to describe what we would hope to accomplish with them and then go from there. If the working groups think this is a better tool than RViz, and there is value in going forward with this, then we can discuss at the next TSC/SPC meeting.
  • Mitsudome-san noted that Foxglove Studio is just a tool that we can use and that we should define the scope of the app that we want to develop using Foxglove -> this could then be included in the MOU.
  • Fatih noted that Foxglove uses TypeScript and that we do not have Autoware developers with that experience. A member company might help if they have such a developer, or if they want a GUI for their own projects and can assist?
  • CJ said that we are not doing much of the development work for eSync (it's mostly being handled by eSync and Excelfore). The alliance is about promoting their solution in the context of the Autoware ecosystem, so perhaps Foxglove could be like that. We ask them to deliver something to us and see what happens. We can tie it in and promote through the CoE network, and may have downstream revenue opportunities for them.
• ACTION: Create a discussion to gather opinions from the Autoware community
• ACTION: Put together an outline to discuss at the next SPC meeting on July 5th/6th

AOB

• Lalith talked about the updates to the Open AD Kit documentation site
  • Armağan commented that having a framework in place would make it easier to create and assign tasks to fill in the missing parts
• CJ asked whether the Logging Simulator (Rosbag Replay Simulator) can be used to test Perception? It works with Web.Auto but can be it used standalone as well (both are possible with the Scenario Simulator)? If we are doing a container update for the demo, it may not be obvious with RViz to show that there has been a performance improvement, but maybe we can show this via Logging Simulator?
  • David: If you were looking at a live cam situation, and there was a bicycle and it didn't show a bounding box around it first time, but did after updating the container.
  • CJ: ConOps/OpsCon talked about incremental improvements in Perception. Do we need a simulation tool to achieve that?
  • Fatih: To show improvement in the perception pipeline, don't need simulators. Output of perception is just bounding boxes and objects being classified. If you can see jumping boxes, or the ground is not removed correctly, you can see the problem obviously and then how that problem is removed with a new container.
  • Mitsudome: Logging Simulator is just a validation tool. If you just want to show that bounding boxes have been improved, then it is needed.
  • CJ: An actual developer will use Logging Simulator to evaluate at a higher-grained level, but for demos we don't need it.
• David noted that MIH demo stretch goal was to use the MIH/Foxconn chassis and Lidar. However, due to space constraints, we cannot use the chassis for the demo, but would a chassis be used for future events?
  • CJ: The chassis will be shown at the October event (v0.5) but the problem is that there are not many chassis available! The AD working group were told they would only have a 2 week window for integration/debugging with the chassis for the demo, which wasn't feasible. The focus will be a static demo (just desktop machines) showing the dev process
  • <Some more stuff between David and CJ, something about smart cameras?>

Actions

• Lalith to confirm who will be providing the x86 image to Ito-san for his Open AD Kit 1.0 test with Universe and if there are any blockages
• Lalith to create a discussion to gather opinions from the Autoware community about Foxglove Studio
• David/Armağan to put together an outline about Foxglove Studio to discuss at the next SPC meeting on July 5th/6th