Tuesday March 17, 2020 15:00
- Geoffrey Biggs (TSC chair)
- Cheng Chen (Autocore)
- Ian Colwell (AutonomouStuff)
- Adam Dabrowski (Robotec.AI)
- Kenji Funaoka (Tier IV)
- David van Geyn (AutonomouStuff)
- Brian Holt (Parkopedia)
- Craig Johnson (AutonomouStuff)
- Jeff Johnson (Mapless.AI)
- Seonman Kim (LG Electronics)
- Stefano Longo (Embotech)
- Kasper Meck (Arm)
- Ryosuke Mitsudome (Tier IV)
- Michał Niezgoda (Robotec.AI)
- Ali Ors (NXP)
- Tsutomu Otake (Macnica)
- Dejan Pangercic (Apex.AI)
- Philipp Robbel (Mapless.AI)
- Brian Shin (LG Electronics)
- Joe Speed (ADLINK)
- Stephane Strahm (Kalray)
- Manabu Tsukada (University of Tokyo)
- Akihiko Tsukuda (eSOL)
- Daisuke Tanaka (AWF)
- Miho Ueno (Tier IV)
- Josh Whitley (AWF)
- Yang Zhang (Linaro)
Minutes Geoffrey Biggs
- Opening remarks and new member introductions (Board)
- Confirmation of previous minutes (All)
- Follow-up Action items from previous meeting (All)
- Admin Election of Technical Steering Committee chair - Results
- Admin Scheduling of the AVP hackathon and demonstration
- Working group report Autonomy Software (@gbiggs, @esteve)
- Working group report Autonomy Hardware (@cheng.chen, @sstrahm)
- The Autoware Foundation’s strategy for the Autoware software (@gbiggs)
- Presentation of proposed architecture for Autoware from Tier IV (@kfunaoka)
- Kick-off of discussion of potential ODDs to set as the next goal after AVP (@gbiggs)
- Sensing architecture proposal (AutonomouStuff)
- Discuss if can provide a maintainer for Autoware.AI
- Prepare first pass of detailed plan for transitioning to Autoware Core/Universe and plan for the development of Autoware
- @gbiggs, @josh
Opening remarks and new member introductions (Board)
- New premium member: Robotec.AI
- New premium member: Mapless.AI
- New academic member: Tokyo University
Confirmation of previous minutes (All)
Action items from previous meeting (All)
- Announce suspension of Maps and Routing working group
- Provide thirty minutes minimum in the March TSC meeting for Tier IV to present their new
- Action taken Added to agenda
- Provide a detailed presentation on Tier IV’s new proposed architecture
- Action taken Provided
- Provide detailed architecture design documentation to the TSC for evaluation with the above presentation or as soon as possible after the March TSC meeting
- Action taken Provided
- Run an election for the TSC chair
- AWF board
- Action taken Election held.
Election of Technical Steering Committee chair - Results
- 1 nomination (Geoffrey Biggs, Tier IV)
- 13 votes:
- Geoffrey Biggs: 13
Scheduling of the AVP hackathon and demonstration
It will come as no surprise to most, but due to the covid-19 situation the AVP hackathon and demonstration planned for April in Palo Alto will be postponed.
At this stage we are looking at a minimum of 2 months’ delay. This will also impact the planned demonstration at IV2020. A new schedule will be set once the covid-19 situation’s resolution becomes clearer.
Working group report: Autonomy Software
Autoware.AI Development Update:
The merge request submission deadline for 1.14.0 has already passed (2020/03/01) and the merge deadline is fast-approaching (2020/04/01). The new features that we expect to see as part of this release include:
- The ability to use the CARLA Simulator with Autoware.AI
- The ability to command reverse gear
- The ability to use the
gnss_localizerwith UTM or other coordinates that are not relative to the Japanese Rectangular Plane system
- Several (<20) bug fixes
We will need help with on-vehicle testing after the merge deadline. 2 months are dedicated for this task (2020/04/01 - 2020/06/01) and the 1.14.0 release will happen on 2020/06/01.
Still looking for more active maintainers to do merge request reviews and approvals.
Milestone 2 has been significantly extended due to performance issues with NDT matching (lidar-based localization) and MPC (local path follower) as well as problems getting focus from the developers at LGSVL. The performance issues with NDT matching and the MPC are currently being worked on but some of the problems are also blocked by issues with the LGSVL simulator. Even if the AVP demo was not being delayed, I expect that we would still need additional time to complete this milestone as many of the features that are being targeted in this milestone are being written for the first time in an Autoware project with a focus on documentation, best-practices, and deterministic execution.
However, much progress has been made. A localization / transform architecture was agreed upon by the developers and the topic names and namespaces have been aligned to the new architecture. A PCD map (lidar point cloud) can now be published and visualized. Trajectories can be recorded and replayed in the simulation environment. A video of the current MPC performance with record-and-replay can be seen here.
Working group report: Autonomy Hardware
Based on last TSC agreement to proceed with proposed plan, we initiated the collection of hardware/system in used by members
The primary idea is to ‘sort’ them by ODD, in order to help adopters in their choice
This is the very first step
You can find the in-progress activity here: https://gitlab.com/autowarefoundation/autoware-foundation/-/wikis/Collection-of-Autoware-Members-Hardware-Configuration
NOTE we are still missing most probably inputs from other members: PLEASE contact us on Discourse to be listed
We will state end of this month (March) the collected hardware and go for the second phase: " Collect specification and user’s guide for completing « hardware recommendation »"
The Autoware Foundation’s strategy for the Autoware software (@gbiggs)
- Q: What is the distinction between Core and Universe, philosophically? What does “end-to-end” mean?
- A: It means that the Core must provide the autonomy capability for the ODDs that the Autoware Foundation claims to support.
- Q: Should we utilise the ROS build farm only for Core, not Universe? We could do a PPA for Universe instead.
- A: This is a valid point; managing keeping Universe building on the ROS build farm would be a large time sink.
- C: To achieve this, we need to have more than one software engineer working full time for the AWF. Just Josh is not enough. As a TSC, we can decide that we want to go in a certain direction, and then the board needs to figure out how to get us the resources to get there.
- That Autoware.AI be scheduled to go into maintenance mode (no new features added, only patch releases) at the end of 2020.
- That Autoware.AI receive only minor updates (1.15, 1.16, etc.) between now and the end of 2020.
- That Autoware.AI be scheduled for end-of-life (no new releases at all, no MRs accepted) at the end of 2022.
- Transition guidance for users will be important.
- An email vote on the above proposal will be held.
Presentation of proposed architecture for Autoware from Tier IV (@kfunaoka)
Autoware Architecture Proposal.pdf (1.6 MB)
- Q: What does “sample implementation” mean?
- A: We want to put out the code we have produced to demonstrate the feasibility of the architecture
- Q: What is the planned life time of that sample implementation?
- A: We don’t have a fixed plan for that, but we don’t want to give the impression that it’s a third Autoware or have a lot of users of it.
- Q: Is the planning part one of the major changes in the architecture?
- A: Yes, while all parts have received major changes, planning has changed the most.
- Q: For camera and LiDAR fusion, what proposals are you making to improve the vision-based perception?
- A: At a minimum the number of camera inputs has been increased; we use six cameras in our testing.
Sensing architecture proposal (AutonomouStuff)
- Skipped due to lack of time; will return to in next TSC meeting
Kick-off of discussion of potential ODDs to set as the next goal after AVP (@gbiggs)
- The TSC will need to decide what operational design domain to target once the AVP work is complete.
- With the delay to AVP caused by the coronavirus, this has become less of an urgent issue, but please start thinking about what you would like to see (in terms of technologial capability) from our next ODD.
- Remember that we cannot target an ODD that is too big a leap from our current one.