Autoware TSC meeting minutes for February 20th, 2019

The Autoware Foundation (AWF) Technical Steering Committee (TSC) Meeting #3

February 20, 2019

Round 1: 0am Tokyo (+1day), 3pm London, 7am California

  • Attendees:
    • Geoffrey Biggs (Tier IV)
    • Kenji Funaoka (Tier IV)
    • Shinpei Kato (Tier IV, AWF board)
    • Brian Holt (Parkopedia)
    • Victor Duan (Linaro)
    • Esteve Fernandez (Apex.AI)
    • Adrian Bedford (StreetDrone)
    • Antonis Skardasis (StreetDrone)
    • Filipe Rinaldi (Arm)
    • Yang Zhang (Linaro)
    • Dejan Pangercic (Apex.AI)
  • Minutes: Geoffrey Biggs (Tier IV)

Round 2: 10am Tokyo (+1day), 1am London (+1day), 5pm California

  • Attendees:
    • Geoffrey Biggs (Tier IV)
    • Antonis Skardasis (StreetDrone)
    • Kenji Funaoka (Tier IV)
    • Seonmamn Kim (LG)
    • Dmitry Zelenkovsky (LG)
    • Nikos Michalakis (TRI-AD)
    • Cheng Carmark
    • Aka Arthur
    • Alexander Patrikalakis (TRI-AD)
    • John Buszek (AutonomousStuff)
    • Shinpei Kato (Tier IV, AWF board)
  • Minutes: Geoffrey Biggs (Tier IV)

Agenda

  1. Opening remarks and new member introductions
  2. 3rd WICD - World Intelligent Driving Challenge (Linaro)
  3. Discourse categories
  4. Autoware.Auto proposal (Tier IV, Apex.AI)
  5. Autoware 2019 survey (Apex.AI)
  6. Replacement for Autoware Slack (Tier IV)
  7. TSC meeting time slot
  8. AutonomousStuff proposed meet-up in Silicon Valley
  9. Project updates
    9.1. Autoware.AI (Tier IV)
    9.2. LGSVL simulator (LG)
    9.3. Computing platform (Linaro)
  10. TRI-AD/Tier IV taxi project and City Pilot Project (Tier IV, TRI-AD)

Action items

  • Add a category to Discourse for discussion of map formats
    • Tier IV
  • Investigate the logistics of getting a car to China to do a demo at the WIDC
    • StreetDrone
  • Determine what a demo at WIDC will look like and report back to TSC
    • Tier IV
    • StreetDrone
  • Draft an announcement for closing the Autoware Google Groups mailing list and shifting discussion to the generic “Autoware” Discourse category
    • Tier IV
  • List the specific skills needed for Autoware.Auto development to aid in members finding people who can contribute
    • Tier IV
  • Investigate logistics and cost of safety-critical software development training
    • Apex.AI
  • Produce a proposal document for the computing platform project
    • Linaro

Opening remarks and new member introductions

  • The foundation is getting increasingly organised, but we still have room for improvement in our projects.
  • The TSC needs to think about how to manage and release and progress Autoware.AI, especially in the time before Autoware.Auto becomes a complete replacement.

3rd World Intelligent Driving Challenge (Linaro)

  • An event held in China by CATARC, which is a company wholly owned by the Chinese government. They have the authority over automotive standards in China, and selling a car in China requires certification from them.
  • Event to be held on May 14th to 17th, 2019 at the Meijiang Convention and Exhibition Center in Tianjin.
  • Deadline for registration is March 23rd. Each foundation member will need to register individually and get insurance, etc. on their own.
    • Despite the “individual” participation, we need to have a consistent message from foundation members.
  • This event reflects their ambition for China to be a leading player in autonomous driving in the world. This is why it is not just a China-focused challenge.
  • Companies and projects from China and abroad will be involved in the challenge, including:
    • Baidu/Apollo
    • Tesla
    • Audi
  • This is a good chance to improve recognition in China and improve relationships with the Chinese authorities related to autonomous driving.
  • The event itself is free but there are costs that must be covered by the participants, such as having insurance coverage.
  • Linaro would like to see the AWF join as a group, not just have a few members join individually.
  • Shinpei Kato has been invited to give a presentation at the event already.
  • If any foundation members are interested in joining the event, please let the foundation know.
  • Tier IV is proposing showing a car demo at the event; one proposal is to use a StreetDrone vehicle.
    • StreetDrone will discuss this idea.
    • Logistics of getting a vehicle there and doing a demo will be discussed offline.
    • Tier IV is considering taking a Tajima EV vehicle for a demo as well.
    • Multiple vehicles in the demo shows the flexibility of Autoware.
  • There are several categories for demos, such as geofenced driving, highway driving, urban driving, etc.
  • TRI-AD is still setting up their team so may not be able to contribute.
  • Will come back to members when the details of the demo are more concrete to find out what members can contribute.

Discourse categories (Tier IV)

  1. https://discourse.ros.org/c/autoware/simulator
  2. https://discourse.ros.org/c/autoware/computing-platform
  3. https://discourse.ros.org/c/autoware/city-pilot
  • These three categories are intended for public discussion related to the above three projects.
  • We need to encourage use of Discourse for discussion as much as possible, particularly for longer design-related and project-management-related discussions.
    • A good rule of thumb for if it should go on Discourse or in a chat is, if it is time critical then use chat, otherwise use Discourse.
  • The Google Groups should be closed in favour of the Discourse general Autoware category.
    • Groups will be closed to new members and new posts.
    • An announcement of the move to Discourse will be made in the Groups mailing list.
    • Geoff to write a draft announcement.
  • The foundation may consider in the future setting up an answers site for Autoware. This needs to be decided based on how many technical support style questions we get in the Discourse.

Autoware.Auto proposal (Tier IV, Apex.AI)

  • Goal: Develop an autonomous driving framework that is certifiable, and certify it to ISO 26262 (without killing the Autoware community).
  • Background:
    • We need a certified version of Autoware for use in products that require safety certification.
    • The work necessary to overcome the technical debt in Autoware.AI and certify it is greater than the work necessary to do it all again from scratch.
  • Proposal:
    • Develop Autoware.Auto from scratch following a certifiable development process.
    • Transition users from Autoware.AI to Autoware.Auto over time, not as a discontinuity.
    • Provide certification evidence to Autoware Foundation members for their use.
    • Maintain openness of the software and provide a non-certified version to allow a low barrier of entry for outside contributions.
  • The long-term vision:
    • Autoware.Auto, a certifiable, safe and secure autonomous driving framework with a controlled development process following safety-critical engineering practices by qualified developers, ready for use in autonomous driving products and services.
    • Autoware.AI, a branding of Autoware.Auto for use as a “sandbox” project by researchers and other contributors who just want to make the Next Cool Algorithm and contribute it to Autoware, and see it actually integrated and available for use without having to learn safety-critical software development.
  • Parkopedia supports the proposal, and wants to contribute what they have got so far. They have a number of safety engineers from a third party that they work with, and they can provide documents that can kick-start the certification documents.
    • Including method of operation documents, requirements documents, architecture designs, FMEAs, hazard and risk assessments, etc.
    • These documents will go up on the project website anyway, but they will provide it directly too.
    • Apart from their proprietary stuff, they are open by default.
  • Parkopedia’s proposed development timeline aligns well with the proposed Autoware.Auto milestones, so they are able to assist in the testing side of things.
  • Can Parkopedia provide a map of a carpark for use in the Autoware.Auto development effort?
  • Parkopedia: We need to think about how Autoware.Auto will interact with road-level maps, carpark maps, etc.
  • Tier IV: We may want to aim for supporting as many map formats as possible. If Autoware only supports a particular set of maps, this is a barrier for new users who might be using another format.
    • Geoff to add a Discourse category for map formats.
  • Parkopedia: The automotive industry has settled on NDS. It is a well-accepted standard that they use. If Autoware tries to support five different map formats, it becomes much harder to maintain as the formats evolve and possibly diverge.
  • Parkopedia is willing to take on responsibility for the map aspects of Autoware.Auto.
    • They are unsure if they should go straight for Autoware.Auto and aim for certifiable, or if they should go via Autoware.AI.
    • Tier IV recommends going into Autoware.Auto if possible because it will be easy to reuse in Autoware.AI, but Autoware.AI is not as easy to reuse in Autoware.Auto.
  • Linaro is concerned about the choice of OS and underlying hardware platform. A decision will be needed relatively soon. Apart from this, they support the proposal
    • Linaro will discuss in the next TSC meeting.
  • TRI-AD is going through some of the same issues that Autoware.Auto.
    • They are interested in being able to do this work in an agile way and hope to collaborate on that point.
    • They have QA teams but they are using old tools. They are trying to bridge that gap with new tools and techniques.
    • TRI-AD is interested in getting the safety-critical development training.
  • When Apex.AI finds out the details of the safety-critical training, we will forward this to foundation members.
  • TRI-AD is supportive of of the plan.
    • Tier IV to go talk to TRI-AD team in person to discuss contributions
  • TRI-AD thinks they can help on the data management side, such as moving data into and out of the car, e.g. maps.
    • TRI-AD to check if they have access to security experts who can help on the security side of things.
  • Antonis (StreetDrone) previously worked on autonomous forklifts so could help with a scenario in that area (when we get there).
  • LG has concerns about the toolchain.
    • We need to make sure everyone contributing uses the same toolchain.
    • Using Docker to provide a validated build environment will be important. Build on the work already done for the ADE by Apex.AI.

Autoware 2019 survey (Apex.AI)

Replacement for Autoware Slack (Tier IV)

  • To be discussed in the TSC Slack channel.
  • Nikos (TRI-AD) will be meeting the CEO of Slack soon, and can raise the concerns we have about using it for an open community.

TSC Meeting time slot

  • Preference is for a fixed time (no rotation)
  • Until daylight savings starts, use the following time:
    • San Francisco, USA 14:00 PST
    • Chicago, USA 16:00 CST
    • Washington DC, USA 17:00 EST
    • Berlin, Germany 23:00 CET
    • Tokyo, Japan 07:00 JST
    • UTC 22:00
  • After daylight savings starts, use one of two options:
    • Option 1:
      • San Francisco, USA 13:00 PST
      • Chicago, USA 15:00 CST
      • Washington DC, USA 16:00 EST
      • Berlin, Germany 22:00 CET
      • Tokyo, Japan 06:00 JST
      • UTC 21:00
    • Option 2:
      • San Francisco, USA 07:00 PST
      • Chicago, USA 09:00 CST
      • Washington DC, USA 10:00 EST
      • Berlin, Germany 16:00 CET
      • Tokyo, Japan 00:00 JST
      • UTC 15:00

AutonomousStuff proposed meet-up in Silicon Valley

  • Not discussed

Project updates

Autoware.AI

  • From a perspective of functionality, Autoware.AI is improving significantly.
  • Release 1.11:
    • Contains many new features.
    • In particular, enables much-needed capabilities such as backwards driving.
    • A detailed feature list can be found in “Autoware.AI Plan - TSC 20th Feb…pdf” in the TSC meeting #3 folder.
  • Release 1.12 will:
    • focus on the package and repository reorganisation,
    • introduce a Kanban task management tool,
    • switch to using vcstools for the installation process, and
    • add the ROS 1/ROS 2 bridge as a dependency and add it to launch scripts.
  • Release 1.13 will:
    • focus on architecture reorganisation, splitting node functionality out into libraries, and conversion to ROS 2 of some packages,
    • add Autoware.Auto as a dependency, and
    • remove the existing Velodyne driver in favour of the high-quality driver in Autoware.Auto.
  • Release 1.14 will:
    • focus on conversion of packages and nodes to ROS 2, and
    • focus on adding tests made available by use of ROS 2 tools.
  • The release cadence will be set to three months.
    • Firm due dates for feature freeze, code freeze, etc. will be set with enough time between each to allow for testing and verification.

LGSVL simulator

  • A simulator made with Unity.
  • Focuses on simulating the sensors, including LIDAR, radar, various cameras, IMU, and GPS.
    • The LIDAR simulation is GPU-powered so it can generate up to 128 beams in real time.
    • The simulator can be used through rosbridge so that sensor information can be used in ROS and viewed in rviz.
  • The use of Unity means that it is very easy to modify the environment. Their recent efforts are focusing on decoupling simulation from environment definition.
  • They are also working on allowing creating a simulator environment from data collected from the real world, i.e. drive around and record LIDAR and camera data and use that to make a simulation environment of the route.
  • They are working on an environment that covers San Francisco.
  • Running a single simulation over multiple computing nodes in parallel is a goal.
  • The simulator has better visuals than other simulators, because this is where they are putting their efforts due to the focus on sensors.
  • They are working on an API for remotely managing and interacting with the simulation.
  • Their release model is:
    • Push code to GitHub weekly.
    • Make tested binary builds monthly.
  • They also provide a tool to do map annotation in the simulation (e.g. “this is a footpath”), and that annotation can be exported to an Autoware map.
  • The LGSVL simulator and the CARLA simulator have their different strengths. We need to clarify these so we can ensure each simulator is used in an appropriate way.
  • Tier IV, LG and TRI to discuss simulators in-person in early March.

Computing platform

  • Linaro is working on the proposal document for the computing platform project.
  • It will be released soon in the Discourse category for open discussion.
  • Apex.AI is finding it difficult to work with NVIDIA because they are not a large OEM, so if this project can provide an alternative computing platform, that is an urgent need now.
    • Linaro feels that this is somewhere the foundation can make a difference because they feel a lack of a readily-available heterogeneous computing platform is becoming very limiting.
    • Tier IV has given up trying to run QNX on NVIDIA’s computing platform because of a lack of support.
  • Tier IV also has contact with Renesas, but has had problems there, too.

TRI-AD/Tier IV taxi project and City Pilot Project

  • TRI-AD is still looking at what the MVP is for their demo project.
  • TRI-AD wants to look at what the requirements are on a vehicle from Toyota, e.g. the Lexus that AutonomousStuff produced.
  • Tier IV to share more information in the next TSC meeting after doing work in USA
  • Tier IV is working with the Jpn Taxi model, so demos in Japan may be able to use a real taxi vehicle. Tier IV is working with AutonomousStuff to make modifications to this vehicle.
    • The goal is to have an Autoware-controlled one by July.
    • Tier IV proposes doing a demo with TRI-AD under the foundation umbrella.
    • TRI-AD and Tier IV agree on using the Jpn Taxi vehicle.
    • TRI-AD needs to make sure that whatever AutonomousStuff is doing is approved by Toyota, because the Toyota name will be visible in the demo and so it needs to not have a bad impact on the brand. It is possible that Toyota may require modifications be done by their own accepted shop than by AutonomousStuff.
  • It is desirable for the LGSVL simulator to be used for the demos.
2 Likes

Ack this.

@Brian_Holt would you be open to team up with some other companies that AWF members have contact to (Carmera, Mandli, Atlatec) to come up with the format or with the abstraction layer for Autoware? I think that the mapping and AD industry really, really needs and now it is the time to do it.

Awesome. Please have a look at https://autowareauto.gitlab.io/AutowareAuto/installation-and-development.html and request if changes are needed.

Please also pull @esteve into this. Him and I have so far looked into Gazebo, LGSVL and he will also talk to Carla folks in the next days. So he will have a really good comparison of the 3 best open source simulators.

@Nikos would you guys also consider an AVP as an MVP: AutowareAuto / AutowareAuto · GitLab? This way we would all work towards the same MVP.

@Buszek
I am actually still not 100% sure. Is CityPilot an AWF project or an AutonomousStuff project? As far as I can understand it is the latter. If the latter - why are we discussing it here? Also, in the spirit of “the strength of the wolf is the pack” it would be great to work on the SAME use cases/demos/MVPs.

What do you guys think?

Ping Shinpei and find out when they are meeting. I imagine Esteve’s experience will be welcome.

I’m not going to speak for TRI-AD’s intentions, but a relevant piece of information is that valet parking is not a big thing in Japan.

It’s not a big thing in Europe either, but AVP is an excellent usecase for Autoware.Auto (and Autoware.AI):

  • Not on public roads, so while safety is still important, we are minimizing risks, and permits and insurance are hopefully easier to obtain.
  • Low speeds.
  • Maps for the most part don’t change.
  • Very few pedestrians.
  • No cyclists.
  • We have partners within the foundation that are already working on AVP as a product/service (e.g. Parkopedia), so we can leverage their expertise.
  • Simpler to simulate.
  • It is a useful and real usecase, it’s not a synthetic demo that won’t have any kind of impact.

@Brian_Holt wrote a great article detailing why AVP is the most likely self-driving technology to be deployed in real cars in the short/medium term Why Autonomous Valet Parking? - AVP

@Brian_Holt would you be open to team up with some other companies that AWF members have contact to (Carmera, Mandli, Atlatec) to come up with the format or with the abstraction layer for Autoware? I think that the mapping and AD industry really, really needs and now it is the time to do it.

@Dejan_Pangercic I’d be happy to discuss maps with other providers, it makes sense for us to try to develop a common standard that Autoware can use.

That all sounds good to me.

Having read that, I can see more of a use case for it in Japan too, now. Outside of the centre of major urban areas like Tokyo, Nagoya and Osaka, there are enough large shopping centres where not having to park your car itself would be nice. Even the centre of a reasonably large city like Hamamatsu has enough large carparks that would benefit from such a feature.

@Brian_Holt, given you wrote that article, I’m interested in your thoughts on this opinion piece.

Please do so and let us know how you get on. I get the feeling that this is one thing we need to move rapidly on. At the very least the AWF needs to be involved in any discussion that happens.