Planning&Control WG Meeting 2023/09/14

[mehmetdogru]

Administrative

• Meeting link
• Goals
• Previous meeting notes

Announcements

• xxx

Updates

• (@mehmetdogru) Comprehensive AWSIM tests has been made by @xmfcx.
  • Multi PC AWSIM + Autoware Tests
  • Related to AWSIM feedback
• (@mehmetdogru) Is there any plan behind visualizing curbstones? [Created by @saka1-s]

Discussion Topics

• (@mehmetdogru) Why use surround_obstacle_checker? [Created by @xmfcx]
• (@armaganarsln): He doesn't like current control architecture in general. Delay compensation, slope compensation, usage of HD-Map in longitudinal control is not proper.
  • (@brkay54): He has some plans to fix the problems.
  • (@armaganarsln): Requests to be discussed overall control architecture.
  • (@TakaHoribe): We are supporting 2 options: IMU pitch and HD-Map slope info.
  • (@mitsudome-r): It makes sense that we have 2 alternatives. Why not using IMU sensor input?
  • (@brkay54): Delay and slope compensation is not working well with IMU sensor input. Need the future(expected) slope info.
  • (@mitsudome-r): Is control really that sensitive to slope change?
  • (@brkay54): Yes needed accurate slope info.
  • (@xmfcx): Maybe we are doing something else wrong. Because robione is driving in the same environment without slope compensation.
  • (@brkay54): Robione is using vel info from traj. They have the same problems.

Action Items

• Address (and create issues/PRs if necessary) planning & control related problems pointed out by M. Fatih Cırıt with Multi PC AWSIM + Autoware Tests
  • (@mehmetdogru) to complete planning related scenarios for test 1: https://github.com/orgs/autowarefoundation/discussions/3813#discussioncomment-6998775
• @brkay54 to create a discussion and share enough information to address the current control problems.
• Disable surround_obstacle_checker by default.

[TakaHoribe]

@mehmetdogru @xmfcx @armaganarsln
Thank you for today's discussion. I didn't expect the AWSIM testing to give us that much information. Today's AWSIM discussion was awesome. Thank you so much for working on this topic.

@xmfcx Recently we've tested with the main branch in the AWSIM, and now all interface-related issues (unconnected traffic light, etc) are fixed. Would you try to test in the main branch? I also asked @mitsudome-r -san to provide a stable branch for AWSIM which is closer to the main.

[TakaHoribe]

@brkay54 Thank you for today's discussion. I have some comments on the slope compensation topic.

You said you need a future slope angle due to the time delay, but I still don't think the change in 0.2~0.3 seconds has a large effect on vehicle control. But I assume it does here.

If it has a large effect on the vehicle, then I recommend more relying on the integral term of the PID.
You can use the sensor information (e.g. IMU) to estimate the slope angle, but you should not 100% trust it. The deviation from the ideal information should be covered by the feedback control, especially an integral term in this case.

[brkay54]

@TakaHoribe Thank you for your comment! Actually, the main problem is that the slope is not static, it is dynamic, and you can think that the slope of the area we are trying to drive is always changing (positive to negative or negative to positive) in a short distance. I tried the integral term a lot but it does not provide stability because the slope is very dynamic. Because of inaccurate slope information increases our error and inflates our integral term to decrease error but same time it causes delay and instability in the feedback controller.

[villaoningfootball]

Hello everyone，I really want to know the conclusion of the first discussion topic，please share me，thank you

[mehmetdogru]

Hi @villaoningfootball

The conclusion is to disable surround_obstacle_checker by default.