Autoware Architecture WG Meeting (planning) 2022/06/01

[mitsudome-r]

Participants

Chaired by Ryohsuke

Goal of this WG

• https://docs.google.com/presentation/d/1AhuCcScktgBwJreAjDspyf2OcKTNFbK-fFI1QT9NRLg/edit?usp=sharing
• Wiki: https://github.com/autowarefoundation/autoware-projects/wiki/Autoware-Architecture-Working-Group

Discussion Topics:

Any follow-up from the previous meeting

#364

Review Planning Interface Design

Reference:

• https://github.com/orgs/autowarefoundation/discussions/4

• Current output of Planning:

  • Trajectory
  • Vehicle Signal Commands
  • Planning status information (engage request, mission status) (TBD)
  • TrafficLightQuery (TBD)

• (If we have time) It might worth reviewing the output of global planner

  • Although it is treated as internal message within planning component, it maybe coming from fleet management system as cloud service in the future.
  • HADMapRoute Msgs
  • I would like to change the name to just Route message

Discussions:

• Trajectory Pointmessage:
  • Do we need time_from_start?
    • We can remove it and assume constant time_step defined in Trajectory.msg
    • What is the general timestep between the points?
    • In general 10-30 points per second.
  • Do we need lateral velocity?
    • If yes, should acceleration have lateral velocity?
    • If we are including fields that are not filled in, we should have a boolean to indicate whether the values are valid.
• Do we need wheel angles?
  • If we check http://docs.ros.org/en/api/geometry_msgs/html/msg/PoseStamped.html, we only have PoseWithStamped array. Velocity can be derived from the position.
  • Fatih: I prefer to limit the message field simple and let the following modules to recalculate any necessary information if needed.

Final image design:

• Global Route message:
• Route message only seems to indicate lane_id. Does it assume that there is a map information synchronized with this message?
  • yes.
• Does lane_ids in RouteSection indicate the lane sequence that vehicle should follow, or does it provide possible lanes that behavior planner can use.
  • It indicates the lanes that the local planner can use. It would be up to the local planner which lane vehicle will drive.
• If we consider very complex lane (e.g., lane that self-intersect), then don't we need more information to indicate goal_pose?
  • Lanelet2 specificiation does not allow a lane that self-intersect.
• How should we handle multistorey carpark?
  • We currently assume that lanelet2 also have 3d information attached to the lanes. We should be able to match closest lane from the given goal pose by searching in z direciton.

Next Week Topic

Vehicle Modules

[sglee-morai]

Example Scenario

The vehicle will start from the 2nd floor and get out of the building.

Concern

Currently, start_pos and end_pos is defined in the (x,y,z) - defined w.r.t. 3d cartesian coordinate system.

• For multi-story buildings, where getting high accuracy z value is difficult, start_pos (or end_pos) defined this way could make it difficult to guess where it actually is.
• For the example scenario, if start_pos is defined in (x,y,z), it's hard to tell whether the car is starting from the 1st, 2nd, or 3rd floor.

Suggestion

Defining start_pos and end_pos w.r.t to (lane_id, s, t)

• Each lane in the different story will have a unique lane_id. Thus telling lane_id first will eliminate the guessing.
• For the example scenario, start_pos might be defined as (lane_id = lane22, s = 10, t = 0)

Would changing this way create another problem? I'm very new to the planning stack of Autoware, so I'm not really whether the concern I raised is important or not.

[mitsudome-r]

Thanks for the nice diagram.

Currently, we assume that Lanelet2 will include the z information of the lanes that matches with the pointcloud map coordinates. If this assumption stands true, then I think it wouldn't be a problem to give start_pose/goal_pose in 3D.
Also, I cannot really understand why providing (lane_id = lane22, s = 10, t = 0) would be easier than providing the 3d information if you don't know where exactly you are in the map?

[Sharrrrk]

I agree with @mitsudome-r, it won't be a problem to get the right lane_id based on given pose. What's more, the pose information is also required for parking scenario in which the target position is beyond lane.

[Hyeongseok-Jeon]

I have comment on the output msg from the planning. Do you consider generating the pose equally spaced by distance not time? In my personal work, planning module outputs the pose of the waypoints spaced by 0.5meter and corresponding desired velocity for each way point. This is for completely separate lateral plan (equally space future way points) and longitudinal plan (desired velocity for each way point). Sorry for the delayed comment.