autoware_ai_planning icon indicating copy to clipboard operation
autoware_ai_planning copied to clipboard

Global path starts far from the current (initial) pose

Open squizz617 opened this issue 3 years ago • 7 comments

Bug report

Required information:

  • Operating system and version: Ubuntu 18.04
  • Autoware installation type: Docker image autoware/autoware:1.14.0-melodic-cuda, as specified in the Dockerfile of https://github.com/carla-simulator/carla-autoware at commit 975b6f3.
  • Autoware version or commit hash - 1.14.0
  • ROS distribution and version:
    • ROS Melodic (pre-installed in the docker image)
  • ROS installation type:
    • Pre-installed in the docker image

Description of the bug

  • I am running Autoware 1.14.0 (docker image: autoware/autoware:1.14.0-melodic-cuda), Carla 0.9.10.1 (docker image: carlasim/carla:0.9.10.1) as a simulator and carla-autoware-bridge at 975b6f3 as a bridge.
  • When a goal is published, generated global path starts from the lane next to the current pose of the vehicle, as shown in the rivz.
  • Then, the vehicle starts moving in the "WaitDriveReady" state, tries to pursue the path but oscillates and deviates from the path.
  • Camera data
  • rosbag
  • Map files can be downloaded at: autoware-contents.

Steps to reproduce the bug

  1. Replay the self-contained rosbag file.

Expected behavior

Global path should start from the lane that the vehicle is on.

Actual behavior

Global path starts from the lane next to the vehicle, and the vehicle clumsily pursue the path, ending up crashing.

Screenshots

Additional information

squizz617 avatar Apr 09 '21 12:04 squizz617

@squizz617 There are two problems here:

  1. The incorrect lane is chosen as the current lane for the vehicle start point. This is a problem with the planner.
  2. The control algorithm is unable to accurately follow the path when the target point is so close to the ego vehicle yet the heading error to reach the point is incredibly large.

In order to help resolve these issues, I need to know the following:

  • Which planner are you using?
  • Which control algorithm are you using?
  • What configuration parameters are you providing to the control algorithm?

JWhitleyWork avatar Apr 09 '21 16:04 JWhitleyWork

Thanks for the comment @JWhitleyWork !

Here's the requested information:

  • Planner: op_global_planner
    • pathDensity = 0.75
    • enableSmoothing = true
    • enableLaneChange = false
    • enableRvizInput = true
    • enableReplan = true
    • velocitySource = 1
    • mapSource = 0
  • Control algo: pure_pursuit
    • config: "{ header: auto, param_flag: 0, velocity: 6.0, lookahead_distance: 4.0, lookahead_ratio: 0.5, minimum_lookahead_distance: 4.0, displacement_threshold: 0.0, relative_angle_threshold: 0.0 }"

squizz617 avatar Apr 12 '21 15:04 squizz617

@hatem-darweesh might be able to help with problem 1 since that's part of the OpenPlanner implementation.

For the control algorithm, you will need to set minimum_lookahead_distance to something closer (try 1.0) in order to handle trajectories that close to the vehicle.

JWhitleyWork avatar Apr 14 '21 20:04 JWhitleyWork

Hi @squizz617 There was a lot of issues for OpenPlanner 1.5 to work with CARLA Maps, 1- how do you convert the map. 2- what format you ended using (Vector Map, lanelet2, KML) 3- what version of OpenPlanner you using.

Currently latest OpenPlanner 2.0 is workin on top of Release 1.13 -> https://github.com/hatem-darweesh we have working group specially for that, check the news and development, also document explain how to use openplanner.1.13

https://app.slack.com/client/T3BKJGA7J/C01PTJT7YTX download the OpenPlannet_presentation.pdf Currently we are merging with Autoware.15 release (latest master)

In the other repository , you are using Autoware 1.13 as it is, only OpenPlanner include the new fixed and developments.

hatem-darweesh avatar Apr 19 '21 05:04 hatem-darweesh

Thanks for the information @hatem-darweesh . 1 - I've used the vector maps provided by CARLA at carla-autoware and autoware-contents. Not sure how they converted them. 2- The planner is using the vector map (Aisan). 3- I ran the autoware/autoware:1.14.0-melodic-cuda container as is. Not sure what's the version of OpenPlanner, but the one included in the container is https://github.com/Autoware-AI/core_planning/tree/1.14.0/op_global_planner, which is behind your fork.

Just to confirm, are the other issues I reported (#45 and #47) also related to the issues with OpenPlanner & CARLA maps that you've mentioned?

squizz617 avatar May 03 '21 05:05 squizz617

Which Town(s) map you have this issue with ? what I can do is to check map connectivity and if it works find with op global planner

hatem-darweesh avatar May 04 '21 05:05 hatem-darweesh

Thank you. The current issue consistently happens in Town04.

squizz617 avatar May 04 '21 18:05 squizz617