Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Bug]: Car doesn't stay in its lane #358

Open
Johannes1098 opened this issue Oct 28, 2024 · 5 comments
Open

[Bug]: Car doesn't stay in its lane #358

Johannes1098 opened this issue Oct 28, 2024 · 5 comments
Assignees
Labels

Comments

@Johannes1098
Copy link
Collaborator

Johannes1098 commented Oct 28, 2024

Current Behavior

The car is moving inside the lane from left to right instead of driving straight in it.
The agent also frequently deviates from the lane, especially after recovering from being stuck. This instability leads to unsteady lane positioning and occasional veering into other lanes or off-road areas.

Expected Behavior

The car should drive straight in the lane.

How to reproduce the issue

Start the simulation and let the car drive for a while.

Effort Estimate

6

Testability

Run tests with the agent in a variety of lane-holding scenarios, including recovery from obstacles, to ensure lane stability.

Dependencies

May depend on localization accuracy and sensor data processing updates.

@Johannes1098 Johannes1098 added the bug Something isn't working label Oct 28, 2024
@Johannes1098
Copy link
Collaborator Author

Possible error source:
GPS-Noise? Kalman?

@vinzenzm
Copy link
Collaborator

vinzenzm commented Nov 1, 2024

After some testing there seem to be multiple reasons for this behavior:

  1. Noise in position estimation:
    • Because the position estimation is noisy we wobble left to right in our lane
    • This can be observed by looking at the green trajectory in rviz (Basically the trajectory is fixed in world while the car uses its gps) Because rviz centers on the car (in current rviz setup) the trajectory seems to move (its the estimated position however)
  2. There seems to be an issue in the motion planner where a lanechange is beeing commanded if the path is blocked. The two images where taken a few minutes apart. In my opinion the motion planning decided to add extra lane-changes in the scenario.
    random_lanechange_in_trajectory
    second_lanechange_to_already_bad_pathing

How to fix these two issues:

  1. Position estimation noise:
    I dont think that the position estimation has to improve in order to fix this. We do however decide on our steering on this global trajectory.
    My idea would be to change the current interface from planning to acting. I suggest to use a local trajectory, in this it would be far more easy to implement more sophisticated collision avoidance.
    "We than want to guide ourselfs by collision avoidance with a general direction given by global trajectory".
  2. Motion Planner lane changes:
    I will add the planning label to this issue.@ Planners: You might want to have a look at this. At the moment I myself dont have a proper understanding about the Planning structure.

@niklasr22
Copy link
Contributor

niklasr22 commented Nov 2, 2024

The trajectory on the second image looks like while performing an overtake maneuver, a second overtake maneuver got added. The second one might be triggered by the unstuck behavior as the car apparently got stuck in the oncoming traffic while performing the first maneuver.

@niklasr22
Copy link
Contributor

Obviously adding another x offset towards the left on the trajectory doesn't help in the situation depicted in the image. We might need to add a mechanism that decides whether to move the trajectory to the right or left.

@vinzenzm
Copy link
Collaborator

vinzenzm commented Nov 5, 2024

For completeness: In this document the unstuck behaviour is well documented:
https://github.com/una-auxme/paf/blob/main/doc/planning/Unstuck_Behavior.md

Its the explanation for the second bahaviour I observed in my comment above.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
Status: No status
Development

No branches or pull requests

4 participants